1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1990, 2010, Oracle and/or its affiliates. All rights reserved.
  24  */
  25 /*
  26  * Copyright (c) 2011 Bayard G. Bell.  All rights reserved.
  27  * Copyright (c) 2012 by Delphix. All rights reserved.
  28  * Copyright 2013 Nexenta Systems, Inc.  All rights reserved.
  29  */
  30 /*
  31  * Copyright 2011 cyril.galibern@opensvc.com
  32  */
  33 
  34 /*
  35  * SCSI disk target driver.
  36  */
  37 #include <sys/scsi/scsi.h>
  38 #include <sys/dkbad.h>
  39 #include <sys/dklabel.h>
  40 #include <sys/dkio.h>
  41 #include <sys/fdio.h>
  42 #include <sys/cdio.h>
  43 #include <sys/mhd.h>
  44 #include <sys/vtoc.h>
  45 #include <sys/dktp/fdisk.h>
  46 #include <sys/kstat.h>
  47 #include <sys/vtrace.h>
  48 #include <sys/note.h>
  49 #include <sys/thread.h>
  50 #include <sys/proc.h>
  51 #include <sys/efi_partition.h>
  52 #include <sys/var.h>
  53 #include <sys/aio_req.h>
  54 
  55 #ifdef __lock_lint
  56 #define _LP64
  57 #define __amd64
  58 #endif
  59 
  60 #if (defined(__fibre))
  61 /* Note: is there a leadville version of the following? */
  62 #include <sys/fc4/fcal_linkapp.h>
  63 #endif
  64 #include <sys/taskq.h>
  65 #include <sys/uuid.h>
  66 #include <sys/byteorder.h>
  67 #include <sys/sdt.h>
  68 
  69 #include "sd_xbuf.h"
  70 
  71 #include <sys/scsi/targets/sddef.h>
  72 #include <sys/cmlb.h>
  73 #include <sys/sysevent/eventdefs.h>
  74 #include <sys/sysevent/dev.h>
  75 
  76 #include <sys/fm/protocol.h>
  77 
  78 /*
  79  * Loadable module info.
  80  */
  81 #if (defined(__fibre))
  82 #define SD_MODULE_NAME  "SCSI SSA/FCAL Disk Driver"
  83 #else /* !__fibre */
  84 #define SD_MODULE_NAME  "SCSI Disk Driver"
  85 #endif /* !__fibre */
  86 
  87 /*
  88  * Define the interconnect type, to allow the driver to distinguish
  89  * between parallel SCSI (sd) and fibre channel (ssd) behaviors.
  90  *
  91  * This is really for backward compatibility. In the future, the driver
  92  * should actually check the "interconnect-type" property as reported by
  93  * the HBA; however at present this property is not defined by all HBAs,
  94  * so we will use this #define (1) to permit the driver to run in
  95  * backward-compatibility mode; and (2) to print a notification message
  96  * if an FC HBA does not support the "interconnect-type" property.  The
  97  * behavior of the driver will be to assume parallel SCSI behaviors unless
  98  * the "interconnect-type" property is defined by the HBA **AND** has a
  99  * value of either INTERCONNECT_FIBRE, INTERCONNECT_SSA, or
 100  * INTERCONNECT_FABRIC, in which case the driver will assume Fibre
 101  * Channel behaviors (as per the old ssd).  (Note that the
 102  * INTERCONNECT_1394 and INTERCONNECT_USB types are not supported and
 103  * will result in the driver assuming parallel SCSI behaviors.)
 104  *
 105  * (see common/sys/scsi/impl/services.h)
 106  *
 107  * Note: For ssd semantics, don't use INTERCONNECT_FABRIC as the default
 108  * since some FC HBAs may already support that, and there is some code in
 109  * the driver that already looks for it.  Using INTERCONNECT_FABRIC as the
 110  * default would confuse that code, and besides things should work fine
 111  * anyways if the FC HBA already reports INTERCONNECT_FABRIC for the
 112  * "interconnect_type" property.
 113  *
 114  */
 115 #if (defined(__fibre))
 116 #define SD_DEFAULT_INTERCONNECT_TYPE    SD_INTERCONNECT_FIBRE
 117 #else
 118 #define SD_DEFAULT_INTERCONNECT_TYPE    SD_INTERCONNECT_PARALLEL
 119 #endif
 120 
 121 /*
 122  * The name of the driver, established from the module name in _init.
 123  */
 124 static  char *sd_label                  = NULL;
 125 
 126 /*
 127  * Driver name is unfortunately prefixed on some driver.conf properties.
 128  */
 129 #if (defined(__fibre))
 130 #define sd_max_xfer_size                ssd_max_xfer_size
 131 #define sd_config_list                  ssd_config_list
 132 static  char *sd_max_xfer_size          = "ssd_max_xfer_size";
 133 static  char *sd_config_list            = "ssd-config-list";
 134 #else
 135 static  char *sd_max_xfer_size          = "sd_max_xfer_size";
 136 static  char *sd_config_list            = "sd-config-list";
 137 #endif
 138 
 139 /*
 140  * Driver global variables
 141  */
 142 
 143 #if (defined(__fibre))
 144 /*
 145  * These #defines are to avoid namespace collisions that occur because this
 146  * code is currently used to compile two separate driver modules: sd and ssd.
 147  * All global variables need to be treated this way (even if declared static)
 148  * in order to allow the debugger to resolve the names properly.
 149  * It is anticipated that in the near future the ssd module will be obsoleted,
 150  * at which time this namespace issue should go away.
 151  */
 152 #define sd_state                        ssd_state
 153 #define sd_io_time                      ssd_io_time
 154 #define sd_failfast_enable              ssd_failfast_enable
 155 #define sd_ua_retry_count               ssd_ua_retry_count
 156 #define sd_report_pfa                   ssd_report_pfa
 157 #define sd_max_throttle                 ssd_max_throttle
 158 #define sd_min_throttle                 ssd_min_throttle
 159 #define sd_rot_delay                    ssd_rot_delay
 160 
 161 #define sd_retry_on_reservation_conflict        \
 162                                         ssd_retry_on_reservation_conflict
 163 #define sd_reinstate_resv_delay         ssd_reinstate_resv_delay
 164 #define sd_resv_conflict_name           ssd_resv_conflict_name
 165 
 166 #define sd_component_mask               ssd_component_mask
 167 #define sd_level_mask                   ssd_level_mask
 168 #define sd_debug_un                     ssd_debug_un
 169 #define sd_error_level                  ssd_error_level
 170 
 171 #define sd_xbuf_active_limit            ssd_xbuf_active_limit
 172 #define sd_xbuf_reserve_limit           ssd_xbuf_reserve_limit
 173 
 174 #define sd_tr                           ssd_tr
 175 #define sd_reset_throttle_timeout       ssd_reset_throttle_timeout
 176 #define sd_qfull_throttle_timeout       ssd_qfull_throttle_timeout
 177 #define sd_qfull_throttle_enable        ssd_qfull_throttle_enable
 178 #define sd_check_media_time             ssd_check_media_time
 179 #define sd_wait_cmds_complete           ssd_wait_cmds_complete
 180 #define sd_label_mutex                  ssd_label_mutex
 181 #define sd_detach_mutex                 ssd_detach_mutex
 182 #define sd_log_buf                      ssd_log_buf
 183 #define sd_log_mutex                    ssd_log_mutex
 184 
 185 #define sd_disk_table                   ssd_disk_table
 186 #define sd_disk_table_size              ssd_disk_table_size
 187 #define sd_sense_mutex                  ssd_sense_mutex
 188 #define sd_cdbtab                       ssd_cdbtab
 189 
 190 #define sd_cb_ops                       ssd_cb_ops
 191 #define sd_ops                          ssd_ops
 192 #define sd_additional_codes             ssd_additional_codes
 193 #define sd_tgops                        ssd_tgops
 194 
 195 #define sd_minor_data                   ssd_minor_data
 196 #define sd_minor_data_efi               ssd_minor_data_efi
 197 
 198 #define sd_tq                           ssd_tq
 199 #define sd_wmr_tq                       ssd_wmr_tq
 200 #define sd_taskq_name                   ssd_taskq_name
 201 #define sd_wmr_taskq_name               ssd_wmr_taskq_name
 202 #define sd_taskq_minalloc               ssd_taskq_minalloc
 203 #define sd_taskq_maxalloc               ssd_taskq_maxalloc
 204 
 205 #define sd_dump_format_string           ssd_dump_format_string
 206 
 207 #define sd_iostart_chain                ssd_iostart_chain
 208 #define sd_iodone_chain                 ssd_iodone_chain
 209 
 210 #define sd_pm_idletime                  ssd_pm_idletime
 211 
 212 #define sd_force_pm_supported           ssd_force_pm_supported
 213 
 214 #define sd_dtype_optical_bind           ssd_dtype_optical_bind
 215 
 216 #define sd_ssc_init                     ssd_ssc_init
 217 #define sd_ssc_send                     ssd_ssc_send
 218 #define sd_ssc_fini                     ssd_ssc_fini
 219 #define sd_ssc_assessment               ssd_ssc_assessment
 220 #define sd_ssc_post                     ssd_ssc_post
 221 #define sd_ssc_print                    ssd_ssc_print
 222 #define sd_ssc_ereport_post             ssd_ssc_ereport_post
 223 #define sd_ssc_set_info                 ssd_ssc_set_info
 224 #define sd_ssc_extract_info             ssd_ssc_extract_info
 225 
 226 #endif
 227 
 228 #ifdef  SDDEBUG
 229 int     sd_force_pm_supported           = 0;
 230 #endif  /* SDDEBUG */
 231 
 232 void *sd_state                          = NULL;
 233 int sd_io_time                          = SD_IO_TIME;
 234 int sd_failfast_enable                  = 1;
 235 int sd_ua_retry_count                   = SD_UA_RETRY_COUNT;
 236 int sd_report_pfa                       = 1;
 237 int sd_max_throttle                     = SD_MAX_THROTTLE;
 238 int sd_min_throttle                     = SD_MIN_THROTTLE;
 239 int sd_rot_delay                        = 4; /* Default 4ms Rotation delay */
 240 int sd_qfull_throttle_enable            = TRUE;
 241 
 242 int sd_retry_on_reservation_conflict    = 1;
 243 int sd_reinstate_resv_delay             = SD_REINSTATE_RESV_DELAY;
 244 _NOTE(SCHEME_PROTECTS_DATA("safe sharing", sd_reinstate_resv_delay))
 245 
 246 static int sd_dtype_optical_bind        = -1;
 247 
 248 /* Note: the following is not a bug, it really is "sd_" and not "ssd_" */
 249 static  char *sd_resv_conflict_name     = "sd_retry_on_reservation_conflict";
 250 
 251 /*
 252  * Global data for debug logging. To enable debug printing, sd_component_mask
 253  * and sd_level_mask should be set to the desired bit patterns as outlined in
 254  * sddef.h.
 255  */
 256 uint_t  sd_component_mask               = 0x0;
 257 uint_t  sd_level_mask                   = 0x0;
 258 struct  sd_lun *sd_debug_un             = NULL;
 259 uint_t  sd_error_level                  = SCSI_ERR_RETRYABLE;
 260 
 261 /* Note: these may go away in the future... */
 262 static uint32_t sd_xbuf_active_limit    = 512;
 263 static uint32_t sd_xbuf_reserve_limit   = 16;
 264 
 265 static struct sd_resv_reclaim_request   sd_tr = { NULL, NULL, NULL, 0, 0, 0 };
 266 
 267 /*
 268  * Timer value used to reset the throttle after it has been reduced
 269  * (typically in response to TRAN_BUSY or STATUS_QFULL)
 270  */
 271 static int sd_reset_throttle_timeout    = SD_RESET_THROTTLE_TIMEOUT;
 272 static int sd_qfull_throttle_timeout    = SD_QFULL_THROTTLE_TIMEOUT;
 273 
 274 /*
 275  * Interval value associated with the media change scsi watch.
 276  */
 277 static int sd_check_media_time          = 3000000;
 278 
 279 /*
 280  * Wait value used for in progress operations during a DDI_SUSPEND
 281  */
 282 static int sd_wait_cmds_complete        = SD_WAIT_CMDS_COMPLETE;
 283 
 284 /*
 285  * sd_label_mutex protects a static buffer used in the disk label
 286  * component of the driver
 287  */
 288 static kmutex_t sd_label_mutex;
 289 
 290 /*
 291  * sd_detach_mutex protects un_layer_count, un_detach_count, and
 292  * un_opens_in_progress in the sd_lun structure.
 293  */
 294 static kmutex_t sd_detach_mutex;
 295 
 296 _NOTE(MUTEX_PROTECTS_DATA(sd_detach_mutex,
 297         sd_lun::{un_layer_count un_detach_count un_opens_in_progress}))
 298 
 299 /*
 300  * Global buffer and mutex for debug logging
 301  */
 302 static char     sd_log_buf[1024];
 303 static kmutex_t sd_log_mutex;
 304 
 305 /*
 306  * Structs and globals for recording attached lun information.
 307  * This maintains a chain. Each node in the chain represents a SCSI controller.
 308  * The structure records the number of luns attached to each target connected
 309  * with the controller.
 310  * For parallel scsi device only.
 311  */
 312 struct sd_scsi_hba_tgt_lun {
 313         struct sd_scsi_hba_tgt_lun      *next;
 314         dev_info_t                      *pdip;
 315         int                             nlun[NTARGETS_WIDE];
 316 };
 317 
 318 /*
 319  * Flag to indicate the lun is attached or detached
 320  */
 321 #define SD_SCSI_LUN_ATTACH      0
 322 #define SD_SCSI_LUN_DETACH      1
 323 
 324 static kmutex_t sd_scsi_target_lun_mutex;
 325 static struct sd_scsi_hba_tgt_lun       *sd_scsi_target_lun_head = NULL;
 326 
 327 _NOTE(MUTEX_PROTECTS_DATA(sd_scsi_target_lun_mutex,
 328     sd_scsi_hba_tgt_lun::next sd_scsi_hba_tgt_lun::pdip))
 329 
 330 _NOTE(MUTEX_PROTECTS_DATA(sd_scsi_target_lun_mutex,
 331     sd_scsi_target_lun_head))
 332 
 333 /*
 334  * "Smart" Probe Caching structs, globals, #defines, etc.
 335  * For parallel scsi and non-self-identify device only.
 336  */
 337 
 338 /*
 339  * The following resources and routines are implemented to support
 340  * "smart" probing, which caches the scsi_probe() results in an array,
 341  * in order to help avoid long probe times.
 342  */
 343 struct sd_scsi_probe_cache {
 344         struct  sd_scsi_probe_cache     *next;
 345         dev_info_t      *pdip;
 346         int             cache[NTARGETS_WIDE];
 347 };
 348 
 349 static kmutex_t sd_scsi_probe_cache_mutex;
 350 static struct   sd_scsi_probe_cache *sd_scsi_probe_cache_head = NULL;
 351 
 352 /*
 353  * Really we only need protection on the head of the linked list, but
 354  * better safe than sorry.
 355  */
 356 _NOTE(MUTEX_PROTECTS_DATA(sd_scsi_probe_cache_mutex,
 357     sd_scsi_probe_cache::next sd_scsi_probe_cache::pdip))
 358 
 359 _NOTE(MUTEX_PROTECTS_DATA(sd_scsi_probe_cache_mutex,
 360     sd_scsi_probe_cache_head))
 361 
 362 /*
 363  * Power attribute table
 364  */
 365 static sd_power_attr_ss sd_pwr_ss = {
 366         { "NAME=spindle-motor", "0=off", "1=on", NULL },
 367         {0, 100},
 368         {30, 0},
 369         {20000, 0}
 370 };
 371 
 372 static sd_power_attr_pc sd_pwr_pc = {
 373         { "NAME=spindle-motor", "0=stopped", "1=standby", "2=idle",
 374                 "3=active", NULL },
 375         {0, 0, 0, 100},
 376         {90, 90, 20, 0},
 377         {15000, 15000, 1000, 0}
 378 };
 379 
 380 /*
 381  * Power level to power condition
 382  */
 383 static int sd_pl2pc[] = {
 384         SD_TARGET_START_VALID,
 385         SD_TARGET_STANDBY,
 386         SD_TARGET_IDLE,
 387         SD_TARGET_ACTIVE
 388 };
 389 
 390 /*
 391  * Vendor specific data name property declarations
 392  */
 393 
 394 #if defined(__fibre) || defined(__i386) ||defined(__amd64)
 395 
 396 static sd_tunables seagate_properties = {
 397         SEAGATE_THROTTLE_VALUE,
 398         0,
 399         0,
 400         0,
 401         0,
 402         0,
 403         0,
 404         0,
 405         0
 406 };
 407 
 408 
 409 static sd_tunables fujitsu_properties = {
 410         FUJITSU_THROTTLE_VALUE,
 411         0,
 412         0,
 413         0,
 414         0,
 415         0,
 416         0,
 417         0,
 418         0
 419 };
 420 
 421 static sd_tunables ibm_properties = {
 422         IBM_THROTTLE_VALUE,
 423         0,
 424         0,
 425         0,
 426         0,
 427         0,
 428         0,
 429         0,
 430         0
 431 };
 432 
 433 static sd_tunables purple_properties = {
 434         PURPLE_THROTTLE_VALUE,
 435         0,
 436         0,
 437         PURPLE_BUSY_RETRIES,
 438         PURPLE_RESET_RETRY_COUNT,
 439         PURPLE_RESERVE_RELEASE_TIME,
 440         0,
 441         0,
 442         0
 443 };
 444 
 445 static sd_tunables sve_properties = {
 446         SVE_THROTTLE_VALUE,
 447         0,
 448         0,
 449         SVE_BUSY_RETRIES,
 450         SVE_RESET_RETRY_COUNT,
 451         SVE_RESERVE_RELEASE_TIME,
 452         SVE_MIN_THROTTLE_VALUE,
 453         SVE_DISKSORT_DISABLED_FLAG,
 454         0
 455 };
 456 
 457 static sd_tunables maserati_properties = {
 458         0,
 459         0,
 460         0,
 461         0,
 462         0,
 463         0,
 464         0,
 465         MASERATI_DISKSORT_DISABLED_FLAG,
 466         MASERATI_LUN_RESET_ENABLED_FLAG
 467 };
 468 
 469 static sd_tunables pirus_properties = {
 470         PIRUS_THROTTLE_VALUE,
 471         0,
 472         PIRUS_NRR_COUNT,
 473         PIRUS_BUSY_RETRIES,
 474         PIRUS_RESET_RETRY_COUNT,
 475         0,
 476         PIRUS_MIN_THROTTLE_VALUE,
 477         PIRUS_DISKSORT_DISABLED_FLAG,
 478         PIRUS_LUN_RESET_ENABLED_FLAG
 479 };
 480 
 481 #endif
 482 
 483 #if (defined(__sparc) && !defined(__fibre)) || \
 484         (defined(__i386) || defined(__amd64))
 485 
 486 
 487 static sd_tunables elite_properties = {
 488         ELITE_THROTTLE_VALUE,
 489         0,
 490         0,
 491         0,
 492         0,
 493         0,
 494         0,
 495         0,
 496         0
 497 };
 498 
 499 static sd_tunables st31200n_properties = {
 500         ST31200N_THROTTLE_VALUE,
 501         0,
 502         0,
 503         0,
 504         0,
 505         0,
 506         0,
 507         0,
 508         0
 509 };
 510 
 511 #endif /* Fibre or not */
 512 
 513 static sd_tunables lsi_properties_scsi = {
 514         LSI_THROTTLE_VALUE,
 515         0,
 516         LSI_NOTREADY_RETRIES,
 517         0,
 518         0,
 519         0,
 520         0,
 521         0,
 522         0
 523 };
 524 
 525 static sd_tunables symbios_properties = {
 526         SYMBIOS_THROTTLE_VALUE,
 527         0,
 528         SYMBIOS_NOTREADY_RETRIES,
 529         0,
 530         0,
 531         0,
 532         0,
 533         0,
 534         0
 535 };
 536 
 537 static sd_tunables lsi_properties = {
 538         0,
 539         0,
 540         LSI_NOTREADY_RETRIES,
 541         0,
 542         0,
 543         0,
 544         0,
 545         0,
 546         0
 547 };
 548 
 549 static sd_tunables lsi_oem_properties = {
 550         0,
 551         0,
 552         LSI_OEM_NOTREADY_RETRIES,
 553         0,
 554         0,
 555         0,
 556         0,
 557         0,
 558         0,
 559         1
 560 };
 561 
 562 
 563 
 564 #if (defined(SD_PROP_TST))
 565 
 566 #define SD_TST_CTYPE_VAL        CTYPE_CDROM
 567 #define SD_TST_THROTTLE_VAL     16
 568 #define SD_TST_NOTREADY_VAL     12
 569 #define SD_TST_BUSY_VAL         60
 570 #define SD_TST_RST_RETRY_VAL    36
 571 #define SD_TST_RSV_REL_TIME     60
 572 
 573 static sd_tunables tst_properties = {
 574         SD_TST_THROTTLE_VAL,
 575         SD_TST_CTYPE_VAL,
 576         SD_TST_NOTREADY_VAL,
 577         SD_TST_BUSY_VAL,
 578         SD_TST_RST_RETRY_VAL,
 579         SD_TST_RSV_REL_TIME,
 580         0,
 581         0,
 582         0
 583 };
 584 #endif
 585 
 586 /* This is similar to the ANSI toupper implementation */
 587 #define SD_TOUPPER(C)   (((C) >= 'a' && (C) <= 'z') ? (C) - 'a' + 'A' : (C))
 588 
 589 /*
 590  * Static Driver Configuration Table
 591  *
 592  * This is the table of disks which need throttle adjustment (or, perhaps
 593  * something else as defined by the flags at a future time.)  device_id
 594  * is a string consisting of concatenated vid (vendor), pid (product/model)
 595  * and revision strings as defined in the scsi_inquiry structure.  Offsets of
 596  * the parts of the string are as defined by the sizes in the scsi_inquiry
 597  * structure.  Device type is searched as far as the device_id string is
 598  * defined.  Flags defines which values are to be set in the driver from the
 599  * properties list.
 600  *
 601  * Entries below which begin and end with a "*" are a special case.
 602  * These do not have a specific vendor, and the string which follows
 603  * can appear anywhere in the 16 byte PID portion of the inquiry data.
 604  *
 605  * Entries below which begin and end with a " " (blank) are a special
 606  * case. The comparison function will treat multiple consecutive blanks
 607  * as equivalent to a single blank. For example, this causes a
 608  * sd_disk_table entry of " NEC CDROM " to match a device's id string
 609  * of  "NEC       CDROM".
 610  *
 611  * Note: The MD21 controller type has been obsoleted.
 612  *       ST318202F is a Legacy device
 613  *       MAM3182FC, MAM3364FC, MAM3738FC do not appear to have ever been
 614  *       made with an FC connection. The entries here are a legacy.
 615  */
 616 static sd_disk_config_t sd_disk_table[] = {
 617 #if defined(__fibre) || defined(__i386) || defined(__amd64)
 618         { "SEAGATE ST34371FC", SD_CONF_BSET_THROTTLE, &seagate_properties },
 619         { "SEAGATE ST19171FC", SD_CONF_BSET_THROTTLE, &seagate_properties },
 620         { "SEAGATE ST39102FC", SD_CONF_BSET_THROTTLE, &seagate_properties },
 621         { "SEAGATE ST39103FC", SD_CONF_BSET_THROTTLE, &seagate_properties },
 622         { "SEAGATE ST118273F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 623         { "SEAGATE ST318202F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 624         { "SEAGATE ST318203F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 625         { "SEAGATE ST136403F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 626         { "SEAGATE ST318304F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 627         { "SEAGATE ST336704F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 628         { "SEAGATE ST373405F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 629         { "SEAGATE ST336605F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 630         { "SEAGATE ST336752F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 631         { "SEAGATE ST318452F", SD_CONF_BSET_THROTTLE, &seagate_properties },
 632         { "FUJITSU MAG3091F",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 633         { "FUJITSU MAG3182F",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 634         { "FUJITSU MAA3182F",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 635         { "FUJITSU MAF3364F",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 636         { "FUJITSU MAL3364F",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 637         { "FUJITSU MAL3738F",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 638         { "FUJITSU MAM3182FC",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 639         { "FUJITSU MAM3364FC",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 640         { "FUJITSU MAM3738FC",  SD_CONF_BSET_THROTTLE, &fujitsu_properties },
 641         { "IBM     DDYFT1835",  SD_CONF_BSET_THROTTLE, &ibm_properties },
 642         { "IBM     DDYFT3695",  SD_CONF_BSET_THROTTLE, &ibm_properties },
 643         { "IBM     IC35LF2D2",  SD_CONF_BSET_THROTTLE, &ibm_properties },
 644         { "IBM     IC35LF2PR",  SD_CONF_BSET_THROTTLE, &ibm_properties },
 645         { "IBM     1724-100",   SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 646         { "IBM     1726-2xx",   SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 647         { "IBM     1726-22x",   SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 648         { "IBM     1726-4xx",   SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 649         { "IBM     1726-42x",   SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 650         { "IBM     1726-3xx",   SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 651         { "IBM     3526",       SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 652         { "IBM     3542",       SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 653         { "IBM     3552",       SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 654         { "IBM     1722",       SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 655         { "IBM     1742",       SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 656         { "IBM     1815",       SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 657         { "IBM     FAStT",      SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 658         { "IBM     1814",       SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 659         { "IBM     1814-200",   SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 660         { "IBM     1818",       SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 661         { "DELL    MD3000",     SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 662         { "DELL    MD3000i",    SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 663         { "LSI     INF",        SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 664         { "ENGENIO INF",        SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 665         { "SGI     TP",         SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 666         { "SGI     IS",         SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 667         { "*CSM100_*",          SD_CONF_BSET_NRR_COUNT |
 668                         SD_CONF_BSET_CACHE_IS_NV, &lsi_oem_properties },
 669         { "*CSM200_*",          SD_CONF_BSET_NRR_COUNT |
 670                         SD_CONF_BSET_CACHE_IS_NV, &lsi_oem_properties },
 671         { "Fujitsu SX300",      SD_CONF_BSET_THROTTLE,  &lsi_oem_properties },
 672         { "LSI",                SD_CONF_BSET_NRR_COUNT, &lsi_properties },
 673         { "SUN     T3", SD_CONF_BSET_THROTTLE |
 674                         SD_CONF_BSET_BSY_RETRY_COUNT|
 675                         SD_CONF_BSET_RST_RETRIES|
 676                         SD_CONF_BSET_RSV_REL_TIME,
 677                 &purple_properties },
 678         { "SUN     SESS01", SD_CONF_BSET_THROTTLE |
 679                 SD_CONF_BSET_BSY_RETRY_COUNT|
 680                 SD_CONF_BSET_RST_RETRIES|
 681                 SD_CONF_BSET_RSV_REL_TIME|
 682                 SD_CONF_BSET_MIN_THROTTLE|
 683                 SD_CONF_BSET_DISKSORT_DISABLED,
 684                 &sve_properties },
 685         { "SUN     T4", SD_CONF_BSET_THROTTLE |
 686                         SD_CONF_BSET_BSY_RETRY_COUNT|
 687                         SD_CONF_BSET_RST_RETRIES|
 688                         SD_CONF_BSET_RSV_REL_TIME,
 689                 &purple_properties },
 690         { "SUN     SVE01", SD_CONF_BSET_DISKSORT_DISABLED |
 691                 SD_CONF_BSET_LUN_RESET_ENABLED,
 692                 &maserati_properties },
 693         { "SUN     SE6920", SD_CONF_BSET_THROTTLE |
 694                 SD_CONF_BSET_NRR_COUNT|
 695                 SD_CONF_BSET_BSY_RETRY_COUNT|
 696                 SD_CONF_BSET_RST_RETRIES|
 697                 SD_CONF_BSET_MIN_THROTTLE|
 698                 SD_CONF_BSET_DISKSORT_DISABLED|
 699                 SD_CONF_BSET_LUN_RESET_ENABLED,
 700                 &pirus_properties },
 701         { "SUN     SE6940", SD_CONF_BSET_THROTTLE |
 702                 SD_CONF_BSET_NRR_COUNT|
 703                 SD_CONF_BSET_BSY_RETRY_COUNT|
 704                 SD_CONF_BSET_RST_RETRIES|
 705                 SD_CONF_BSET_MIN_THROTTLE|
 706                 SD_CONF_BSET_DISKSORT_DISABLED|
 707                 SD_CONF_BSET_LUN_RESET_ENABLED,
 708                 &pirus_properties },
 709         { "SUN     StorageTek 6920", SD_CONF_BSET_THROTTLE |
 710                 SD_CONF_BSET_NRR_COUNT|
 711                 SD_CONF_BSET_BSY_RETRY_COUNT|
 712                 SD_CONF_BSET_RST_RETRIES|
 713                 SD_CONF_BSET_MIN_THROTTLE|
 714                 SD_CONF_BSET_DISKSORT_DISABLED|
 715                 SD_CONF_BSET_LUN_RESET_ENABLED,
 716                 &pirus_properties },
 717         { "SUN     StorageTek 6940", SD_CONF_BSET_THROTTLE |
 718                 SD_CONF_BSET_NRR_COUNT|
 719                 SD_CONF_BSET_BSY_RETRY_COUNT|
 720                 SD_CONF_BSET_RST_RETRIES|
 721                 SD_CONF_BSET_MIN_THROTTLE|
 722                 SD_CONF_BSET_DISKSORT_DISABLED|
 723                 SD_CONF_BSET_LUN_RESET_ENABLED,
 724                 &pirus_properties },
 725         { "SUN     PSX1000", SD_CONF_BSET_THROTTLE |
 726                 SD_CONF_BSET_NRR_COUNT|
 727                 SD_CONF_BSET_BSY_RETRY_COUNT|
 728                 SD_CONF_BSET_RST_RETRIES|
 729                 SD_CONF_BSET_MIN_THROTTLE|
 730                 SD_CONF_BSET_DISKSORT_DISABLED|
 731                 SD_CONF_BSET_LUN_RESET_ENABLED,
 732                 &pirus_properties },
 733         { "SUN     SE6330", SD_CONF_BSET_THROTTLE |
 734                 SD_CONF_BSET_NRR_COUNT|
 735                 SD_CONF_BSET_BSY_RETRY_COUNT|
 736                 SD_CONF_BSET_RST_RETRIES|
 737                 SD_CONF_BSET_MIN_THROTTLE|
 738                 SD_CONF_BSET_DISKSORT_DISABLED|
 739                 SD_CONF_BSET_LUN_RESET_ENABLED,
 740                 &pirus_properties },
 741         { "SUN     STK6580_6780", SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 742         { "SUN     SUN_6180", SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 743         { "STK     OPENstorage", SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 744         { "STK     OpenStorage", SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 745         { "STK     BladeCtlr",  SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 746         { "STK     FLEXLINE",   SD_CONF_BSET_NRR_COUNT, &lsi_oem_properties },
 747         { "SYMBIOS", SD_CONF_BSET_NRR_COUNT, &symbios_properties },
 748 #endif /* fibre or NON-sparc platforms */
 749 #if ((defined(__sparc) && !defined(__fibre)) ||\
 750         (defined(__i386) || defined(__amd64)))
 751         { "SEAGATE ST42400N", SD_CONF_BSET_THROTTLE, &elite_properties },
 752         { "SEAGATE ST31200N", SD_CONF_BSET_THROTTLE, &st31200n_properties },
 753         { "SEAGATE ST41600N", SD_CONF_BSET_TUR_CHECK, NULL },
 754         { "CONNER  CP30540",  SD_CONF_BSET_NOCACHE,  NULL },
 755         { "*SUN0104*", SD_CONF_BSET_FAB_DEVID, NULL },
 756         { "*SUN0207*", SD_CONF_BSET_FAB_DEVID, NULL },
 757         { "*SUN0327*", SD_CONF_BSET_FAB_DEVID, NULL },
 758         { "*SUN0340*", SD_CONF_BSET_FAB_DEVID, NULL },
 759         { "*SUN0424*", SD_CONF_BSET_FAB_DEVID, NULL },
 760         { "*SUN0669*", SD_CONF_BSET_FAB_DEVID, NULL },
 761         { "*SUN1.0G*", SD_CONF_BSET_FAB_DEVID, NULL },
 762         { "SYMBIOS INF-01-00       ", SD_CONF_BSET_FAB_DEVID, NULL },
 763         { "SYMBIOS", SD_CONF_BSET_THROTTLE|SD_CONF_BSET_NRR_COUNT,
 764             &symbios_properties },
 765         { "LSI", SD_CONF_BSET_THROTTLE | SD_CONF_BSET_NRR_COUNT,
 766             &lsi_properties_scsi },
 767 #if defined(__i386) || defined(__amd64)
 768         { " NEC CD-ROM DRIVE:260 ", (SD_CONF_BSET_PLAYMSF_BCD
 769                                     | SD_CONF_BSET_READSUB_BCD
 770                                     | SD_CONF_BSET_READ_TOC_ADDR_BCD
 771                                     | SD_CONF_BSET_NO_READ_HEADER
 772                                     | SD_CONF_BSET_READ_CD_XD4), NULL },
 773 
 774         { " NEC CD-ROM DRIVE:270 ", (SD_CONF_BSET_PLAYMSF_BCD
 775                                     | SD_CONF_BSET_READSUB_BCD
 776                                     | SD_CONF_BSET_READ_TOC_ADDR_BCD
 777                                     | SD_CONF_BSET_NO_READ_HEADER
 778                                     | SD_CONF_BSET_READ_CD_XD4), NULL },
 779 #endif /* __i386 || __amd64 */
 780 #endif /* sparc NON-fibre or NON-sparc platforms */
 781 
 782 #if (defined(SD_PROP_TST))
 783         { "VENDOR  PRODUCT ", (SD_CONF_BSET_THROTTLE
 784                                 | SD_CONF_BSET_CTYPE
 785                                 | SD_CONF_BSET_NRR_COUNT
 786                                 | SD_CONF_BSET_FAB_DEVID
 787                                 | SD_CONF_BSET_NOCACHE
 788                                 | SD_CONF_BSET_BSY_RETRY_COUNT
 789                                 | SD_CONF_BSET_PLAYMSF_BCD
 790                                 | SD_CONF_BSET_READSUB_BCD
 791                                 | SD_CONF_BSET_READ_TOC_TRK_BCD
 792                                 | SD_CONF_BSET_READ_TOC_ADDR_BCD
 793                                 | SD_CONF_BSET_NO_READ_HEADER
 794                                 | SD_CONF_BSET_READ_CD_XD4
 795                                 | SD_CONF_BSET_RST_RETRIES
 796                                 | SD_CONF_BSET_RSV_REL_TIME
 797                                 | SD_CONF_BSET_TUR_CHECK), &tst_properties},
 798 #endif
 799 };
 800 
 801 static const int sd_disk_table_size =
 802         sizeof (sd_disk_table)/ sizeof (sd_disk_config_t);
 803 
 804 /*
 805  * Emulation mode disk drive VID/PID table
 806  */
 807 static char sd_flash_dev_table[][25] = {
 808         "ATA     MARVELL SD88SA02",
 809         "MARVELL SD88SA02",
 810         "TOSHIBA THNSNV05",
 811 };
 812 
 813 static const int sd_flash_dev_table_size =
 814         sizeof (sd_flash_dev_table) / sizeof (sd_flash_dev_table[0]);
 815 
 816 #define SD_INTERCONNECT_PARALLEL        0
 817 #define SD_INTERCONNECT_FABRIC          1
 818 #define SD_INTERCONNECT_FIBRE           2
 819 #define SD_INTERCONNECT_SSA             3
 820 #define SD_INTERCONNECT_SATA            4
 821 #define SD_INTERCONNECT_SAS             5
 822 
 823 #define SD_IS_PARALLEL_SCSI(un)         \
 824         ((un)->un_interconnect_type == SD_INTERCONNECT_PARALLEL)
 825 #define SD_IS_SERIAL(un)                \
 826         (((un)->un_interconnect_type == SD_INTERCONNECT_SATA) ||\
 827         ((un)->un_interconnect_type == SD_INTERCONNECT_SAS))
 828 
 829 /*
 830  * Definitions used by device id registration routines
 831  */
 832 #define VPD_HEAD_OFFSET         3       /* size of head for vpd page */
 833 #define VPD_PAGE_LENGTH         3       /* offset for pge length data */
 834 #define VPD_MODE_PAGE           1       /* offset into vpd pg for "page code" */
 835 
 836 static kmutex_t sd_sense_mutex = {0};
 837 
 838 /*
 839  * Macros for updates of the driver state
 840  */
 841 #define New_state(un, s)        \
 842         (un)->un_last_state = (un)->un_state, (un)->un_state = (s)
 843 #define Restore_state(un)       \
 844         { uchar_t tmp = (un)->un_last_state; New_state((un), tmp); }
 845 
 846 static struct sd_cdbinfo sd_cdbtab[] = {
 847         { CDB_GROUP0, 0x00,        0x1FFFFF,   0xFF,        },
 848         { CDB_GROUP1, SCMD_GROUP1, 0xFFFFFFFF, 0xFFFF,      },
 849         { CDB_GROUP5, SCMD_GROUP5, 0xFFFFFFFF, 0xFFFFFFFF,  },
 850         { CDB_GROUP4, SCMD_GROUP4, 0xFFFFFFFFFFFFFFFF, 0xFFFFFFFF, },
 851 };
 852 
 853 /*
 854  * Specifies the number of seconds that must have elapsed since the last
 855  * cmd. has completed for a device to be declared idle to the PM framework.
 856  */
 857 static int sd_pm_idletime = 1;
 858 
 859 /*
 860  * Internal function prototypes
 861  */
 862 
 863 #if (defined(__fibre))
 864 /*
 865  * These #defines are to avoid namespace collisions that occur because this
 866  * code is currently used to compile two separate driver modules: sd and ssd.
 867  * All function names need to be treated this way (even if declared static)
 868  * in order to allow the debugger to resolve the names properly.
 869  * It is anticipated that in the near future the ssd module will be obsoleted,
 870  * at which time this ugliness should go away.
 871  */
 872 #define sd_log_trace                    ssd_log_trace
 873 #define sd_log_info                     ssd_log_info
 874 #define sd_log_err                      ssd_log_err
 875 #define sdprobe                         ssdprobe
 876 #define sdinfo                          ssdinfo
 877 #define sd_prop_op                      ssd_prop_op
 878 #define sd_scsi_probe_cache_init        ssd_scsi_probe_cache_init
 879 #define sd_scsi_probe_cache_fini        ssd_scsi_probe_cache_fini
 880 #define sd_scsi_clear_probe_cache       ssd_scsi_clear_probe_cache
 881 #define sd_scsi_probe_with_cache        ssd_scsi_probe_with_cache
 882 #define sd_scsi_target_lun_init         ssd_scsi_target_lun_init
 883 #define sd_scsi_target_lun_fini         ssd_scsi_target_lun_fini
 884 #define sd_scsi_get_target_lun_count    ssd_scsi_get_target_lun_count
 885 #define sd_scsi_update_lun_on_target    ssd_scsi_update_lun_on_target
 886 #define sd_spin_up_unit                 ssd_spin_up_unit
 887 #define sd_enable_descr_sense           ssd_enable_descr_sense
 888 #define sd_reenable_dsense_task         ssd_reenable_dsense_task
 889 #define sd_set_mmc_caps                 ssd_set_mmc_caps
 890 #define sd_read_unit_properties         ssd_read_unit_properties
 891 #define sd_process_sdconf_file          ssd_process_sdconf_file
 892 #define sd_process_sdconf_table         ssd_process_sdconf_table
 893 #define sd_sdconf_id_match              ssd_sdconf_id_match
 894 #define sd_blank_cmp                    ssd_blank_cmp
 895 #define sd_chk_vers1_data               ssd_chk_vers1_data
 896 #define sd_set_vers1_properties         ssd_set_vers1_properties
 897 #define sd_check_solid_state            ssd_check_solid_state
 898 #define sd_check_emulation_mode         ssd_check_emulation_mode
 899 
 900 #define sd_get_physical_geometry        ssd_get_physical_geometry
 901 #define sd_get_virtual_geometry         ssd_get_virtual_geometry
 902 #define sd_update_block_info            ssd_update_block_info
 903 #define sd_register_devid               ssd_register_devid
 904 #define sd_get_devid                    ssd_get_devid
 905 #define sd_create_devid                 ssd_create_devid
 906 #define sd_write_deviceid               ssd_write_deviceid
 907 #define sd_check_vpd_page_support       ssd_check_vpd_page_support
 908 #define sd_setup_pm                     ssd_setup_pm
 909 #define sd_create_pm_components         ssd_create_pm_components
 910 #define sd_ddi_suspend                  ssd_ddi_suspend
 911 #define sd_ddi_resume                   ssd_ddi_resume
 912 #define sd_pm_state_change              ssd_pm_state_change
 913 #define sdpower                         ssdpower
 914 #define sdattach                        ssdattach
 915 #define sddetach                        ssddetach
 916 #define sd_unit_attach                  ssd_unit_attach
 917 #define sd_unit_detach                  ssd_unit_detach
 918 #define sd_set_unit_attributes          ssd_set_unit_attributes
 919 #define sd_create_errstats              ssd_create_errstats
 920 #define sd_set_errstats                 ssd_set_errstats
 921 #define sd_set_pstats                   ssd_set_pstats
 922 #define sddump                          ssddump
 923 #define sd_scsi_poll                    ssd_scsi_poll
 924 #define sd_send_polled_RQS              ssd_send_polled_RQS
 925 #define sd_ddi_scsi_poll                ssd_ddi_scsi_poll
 926 #define sd_init_event_callbacks         ssd_init_event_callbacks
 927 #define sd_event_callback               ssd_event_callback
 928 #define sd_cache_control                ssd_cache_control
 929 #define sd_get_write_cache_enabled      ssd_get_write_cache_enabled
 930 #define sd_get_nv_sup                   ssd_get_nv_sup
 931 #define sd_make_device                  ssd_make_device
 932 #define sdopen                          ssdopen
 933 #define sdclose                         ssdclose
 934 #define sd_ready_and_valid              ssd_ready_and_valid
 935 #define sdmin                           ssdmin
 936 #define sdread                          ssdread
 937 #define sdwrite                         ssdwrite
 938 #define sdaread                         ssdaread
 939 #define sdawrite                        ssdawrite
 940 #define sdstrategy                      ssdstrategy
 941 #define sdioctl                         ssdioctl
 942 #define sd_mapblockaddr_iostart         ssd_mapblockaddr_iostart
 943 #define sd_mapblocksize_iostart         ssd_mapblocksize_iostart
 944 #define sd_checksum_iostart             ssd_checksum_iostart
 945 #define sd_checksum_uscsi_iostart       ssd_checksum_uscsi_iostart
 946 #define sd_pm_iostart                   ssd_pm_iostart
 947 #define sd_core_iostart                 ssd_core_iostart
 948 #define sd_mapblockaddr_iodone          ssd_mapblockaddr_iodone
 949 #define sd_mapblocksize_iodone          ssd_mapblocksize_iodone
 950 #define sd_checksum_iodone              ssd_checksum_iodone
 951 #define sd_checksum_uscsi_iodone        ssd_checksum_uscsi_iodone
 952 #define sd_pm_iodone                    ssd_pm_iodone
 953 #define sd_initpkt_for_buf              ssd_initpkt_for_buf
 954 #define sd_destroypkt_for_buf           ssd_destroypkt_for_buf
 955 #define sd_setup_rw_pkt                 ssd_setup_rw_pkt
 956 #define sd_setup_next_rw_pkt            ssd_setup_next_rw_pkt
 957 #define sd_buf_iodone                   ssd_buf_iodone
 958 #define sd_uscsi_strategy               ssd_uscsi_strategy
 959 #define sd_initpkt_for_uscsi            ssd_initpkt_for_uscsi
 960 #define sd_destroypkt_for_uscsi         ssd_destroypkt_for_uscsi
 961 #define sd_uscsi_iodone                 ssd_uscsi_iodone
 962 #define sd_xbuf_strategy                ssd_xbuf_strategy
 963 #define sd_xbuf_init                    ssd_xbuf_init
 964 #define sd_pm_entry                     ssd_pm_entry
 965 #define sd_pm_exit                      ssd_pm_exit
 966 
 967 #define sd_pm_idletimeout_handler       ssd_pm_idletimeout_handler
 968 #define sd_pm_timeout_handler           ssd_pm_timeout_handler
 969 
 970 #define sd_add_buf_to_waitq             ssd_add_buf_to_waitq
 971 #define sdintr                          ssdintr
 972 #define sd_start_cmds                   ssd_start_cmds
 973 #define sd_send_scsi_cmd                ssd_send_scsi_cmd
 974 #define sd_bioclone_alloc               ssd_bioclone_alloc
 975 #define sd_bioclone_free                ssd_bioclone_free
 976 #define sd_shadow_buf_alloc             ssd_shadow_buf_alloc
 977 #define sd_shadow_buf_free              ssd_shadow_buf_free
 978 #define sd_print_transport_rejected_message     \
 979                                         ssd_print_transport_rejected_message
 980 #define sd_retry_command                ssd_retry_command
 981 #define sd_set_retry_bp                 ssd_set_retry_bp
 982 #define sd_send_request_sense_command   ssd_send_request_sense_command
 983 #define sd_start_retry_command          ssd_start_retry_command
 984 #define sd_start_direct_priority_command        \
 985                                         ssd_start_direct_priority_command
 986 #define sd_return_failed_command        ssd_return_failed_command
 987 #define sd_return_failed_command_no_restart     \
 988                                         ssd_return_failed_command_no_restart
 989 #define sd_return_command               ssd_return_command
 990 #define sd_sync_with_callback           ssd_sync_with_callback
 991 #define sdrunout                        ssdrunout
 992 #define sd_mark_rqs_busy                ssd_mark_rqs_busy
 993 #define sd_mark_rqs_idle                ssd_mark_rqs_idle
 994 #define sd_reduce_throttle              ssd_reduce_throttle
 995 #define sd_restore_throttle             ssd_restore_throttle
 996 #define sd_print_incomplete_msg         ssd_print_incomplete_msg
 997 #define sd_init_cdb_limits              ssd_init_cdb_limits
 998 #define sd_pkt_status_good              ssd_pkt_status_good
 999 #define sd_pkt_status_check_condition   ssd_pkt_status_check_condition
1000 #define sd_pkt_status_busy              ssd_pkt_status_busy
1001 #define sd_pkt_status_reservation_conflict      \
1002                                         ssd_pkt_status_reservation_conflict
1003 #define sd_pkt_status_qfull             ssd_pkt_status_qfull
1004 #define sd_handle_request_sense         ssd_handle_request_sense
1005 #define sd_handle_auto_request_sense    ssd_handle_auto_request_sense
1006 #define sd_print_sense_failed_msg       ssd_print_sense_failed_msg
1007 #define sd_validate_sense_data          ssd_validate_sense_data
1008 #define sd_decode_sense                 ssd_decode_sense
1009 #define sd_print_sense_msg              ssd_print_sense_msg
1010 #define sd_sense_key_no_sense           ssd_sense_key_no_sense
1011 #define sd_sense_key_recoverable_error  ssd_sense_key_recoverable_error
1012 #define sd_sense_key_not_ready          ssd_sense_key_not_ready
1013 #define sd_sense_key_medium_or_hardware_error   \
1014                                         ssd_sense_key_medium_or_hardware_error
1015 #define sd_sense_key_illegal_request    ssd_sense_key_illegal_request
1016 #define sd_sense_key_unit_attention     ssd_sense_key_unit_attention
1017 #define sd_sense_key_fail_command       ssd_sense_key_fail_command
1018 #define sd_sense_key_blank_check        ssd_sense_key_blank_check
1019 #define sd_sense_key_aborted_command    ssd_sense_key_aborted_command
1020 #define sd_sense_key_default            ssd_sense_key_default
1021 #define sd_print_retry_msg              ssd_print_retry_msg
1022 #define sd_print_cmd_incomplete_msg     ssd_print_cmd_incomplete_msg
1023 #define sd_pkt_reason_cmd_incomplete    ssd_pkt_reason_cmd_incomplete
1024 #define sd_pkt_reason_cmd_tran_err      ssd_pkt_reason_cmd_tran_err
1025 #define sd_pkt_reason_cmd_reset         ssd_pkt_reason_cmd_reset
1026 #define sd_pkt_reason_cmd_aborted       ssd_pkt_reason_cmd_aborted
1027 #define sd_pkt_reason_cmd_timeout       ssd_pkt_reason_cmd_timeout
1028 #define sd_pkt_reason_cmd_unx_bus_free  ssd_pkt_reason_cmd_unx_bus_free
1029 #define sd_pkt_reason_cmd_tag_reject    ssd_pkt_reason_cmd_tag_reject
1030 #define sd_pkt_reason_default           ssd_pkt_reason_default
1031 #define sd_reset_target                 ssd_reset_target
1032 #define sd_start_stop_unit_callback     ssd_start_stop_unit_callback
1033 #define sd_start_stop_unit_task         ssd_start_stop_unit_task
1034 #define sd_taskq_create                 ssd_taskq_create
1035 #define sd_taskq_delete                 ssd_taskq_delete
1036 #define sd_target_change_task           ssd_target_change_task
1037 #define sd_log_dev_status_event         ssd_log_dev_status_event
1038 #define sd_log_lun_expansion_event      ssd_log_lun_expansion_event
1039 #define sd_log_eject_request_event      ssd_log_eject_request_event
1040 #define sd_media_change_task            ssd_media_change_task
1041 #define sd_handle_mchange               ssd_handle_mchange
1042 #define sd_send_scsi_DOORLOCK           ssd_send_scsi_DOORLOCK
1043 #define sd_send_scsi_READ_CAPACITY      ssd_send_scsi_READ_CAPACITY
1044 #define sd_send_scsi_READ_CAPACITY_16   ssd_send_scsi_READ_CAPACITY_16
1045 #define sd_send_scsi_GET_CONFIGURATION  ssd_send_scsi_GET_CONFIGURATION
1046 #define sd_send_scsi_feature_GET_CONFIGURATION  \
1047                                         sd_send_scsi_feature_GET_CONFIGURATION
1048 #define sd_send_scsi_START_STOP_UNIT    ssd_send_scsi_START_STOP_UNIT
1049 #define sd_send_scsi_INQUIRY            ssd_send_scsi_INQUIRY
1050 #define sd_send_scsi_TEST_UNIT_READY    ssd_send_scsi_TEST_UNIT_READY
1051 #define sd_send_scsi_PERSISTENT_RESERVE_IN      \
1052                                         ssd_send_scsi_PERSISTENT_RESERVE_IN
1053 #define sd_send_scsi_PERSISTENT_RESERVE_OUT     \
1054                                         ssd_send_scsi_PERSISTENT_RESERVE_OUT
1055 #define sd_send_scsi_SYNCHRONIZE_CACHE  ssd_send_scsi_SYNCHRONIZE_CACHE
1056 #define sd_send_scsi_SYNCHRONIZE_CACHE_biodone  \
1057                                         ssd_send_scsi_SYNCHRONIZE_CACHE_biodone
1058 #define sd_send_scsi_MODE_SENSE         ssd_send_scsi_MODE_SENSE
1059 #define sd_send_scsi_MODE_SELECT        ssd_send_scsi_MODE_SELECT
1060 #define sd_send_scsi_RDWR               ssd_send_scsi_RDWR
1061 #define sd_send_scsi_LOG_SENSE          ssd_send_scsi_LOG_SENSE
1062 #define sd_send_scsi_GET_EVENT_STATUS_NOTIFICATION      \
1063                                 ssd_send_scsi_GET_EVENT_STATUS_NOTIFICATION
1064 #define sd_gesn_media_data_valid        ssd_gesn_media_data_valid
1065 #define sd_alloc_rqs                    ssd_alloc_rqs
1066 #define sd_free_rqs                     ssd_free_rqs
1067 #define sd_dump_memory                  ssd_dump_memory
1068 #define sd_get_media_info_com           ssd_get_media_info_com
1069 #define sd_get_media_info               ssd_get_media_info
1070 #define sd_get_media_info_ext           ssd_get_media_info_ext
1071 #define sd_dkio_ctrl_info               ssd_dkio_ctrl_info
1072 #define sd_nvpair_str_decode            ssd_nvpair_str_decode
1073 #define sd_strtok_r                     ssd_strtok_r
1074 #define sd_set_properties               ssd_set_properties
1075 #define sd_get_tunables_from_conf       ssd_get_tunables_from_conf
1076 #define sd_setup_next_xfer              ssd_setup_next_xfer
1077 #define sd_dkio_get_temp                ssd_dkio_get_temp
1078 #define sd_check_mhd                    ssd_check_mhd
1079 #define sd_mhd_watch_cb                 ssd_mhd_watch_cb
1080 #define sd_mhd_watch_incomplete         ssd_mhd_watch_incomplete
1081 #define sd_sname                        ssd_sname
1082 #define sd_mhd_resvd_recover            ssd_mhd_resvd_recover
1083 #define sd_resv_reclaim_thread          ssd_resv_reclaim_thread
1084 #define sd_take_ownership               ssd_take_ownership
1085 #define sd_reserve_release              ssd_reserve_release
1086 #define sd_rmv_resv_reclaim_req         ssd_rmv_resv_reclaim_req
1087 #define sd_mhd_reset_notify_cb          ssd_mhd_reset_notify_cb
1088 #define sd_persistent_reservation_in_read_keys  \
1089                                         ssd_persistent_reservation_in_read_keys
1090 #define sd_persistent_reservation_in_read_resv  \
1091                                         ssd_persistent_reservation_in_read_resv
1092 #define sd_mhdioc_takeown               ssd_mhdioc_takeown
1093 #define sd_mhdioc_failfast              ssd_mhdioc_failfast
1094 #define sd_mhdioc_release               ssd_mhdioc_release
1095 #define sd_mhdioc_register_devid        ssd_mhdioc_register_devid
1096 #define sd_mhdioc_inkeys                ssd_mhdioc_inkeys
1097 #define sd_mhdioc_inresv                ssd_mhdioc_inresv
1098 #define sr_change_blkmode               ssr_change_blkmode
1099 #define sr_change_speed                 ssr_change_speed
1100 #define sr_atapi_change_speed           ssr_atapi_change_speed
1101 #define sr_pause_resume                 ssr_pause_resume
1102 #define sr_play_msf                     ssr_play_msf
1103 #define sr_play_trkind                  ssr_play_trkind
1104 #define sr_read_all_subcodes            ssr_read_all_subcodes
1105 #define sr_read_subchannel              ssr_read_subchannel
1106 #define sr_read_tocentry                ssr_read_tocentry
1107 #define sr_read_tochdr                  ssr_read_tochdr
1108 #define sr_read_cdda                    ssr_read_cdda
1109 #define sr_read_cdxa                    ssr_read_cdxa
1110 #define sr_read_mode1                   ssr_read_mode1
1111 #define sr_read_mode2                   ssr_read_mode2
1112 #define sr_read_cd_mode2                ssr_read_cd_mode2
1113 #define sr_sector_mode                  ssr_sector_mode
1114 #define sr_eject                        ssr_eject
1115 #define sr_ejected                      ssr_ejected
1116 #define sr_check_wp                     ssr_check_wp
1117 #define sd_watch_request_submit         ssd_watch_request_submit
1118 #define sd_check_media                  ssd_check_media
1119 #define sd_media_watch_cb               ssd_media_watch_cb
1120 #define sd_delayed_cv_broadcast         ssd_delayed_cv_broadcast
1121 #define sr_volume_ctrl                  ssr_volume_ctrl
1122 #define sr_read_sony_session_offset     ssr_read_sony_session_offset
1123 #define sd_log_page_supported           ssd_log_page_supported
1124 #define sd_check_for_writable_cd        ssd_check_for_writable_cd
1125 #define sd_wm_cache_constructor         ssd_wm_cache_constructor
1126 #define sd_wm_cache_destructor          ssd_wm_cache_destructor
1127 #define sd_range_lock                   ssd_range_lock
1128 #define sd_get_range                    ssd_get_range
1129 #define sd_free_inlist_wmap             ssd_free_inlist_wmap
1130 #define sd_range_unlock                 ssd_range_unlock
1131 #define sd_read_modify_write_task       ssd_read_modify_write_task
1132 #define sddump_do_read_of_rmw           ssddump_do_read_of_rmw
1133 
1134 #define sd_iostart_chain                ssd_iostart_chain
1135 #define sd_iodone_chain                 ssd_iodone_chain
1136 #define sd_initpkt_map                  ssd_initpkt_map
1137 #define sd_destroypkt_map               ssd_destroypkt_map
1138 #define sd_chain_type_map               ssd_chain_type_map
1139 #define sd_chain_index_map              ssd_chain_index_map
1140 
1141 #define sd_failfast_flushctl            ssd_failfast_flushctl
1142 #define sd_failfast_flushq              ssd_failfast_flushq
1143 #define sd_failfast_flushq_callback     ssd_failfast_flushq_callback
1144 
1145 #define sd_is_lsi                       ssd_is_lsi
1146 #define sd_tg_rdwr                      ssd_tg_rdwr
1147 #define sd_tg_getinfo                   ssd_tg_getinfo
1148 #define sd_rmw_msg_print_handler        ssd_rmw_msg_print_handler
1149 
1150 #endif  /* #if (defined(__fibre)) */
1151 
1152 
1153 int _init(void);
1154 int _fini(void);
1155 int _info(struct modinfo *modinfop);
1156 
1157 /*PRINTFLIKE3*/
1158 static void sd_log_trace(uint_t comp, struct sd_lun *un, const char *fmt, ...);
1159 /*PRINTFLIKE3*/
1160 static void sd_log_info(uint_t comp, struct sd_lun *un, const char *fmt, ...);
1161 /*PRINTFLIKE3*/
1162 static void sd_log_err(uint_t comp, struct sd_lun *un, const char *fmt, ...);
1163 
1164 static int sdprobe(dev_info_t *devi);
1165 static int sdinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg,
1166     void **result);
1167 static int sd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op,
1168     int mod_flags, char *name, caddr_t valuep, int *lengthp);
1169 
1170 /*
1171  * Smart probe for parallel scsi
1172  */
1173 static void sd_scsi_probe_cache_init(void);
1174 static void sd_scsi_probe_cache_fini(void);
1175 static void sd_scsi_clear_probe_cache(void);
1176 static int  sd_scsi_probe_with_cache(struct scsi_device *devp, int (*fn)());
1177 
1178 /*
1179  * Attached luns on target for parallel scsi
1180  */
1181 static void sd_scsi_target_lun_init(void);
1182 static void sd_scsi_target_lun_fini(void);
1183 static int  sd_scsi_get_target_lun_count(dev_info_t *dip, int target);
1184 static void sd_scsi_update_lun_on_target(dev_info_t *dip, int target, int flag);
1185 
1186 static int      sd_spin_up_unit(sd_ssc_t *ssc);
1187 
1188 /*
1189  * Using sd_ssc_init to establish sd_ssc_t struct
1190  * Using sd_ssc_send to send uscsi internal command
1191  * Using sd_ssc_fini to free sd_ssc_t struct
1192  */
1193 static sd_ssc_t *sd_ssc_init(struct sd_lun *un);
1194 static int sd_ssc_send(sd_ssc_t *ssc, struct uscsi_cmd *incmd,
1195     int flag, enum uio_seg dataspace, int path_flag);
1196 static void sd_ssc_fini(sd_ssc_t *ssc);
1197 
1198 /*
1199  * Using sd_ssc_assessment to set correct type-of-assessment
1200  * Using sd_ssc_post to post ereport & system log
1201  *       sd_ssc_post will call sd_ssc_print to print system log
1202  *       sd_ssc_post will call sd_ssd_ereport_post to post ereport
1203  */
1204 static void sd_ssc_assessment(sd_ssc_t *ssc,
1205     enum sd_type_assessment tp_assess);
1206 
1207 static void sd_ssc_post(sd_ssc_t *ssc, enum sd_driver_assessment sd_assess);
1208 static void sd_ssc_print(sd_ssc_t *ssc, int sd_severity);
1209 static void sd_ssc_ereport_post(sd_ssc_t *ssc,
1210     enum sd_driver_assessment drv_assess);
1211 
1212 /*
1213  * Using sd_ssc_set_info to mark an un-decodable-data error.
1214  * Using sd_ssc_extract_info to transfer information from internal
1215  *       data structures to sd_ssc_t.
1216  */
1217 static void sd_ssc_set_info(sd_ssc_t *ssc, int ssc_flags, uint_t comp,
1218     const char *fmt, ...);
1219 static void sd_ssc_extract_info(sd_ssc_t *ssc, struct sd_lun *un,
1220     struct scsi_pkt *pktp, struct buf *bp, struct sd_xbuf *xp);
1221 
1222 static int sd_send_scsi_cmd(dev_t dev, struct uscsi_cmd *incmd, int flag,
1223     enum uio_seg dataspace, int path_flag);
1224 
1225 #ifdef _LP64
1226 static void     sd_enable_descr_sense(sd_ssc_t *ssc);
1227 static void     sd_reenable_dsense_task(void *arg);
1228 #endif /* _LP64 */
1229 
1230 static void     sd_set_mmc_caps(sd_ssc_t *ssc);
1231 
1232 static void sd_read_unit_properties(struct sd_lun *un);
1233 static int  sd_process_sdconf_file(struct sd_lun *un);
1234 static void sd_nvpair_str_decode(struct sd_lun *un, char *nvpair_str);
1235 static char *sd_strtok_r(char *string, const char *sepset, char **lasts);
1236 static void sd_set_properties(struct sd_lun *un, char *name, char *value);
1237 static void sd_get_tunables_from_conf(struct sd_lun *un, int flags,
1238     int *data_list, sd_tunables *values);
1239 static void sd_process_sdconf_table(struct sd_lun *un);
1240 static int  sd_sdconf_id_match(struct sd_lun *un, char *id, int idlen);
1241 static int  sd_blank_cmp(struct sd_lun *un, char *id, int idlen);
1242 static int  sd_chk_vers1_data(struct sd_lun *un, int flags, int *prop_list,
1243         int list_len, char *dataname_ptr);
1244 static void sd_set_vers1_properties(struct sd_lun *un, int flags,
1245     sd_tunables *prop_list);
1246 
1247 static void sd_register_devid(sd_ssc_t *ssc, dev_info_t *devi,
1248     int reservation_flag);
1249 static int  sd_get_devid(sd_ssc_t *ssc);
1250 static ddi_devid_t sd_create_devid(sd_ssc_t *ssc);
1251 static int  sd_write_deviceid(sd_ssc_t *ssc);
1252 static int  sd_get_devid_page(struct sd_lun *un, uchar_t *wwn, int *len);
1253 static int  sd_check_vpd_page_support(sd_ssc_t *ssc);
1254 
1255 static void sd_setup_pm(sd_ssc_t *ssc, dev_info_t *devi);
1256 static void sd_create_pm_components(dev_info_t *devi, struct sd_lun *un);
1257 
1258 static int  sd_ddi_suspend(dev_info_t *devi);
1259 static int  sd_ddi_resume(dev_info_t *devi);
1260 static int  sd_pm_state_change(struct sd_lun *un, int level, int flag);
1261 static int  sdpower(dev_info_t *devi, int component, int level);
1262 
1263 static int  sdattach(dev_info_t *devi, ddi_attach_cmd_t cmd);
1264 static int  sddetach(dev_info_t *devi, ddi_detach_cmd_t cmd);
1265 static int  sd_unit_attach(dev_info_t *devi);
1266 static int  sd_unit_detach(dev_info_t *devi);
1267 
1268 static void sd_set_unit_attributes(struct sd_lun *un, dev_info_t *devi);
1269 static void sd_create_errstats(struct sd_lun *un, int instance);
1270 static void sd_set_errstats(struct sd_lun *un);
1271 static void sd_set_pstats(struct sd_lun *un);
1272 
1273 static int  sddump(dev_t dev, caddr_t addr, daddr_t blkno, int nblk);
1274 static int  sd_scsi_poll(struct sd_lun *un, struct scsi_pkt *pkt);
1275 static int  sd_send_polled_RQS(struct sd_lun *un);
1276 static int  sd_ddi_scsi_poll(struct scsi_pkt *pkt);
1277 
1278 #if (defined(__fibre))
1279 /*
1280  * Event callbacks (photon)
1281  */
1282 static void sd_init_event_callbacks(struct sd_lun *un);
1283 static void  sd_event_callback(dev_info_t *, ddi_eventcookie_t, void *, void *);
1284 #endif
1285 
1286 /*
1287  * Defines for sd_cache_control
1288  */
1289 
1290 #define SD_CACHE_ENABLE         1
1291 #define SD_CACHE_DISABLE        0
1292 #define SD_CACHE_NOCHANGE       -1
1293 
1294 static int   sd_cache_control(sd_ssc_t *ssc, int rcd_flag, int wce_flag);
1295 static int   sd_get_write_cache_enabled(sd_ssc_t *ssc, int *is_enabled);
1296 static void  sd_get_nv_sup(sd_ssc_t *ssc);
1297 static dev_t sd_make_device(dev_info_t *devi);
1298 static void  sd_check_solid_state(sd_ssc_t *ssc);
1299 static void  sd_check_emulation_mode(sd_ssc_t *ssc);
1300 static void  sd_update_block_info(struct sd_lun *un, uint32_t lbasize,
1301         uint64_t capacity);
1302 
1303 /*
1304  * Driver entry point functions.
1305  */
1306 static int  sdopen(dev_t *dev_p, int flag, int otyp, cred_t *cred_p);
1307 static int  sdclose(dev_t dev, int flag, int otyp, cred_t *cred_p);
1308 static int  sd_ready_and_valid(sd_ssc_t *ssc, int part);
1309 
1310 static void sdmin(struct buf *bp);
1311 static int sdread(dev_t dev, struct uio *uio, cred_t *cred_p);
1312 static int sdwrite(dev_t dev, struct uio *uio, cred_t *cred_p);
1313 static int sdaread(dev_t dev, struct aio_req *aio, cred_t *cred_p);
1314 static int sdawrite(dev_t dev, struct aio_req *aio, cred_t *cred_p);
1315 
1316 static int sdstrategy(struct buf *bp);
1317 static int sdioctl(dev_t, int, intptr_t, int, cred_t *, int *);
1318 
1319 /*
1320  * Function prototypes for layering functions in the iostart chain.
1321  */
1322 static void sd_mapblockaddr_iostart(int index, struct sd_lun *un,
1323         struct buf *bp);
1324 static void sd_mapblocksize_iostart(int index, struct sd_lun *un,
1325         struct buf *bp);
1326 static void sd_checksum_iostart(int index, struct sd_lun *un, struct buf *bp);
1327 static void sd_checksum_uscsi_iostart(int index, struct sd_lun *un,
1328         struct buf *bp);
1329 static void sd_pm_iostart(int index, struct sd_lun *un, struct buf *bp);
1330 static void sd_core_iostart(int index, struct sd_lun *un, struct buf *bp);
1331 
1332 /*
1333  * Function prototypes for layering functions in the iodone chain.
1334  */
1335 static void sd_buf_iodone(int index, struct sd_lun *un, struct buf *bp);
1336 static void sd_uscsi_iodone(int index, struct sd_lun *un, struct buf *bp);
1337 static void sd_mapblockaddr_iodone(int index, struct sd_lun *un,
1338         struct buf *bp);
1339 static void sd_mapblocksize_iodone(int index, struct sd_lun *un,
1340         struct buf *bp);
1341 static void sd_checksum_iodone(int index, struct sd_lun *un, struct buf *bp);
1342 static void sd_checksum_uscsi_iodone(int index, struct sd_lun *un,
1343         struct buf *bp);
1344 static void sd_pm_iodone(int index, struct sd_lun *un, struct buf *bp);
1345 
1346 /*
1347  * Prototypes for functions to support buf(9S) based IO.
1348  */
1349 static void sd_xbuf_strategy(struct buf *bp, ddi_xbuf_t xp, void *arg);
1350 static int sd_initpkt_for_buf(struct buf *, struct scsi_pkt **);
1351 static void sd_destroypkt_for_buf(struct buf *);
1352 static int sd_setup_rw_pkt(struct sd_lun *un, struct scsi_pkt **pktpp,
1353         struct buf *bp, int flags,
1354         int (*callback)(caddr_t), caddr_t callback_arg,
1355         diskaddr_t lba, uint32_t blockcount);
1356 static int sd_setup_next_rw_pkt(struct sd_lun *un, struct scsi_pkt *pktp,
1357         struct buf *bp, diskaddr_t lba, uint32_t blockcount);
1358 
1359 /*
1360  * Prototypes for functions to support USCSI IO.
1361  */
1362 static int sd_uscsi_strategy(struct buf *bp);
1363 static int sd_initpkt_for_uscsi(struct buf *, struct scsi_pkt **);
1364 static void sd_destroypkt_for_uscsi(struct buf *);
1365 
1366 static void sd_xbuf_init(struct sd_lun *un, struct buf *bp, struct sd_xbuf *xp,
1367         uchar_t chain_type, void *pktinfop);
1368 
1369 static int  sd_pm_entry(struct sd_lun *un);
1370 static void sd_pm_exit(struct sd_lun *un);
1371 
1372 static void sd_pm_idletimeout_handler(void *arg);
1373 
1374 /*
1375  * sd_core internal functions (used at the sd_core_io layer).
1376  */
1377 static void sd_add_buf_to_waitq(struct sd_lun *un, struct buf *bp);
1378 static void sdintr(struct scsi_pkt *pktp);
1379 static void sd_start_cmds(struct sd_lun *un, struct buf *immed_bp);
1380 
1381 static int sd_send_scsi_cmd(dev_t dev, struct uscsi_cmd *incmd, int flag,
1382         enum uio_seg dataspace, int path_flag);
1383 
1384 static struct buf *sd_bioclone_alloc(struct buf *bp, size_t datalen,
1385         daddr_t blkno, int (*func)(struct buf *));
1386 static struct buf *sd_shadow_buf_alloc(struct buf *bp, size_t datalen,
1387         uint_t bflags, daddr_t blkno, int (*func)(struct buf *));
1388 static void sd_bioclone_free(struct buf *bp);
1389 static void sd_shadow_buf_free(struct buf *bp);
1390 
1391 static void sd_print_transport_rejected_message(struct sd_lun *un,
1392         struct sd_xbuf *xp, int code);
1393 static void sd_print_incomplete_msg(struct sd_lun *un, struct buf *bp,
1394     void *arg, int code);
1395 static void sd_print_sense_failed_msg(struct sd_lun *un, struct buf *bp,
1396     void *arg, int code);
1397 static void sd_print_cmd_incomplete_msg(struct sd_lun *un, struct buf *bp,
1398     void *arg, int code);
1399 
1400 static void sd_retry_command(struct sd_lun *un, struct buf *bp,
1401         int retry_check_flag,
1402         void (*user_funcp)(struct sd_lun *un, struct buf *bp, void *argp,
1403                 int c),
1404         void *user_arg, int failure_code,  clock_t retry_delay,
1405         void (*statp)(kstat_io_t *));
1406 
1407 static void sd_set_retry_bp(struct sd_lun *un, struct buf *bp,
1408         clock_t retry_delay, void (*statp)(kstat_io_t *));
1409 
1410 static void sd_send_request_sense_command(struct sd_lun *un, struct buf *bp,
1411         struct scsi_pkt *pktp);
1412 static void sd_start_retry_command(void *arg);
1413 static void sd_start_direct_priority_command(void *arg);
1414 static void sd_return_failed_command(struct sd_lun *un, struct buf *bp,
1415         int errcode);
1416 static void sd_return_failed_command_no_restart(struct sd_lun *un,
1417         struct buf *bp, int errcode);
1418 static void sd_return_command(struct sd_lun *un, struct buf *bp);
1419 static void sd_sync_with_callback(struct sd_lun *un);
1420 static int sdrunout(caddr_t arg);
1421 
1422 static void sd_mark_rqs_busy(struct sd_lun *un, struct buf *bp);
1423 static struct buf *sd_mark_rqs_idle(struct sd_lun *un, struct sd_xbuf *xp);
1424 
1425 static void sd_reduce_throttle(struct sd_lun *un, int throttle_type);
1426 static void sd_restore_throttle(void *arg);
1427 
1428 static void sd_init_cdb_limits(struct sd_lun *un);
1429 
1430 static void sd_pkt_status_good(struct sd_lun *un, struct buf *bp,
1431         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1432 
1433 /*
1434  * Error handling functions
1435  */
1436 static void sd_pkt_status_check_condition(struct sd_lun *un, struct buf *bp,
1437         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1438 static void sd_pkt_status_busy(struct sd_lun *un, struct buf *bp,
1439         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1440 static void sd_pkt_status_reservation_conflict(struct sd_lun *un,
1441         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp);
1442 static void sd_pkt_status_qfull(struct sd_lun *un, struct buf *bp,
1443         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1444 
1445 static void sd_handle_request_sense(struct sd_lun *un, struct buf *bp,
1446         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1447 static void sd_handle_auto_request_sense(struct sd_lun *un, struct buf *bp,
1448         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1449 static int sd_validate_sense_data(struct sd_lun *un, struct buf *bp,
1450         struct sd_xbuf *xp, size_t actual_len);
1451 static void sd_decode_sense(struct sd_lun *un, struct buf *bp,
1452         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1453 
1454 static void sd_print_sense_msg(struct sd_lun *un, struct buf *bp,
1455         void *arg, int code);
1456 
1457 static void sd_sense_key_no_sense(struct sd_lun *un, struct buf *bp,
1458         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1459 static void sd_sense_key_recoverable_error(struct sd_lun *un,
1460         uint8_t *sense_datap,
1461         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp);
1462 static void sd_sense_key_not_ready(struct sd_lun *un,
1463         uint8_t *sense_datap,
1464         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp);
1465 static void sd_sense_key_medium_or_hardware_error(struct sd_lun *un,
1466         uint8_t *sense_datap,
1467         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp);
1468 static void sd_sense_key_illegal_request(struct sd_lun *un, struct buf *bp,
1469         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1470 static void sd_sense_key_unit_attention(struct sd_lun *un,
1471         uint8_t *sense_datap,
1472         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp);
1473 static void sd_sense_key_fail_command(struct sd_lun *un, struct buf *bp,
1474         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1475 static void sd_sense_key_blank_check(struct sd_lun *un, struct buf *bp,
1476         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1477 static void sd_sense_key_aborted_command(struct sd_lun *un, struct buf *bp,
1478         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1479 static void sd_sense_key_default(struct sd_lun *un,
1480         uint8_t *sense_datap,
1481         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp);
1482 
1483 static void sd_print_retry_msg(struct sd_lun *un, struct buf *bp,
1484         void *arg, int flag);
1485 
1486 static void sd_pkt_reason_cmd_incomplete(struct sd_lun *un, struct buf *bp,
1487         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1488 static void sd_pkt_reason_cmd_tran_err(struct sd_lun *un, struct buf *bp,
1489         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1490 static void sd_pkt_reason_cmd_reset(struct sd_lun *un, struct buf *bp,
1491         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1492 static void sd_pkt_reason_cmd_aborted(struct sd_lun *un, struct buf *bp,
1493         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1494 static void sd_pkt_reason_cmd_timeout(struct sd_lun *un, struct buf *bp,
1495         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1496 static void sd_pkt_reason_cmd_unx_bus_free(struct sd_lun *un, struct buf *bp,
1497         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1498 static void sd_pkt_reason_cmd_tag_reject(struct sd_lun *un, struct buf *bp,
1499         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1500 static void sd_pkt_reason_default(struct sd_lun *un, struct buf *bp,
1501         struct sd_xbuf *xp, struct scsi_pkt *pktp);
1502 
1503 static void sd_reset_target(struct sd_lun *un, struct scsi_pkt *pktp);
1504 
1505 static void sd_start_stop_unit_callback(void *arg);
1506 static void sd_start_stop_unit_task(void *arg);
1507 
1508 static void sd_taskq_create(void);
1509 static void sd_taskq_delete(void);
1510 static void sd_target_change_task(void *arg);
1511 static void sd_log_dev_status_event(struct sd_lun *un, char *esc, int km_flag);
1512 static void sd_log_lun_expansion_event(struct sd_lun *un, int km_flag);
1513 static void sd_log_eject_request_event(struct sd_lun *un, int km_flag);
1514 static void sd_media_change_task(void *arg);
1515 
1516 static int sd_handle_mchange(struct sd_lun *un);
1517 static int sd_send_scsi_DOORLOCK(sd_ssc_t *ssc, int flag, int path_flag);
1518 static int sd_send_scsi_READ_CAPACITY(sd_ssc_t *ssc, uint64_t *capp,
1519         uint32_t *lbap, int path_flag);
1520 static int sd_send_scsi_READ_CAPACITY_16(sd_ssc_t *ssc, uint64_t *capp,
1521         uint32_t *lbap, uint32_t *psp, int path_flag);
1522 static int sd_send_scsi_START_STOP_UNIT(sd_ssc_t *ssc, int pc_flag,
1523         int flag, int path_flag);
1524 static int sd_send_scsi_INQUIRY(sd_ssc_t *ssc, uchar_t *bufaddr,
1525         size_t buflen, uchar_t evpd, uchar_t page_code, size_t *residp);
1526 static int sd_send_scsi_TEST_UNIT_READY(sd_ssc_t *ssc, int flag);
1527 static int sd_send_scsi_PERSISTENT_RESERVE_IN(sd_ssc_t *ssc,
1528         uchar_t usr_cmd, uint16_t data_len, uchar_t *data_bufp);
1529 static int sd_send_scsi_PERSISTENT_RESERVE_OUT(sd_ssc_t *ssc,
1530         uchar_t usr_cmd, uchar_t *usr_bufp);
1531 static int sd_send_scsi_SYNCHRONIZE_CACHE(struct sd_lun *un,
1532         struct dk_callback *dkc);
1533 static int sd_send_scsi_SYNCHRONIZE_CACHE_biodone(struct buf *bp);
1534 static int sd_send_scsi_GET_CONFIGURATION(sd_ssc_t *ssc,
1535         struct uscsi_cmd *ucmdbuf, uchar_t *rqbuf, uint_t rqbuflen,
1536         uchar_t *bufaddr, uint_t buflen, int path_flag);
1537 static int sd_send_scsi_feature_GET_CONFIGURATION(sd_ssc_t *ssc,
1538         struct uscsi_cmd *ucmdbuf, uchar_t *rqbuf, uint_t rqbuflen,
1539         uchar_t *bufaddr, uint_t buflen, char feature, int path_flag);
1540 static int sd_send_scsi_MODE_SENSE(sd_ssc_t *ssc, int cdbsize,
1541         uchar_t *bufaddr, size_t buflen, uchar_t page_code, int path_flag);
1542 static int sd_send_scsi_MODE_SELECT(sd_ssc_t *ssc, int cdbsize,
1543         uchar_t *bufaddr, size_t buflen, uchar_t save_page, int path_flag);
1544 static int sd_send_scsi_RDWR(sd_ssc_t *ssc, uchar_t cmd, void *bufaddr,
1545         size_t buflen, daddr_t start_block, int path_flag);
1546 #define sd_send_scsi_READ(ssc, bufaddr, buflen, start_block, path_flag) \
1547         sd_send_scsi_RDWR(ssc, SCMD_READ, bufaddr, buflen, start_block, \
1548         path_flag)
1549 #define sd_send_scsi_WRITE(ssc, bufaddr, buflen, start_block, path_flag)\
1550         sd_send_scsi_RDWR(ssc, SCMD_WRITE, bufaddr, buflen, start_block,\
1551         path_flag)
1552 
1553 static int sd_send_scsi_LOG_SENSE(sd_ssc_t *ssc, uchar_t *bufaddr,
1554         uint16_t buflen, uchar_t page_code, uchar_t page_control,
1555         uint16_t param_ptr, int path_flag);
1556 static int sd_send_scsi_GET_EVENT_STATUS_NOTIFICATION(sd_ssc_t *ssc,
1557         uchar_t *bufaddr, size_t buflen, uchar_t class_req);
1558 static boolean_t sd_gesn_media_data_valid(uchar_t *data);
1559 
1560 static int  sd_alloc_rqs(struct scsi_device *devp, struct sd_lun *un);
1561 static void sd_free_rqs(struct sd_lun *un);
1562 
1563 static void sd_dump_memory(struct sd_lun *un, uint_t comp, char *title,
1564         uchar_t *data, int len, int fmt);
1565 static void sd_panic_for_res_conflict(struct sd_lun *un);
1566 
1567 /*
1568  * Disk Ioctl Function Prototypes
1569  */
1570 static int sd_get_media_info(dev_t dev, caddr_t arg, int flag);
1571 static int sd_get_media_info_ext(dev_t dev, caddr_t arg, int flag);
1572 static int sd_dkio_ctrl_info(dev_t dev, caddr_t arg, int flag);
1573 static int sd_dkio_get_temp(dev_t dev, caddr_t arg, int flag);
1574 
1575 /*
1576  * Multi-host Ioctl Prototypes
1577  */
1578 static int sd_check_mhd(dev_t dev, int interval);
1579 static int sd_mhd_watch_cb(caddr_t arg, struct scsi_watch_result *resultp);
1580 static void sd_mhd_watch_incomplete(struct sd_lun *un, struct scsi_pkt *pkt);
1581 static char *sd_sname(uchar_t status);
1582 static void sd_mhd_resvd_recover(void *arg);
1583 static void sd_resv_reclaim_thread();
1584 static int sd_take_ownership(dev_t dev, struct mhioctkown *p);
1585 static int sd_reserve_release(dev_t dev, int cmd);
1586 static void sd_rmv_resv_reclaim_req(dev_t dev);
1587 static void sd_mhd_reset_notify_cb(caddr_t arg);
1588 static int sd_persistent_reservation_in_read_keys(struct sd_lun *un,
1589         mhioc_inkeys_t *usrp, int flag);
1590 static int sd_persistent_reservation_in_read_resv(struct sd_lun *un,
1591         mhioc_inresvs_t *usrp, int flag);
1592 static int sd_mhdioc_takeown(dev_t dev, caddr_t arg, int flag);
1593 static int sd_mhdioc_failfast(dev_t dev, caddr_t arg, int flag);
1594 static int sd_mhdioc_release(dev_t dev);
1595 static int sd_mhdioc_register_devid(dev_t dev);
1596 static int sd_mhdioc_inkeys(dev_t dev, caddr_t arg, int flag);
1597 static int sd_mhdioc_inresv(dev_t dev, caddr_t arg, int flag);
1598 
1599 /*
1600  * SCSI removable prototypes
1601  */
1602 static int sr_change_blkmode(dev_t dev, int cmd, intptr_t data, int flag);
1603 static int sr_change_speed(dev_t dev, int cmd, intptr_t data, int flag);
1604 static int sr_atapi_change_speed(dev_t dev, int cmd, intptr_t data, int flag);
1605 static int sr_pause_resume(dev_t dev, int mode);
1606 static int sr_play_msf(dev_t dev, caddr_t data, int flag);
1607 static int sr_play_trkind(dev_t dev, caddr_t data, int flag);
1608 static int sr_read_all_subcodes(dev_t dev, caddr_t data, int flag);
1609 static int sr_read_subchannel(dev_t dev, caddr_t data, int flag);
1610 static int sr_read_tocentry(dev_t dev, caddr_t data, int flag);
1611 static int sr_read_tochdr(dev_t dev, caddr_t data, int flag);
1612 static int sr_read_cdda(dev_t dev, caddr_t data, int flag);
1613 static int sr_read_cdxa(dev_t dev, caddr_t data, int flag);
1614 static int sr_read_mode1(dev_t dev, caddr_t data, int flag);
1615 static int sr_read_mode2(dev_t dev, caddr_t data, int flag);
1616 static int sr_read_cd_mode2(dev_t dev, caddr_t data, int flag);
1617 static int sr_sector_mode(dev_t dev, uint32_t blksize);
1618 static int sr_eject(dev_t dev);
1619 static void sr_ejected(register struct sd_lun *un);
1620 static int sr_check_wp(dev_t dev);
1621 static opaque_t sd_watch_request_submit(struct sd_lun *un);
1622 static int sd_check_media(dev_t dev, enum dkio_state state);
1623 static int sd_media_watch_cb(caddr_t arg, struct scsi_watch_result *resultp);
1624 static void sd_delayed_cv_broadcast(void *arg);
1625 static int sr_volume_ctrl(dev_t dev, caddr_t data, int flag);
1626 static int sr_read_sony_session_offset(dev_t dev, caddr_t data, int flag);
1627 
1628 static int sd_log_page_supported(sd_ssc_t *ssc, int log_page);
1629 
1630 /*
1631  * Function Prototype for the non-512 support (DVDRAM, MO etc.) functions.
1632  */
1633 static void sd_check_for_writable_cd(sd_ssc_t *ssc, int path_flag);
1634 static int sd_wm_cache_constructor(void *wm, void *un, int flags);
1635 static void sd_wm_cache_destructor(void *wm, void *un);
1636 static struct sd_w_map *sd_range_lock(struct sd_lun *un, daddr_t startb,
1637         daddr_t endb, ushort_t typ);
1638 static struct sd_w_map *sd_get_range(struct sd_lun *un, daddr_t startb,
1639         daddr_t endb);
1640 static void sd_free_inlist_wmap(struct sd_lun *un, struct sd_w_map *wmp);
1641 static void sd_range_unlock(struct sd_lun *un, struct sd_w_map *wm);
1642 static void sd_read_modify_write_task(void * arg);
1643 static int
1644 sddump_do_read_of_rmw(struct sd_lun *un, uint64_t blkno, uint64_t nblk,
1645         struct buf **bpp);
1646 
1647 
1648 /*
1649  * Function prototypes for failfast support.
1650  */
1651 static void sd_failfast_flushq(struct sd_lun *un);
1652 static int sd_failfast_flushq_callback(struct buf *bp);
1653 
1654 /*
1655  * Function prototypes to check for lsi devices
1656  */
1657 static void sd_is_lsi(struct sd_lun *un);
1658 
1659 /*
1660  * Function prototypes for partial DMA support
1661  */
1662 static int sd_setup_next_xfer(struct sd_lun *un, struct buf *bp,
1663                 struct scsi_pkt *pkt, struct sd_xbuf *xp);
1664 
1665 
1666 /* Function prototypes for cmlb */
1667 static int sd_tg_rdwr(dev_info_t *devi, uchar_t cmd, void *bufaddr,
1668     diskaddr_t start_block, size_t reqlength, void *tg_cookie);
1669 
1670 static int sd_tg_getinfo(dev_info_t *devi, int cmd, void *arg, void *tg_cookie);
1671 
1672 /*
1673  * For printing RMW warning message timely
1674  */
1675 static void sd_rmw_msg_print_handler(void *arg);
1676 
1677 /*
1678  * Constants for failfast support:
1679  *
1680  * SD_FAILFAST_INACTIVE: Instance is currently in a normal state, with NO
1681  * failfast processing being performed.
1682  *
1683  * SD_FAILFAST_ACTIVE: Instance is in the failfast state and is performing
1684  * failfast processing on all bufs with B_FAILFAST set.
1685  */
1686 
1687 #define SD_FAILFAST_INACTIVE            0
1688 #define SD_FAILFAST_ACTIVE              1
1689 
1690 /*
1691  * Bitmask to control behavior of buf(9S) flushes when a transition to
1692  * the failfast state occurs. Optional bits include:
1693  *
1694  * SD_FAILFAST_FLUSH_ALL_BUFS: When set, flush ALL bufs including those that
1695  * do NOT have B_FAILFAST set. When clear, only bufs with B_FAILFAST will
1696  * be flushed.
1697  *
1698  * SD_FAILFAST_FLUSH_ALL_QUEUES: When set, flush any/all other queues in the
1699  * driver, in addition to the regular wait queue. This includes the xbuf
1700  * queues. When clear, only the driver's wait queue will be flushed.
1701  */
1702 #define SD_FAILFAST_FLUSH_ALL_BUFS      0x01
1703 #define SD_FAILFAST_FLUSH_ALL_QUEUES    0x02
1704 
1705 /*
1706  * The default behavior is to only flush bufs that have B_FAILFAST set, but
1707  * to flush all queues within the driver.
1708  */
1709 static int sd_failfast_flushctl = SD_FAILFAST_FLUSH_ALL_QUEUES;
1710 
1711 
1712 /*
1713  * SD Testing Fault Injection
1714  */
1715 #ifdef SD_FAULT_INJECTION
1716 static void sd_faultinjection_ioctl(int cmd, intptr_t arg, struct sd_lun *un);
1717 static void sd_faultinjection(struct scsi_pkt *pktp);
1718 static void sd_injection_log(char *buf, struct sd_lun *un);
1719 #endif
1720 
1721 /*
1722  * Device driver ops vector
1723  */
1724 static struct cb_ops sd_cb_ops = {
1725         sdopen,                 /* open */
1726         sdclose,                /* close */
1727         sdstrategy,             /* strategy */
1728         nodev,                  /* print */
1729         sddump,                 /* dump */
1730         sdread,                 /* read */
1731         sdwrite,                /* write */
1732         sdioctl,                /* ioctl */
1733         nodev,                  /* devmap */
1734         nodev,                  /* mmap */
1735         nodev,                  /* segmap */
1736         nochpoll,               /* poll */
1737         sd_prop_op,             /* cb_prop_op */
1738         0,                      /* streamtab  */
1739         D_64BIT | D_MP | D_NEW | D_HOTPLUG, /* Driver compatibility flags */
1740         CB_REV,                 /* cb_rev */
1741         sdaread,                /* async I/O read entry point */
1742         sdawrite                /* async I/O write entry point */
1743 };
1744 
1745 struct dev_ops sd_ops = {
1746         DEVO_REV,               /* devo_rev, */
1747         0,                      /* refcnt  */
1748         sdinfo,                 /* info */
1749         nulldev,                /* identify */
1750         sdprobe,                /* probe */
1751         sdattach,               /* attach */
1752         sddetach,               /* detach */
1753         nodev,                  /* reset */
1754         &sd_cb_ops,         /* driver operations */
1755         NULL,                   /* bus operations */
1756         sdpower,                /* power */
1757         ddi_quiesce_not_needed,         /* quiesce */
1758 };
1759 
1760 /*
1761  * This is the loadable module wrapper.
1762  */
1763 #include <sys/modctl.h>
1764 
1765 #ifndef XPV_HVM_DRIVER
1766 static struct modldrv modldrv = {
1767         &mod_driverops,             /* Type of module. This one is a driver */
1768         SD_MODULE_NAME,         /* Module name. */
1769         &sd_ops                     /* driver ops */
1770 };
1771 
1772 static struct modlinkage modlinkage = {
1773         MODREV_1, &modldrv, NULL
1774 };
1775 
1776 #else /* XPV_HVM_DRIVER */
1777 static struct modlmisc modlmisc = {
1778         &mod_miscops,               /* Type of module. This one is a misc */
1779         "HVM " SD_MODULE_NAME,          /* Module name. */
1780 };
1781 
1782 static struct modlinkage modlinkage = {
1783         MODREV_1, &modlmisc, NULL
1784 };
1785 
1786 #endif /* XPV_HVM_DRIVER */
1787 
1788 static cmlb_tg_ops_t sd_tgops = {
1789         TG_DK_OPS_VERSION_1,
1790         sd_tg_rdwr,
1791         sd_tg_getinfo
1792 };
1793 
1794 static struct scsi_asq_key_strings sd_additional_codes[] = {
1795         0x81, 0, "Logical Unit is Reserved",
1796         0x85, 0, "Audio Address Not Valid",
1797         0xb6, 0, "Media Load Mechanism Failed",
1798         0xB9, 0, "Audio Play Operation Aborted",
1799         0xbf, 0, "Buffer Overflow for Read All Subcodes Command",
1800         0x53, 2, "Medium removal prevented",
1801         0x6f, 0, "Authentication failed during key exchange",
1802         0x6f, 1, "Key not present",
1803         0x6f, 2, "Key not established",
1804         0x6f, 3, "Read without proper authentication",
1805         0x6f, 4, "Mismatched region to this logical unit",
1806         0x6f, 5, "Region reset count error",
1807         0xffff, 0x0, NULL
1808 };
1809 
1810 
1811 /*
1812  * Struct for passing printing information for sense data messages
1813  */
1814 struct sd_sense_info {
1815         int     ssi_severity;
1816         int     ssi_pfa_flag;
1817 };
1818 
1819 /*
1820  * Table of function pointers for iostart-side routines. Separate "chains"
1821  * of layered function calls are formed by placing the function pointers
1822  * sequentially in the desired order. Functions are called according to an
1823  * incrementing table index ordering. The last function in each chain must
1824  * be sd_core_iostart(). The corresponding iodone-side routines are expected
1825  * in the sd_iodone_chain[] array.
1826  *
1827  * Note: It may seem more natural to organize both the iostart and iodone
1828  * functions together, into an array of structures (or some similar
1829  * organization) with a common index, rather than two separate arrays which
1830  * must be maintained in synchronization. The purpose of this division is
1831  * to achieve improved performance: individual arrays allows for more
1832  * effective cache line utilization on certain platforms.
1833  */
1834 
1835 typedef void (*sd_chain_t)(int index, struct sd_lun *un, struct buf *bp);
1836 
1837 
1838 static sd_chain_t sd_iostart_chain[] = {
1839 
1840         /* Chain for buf IO for disk drive targets (PM enabled) */
1841         sd_mapblockaddr_iostart,        /* Index: 0 */
1842         sd_pm_iostart,                  /* Index: 1 */
1843         sd_core_iostart,                /* Index: 2 */
1844 
1845         /* Chain for buf IO for disk drive targets (PM disabled) */
1846         sd_mapblockaddr_iostart,        /* Index: 3 */
1847         sd_core_iostart,                /* Index: 4 */
1848 
1849         /*
1850          * Chain for buf IO for removable-media or large sector size
1851          * disk drive targets with RMW needed (PM enabled)
1852          */
1853         sd_mapblockaddr_iostart,        /* Index: 5 */
1854         sd_mapblocksize_iostart,        /* Index: 6 */
1855         sd_pm_iostart,                  /* Index: 7 */
1856         sd_core_iostart,                /* Index: 8 */
1857 
1858         /*
1859          * Chain for buf IO for removable-media or large sector size
1860          * disk drive targets with RMW needed (PM disabled)
1861          */
1862         sd_mapblockaddr_iostart,        /* Index: 9 */
1863         sd_mapblocksize_iostart,        /* Index: 10 */
1864         sd_core_iostart,                /* Index: 11 */
1865 
1866         /* Chain for buf IO for disk drives with checksumming (PM enabled) */
1867         sd_mapblockaddr_iostart,        /* Index: 12 */
1868         sd_checksum_iostart,            /* Index: 13 */
1869         sd_pm_iostart,                  /* Index: 14 */
1870         sd_core_iostart,                /* Index: 15 */
1871 
1872         /* Chain for buf IO for disk drives with checksumming (PM disabled) */
1873         sd_mapblockaddr_iostart,        /* Index: 16 */
1874         sd_checksum_iostart,            /* Index: 17 */
1875         sd_core_iostart,                /* Index: 18 */
1876 
1877         /* Chain for USCSI commands (all targets) */
1878         sd_pm_iostart,                  /* Index: 19 */
1879         sd_core_iostart,                /* Index: 20 */
1880 
1881         /* Chain for checksumming USCSI commands (all targets) */
1882         sd_checksum_uscsi_iostart,      /* Index: 21 */
1883         sd_pm_iostart,                  /* Index: 22 */
1884         sd_core_iostart,                /* Index: 23 */
1885 
1886         /* Chain for "direct" USCSI commands (all targets) */
1887         sd_core_iostart,                /* Index: 24 */
1888 
1889         /* Chain for "direct priority" USCSI commands (all targets) */
1890         sd_core_iostart,                /* Index: 25 */
1891 
1892         /*
1893          * Chain for buf IO for large sector size disk drive targets
1894          * with RMW needed with checksumming (PM enabled)
1895          */
1896         sd_mapblockaddr_iostart,        /* Index: 26 */
1897         sd_mapblocksize_iostart,        /* Index: 27 */
1898         sd_checksum_iostart,            /* Index: 28 */
1899         sd_pm_iostart,                  /* Index: 29 */
1900         sd_core_iostart,                /* Index: 30 */
1901 
1902         /*
1903          * Chain for buf IO for large sector size disk drive targets
1904          * with RMW needed with checksumming (PM disabled)
1905          */
1906         sd_mapblockaddr_iostart,        /* Index: 31 */
1907         sd_mapblocksize_iostart,        /* Index: 32 */
1908         sd_checksum_iostart,            /* Index: 33 */
1909         sd_core_iostart,                /* Index: 34 */
1910 
1911 };
1912 
1913 /*
1914  * Macros to locate the first function of each iostart chain in the
1915  * sd_iostart_chain[] array. These are located by the index in the array.
1916  */
1917 #define SD_CHAIN_DISK_IOSTART                   0
1918 #define SD_CHAIN_DISK_IOSTART_NO_PM             3
1919 #define SD_CHAIN_MSS_DISK_IOSTART               5
1920 #define SD_CHAIN_RMMEDIA_IOSTART                5
1921 #define SD_CHAIN_MSS_DISK_IOSTART_NO_PM         9
1922 #define SD_CHAIN_RMMEDIA_IOSTART_NO_PM          9
1923 #define SD_CHAIN_CHKSUM_IOSTART                 12
1924 #define SD_CHAIN_CHKSUM_IOSTART_NO_PM           16
1925 #define SD_CHAIN_USCSI_CMD_IOSTART              19
1926 #define SD_CHAIN_USCSI_CHKSUM_IOSTART           21
1927 #define SD_CHAIN_DIRECT_CMD_IOSTART             24
1928 #define SD_CHAIN_PRIORITY_CMD_IOSTART           25
1929 #define SD_CHAIN_MSS_CHKSUM_IOSTART             26
1930 #define SD_CHAIN_MSS_CHKSUM_IOSTART_NO_PM       31
1931 
1932 
1933 /*
1934  * Table of function pointers for the iodone-side routines for the driver-
1935  * internal layering mechanism.  The calling sequence for iodone routines
1936  * uses a decrementing table index, so the last routine called in a chain
1937  * must be at the lowest array index location for that chain.  The last
1938  * routine for each chain must be either sd_buf_iodone() (for buf(9S) IOs)
1939  * or sd_uscsi_iodone() (for uscsi IOs).  Other than this, the ordering
1940  * of the functions in an iodone side chain must correspond to the ordering
1941  * of the iostart routines for that chain.  Note that there is no iodone
1942  * side routine that corresponds to sd_core_iostart(), so there is no
1943  * entry in the table for this.
1944  */
1945 
1946 static sd_chain_t sd_iodone_chain[] = {
1947 
1948         /* Chain for buf IO for disk drive targets (PM enabled) */
1949         sd_buf_iodone,                  /* Index: 0 */
1950         sd_mapblockaddr_iodone,         /* Index: 1 */
1951         sd_pm_iodone,                   /* Index: 2 */
1952 
1953         /* Chain for buf IO for disk drive targets (PM disabled) */
1954         sd_buf_iodone,                  /* Index: 3 */
1955         sd_mapblockaddr_iodone,         /* Index: 4 */
1956 
1957         /*
1958          * Chain for buf IO for removable-media or large sector size
1959          * disk drive targets with RMW needed (PM enabled)
1960          */
1961         sd_buf_iodone,                  /* Index: 5 */
1962         sd_mapblockaddr_iodone,         /* Index: 6 */
1963         sd_mapblocksize_iodone,         /* Index: 7 */
1964         sd_pm_iodone,                   /* Index: 8 */
1965 
1966         /*
1967          * Chain for buf IO for removable-media or large sector size
1968          * disk drive targets with RMW needed (PM disabled)
1969          */
1970         sd_buf_iodone,                  /* Index: 9 */
1971         sd_mapblockaddr_iodone,         /* Index: 10 */
1972         sd_mapblocksize_iodone,         /* Index: 11 */
1973 
1974         /* Chain for buf IO for disk drives with checksumming (PM enabled) */
1975         sd_buf_iodone,                  /* Index: 12 */
1976         sd_mapblockaddr_iodone,         /* Index: 13 */
1977         sd_checksum_iodone,             /* Index: 14 */
1978         sd_pm_iodone,                   /* Index: 15 */
1979 
1980         /* Chain for buf IO for disk drives with checksumming (PM disabled) */
1981         sd_buf_iodone,                  /* Index: 16 */
1982         sd_mapblockaddr_iodone,         /* Index: 17 */
1983         sd_checksum_iodone,             /* Index: 18 */
1984 
1985         /* Chain for USCSI commands (non-checksum targets) */
1986         sd_uscsi_iodone,                /* Index: 19 */
1987         sd_pm_iodone,                   /* Index: 20 */
1988 
1989         /* Chain for USCSI commands (checksum targets) */
1990         sd_uscsi_iodone,                /* Index: 21 */
1991         sd_checksum_uscsi_iodone,       /* Index: 22 */
1992         sd_pm_iodone,                   /* Index: 22 */
1993 
1994         /* Chain for "direct" USCSI commands (all targets) */
1995         sd_uscsi_iodone,                /* Index: 24 */
1996 
1997         /* Chain for "direct priority" USCSI commands (all targets) */
1998         sd_uscsi_iodone,                /* Index: 25 */
1999 
2000         /*
2001          * Chain for buf IO for large sector size disk drive targets
2002          * with checksumming (PM enabled)
2003          */
2004         sd_buf_iodone,                  /* Index: 26 */
2005         sd_mapblockaddr_iodone,         /* Index: 27 */
2006         sd_mapblocksize_iodone,         /* Index: 28 */
2007         sd_checksum_iodone,             /* Index: 29 */
2008         sd_pm_iodone,                   /* Index: 30 */
2009 
2010         /*
2011          * Chain for buf IO for large sector size disk drive targets
2012          * with checksumming (PM disabled)
2013          */
2014         sd_buf_iodone,                  /* Index: 31 */
2015         sd_mapblockaddr_iodone,         /* Index: 32 */
2016         sd_mapblocksize_iodone,         /* Index: 33 */
2017         sd_checksum_iodone,             /* Index: 34 */
2018 };
2019 
2020 
2021 /*
2022  * Macros to locate the "first" function in the sd_iodone_chain[] array for
2023  * each iodone-side chain. These are located by the array index, but as the
2024  * iodone side functions are called in a decrementing-index order, the
2025  * highest index number in each chain must be specified (as these correspond
2026  * to the first function in the iodone chain that will be called by the core
2027  * at IO completion time).
2028  */
2029 
2030 #define SD_CHAIN_DISK_IODONE                    2
2031 #define SD_CHAIN_DISK_IODONE_NO_PM              4
2032 #define SD_CHAIN_RMMEDIA_IODONE                 8
2033 #define SD_CHAIN_MSS_DISK_IODONE                8
2034 #define SD_CHAIN_RMMEDIA_IODONE_NO_PM           11
2035 #define SD_CHAIN_MSS_DISK_IODONE_NO_PM          11
2036 #define SD_CHAIN_CHKSUM_IODONE                  15
2037 #define SD_CHAIN_CHKSUM_IODONE_NO_PM            18
2038 #define SD_CHAIN_USCSI_CMD_IODONE               20
2039 #define SD_CHAIN_USCSI_CHKSUM_IODONE            22
2040 #define SD_CHAIN_DIRECT_CMD_IODONE              24
2041 #define SD_CHAIN_PRIORITY_CMD_IODONE            25
2042 #define SD_CHAIN_MSS_CHKSUM_IODONE              30
2043 #define SD_CHAIN_MSS_CHKSUM_IODONE_NO_PM        34
2044 
2045 
2046 
2047 /*
2048  * Array to map a layering chain index to the appropriate initpkt routine.
2049  * The redundant entries are present so that the index used for accessing
2050  * the above sd_iostart_chain and sd_iodone_chain tables can be used directly
2051  * with this table as well.
2052  */
2053 typedef int (*sd_initpkt_t)(struct buf *, struct scsi_pkt **);
2054 
2055 static sd_initpkt_t     sd_initpkt_map[] = {
2056 
2057         /* Chain for buf IO for disk drive targets (PM enabled) */
2058         sd_initpkt_for_buf,             /* Index: 0 */
2059         sd_initpkt_for_buf,             /* Index: 1 */
2060         sd_initpkt_for_buf,             /* Index: 2 */
2061 
2062         /* Chain for buf IO for disk drive targets (PM disabled) */
2063         sd_initpkt_for_buf,             /* Index: 3 */
2064         sd_initpkt_for_buf,             /* Index: 4 */
2065 
2066         /*
2067          * Chain for buf IO for removable-media or large sector size
2068          * disk drive targets (PM enabled)
2069          */
2070         sd_initpkt_for_buf,             /* Index: 5 */
2071         sd_initpkt_for_buf,             /* Index: 6 */
2072         sd_initpkt_for_buf,             /* Index: 7 */
2073         sd_initpkt_for_buf,             /* Index: 8 */
2074 
2075         /*
2076          * Chain for buf IO for removable-media or large sector size
2077          * disk drive targets (PM disabled)
2078          */
2079         sd_initpkt_for_buf,             /* Index: 9 */
2080         sd_initpkt_for_buf,             /* Index: 10 */
2081         sd_initpkt_for_buf,             /* Index: 11 */
2082 
2083         /* Chain for buf IO for disk drives with checksumming (PM enabled) */
2084         sd_initpkt_for_buf,             /* Index: 12 */
2085         sd_initpkt_for_buf,             /* Index: 13 */
2086         sd_initpkt_for_buf,             /* Index: 14 */
2087         sd_initpkt_for_buf,             /* Index: 15 */
2088 
2089         /* Chain for buf IO for disk drives with checksumming (PM disabled) */
2090         sd_initpkt_for_buf,             /* Index: 16 */
2091         sd_initpkt_for_buf,             /* Index: 17 */
2092         sd_initpkt_for_buf,             /* Index: 18 */
2093 
2094         /* Chain for USCSI commands (non-checksum targets) */
2095         sd_initpkt_for_uscsi,           /* Index: 19 */
2096         sd_initpkt_for_uscsi,           /* Index: 20 */
2097 
2098         /* Chain for USCSI commands (checksum targets) */
2099         sd_initpkt_for_uscsi,           /* Index: 21 */
2100         sd_initpkt_for_uscsi,           /* Index: 22 */
2101         sd_initpkt_for_uscsi,           /* Index: 22 */
2102 
2103         /* Chain for "direct" USCSI commands (all targets) */
2104         sd_initpkt_for_uscsi,           /* Index: 24 */
2105 
2106         /* Chain for "direct priority" USCSI commands (all targets) */
2107         sd_initpkt_for_uscsi,           /* Index: 25 */
2108 
2109         /*
2110          * Chain for buf IO for large sector size disk drive targets
2111          * with checksumming (PM enabled)
2112          */
2113         sd_initpkt_for_buf,             /* Index: 26 */
2114         sd_initpkt_for_buf,             /* Index: 27 */
2115         sd_initpkt_for_buf,             /* Index: 28 */
2116         sd_initpkt_for_buf,             /* Index: 29 */
2117         sd_initpkt_for_buf,             /* Index: 30 */
2118 
2119         /*
2120          * Chain for buf IO for large sector size disk drive targets
2121          * with checksumming (PM disabled)
2122          */
2123         sd_initpkt_for_buf,             /* Index: 31 */
2124         sd_initpkt_for_buf,             /* Index: 32 */
2125         sd_initpkt_for_buf,             /* Index: 33 */
2126         sd_initpkt_for_buf,             /* Index: 34 */
2127 };
2128 
2129 
2130 /*
2131  * Array to map a layering chain index to the appropriate destroypktpkt routine.
2132  * The redundant entries are present so that the index used for accessing
2133  * the above sd_iostart_chain and sd_iodone_chain tables can be used directly
2134  * with this table as well.
2135  */
2136 typedef void (*sd_destroypkt_t)(struct buf *);
2137 
2138 static sd_destroypkt_t  sd_destroypkt_map[] = {
2139 
2140         /* Chain for buf IO for disk drive targets (PM enabled) */
2141         sd_destroypkt_for_buf,          /* Index: 0 */
2142         sd_destroypkt_for_buf,          /* Index: 1 */
2143         sd_destroypkt_for_buf,          /* Index: 2 */
2144 
2145         /* Chain for buf IO for disk drive targets (PM disabled) */
2146         sd_destroypkt_for_buf,          /* Index: 3 */
2147         sd_destroypkt_for_buf,          /* Index: 4 */
2148 
2149         /*
2150          * Chain for buf IO for removable-media or large sector size
2151          * disk drive targets (PM enabled)
2152          */
2153         sd_destroypkt_for_buf,          /* Index: 5 */
2154         sd_destroypkt_for_buf,          /* Index: 6 */
2155         sd_destroypkt_for_buf,          /* Index: 7 */
2156         sd_destroypkt_for_buf,          /* Index: 8 */
2157 
2158         /*
2159          * Chain for buf IO for removable-media or large sector size
2160          * disk drive targets (PM disabled)
2161          */
2162         sd_destroypkt_for_buf,          /* Index: 9 */
2163         sd_destroypkt_for_buf,          /* Index: 10 */
2164         sd_destroypkt_for_buf,          /* Index: 11 */
2165 
2166         /* Chain for buf IO for disk drives with checksumming (PM enabled) */
2167         sd_destroypkt_for_buf,          /* Index: 12 */
2168         sd_destroypkt_for_buf,          /* Index: 13 */
2169         sd_destroypkt_for_buf,          /* Index: 14 */
2170         sd_destroypkt_for_buf,          /* Index: 15 */
2171 
2172         /* Chain for buf IO for disk drives with checksumming (PM disabled) */
2173         sd_destroypkt_for_buf,          /* Index: 16 */
2174         sd_destroypkt_for_buf,          /* Index: 17 */
2175         sd_destroypkt_for_buf,          /* Index: 18 */
2176 
2177         /* Chain for USCSI commands (non-checksum targets) */
2178         sd_destroypkt_for_uscsi,        /* Index: 19 */
2179         sd_destroypkt_for_uscsi,        /* Index: 20 */
2180 
2181         /* Chain for USCSI commands (checksum targets) */
2182         sd_destroypkt_for_uscsi,        /* Index: 21 */
2183         sd_destroypkt_for_uscsi,        /* Index: 22 */
2184         sd_destroypkt_for_uscsi,        /* Index: 22 */
2185 
2186         /* Chain for "direct" USCSI commands (all targets) */
2187         sd_destroypkt_for_uscsi,        /* Index: 24 */
2188 
2189         /* Chain for "direct priority" USCSI commands (all targets) */
2190         sd_destroypkt_for_uscsi,        /* Index: 25 */
2191 
2192         /*
2193          * Chain for buf IO for large sector size disk drive targets
2194          * with checksumming (PM disabled)
2195          */
2196         sd_destroypkt_for_buf,          /* Index: 26 */
2197         sd_destroypkt_for_buf,          /* Index: 27 */
2198         sd_destroypkt_for_buf,          /* Index: 28 */
2199         sd_destroypkt_for_buf,          /* Index: 29 */
2200         sd_destroypkt_for_buf,          /* Index: 30 */
2201 
2202         /*
2203          * Chain for buf IO for large sector size disk drive targets
2204          * with checksumming (PM enabled)
2205          */
2206         sd_destroypkt_for_buf,          /* Index: 31 */
2207         sd_destroypkt_for_buf,          /* Index: 32 */
2208         sd_destroypkt_for_buf,          /* Index: 33 */
2209         sd_destroypkt_for_buf,          /* Index: 34 */
2210 };
2211 
2212 
2213 
2214 /*
2215  * Array to map a layering chain index to the appropriate chain "type".
2216  * The chain type indicates a specific property/usage of the chain.
2217  * The redundant entries are present so that the index used for accessing
2218  * the above sd_iostart_chain and sd_iodone_chain tables can be used directly
2219  * with this table as well.
2220  */
2221 
2222 #define SD_CHAIN_NULL                   0       /* for the special RQS cmd */
2223 #define SD_CHAIN_BUFIO                  1       /* regular buf IO */
2224 #define SD_CHAIN_USCSI                  2       /* regular USCSI commands */
2225 #define SD_CHAIN_DIRECT                 3       /* uscsi, w/ bypass power mgt */
2226 #define SD_CHAIN_DIRECT_PRIORITY        4       /* uscsi, w/ bypass power mgt */
2227                                                 /* (for error recovery) */
2228 
2229 static int sd_chain_type_map[] = {
2230 
2231         /* Chain for buf IO for disk drive targets (PM enabled) */
2232         SD_CHAIN_BUFIO,                 /* Index: 0 */
2233         SD_CHAIN_BUFIO,                 /* Index: 1 */
2234         SD_CHAIN_BUFIO,                 /* Index: 2 */
2235 
2236         /* Chain for buf IO for disk drive targets (PM disabled) */
2237         SD_CHAIN_BUFIO,                 /* Index: 3 */
2238         SD_CHAIN_BUFIO,                 /* Index: 4 */
2239 
2240         /*
2241          * Chain for buf IO for removable-media or large sector size
2242          * disk drive targets (PM enabled)
2243          */
2244         SD_CHAIN_BUFIO,                 /* Index: 5 */
2245         SD_CHAIN_BUFIO,                 /* Index: 6 */
2246         SD_CHAIN_BUFIO,                 /* Index: 7 */
2247         SD_CHAIN_BUFIO,                 /* Index: 8 */
2248 
2249         /*
2250          * Chain for buf IO for removable-media or large sector size
2251          * disk drive targets (PM disabled)
2252          */
2253         SD_CHAIN_BUFIO,                 /* Index: 9 */
2254         SD_CHAIN_BUFIO,                 /* Index: 10 */
2255         SD_CHAIN_BUFIO,                 /* Index: 11 */
2256 
2257         /* Chain for buf IO for disk drives with checksumming (PM enabled) */
2258         SD_CHAIN_BUFIO,                 /* Index: 12 */
2259         SD_CHAIN_BUFIO,                 /* Index: 13 */
2260         SD_CHAIN_BUFIO,                 /* Index: 14 */
2261         SD_CHAIN_BUFIO,                 /* Index: 15 */
2262 
2263         /* Chain for buf IO for disk drives with checksumming (PM disabled) */
2264         SD_CHAIN_BUFIO,                 /* Index: 16 */
2265         SD_CHAIN_BUFIO,                 /* Index: 17 */
2266         SD_CHAIN_BUFIO,                 /* Index: 18 */
2267 
2268         /* Chain for USCSI commands (non-checksum targets) */
2269         SD_CHAIN_USCSI,                 /* Index: 19 */
2270         SD_CHAIN_USCSI,                 /* Index: 20 */
2271 
2272         /* Chain for USCSI commands (checksum targets) */
2273         SD_CHAIN_USCSI,                 /* Index: 21 */
2274         SD_CHAIN_USCSI,                 /* Index: 22 */
2275         SD_CHAIN_USCSI,                 /* Index: 23 */
2276 
2277         /* Chain for "direct" USCSI commands (all targets) */
2278         SD_CHAIN_DIRECT,                /* Index: 24 */
2279 
2280         /* Chain for "direct priority" USCSI commands (all targets) */
2281         SD_CHAIN_DIRECT_PRIORITY,       /* Index: 25 */
2282 
2283         /*
2284          * Chain for buf IO for large sector size disk drive targets
2285          * with checksumming (PM enabled)
2286          */
2287         SD_CHAIN_BUFIO,                 /* Index: 26 */
2288         SD_CHAIN_BUFIO,                 /* Index: 27 */
2289         SD_CHAIN_BUFIO,                 /* Index: 28 */
2290         SD_CHAIN_BUFIO,                 /* Index: 29 */
2291         SD_CHAIN_BUFIO,                 /* Index: 30 */
2292 
2293         /*
2294          * Chain for buf IO for large sector size disk drive targets
2295          * with checksumming (PM disabled)
2296          */
2297         SD_CHAIN_BUFIO,                 /* Index: 31 */
2298         SD_CHAIN_BUFIO,                 /* Index: 32 */
2299         SD_CHAIN_BUFIO,                 /* Index: 33 */
2300         SD_CHAIN_BUFIO,                 /* Index: 34 */
2301 };
2302 
2303 
2304 /* Macro to return TRUE if the IO has come from the sd_buf_iostart() chain. */
2305 #define SD_IS_BUFIO(xp)                 \
2306         (sd_chain_type_map[(xp)->xb_chain_iostart] == SD_CHAIN_BUFIO)
2307 
2308 /* Macro to return TRUE if the IO has come from the "direct priority" chain. */
2309 #define SD_IS_DIRECT_PRIORITY(xp)       \
2310         (sd_chain_type_map[(xp)->xb_chain_iostart] == SD_CHAIN_DIRECT_PRIORITY)
2311 
2312 
2313 
2314 /*
2315  * Struct, array, and macros to map a specific chain to the appropriate
2316  * layering indexes in the sd_iostart_chain[] and sd_iodone_chain[] arrays.
2317  *
2318  * The sd_chain_index_map[] array is used at attach time to set the various
2319  * un_xxx_chain type members of the sd_lun softstate to the specific layering
2320  * chain to be used with the instance. This allows different instances to use
2321  * different chain for buf IO, uscsi IO, etc.. Also, since the xb_chain_iostart
2322  * and xb_chain_iodone index values in the sd_xbuf are initialized to these
2323  * values at sd_xbuf init time, this allows (1) layering chains may be changed
2324  * dynamically & without the use of locking; and (2) a layer may update the
2325  * xb_chain_io[start|done] member in a given xbuf with its current index value,
2326  * to allow for deferred processing of an IO within the same chain from a
2327  * different execution context.
2328  */
2329 
2330 struct sd_chain_index {
2331         int     sci_iostart_index;
2332         int     sci_iodone_index;
2333 };
2334 
2335 static struct sd_chain_index    sd_chain_index_map[] = {
2336         { SD_CHAIN_DISK_IOSTART,                SD_CHAIN_DISK_IODONE },
2337         { SD_CHAIN_DISK_IOSTART_NO_PM,          SD_CHAIN_DISK_IODONE_NO_PM },
2338         { SD_CHAIN_RMMEDIA_IOSTART,             SD_CHAIN_RMMEDIA_IODONE },
2339         { SD_CHAIN_RMMEDIA_IOSTART_NO_PM,       SD_CHAIN_RMMEDIA_IODONE_NO_PM },
2340         { SD_CHAIN_CHKSUM_IOSTART,              SD_CHAIN_CHKSUM_IODONE },
2341         { SD_CHAIN_CHKSUM_IOSTART_NO_PM,        SD_CHAIN_CHKSUM_IODONE_NO_PM },
2342         { SD_CHAIN_USCSI_CMD_IOSTART,           SD_CHAIN_USCSI_CMD_IODONE },
2343         { SD_CHAIN_USCSI_CHKSUM_IOSTART,        SD_CHAIN_USCSI_CHKSUM_IODONE },
2344         { SD_CHAIN_DIRECT_CMD_IOSTART,          SD_CHAIN_DIRECT_CMD_IODONE },
2345         { SD_CHAIN_PRIORITY_CMD_IOSTART,        SD_CHAIN_PRIORITY_CMD_IODONE },
2346         { SD_CHAIN_MSS_CHKSUM_IOSTART,          SD_CHAIN_MSS_CHKSUM_IODONE },
2347         { SD_CHAIN_MSS_CHKSUM_IOSTART_NO_PM, SD_CHAIN_MSS_CHKSUM_IODONE_NO_PM },
2348 
2349 };
2350 
2351 
2352 /*
2353  * The following are indexes into the sd_chain_index_map[] array.
2354  */
2355 
2356 /* un->un_buf_chain_type must be set to one of these */
2357 #define SD_CHAIN_INFO_DISK              0
2358 #define SD_CHAIN_INFO_DISK_NO_PM        1
2359 #define SD_CHAIN_INFO_RMMEDIA           2
2360 #define SD_CHAIN_INFO_MSS_DISK          2
2361 #define SD_CHAIN_INFO_RMMEDIA_NO_PM     3
2362 #define SD_CHAIN_INFO_MSS_DSK_NO_PM     3
2363 #define SD_CHAIN_INFO_CHKSUM            4
2364 #define SD_CHAIN_INFO_CHKSUM_NO_PM      5
2365 #define SD_CHAIN_INFO_MSS_DISK_CHKSUM   10
2366 #define SD_CHAIN_INFO_MSS_DISK_CHKSUM_NO_PM     11
2367 
2368 /* un->un_uscsi_chain_type must be set to one of these */
2369 #define SD_CHAIN_INFO_USCSI_CMD         6
2370 /* USCSI with PM disabled is the same as DIRECT */
2371 #define SD_CHAIN_INFO_USCSI_CMD_NO_PM   8
2372 #define SD_CHAIN_INFO_USCSI_CHKSUM      7
2373 
2374 /* un->un_direct_chain_type must be set to one of these */
2375 #define SD_CHAIN_INFO_DIRECT_CMD        8
2376 
2377 /* un->un_priority_chain_type must be set to one of these */
2378 #define SD_CHAIN_INFO_PRIORITY_CMD      9
2379 
2380 /* size for devid inquiries */
2381 #define MAX_INQUIRY_SIZE                0xF0
2382 
2383 /*
2384  * Macros used by functions to pass a given buf(9S) struct along to the
2385  * next function in the layering chain for further processing.
2386  *
2387  * In the following macros, passing more than three arguments to the called
2388  * routines causes the optimizer for the SPARC compiler to stop doing tail
2389  * call elimination which results in significant performance degradation.
2390  */
2391 #define SD_BEGIN_IOSTART(index, un, bp) \
2392         ((*(sd_iostart_chain[index]))(index, un, bp))
2393 
2394 #define SD_BEGIN_IODONE(index, un, bp)  \
2395         ((*(sd_iodone_chain[index]))(index, un, bp))
2396 
2397 #define SD_NEXT_IOSTART(index, un, bp)                          \
2398         ((*(sd_iostart_chain[(index) + 1]))((index) + 1, un, bp))
2399 
2400 #define SD_NEXT_IODONE(index, un, bp)                           \
2401         ((*(sd_iodone_chain[(index) - 1]))((index) - 1, un, bp))
2402 
2403 /*
2404  *    Function: _init
2405  *
2406  * Description: This is the driver _init(9E) entry point.
2407  *
2408  * Return Code: Returns the value from mod_install(9F) or
2409  *              ddi_soft_state_init(9F) as appropriate.
2410  *
2411  *     Context: Called when driver module loaded.
2412  */
2413 
2414 int
2415 _init(void)
2416 {
2417         int     err;
2418 
2419         /* establish driver name from module name */
2420         sd_label = (char *)mod_modname(&modlinkage);
2421 
2422 #ifndef XPV_HVM_DRIVER
2423         err = ddi_soft_state_init(&sd_state, sizeof (struct sd_lun),
2424             SD_MAXUNIT);
2425         if (err != 0) {
2426                 return (err);
2427         }
2428 
2429 #else /* XPV_HVM_DRIVER */
2430         /* Remove the leading "hvm_" from the module name */
2431         ASSERT(strncmp(sd_label, "hvm_", strlen("hvm_")) == 0);
2432         sd_label += strlen("hvm_");
2433 
2434 #endif /* XPV_HVM_DRIVER */
2435 
2436         mutex_init(&sd_detach_mutex, NULL, MUTEX_DRIVER, NULL);
2437         mutex_init(&sd_log_mutex,    NULL, MUTEX_DRIVER, NULL);
2438         mutex_init(&sd_label_mutex,  NULL, MUTEX_DRIVER, NULL);
2439 
2440         mutex_init(&sd_tr.srq_resv_reclaim_mutex, NULL, MUTEX_DRIVER, NULL);
2441         cv_init(&sd_tr.srq_resv_reclaim_cv, NULL, CV_DRIVER, NULL);
2442         cv_init(&sd_tr.srq_inprocess_cv, NULL, CV_DRIVER, NULL);
2443 
2444         /*
2445          * it's ok to init here even for fibre device
2446          */
2447         sd_scsi_probe_cache_init();
2448 
2449         sd_scsi_target_lun_init();
2450 
2451         /*
2452          * Creating taskq before mod_install ensures that all callers (threads)
2453          * that enter the module after a successful mod_install encounter
2454          * a valid taskq.
2455          */
2456         sd_taskq_create();
2457 
2458         err = mod_install(&modlinkage);
2459         if (err != 0) {
2460                 /* delete taskq if install fails */
2461                 sd_taskq_delete();
2462 
2463                 mutex_destroy(&sd_detach_mutex);
2464                 mutex_destroy(&sd_log_mutex);
2465                 mutex_destroy(&sd_label_mutex);
2466 
2467                 mutex_destroy(&sd_tr.srq_resv_reclaim_mutex);
2468                 cv_destroy(&sd_tr.srq_resv_reclaim_cv);
2469                 cv_destroy(&sd_tr.srq_inprocess_cv);
2470 
2471                 sd_scsi_probe_cache_fini();
2472 
2473                 sd_scsi_target_lun_fini();
2474 
2475 #ifndef XPV_HVM_DRIVER
2476                 ddi_soft_state_fini(&sd_state);
2477 #endif /* !XPV_HVM_DRIVER */
2478                 return (err);
2479         }
2480 
2481         return (err);
2482 }
2483 
2484 
2485 /*
2486  *    Function: _fini
2487  *
2488  * Description: This is the driver _fini(9E) entry point.
2489  *
2490  * Return Code: Returns the value from mod_remove(9F)
2491  *
2492  *     Context: Called when driver module is unloaded.
2493  */
2494 
2495 int
2496 _fini(void)
2497 {
2498         int err;
2499 
2500         if ((err = mod_remove(&modlinkage)) != 0) {
2501                 return (err);
2502         }
2503 
2504         sd_taskq_delete();
2505 
2506         mutex_destroy(&sd_detach_mutex);
2507         mutex_destroy(&sd_log_mutex);
2508         mutex_destroy(&sd_label_mutex);
2509         mutex_destroy(&sd_tr.srq_resv_reclaim_mutex);
2510 
2511         sd_scsi_probe_cache_fini();
2512 
2513         sd_scsi_target_lun_fini();
2514 
2515         cv_destroy(&sd_tr.srq_resv_reclaim_cv);
2516         cv_destroy(&sd_tr.srq_inprocess_cv);
2517 
2518 #ifndef XPV_HVM_DRIVER
2519         ddi_soft_state_fini(&sd_state);
2520 #endif /* !XPV_HVM_DRIVER */
2521 
2522         return (err);
2523 }
2524 
2525 
2526 /*
2527  *    Function: _info
2528  *
2529  * Description: This is the driver _info(9E) entry point.
2530  *
2531  *   Arguments: modinfop - pointer to the driver modinfo structure
2532  *
2533  * Return Code: Returns the value from mod_info(9F).
2534  *
2535  *     Context: Kernel thread context
2536  */
2537 
2538 int
2539 _info(struct modinfo *modinfop)
2540 {
2541         return (mod_info(&modlinkage, modinfop));
2542 }
2543 
2544 
2545 /*
2546  * The following routines implement the driver message logging facility.
2547  * They provide component- and level- based debug output filtering.
2548  * Output may also be restricted to messages for a single instance by
2549  * specifying a soft state pointer in sd_debug_un. If sd_debug_un is set
2550  * to NULL, then messages for all instances are printed.
2551  *
2552  * These routines have been cloned from each other due to the language
2553  * constraints of macros and variable argument list processing.
2554  */
2555 
2556 
2557 /*
2558  *    Function: sd_log_err
2559  *
2560  * Description: This routine is called by the SD_ERROR macro for debug
2561  *              logging of error conditions.
2562  *
2563  *   Arguments: comp - driver component being logged
2564  *              dev  - pointer to driver info structure
2565  *              fmt  - error string and format to be logged
2566  */
2567 
2568 static void
2569 sd_log_err(uint_t comp, struct sd_lun *un, const char *fmt, ...)
2570 {
2571         va_list         ap;
2572         dev_info_t      *dev;
2573 
2574         ASSERT(un != NULL);
2575         dev = SD_DEVINFO(un);
2576         ASSERT(dev != NULL);
2577 
2578         /*
2579          * Filter messages based on the global component and level masks.
2580          * Also print if un matches the value of sd_debug_un, or if
2581          * sd_debug_un is set to NULL.
2582          */
2583         if ((sd_component_mask & comp) && (sd_level_mask & SD_LOGMASK_ERROR) &&
2584             ((sd_debug_un == NULL) || (sd_debug_un == un))) {
2585                 mutex_enter(&sd_log_mutex);
2586                 va_start(ap, fmt);
2587                 (void) vsprintf(sd_log_buf, fmt, ap);
2588                 va_end(ap);
2589                 scsi_log(dev, sd_label, CE_CONT, "%s", sd_log_buf);
2590                 mutex_exit(&sd_log_mutex);
2591         }
2592 #ifdef SD_FAULT_INJECTION
2593         _NOTE(DATA_READABLE_WITHOUT_LOCK(sd_lun::sd_injection_mask));
2594         if (un->sd_injection_mask & comp) {
2595                 mutex_enter(&sd_log_mutex);
2596                 va_start(ap, fmt);
2597                 (void) vsprintf(sd_log_buf, fmt, ap);
2598                 va_end(ap);
2599                 sd_injection_log(sd_log_buf, un);
2600                 mutex_exit(&sd_log_mutex);
2601         }
2602 #endif
2603 }
2604 
2605 
2606 /*
2607  *    Function: sd_log_info
2608  *
2609  * Description: This routine is called by the SD_INFO macro for debug
2610  *              logging of general purpose informational conditions.
2611  *
2612  *   Arguments: comp - driver component being logged
2613  *              dev  - pointer to driver info structure
2614  *              fmt  - info string and format to be logged
2615  */
2616 
2617 static void
2618 sd_log_info(uint_t component, struct sd_lun *un, const char *fmt, ...)
2619 {
2620         va_list         ap;
2621         dev_info_t      *dev;
2622 
2623         ASSERT(un != NULL);
2624         dev = SD_DEVINFO(un);
2625         ASSERT(dev != NULL);
2626 
2627         /*
2628          * Filter messages based on the global component and level masks.
2629          * Also print if un matches the value of sd_debug_un, or if
2630          * sd_debug_un is set to NULL.
2631          */
2632         if ((sd_component_mask & component) &&
2633             (sd_level_mask & SD_LOGMASK_INFO) &&
2634             ((sd_debug_un == NULL) || (sd_debug_un == un))) {
2635                 mutex_enter(&sd_log_mutex);
2636                 va_start(ap, fmt);
2637                 (void) vsprintf(sd_log_buf, fmt, ap);
2638                 va_end(ap);
2639                 scsi_log(dev, sd_label, CE_CONT, "%s", sd_log_buf);
2640                 mutex_exit(&sd_log_mutex);
2641         }
2642 #ifdef SD_FAULT_INJECTION
2643         _NOTE(DATA_READABLE_WITHOUT_LOCK(sd_lun::sd_injection_mask));
2644         if (un->sd_injection_mask & component) {
2645                 mutex_enter(&sd_log_mutex);
2646                 va_start(ap, fmt);
2647                 (void) vsprintf(sd_log_buf, fmt, ap);
2648                 va_end(ap);
2649                 sd_injection_log(sd_log_buf, un);
2650                 mutex_exit(&sd_log_mutex);
2651         }
2652 #endif
2653 }
2654 
2655 
2656 /*
2657  *    Function: sd_log_trace
2658  *
2659  * Description: This routine is called by the SD_TRACE macro for debug
2660  *              logging of trace conditions (i.e. function entry/exit).
2661  *
2662  *   Arguments: comp - driver component being logged
2663  *              dev  - pointer to driver info structure
2664  *              fmt  - trace string and format to be logged
2665  */
2666 
2667 static void
2668 sd_log_trace(uint_t component, struct sd_lun *un, const char *fmt, ...)
2669 {
2670         va_list         ap;
2671         dev_info_t      *dev;
2672 
2673         ASSERT(un != NULL);
2674         dev = SD_DEVINFO(un);
2675         ASSERT(dev != NULL);
2676 
2677         /*
2678          * Filter messages based on the global component and level masks.
2679          * Also print if un matches the value of sd_debug_un, or if
2680          * sd_debug_un is set to NULL.
2681          */
2682         if ((sd_component_mask & component) &&
2683             (sd_level_mask & SD_LOGMASK_TRACE) &&
2684             ((sd_debug_un == NULL) || (sd_debug_un == un))) {
2685                 mutex_enter(&sd_log_mutex);
2686                 va_start(ap, fmt);
2687                 (void) vsprintf(sd_log_buf, fmt, ap);
2688                 va_end(ap);
2689                 scsi_log(dev, sd_label, CE_CONT, "%s", sd_log_buf);
2690                 mutex_exit(&sd_log_mutex);
2691         }
2692 #ifdef SD_FAULT_INJECTION
2693         _NOTE(DATA_READABLE_WITHOUT_LOCK(sd_lun::sd_injection_mask));
2694         if (un->sd_injection_mask & component) {
2695                 mutex_enter(&sd_log_mutex);
2696                 va_start(ap, fmt);
2697                 (void) vsprintf(sd_log_buf, fmt, ap);
2698                 va_end(ap);
2699                 sd_injection_log(sd_log_buf, un);
2700                 mutex_exit(&sd_log_mutex);
2701         }
2702 #endif
2703 }
2704 
2705 
2706 /*
2707  *    Function: sdprobe
2708  *
2709  * Description: This is the driver probe(9e) entry point function.
2710  *
2711  *   Arguments: devi - opaque device info handle
2712  *
2713  * Return Code: DDI_PROBE_SUCCESS: If the probe was successful.
2714  *              DDI_PROBE_FAILURE: If the probe failed.
2715  *              DDI_PROBE_PARTIAL: If the instance is not present now,
2716  *                                 but may be present in the future.
2717  */
2718 
2719 static int
2720 sdprobe(dev_info_t *devi)
2721 {
2722         struct scsi_device      *devp;
2723         int                     rval;
2724 #ifndef XPV_HVM_DRIVER
2725         int                     instance = ddi_get_instance(devi);
2726 #endif /* !XPV_HVM_DRIVER */
2727 
2728         /*
2729          * if it wasn't for pln, sdprobe could actually be nulldev
2730          * in the "__fibre" case.
2731          */
2732         if (ddi_dev_is_sid(devi) == DDI_SUCCESS) {
2733                 return (DDI_PROBE_DONTCARE);
2734         }
2735 
2736         devp = ddi_get_driver_private(devi);
2737 
2738         if (devp == NULL) {
2739                 /* Ooops... nexus driver is mis-configured... */
2740                 return (DDI_PROBE_FAILURE);
2741         }
2742 
2743 #ifndef XPV_HVM_DRIVER
2744         if (ddi_get_soft_state(sd_state, instance) != NULL) {
2745                 return (DDI_PROBE_PARTIAL);
2746         }
2747 #endif /* !XPV_HVM_DRIVER */
2748 
2749         /*
2750          * Call the SCSA utility probe routine to see if we actually
2751          * have a target at this SCSI nexus.
2752          */
2753         switch (sd_scsi_probe_with_cache(devp, NULL_FUNC)) {
2754         case SCSIPROBE_EXISTS:
2755                 switch (devp->sd_inq->inq_dtype) {
2756                 case DTYPE_DIRECT:
2757                         rval = DDI_PROBE_SUCCESS;
2758                         break;
2759                 case DTYPE_RODIRECT:
2760                         /* CDs etc. Can be removable media */
2761                         rval = DDI_PROBE_SUCCESS;
2762                         break;
2763                 case DTYPE_OPTICAL:
2764                         /*
2765                          * Rewritable optical driver HP115AA
2766                          * Can also be removable media
2767                          */
2768 
2769                         /*
2770                          * Do not attempt to bind to  DTYPE_OPTICAL if
2771                          * pre solaris 9 sparc sd behavior is required
2772                          *
2773                          * If first time through and sd_dtype_optical_bind
2774                          * has not been set in /etc/system check properties
2775                          */
2776 
2777                         if (sd_dtype_optical_bind  < 0) {
2778                                 sd_dtype_optical_bind = ddi_prop_get_int
2779                                     (DDI_DEV_T_ANY, devi, 0,
2780                                     "optical-device-bind", 1);
2781                         }
2782 
2783                         if (sd_dtype_optical_bind == 0) {
2784                                 rval = DDI_PROBE_FAILURE;
2785                         } else {
2786                                 rval = DDI_PROBE_SUCCESS;
2787                         }
2788                         break;
2789 
2790                 case DTYPE_NOTPRESENT:
2791                 default:
2792                         rval = DDI_PROBE_FAILURE;
2793                         break;
2794                 }
2795                 break;
2796         default:
2797                 rval = DDI_PROBE_PARTIAL;
2798                 break;
2799         }
2800 
2801         /*
2802          * This routine checks for resource allocation prior to freeing,
2803          * so it will take care of the "smart probing" case where a
2804          * scsi_probe() may or may not have been issued and will *not*
2805          * free previously-freed resources.
2806          */
2807         scsi_unprobe(devp);
2808         return (rval);
2809 }
2810 
2811 
2812 /*
2813  *    Function: sdinfo
2814  *
2815  * Description: This is the driver getinfo(9e) entry point function.
2816  *              Given the device number, return the devinfo pointer from
2817  *              the scsi_device structure or the instance number
2818  *              associated with the dev_t.
2819  *
2820  *   Arguments: dip     - pointer to device info structure
2821  *              infocmd - command argument (DDI_INFO_DEVT2DEVINFO,
2822  *                        DDI_INFO_DEVT2INSTANCE)
2823  *              arg     - driver dev_t
2824  *              resultp - user buffer for request response
2825  *
2826  * Return Code: DDI_SUCCESS
2827  *              DDI_FAILURE
2828  */
2829 /* ARGSUSED */
2830 static int
2831 sdinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
2832 {
2833         struct sd_lun   *un;
2834         dev_t           dev;
2835         int             instance;
2836         int             error;
2837 
2838         switch (infocmd) {
2839         case DDI_INFO_DEVT2DEVINFO:
2840                 dev = (dev_t)arg;
2841                 instance = SDUNIT(dev);
2842                 if ((un = ddi_get_soft_state(sd_state, instance)) == NULL) {
2843                         return (DDI_FAILURE);
2844                 }
2845                 *result = (void *) SD_DEVINFO(un);
2846                 error = DDI_SUCCESS;
2847                 break;
2848         case DDI_INFO_DEVT2INSTANCE:
2849                 dev = (dev_t)arg;
2850                 instance = SDUNIT(dev);
2851                 *result = (void *)(uintptr_t)instance;
2852                 error = DDI_SUCCESS;
2853                 break;
2854         default:
2855                 error = DDI_FAILURE;
2856         }
2857         return (error);
2858 }
2859 
2860 /*
2861  *    Function: sd_prop_op
2862  *
2863  * Description: This is the driver prop_op(9e) entry point function.
2864  *              Return the number of blocks for the partition in question
2865  *              or forward the request to the property facilities.
2866  *
2867  *   Arguments: dev       - device number
2868  *              dip       - pointer to device info structure
2869  *              prop_op   - property operator
2870  *              mod_flags - DDI_PROP_DONTPASS, don't pass to parent
2871  *              name      - pointer to property name
2872  *              valuep    - pointer or address of the user buffer
2873  *              lengthp   - property length
2874  *
2875  * Return Code: DDI_PROP_SUCCESS
2876  *              DDI_PROP_NOT_FOUND
2877  *              DDI_PROP_UNDEFINED
2878  *              DDI_PROP_NO_MEMORY
2879  *              DDI_PROP_BUF_TOO_SMALL
2880  */
2881 
2882 static int
2883 sd_prop_op(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int mod_flags,
2884         char *name, caddr_t valuep, int *lengthp)
2885 {
2886         struct sd_lun   *un;
2887 
2888         if ((un = ddi_get_soft_state(sd_state, ddi_get_instance(dip))) == NULL)
2889                 return (ddi_prop_op(dev, dip, prop_op, mod_flags,
2890                     name, valuep, lengthp));
2891 
2892         return (cmlb_prop_op(un->un_cmlbhandle,
2893             dev, dip, prop_op, mod_flags, name, valuep, lengthp,
2894             SDPART(dev), (void *)SD_PATH_DIRECT));
2895 }
2896 
2897 /*
2898  * The following functions are for smart probing:
2899  * sd_scsi_probe_cache_init()
2900  * sd_scsi_probe_cache_fini()
2901  * sd_scsi_clear_probe_cache()
2902  * sd_scsi_probe_with_cache()
2903  */
2904 
2905 /*
2906  *    Function: sd_scsi_probe_cache_init
2907  *
2908  * Description: Initializes the probe response cache mutex and head pointer.
2909  *
2910  *     Context: Kernel thread context
2911  */
2912 
2913 static void
2914 sd_scsi_probe_cache_init(void)
2915 {
2916         mutex_init(&sd_scsi_probe_cache_mutex, NULL, MUTEX_DRIVER, NULL);
2917         sd_scsi_probe_cache_head = NULL;
2918 }
2919 
2920 
2921 /*
2922  *    Function: sd_scsi_probe_cache_fini
2923  *
2924  * Description: Frees all resources associated with the probe response cache.
2925  *
2926  *     Context: Kernel thread context
2927  */
2928 
2929 static void
2930 sd_scsi_probe_cache_fini(void)
2931 {
2932         struct sd_scsi_probe_cache *cp;
2933         struct sd_scsi_probe_cache *ncp;
2934 
2935         /* Clean up our smart probing linked list */
2936         for (cp = sd_scsi_probe_cache_head; cp != NULL; cp = ncp) {
2937                 ncp = cp->next;
2938                 kmem_free(cp, sizeof (struct sd_scsi_probe_cache));
2939         }
2940         sd_scsi_probe_cache_head = NULL;
2941         mutex_destroy(&sd_scsi_probe_cache_mutex);
2942 }
2943 
2944 
2945 /*
2946  *    Function: sd_scsi_clear_probe_cache
2947  *
2948  * Description: This routine clears the probe response cache. This is
2949  *              done when open() returns ENXIO so that when deferred
2950  *              attach is attempted (possibly after a device has been
2951  *              turned on) we will retry the probe. Since we don't know
2952  *              which target we failed to open, we just clear the
2953  *              entire cache.
2954  *
2955  *     Context: Kernel thread context
2956  */
2957 
2958 static void
2959 sd_scsi_clear_probe_cache(void)
2960 {
2961         struct sd_scsi_probe_cache      *cp;
2962         int                             i;
2963 
2964         mutex_enter(&sd_scsi_probe_cache_mutex);
2965         for (cp = sd_scsi_probe_cache_head; cp != NULL; cp = cp->next) {
2966                 /*
2967                  * Reset all entries to SCSIPROBE_EXISTS.  This will
2968                  * force probing to be performed the next time
2969                  * sd_scsi_probe_with_cache is called.
2970                  */
2971                 for (i = 0; i < NTARGETS_WIDE; i++) {
2972                         cp->cache[i] = SCSIPROBE_EXISTS;
2973                 }
2974         }
2975         mutex_exit(&sd_scsi_probe_cache_mutex);
2976 }
2977 
2978 
2979 /*
2980  *    Function: sd_scsi_probe_with_cache
2981  *
2982  * Description: This routine implements support for a scsi device probe
2983  *              with cache. The driver maintains a cache of the target
2984  *              responses to scsi probes. If we get no response from a
2985  *              target during a probe inquiry, we remember that, and we
2986  *              avoid additional calls to scsi_probe on non-zero LUNs
2987  *              on the same target until the cache is cleared. By doing
2988  *              so we avoid the 1/4 sec selection timeout for nonzero
2989  *              LUNs. lun0 of a target is always probed.
2990  *
2991  *   Arguments: devp     - Pointer to a scsi_device(9S) structure
2992  *              waitfunc - indicates what the allocator routines should
2993  *                         do when resources are not available. This value
2994  *                         is passed on to scsi_probe() when that routine
2995  *                         is called.
2996  *
2997  * Return Code: SCSIPROBE_NORESP if a NORESP in probe response cache;
2998  *              otherwise the value returned by scsi_probe(9F).
2999  *
3000  *     Context: Kernel thread context
3001  */
3002 
3003 static int
3004 sd_scsi_probe_with_cache(struct scsi_device *devp, int (*waitfn)())
3005 {
3006         struct sd_scsi_probe_cache      *cp;
3007         dev_info_t      *pdip = ddi_get_parent(devp->sd_dev);
3008         int             lun, tgt;
3009 
3010         lun = ddi_prop_get_int(DDI_DEV_T_ANY, devp->sd_dev, DDI_PROP_DONTPASS,
3011             SCSI_ADDR_PROP_LUN, 0);
3012         tgt = ddi_prop_get_int(DDI_DEV_T_ANY, devp->sd_dev, DDI_PROP_DONTPASS,
3013             SCSI_ADDR_PROP_TARGET, -1);
3014 
3015         /* Make sure caching enabled and target in range */
3016         if ((tgt < 0) || (tgt >= NTARGETS_WIDE)) {
3017                 /* do it the old way (no cache) */
3018                 return (scsi_probe(devp, waitfn));
3019         }
3020 
3021         mutex_enter(&sd_scsi_probe_cache_mutex);
3022 
3023         /* Find the cache for this scsi bus instance */
3024         for (cp = sd_scsi_probe_cache_head; cp != NULL; cp = cp->next) {
3025                 if (cp->pdip == pdip) {
3026                         break;
3027                 }
3028         }
3029 
3030         /* If we can't find a cache for this pdip, create one */
3031         if (cp == NULL) {
3032                 int i;
3033 
3034                 cp = kmem_zalloc(sizeof (struct sd_scsi_probe_cache),
3035                     KM_SLEEP);
3036                 cp->pdip = pdip;
3037                 cp->next = sd_scsi_probe_cache_head;
3038                 sd_scsi_probe_cache_head = cp;
3039                 for (i = 0; i < NTARGETS_WIDE; i++) {
3040                         cp->cache[i] = SCSIPROBE_EXISTS;
3041                 }
3042         }
3043 
3044         mutex_exit(&sd_scsi_probe_cache_mutex);
3045 
3046         /* Recompute the cache for this target if LUN zero */
3047         if (lun == 0) {
3048                 cp->cache[tgt] = SCSIPROBE_EXISTS;
3049         }
3050 
3051         /* Don't probe if cache remembers a NORESP from a previous LUN. */
3052         if (cp->cache[tgt] != SCSIPROBE_EXISTS) {
3053                 return (SCSIPROBE_NORESP);
3054         }
3055 
3056         /* Do the actual probe; save & return the result */
3057         return (cp->cache[tgt] = scsi_probe(devp, waitfn));
3058 }
3059 
3060 
3061 /*
3062  *    Function: sd_scsi_target_lun_init
3063  *
3064  * Description: Initializes the attached lun chain mutex and head pointer.
3065  *
3066  *     Context: Kernel thread context
3067  */
3068 
3069 static void
3070 sd_scsi_target_lun_init(void)
3071 {
3072         mutex_init(&sd_scsi_target_lun_mutex, NULL, MUTEX_DRIVER, NULL);
3073         sd_scsi_target_lun_head = NULL;
3074 }
3075 
3076 
3077 /*
3078  *    Function: sd_scsi_target_lun_fini
3079  *
3080  * Description: Frees all resources associated with the attached lun
3081  *              chain
3082  *
3083  *     Context: Kernel thread context
3084  */
3085 
3086 static void
3087 sd_scsi_target_lun_fini(void)
3088 {
3089         struct sd_scsi_hba_tgt_lun      *cp;
3090         struct sd_scsi_hba_tgt_lun      *ncp;
3091 
3092         for (cp = sd_scsi_target_lun_head; cp != NULL; cp = ncp) {
3093                 ncp = cp->next;
3094                 kmem_free(cp, sizeof (struct sd_scsi_hba_tgt_lun));
3095         }
3096         sd_scsi_target_lun_head = NULL;
3097         mutex_destroy(&sd_scsi_target_lun_mutex);
3098 }
3099 
3100 
3101 /*
3102  *    Function: sd_scsi_get_target_lun_count
3103  *
3104  * Description: This routine will check in the attached lun chain to see
3105  *              how many luns are attached on the required SCSI controller
3106  *              and target. Currently, some capabilities like tagged queue
3107  *              are supported per target based by HBA. So all luns in a
3108  *              target have the same capabilities. Based on this assumption,
3109  *              sd should only set these capabilities once per target. This
3110  *              function is called when sd needs to decide how many luns
3111  *              already attached on a target.
3112  *
3113  *   Arguments: dip     - Pointer to the system's dev_info_t for the SCSI
3114  *                        controller device.
3115  *              target  - The target ID on the controller's SCSI bus.
3116  *
3117  * Return Code: The number of luns attached on the required target and
3118  *              controller.
3119  *              -1 if target ID is not in parallel SCSI scope or the given
3120  *              dip is not in the chain.
3121  *
3122  *     Context: Kernel thread context
3123  */
3124 
3125 static int
3126 sd_scsi_get_target_lun_count(dev_info_t *dip, int target)
3127 {
3128         struct sd_scsi_hba_tgt_lun      *cp;
3129 
3130         if ((target < 0) || (target >= NTARGETS_WIDE)) {
3131                 return (-1);
3132         }
3133 
3134         mutex_enter(&sd_scsi_target_lun_mutex);
3135 
3136         for (cp = sd_scsi_target_lun_head; cp != NULL; cp = cp->next) {
3137                 if (cp->pdip == dip) {
3138                         break;
3139                 }
3140         }
3141 
3142         mutex_exit(&sd_scsi_target_lun_mutex);
3143 
3144         if (cp == NULL) {
3145                 return (-1);
3146         }
3147 
3148         return (cp->nlun[target]);
3149 }
3150 
3151 
3152 /*
3153  *    Function: sd_scsi_update_lun_on_target
3154  *
3155  * Description: This routine is used to update the attached lun chain when a
3156  *              lun is attached or detached on a target.
3157  *
3158  *   Arguments: dip     - Pointer to the system's dev_info_t for the SCSI
3159  *                        controller device.
3160  *              target  - The target ID on the controller's SCSI bus.
3161  *              flag    - Indicate the lun is attached or detached.
3162  *
3163  *     Context: Kernel thread context
3164  */
3165 
3166 static void
3167 sd_scsi_update_lun_on_target(dev_info_t *dip, int target, int flag)
3168 {
3169         struct sd_scsi_hba_tgt_lun      *cp;
3170 
3171         mutex_enter(&sd_scsi_target_lun_mutex);
3172 
3173         for (cp = sd_scsi_target_lun_head; cp != NULL; cp = cp->next) {
3174                 if (cp->pdip == dip) {
3175                         break;
3176                 }
3177         }
3178 
3179         if ((cp == NULL) && (flag == SD_SCSI_LUN_ATTACH)) {
3180                 cp = kmem_zalloc(sizeof (struct sd_scsi_hba_tgt_lun),
3181                     KM_SLEEP);
3182                 cp->pdip = dip;
3183                 cp->next = sd_scsi_target_lun_head;
3184                 sd_scsi_target_lun_head = cp;
3185         }
3186 
3187         mutex_exit(&sd_scsi_target_lun_mutex);
3188 
3189         if (cp != NULL) {
3190                 if (flag == SD_SCSI_LUN_ATTACH) {
3191                         cp->nlun[target] ++;
3192                 } else {
3193                         cp->nlun[target] --;
3194                 }
3195         }
3196 }
3197 
3198 
3199 /*
3200  *    Function: sd_spin_up_unit
3201  *
3202  * Description: Issues the following commands to spin-up the device:
3203  *              START STOP UNIT, and INQUIRY.
3204  *
3205  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
3206  *                      structure for this target.
3207  *
3208  * Return Code: 0 - success
3209  *              EIO - failure
3210  *              EACCES - reservation conflict
3211  *
3212  *     Context: Kernel thread context
3213  */
3214 
3215 static int
3216 sd_spin_up_unit(sd_ssc_t *ssc)
3217 {
3218         size_t  resid           = 0;
3219         int     has_conflict    = FALSE;
3220         uchar_t *bufaddr;
3221         int     status;
3222         struct sd_lun   *un;
3223 
3224         ASSERT(ssc != NULL);
3225         un = ssc->ssc_un;
3226         ASSERT(un != NULL);
3227 
3228         /*
3229          * Send a throwaway START UNIT command.
3230          *
3231          * If we fail on this, we don't care presently what precisely
3232          * is wrong.  EMC's arrays will also fail this with a check
3233          * condition (0x2/0x4/0x3) if the device is "inactive," but
3234          * we don't want to fail the attach because it may become
3235          * "active" later.
3236          * We don't know if power condition is supported or not at
3237          * this stage, use START STOP bit.
3238          */
3239         status = sd_send_scsi_START_STOP_UNIT(ssc, SD_START_STOP,
3240             SD_TARGET_START, SD_PATH_DIRECT);
3241 
3242         if (status != 0) {
3243                 if (status == EACCES)
3244                         has_conflict = TRUE;
3245                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3246         }
3247 
3248         /*
3249          * Send another INQUIRY command to the target. This is necessary for
3250          * non-removable media direct access devices because their INQUIRY data
3251          * may not be fully qualified until they are spun up (perhaps via the
3252          * START command above).  Note: This seems to be needed for some
3253          * legacy devices only.) The INQUIRY command should succeed even if a
3254          * Reservation Conflict is present.
3255          */
3256         bufaddr = kmem_zalloc(SUN_INQSIZE, KM_SLEEP);
3257 
3258         if (sd_send_scsi_INQUIRY(ssc, bufaddr, SUN_INQSIZE, 0, 0, &resid)
3259             != 0) {
3260                 kmem_free(bufaddr, SUN_INQSIZE);
3261                 sd_ssc_assessment(ssc, SD_FMT_STATUS_CHECK);
3262                 return (EIO);
3263         }
3264 
3265         /*
3266          * If we got enough INQUIRY data, copy it over the old INQUIRY data.
3267          * Note that this routine does not return a failure here even if the
3268          * INQUIRY command did not return any data.  This is a legacy behavior.
3269          */
3270         if ((SUN_INQSIZE - resid) >= SUN_MIN_INQLEN) {
3271                 bcopy(bufaddr, SD_INQUIRY(un), SUN_INQSIZE);
3272         }
3273 
3274         kmem_free(bufaddr, SUN_INQSIZE);
3275 
3276         /* If we hit a reservation conflict above, tell the caller. */
3277         if (has_conflict == TRUE) {
3278                 return (EACCES);
3279         }
3280 
3281         return (0);
3282 }
3283 
3284 #ifdef _LP64
3285 /*
3286  *    Function: sd_enable_descr_sense
3287  *
3288  * Description: This routine attempts to select descriptor sense format
3289  *              using the Control mode page.  Devices that support 64 bit
3290  *              LBAs (for >2TB luns) should also implement descriptor
3291  *              sense data so we will call this function whenever we see
3292  *              a lun larger than 2TB.  If for some reason the device
3293  *              supports 64 bit LBAs but doesn't support descriptor sense
3294  *              presumably the mode select will fail.  Everything will
3295  *              continue to work normally except that we will not get
3296  *              complete sense data for commands that fail with an LBA
3297  *              larger than 32 bits.
3298  *
3299  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
3300  *                      structure for this target.
3301  *
3302  *     Context: Kernel thread context only
3303  */
3304 
3305 static void
3306 sd_enable_descr_sense(sd_ssc_t *ssc)
3307 {
3308         uchar_t                 *header;
3309         struct mode_control_scsi3 *ctrl_bufp;
3310         size_t                  buflen;
3311         size_t                  bd_len;
3312         int                     status;
3313         struct sd_lun           *un;
3314 
3315         ASSERT(ssc != NULL);
3316         un = ssc->ssc_un;
3317         ASSERT(un != NULL);
3318 
3319         /*
3320          * Read MODE SENSE page 0xA, Control Mode Page
3321          */
3322         buflen = MODE_HEADER_LENGTH + MODE_BLK_DESC_LENGTH +
3323             sizeof (struct mode_control_scsi3);
3324         header = kmem_zalloc(buflen, KM_SLEEP);
3325 
3326         status = sd_send_scsi_MODE_SENSE(ssc, CDB_GROUP0, header, buflen,
3327             MODEPAGE_CTRL_MODE, SD_PATH_DIRECT);
3328 
3329         if (status != 0) {
3330                 SD_ERROR(SD_LOG_COMMON, un,
3331                     "sd_enable_descr_sense: mode sense ctrl page failed\n");
3332                 goto eds_exit;
3333         }
3334 
3335         /*
3336          * Determine size of Block Descriptors in order to locate
3337          * the mode page data. ATAPI devices return 0, SCSI devices
3338          * should return MODE_BLK_DESC_LENGTH.
3339          */
3340         bd_len  = ((struct mode_header *)header)->bdesc_length;
3341 
3342         /* Clear the mode data length field for MODE SELECT */
3343         ((struct mode_header *)header)->length = 0;
3344 
3345         ctrl_bufp = (struct mode_control_scsi3 *)
3346             (header + MODE_HEADER_LENGTH + bd_len);
3347 
3348         /*
3349          * If the page length is smaller than the expected value,
3350          * the target device doesn't support D_SENSE. Bail out here.
3351          */
3352         if (ctrl_bufp->mode_page.length <
3353             sizeof (struct mode_control_scsi3) - 2) {
3354                 SD_ERROR(SD_LOG_COMMON, un,
3355                     "sd_enable_descr_sense: enable D_SENSE failed\n");
3356                 goto eds_exit;
3357         }
3358 
3359         /*
3360          * Clear PS bit for MODE SELECT
3361          */
3362         ctrl_bufp->mode_page.ps = 0;
3363 
3364         /*
3365          * Set D_SENSE to enable descriptor sense format.
3366          */
3367         ctrl_bufp->d_sense = 1;
3368 
3369         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3370 
3371         /*
3372          * Use MODE SELECT to commit the change to the D_SENSE bit
3373          */
3374         status = sd_send_scsi_MODE_SELECT(ssc, CDB_GROUP0, header,
3375             buflen, SD_DONTSAVE_PAGE, SD_PATH_DIRECT);
3376 
3377         if (status != 0) {
3378                 SD_INFO(SD_LOG_COMMON, un,
3379                     "sd_enable_descr_sense: mode select ctrl page failed\n");
3380         } else {
3381                 kmem_free(header, buflen);
3382                 return;
3383         }
3384 
3385 eds_exit:
3386         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3387         kmem_free(header, buflen);
3388 }
3389 
3390 /*
3391  *    Function: sd_reenable_dsense_task
3392  *
3393  * Description: Re-enable descriptor sense after device or bus reset
3394  *
3395  *     Context: Executes in a taskq() thread context
3396  */
3397 static void
3398 sd_reenable_dsense_task(void *arg)
3399 {
3400         struct  sd_lun  *un = arg;
3401         sd_ssc_t        *ssc;
3402 
3403         ASSERT(un != NULL);
3404 
3405         ssc = sd_ssc_init(un);
3406         sd_enable_descr_sense(ssc);
3407         sd_ssc_fini(ssc);
3408 }
3409 #endif /* _LP64 */
3410 
3411 /*
3412  *    Function: sd_set_mmc_caps
3413  *
3414  * Description: This routine determines if the device is MMC compliant and if
3415  *              the device supports CDDA via a mode sense of the CDVD
3416  *              capabilities mode page. Also checks if the device is a
3417  *              dvdram writable device.
3418  *
3419  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
3420  *                      structure for this target.
3421  *
3422  *     Context: Kernel thread context only
3423  */
3424 
3425 static void
3426 sd_set_mmc_caps(sd_ssc_t *ssc)
3427 {
3428         struct mode_header_grp2         *sense_mhp;
3429         uchar_t                         *sense_page;
3430         caddr_t                         buf;
3431         int                             bd_len;
3432         int                             status;
3433         struct uscsi_cmd                com;
3434         int                             rtn;
3435         uchar_t                         *out_data_rw, *out_data_hd;
3436         uchar_t                         *rqbuf_rw, *rqbuf_hd;
3437         uchar_t                         *out_data_gesn;
3438         int                             gesn_len;
3439         struct sd_lun                   *un;
3440 
3441         ASSERT(ssc != NULL);
3442         un = ssc->ssc_un;
3443         ASSERT(un != NULL);
3444 
3445         /*
3446          * The flags which will be set in this function are - mmc compliant,
3447          * dvdram writable device, cdda support. Initialize them to FALSE
3448          * and if a capability is detected - it will be set to TRUE.
3449          */
3450         un->un_f_mmc_cap = FALSE;
3451         un->un_f_dvdram_writable_device = FALSE;
3452         un->un_f_cfg_cdda = FALSE;
3453 
3454         buf = kmem_zalloc(BUFLEN_MODE_CDROM_CAP, KM_SLEEP);
3455         status = sd_send_scsi_MODE_SENSE(ssc, CDB_GROUP1, (uchar_t *)buf,
3456             BUFLEN_MODE_CDROM_CAP, MODEPAGE_CDROM_CAP, SD_PATH_DIRECT);
3457 
3458         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3459 
3460         if (status != 0) {
3461                 /* command failed; just return */
3462                 kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3463                 return;
3464         }
3465         /*
3466          * If the mode sense request for the CDROM CAPABILITIES
3467          * page (0x2A) succeeds the device is assumed to be MMC.
3468          */
3469         un->un_f_mmc_cap = TRUE;
3470 
3471         /* See if GET STATUS EVENT NOTIFICATION is supported */
3472         if (un->un_f_mmc_gesn_polling) {
3473                 gesn_len = SD_GESN_HEADER_LEN + SD_GESN_MEDIA_DATA_LEN;
3474                 out_data_gesn = kmem_zalloc(gesn_len, KM_SLEEP);
3475 
3476                 rtn = sd_send_scsi_GET_EVENT_STATUS_NOTIFICATION(ssc,
3477                     out_data_gesn, gesn_len, 1 << SD_GESN_MEDIA_CLASS);
3478 
3479                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3480 
3481                 if ((rtn != 0) || !sd_gesn_media_data_valid(out_data_gesn)) {
3482                         un->un_f_mmc_gesn_polling = FALSE;
3483                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
3484                             "sd_set_mmc_caps: gesn not supported "
3485                             "%d %x %x %x %x\n", rtn,
3486                             out_data_gesn[0], out_data_gesn[1],
3487                             out_data_gesn[2], out_data_gesn[3]);
3488                 }
3489 
3490                 kmem_free(out_data_gesn, gesn_len);
3491         }
3492 
3493         /* Get to the page data */
3494         sense_mhp = (struct mode_header_grp2 *)buf;
3495         bd_len = (sense_mhp->bdesc_length_hi << 8) |
3496             sense_mhp->bdesc_length_lo;
3497         if (bd_len > MODE_BLK_DESC_LENGTH) {
3498                 /*
3499                  * We did not get back the expected block descriptor
3500                  * length so we cannot determine if the device supports
3501                  * CDDA. However, we still indicate the device is MMC
3502                  * according to the successful response to the page
3503                  * 0x2A mode sense request.
3504                  */
3505                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
3506                     "sd_set_mmc_caps: Mode Sense returned "
3507                     "invalid block descriptor length\n");
3508                 kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3509                 return;
3510         }
3511 
3512         /* See if read CDDA is supported */
3513         sense_page = (uchar_t *)(buf + MODE_HEADER_LENGTH_GRP2 +
3514             bd_len);
3515         un->un_f_cfg_cdda = (sense_page[5] & 0x01) ? TRUE : FALSE;
3516 
3517         /* See if writing DVD RAM is supported. */
3518         un->un_f_dvdram_writable_device = (sense_page[3] & 0x20) ? TRUE : FALSE;
3519         if (un->un_f_dvdram_writable_device == TRUE) {
3520                 kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3521                 return;
3522         }
3523 
3524         /*
3525          * If the device presents DVD or CD capabilities in the mode
3526          * page, we can return here since a RRD will not have
3527          * these capabilities.
3528          */
3529         if ((sense_page[2] & 0x3f) || (sense_page[3] & 0x3f)) {
3530                 kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3531                 return;
3532         }
3533         kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3534 
3535         /*
3536          * If un->un_f_dvdram_writable_device is still FALSE,
3537          * check for a Removable Rigid Disk (RRD).  A RRD
3538          * device is identified by the features RANDOM_WRITABLE and
3539          * HARDWARE_DEFECT_MANAGEMENT.
3540          */
3541         out_data_rw = kmem_zalloc(SD_CURRENT_FEATURE_LEN, KM_SLEEP);
3542         rqbuf_rw = kmem_zalloc(SENSE_LENGTH, KM_SLEEP);
3543 
3544         rtn = sd_send_scsi_feature_GET_CONFIGURATION(ssc, &com, rqbuf_rw,
3545             SENSE_LENGTH, out_data_rw, SD_CURRENT_FEATURE_LEN,
3546             RANDOM_WRITABLE, SD_PATH_STANDARD);
3547 
3548         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3549 
3550         if (rtn != 0) {
3551                 kmem_free(out_data_rw, SD_CURRENT_FEATURE_LEN);
3552                 kmem_free(rqbuf_rw, SENSE_LENGTH);
3553                 return;
3554         }
3555 
3556         out_data_hd = kmem_zalloc(SD_CURRENT_FEATURE_LEN, KM_SLEEP);
3557         rqbuf_hd = kmem_zalloc(SENSE_LENGTH, KM_SLEEP);
3558 
3559         rtn = sd_send_scsi_feature_GET_CONFIGURATION(ssc, &com, rqbuf_hd,
3560             SENSE_LENGTH, out_data_hd, SD_CURRENT_FEATURE_LEN,
3561             HARDWARE_DEFECT_MANAGEMENT, SD_PATH_STANDARD);
3562 
3563         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3564 
3565         if (rtn == 0) {
3566                 /*
3567                  * We have good information, check for random writable
3568                  * and hardware defect features.
3569                  */
3570                 if ((out_data_rw[9] & RANDOM_WRITABLE) &&
3571                     (out_data_hd[9] & HARDWARE_DEFECT_MANAGEMENT)) {
3572                         un->un_f_dvdram_writable_device = TRUE;
3573                 }
3574         }
3575 
3576         kmem_free(out_data_rw, SD_CURRENT_FEATURE_LEN);
3577         kmem_free(rqbuf_rw, SENSE_LENGTH);
3578         kmem_free(out_data_hd, SD_CURRENT_FEATURE_LEN);
3579         kmem_free(rqbuf_hd, SENSE_LENGTH);
3580 }
3581 
3582 /*
3583  *    Function: sd_check_for_writable_cd
3584  *
3585  * Description: This routine determines if the media in the device is
3586  *              writable or not. It uses the get configuration command (0x46)
3587  *              to determine if the media is writable
3588  *
3589  *   Arguments: un - driver soft state (unit) structure
3590  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct"
3591  *                           chain and the normal command waitq, or
3592  *                           SD_PATH_DIRECT_PRIORITY to use the USCSI
3593  *                           "direct" chain and bypass the normal command
3594  *                           waitq.
3595  *
3596  *     Context: Never called at interrupt context.
3597  */
3598 
3599 static void
3600 sd_check_for_writable_cd(sd_ssc_t *ssc, int path_flag)
3601 {
3602         struct uscsi_cmd                com;
3603         uchar_t                         *out_data;
3604         uchar_t                         *rqbuf;
3605         int                             rtn;
3606         uchar_t                         *out_data_rw, *out_data_hd;
3607         uchar_t                         *rqbuf_rw, *rqbuf_hd;
3608         struct mode_header_grp2         *sense_mhp;
3609         uchar_t                         *sense_page;
3610         caddr_t                         buf;
3611         int                             bd_len;
3612         int                             status;
3613         struct sd_lun                   *un;
3614 
3615         ASSERT(ssc != NULL);
3616         un = ssc->ssc_un;
3617         ASSERT(un != NULL);
3618         ASSERT(mutex_owned(SD_MUTEX(un)));
3619 
3620         /*
3621          * Initialize the writable media to false, if configuration info.
3622          * tells us otherwise then only we will set it.
3623          */
3624         un->un_f_mmc_writable_media = FALSE;
3625         mutex_exit(SD_MUTEX(un));
3626 
3627         out_data = kmem_zalloc(SD_PROFILE_HEADER_LEN, KM_SLEEP);
3628         rqbuf = kmem_zalloc(SENSE_LENGTH, KM_SLEEP);
3629 
3630         rtn = sd_send_scsi_GET_CONFIGURATION(ssc, &com, rqbuf, SENSE_LENGTH,
3631             out_data, SD_PROFILE_HEADER_LEN, path_flag);
3632 
3633         if (rtn != 0)
3634                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3635 
3636         mutex_enter(SD_MUTEX(un));
3637         if (rtn == 0) {
3638                 /*
3639                  * We have good information, check for writable DVD.
3640                  */
3641                 if ((out_data[6] == 0) && (out_data[7] == 0x12)) {
3642                         un->un_f_mmc_writable_media = TRUE;
3643                         kmem_free(out_data, SD_PROFILE_HEADER_LEN);
3644                         kmem_free(rqbuf, SENSE_LENGTH);
3645                         return;
3646                 }
3647         }
3648 
3649         kmem_free(out_data, SD_PROFILE_HEADER_LEN);
3650         kmem_free(rqbuf, SENSE_LENGTH);
3651 
3652         /*
3653          * Determine if this is a RRD type device.
3654          */
3655         mutex_exit(SD_MUTEX(un));
3656         buf = kmem_zalloc(BUFLEN_MODE_CDROM_CAP, KM_SLEEP);
3657         status = sd_send_scsi_MODE_SENSE(ssc, CDB_GROUP1, (uchar_t *)buf,
3658             BUFLEN_MODE_CDROM_CAP, MODEPAGE_CDROM_CAP, path_flag);
3659 
3660         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3661 
3662         mutex_enter(SD_MUTEX(un));
3663         if (status != 0) {
3664                 /* command failed; just return */
3665                 kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3666                 return;
3667         }
3668 
3669         /* Get to the page data */
3670         sense_mhp = (struct mode_header_grp2 *)buf;
3671         bd_len = (sense_mhp->bdesc_length_hi << 8) | sense_mhp->bdesc_length_lo;
3672         if (bd_len > MODE_BLK_DESC_LENGTH) {
3673                 /*
3674                  * We did not get back the expected block descriptor length so
3675                  * we cannot check the mode page.
3676                  */
3677                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
3678                     "sd_check_for_writable_cd: Mode Sense returned "
3679                     "invalid block descriptor length\n");
3680                 kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3681                 return;
3682         }
3683 
3684         /*
3685          * If the device presents DVD or CD capabilities in the mode
3686          * page, we can return here since a RRD device will not have
3687          * these capabilities.
3688          */
3689         sense_page = (uchar_t *)(buf + MODE_HEADER_LENGTH_GRP2 + bd_len);
3690         if ((sense_page[2] & 0x3f) || (sense_page[3] & 0x3f)) {
3691                 kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3692                 return;
3693         }
3694         kmem_free(buf, BUFLEN_MODE_CDROM_CAP);
3695 
3696         /*
3697          * If un->un_f_mmc_writable_media is still FALSE,
3698          * check for RRD type media.  A RRD device is identified
3699          * by the features RANDOM_WRITABLE and HARDWARE_DEFECT_MANAGEMENT.
3700          */
3701         mutex_exit(SD_MUTEX(un));
3702         out_data_rw = kmem_zalloc(SD_CURRENT_FEATURE_LEN, KM_SLEEP);
3703         rqbuf_rw = kmem_zalloc(SENSE_LENGTH, KM_SLEEP);
3704 
3705         rtn = sd_send_scsi_feature_GET_CONFIGURATION(ssc, &com, rqbuf_rw,
3706             SENSE_LENGTH, out_data_rw, SD_CURRENT_FEATURE_LEN,
3707             RANDOM_WRITABLE, path_flag);
3708 
3709         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3710         if (rtn != 0) {
3711                 kmem_free(out_data_rw, SD_CURRENT_FEATURE_LEN);
3712                 kmem_free(rqbuf_rw, SENSE_LENGTH);
3713                 mutex_enter(SD_MUTEX(un));
3714                 return;
3715         }
3716 
3717         out_data_hd = kmem_zalloc(SD_CURRENT_FEATURE_LEN, KM_SLEEP);
3718         rqbuf_hd = kmem_zalloc(SENSE_LENGTH, KM_SLEEP);
3719 
3720         rtn = sd_send_scsi_feature_GET_CONFIGURATION(ssc, &com, rqbuf_hd,
3721             SENSE_LENGTH, out_data_hd, SD_CURRENT_FEATURE_LEN,
3722             HARDWARE_DEFECT_MANAGEMENT, path_flag);
3723 
3724         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
3725         mutex_enter(SD_MUTEX(un));
3726         if (rtn == 0) {
3727                 /*
3728                  * We have good information, check for random writable
3729                  * and hardware defect features as current.
3730                  */
3731                 if ((out_data_rw[9] & RANDOM_WRITABLE) &&
3732                     (out_data_rw[10] & 0x1) &&
3733                     (out_data_hd[9] & HARDWARE_DEFECT_MANAGEMENT) &&
3734                     (out_data_hd[10] & 0x1)) {
3735                         un->un_f_mmc_writable_media = TRUE;
3736                 }
3737         }
3738 
3739         kmem_free(out_data_rw, SD_CURRENT_FEATURE_LEN);
3740         kmem_free(rqbuf_rw, SENSE_LENGTH);
3741         kmem_free(out_data_hd, SD_CURRENT_FEATURE_LEN);
3742         kmem_free(rqbuf_hd, SENSE_LENGTH);
3743 }
3744 
3745 /*
3746  *    Function: sd_read_unit_properties
3747  *
3748  * Description: The following implements a property lookup mechanism.
3749  *              Properties for particular disks (keyed on vendor, model
3750  *              and rev numbers) are sought in the sd.conf file via
3751  *              sd_process_sdconf_file(), and if not found there, are
3752  *              looked for in a list hardcoded in this driver via
3753  *              sd_process_sdconf_table() Once located the properties
3754  *              are used to update the driver unit structure.
3755  *
3756  *   Arguments: un - driver soft state (unit) structure
3757  */
3758 
3759 static void
3760 sd_read_unit_properties(struct sd_lun *un)
3761 {
3762         /*
3763          * sd_process_sdconf_file returns SD_FAILURE if it cannot find
3764          * the "sd-config-list" property (from the sd.conf file) or if
3765          * there was not a match for the inquiry vid/pid. If this event
3766          * occurs the static driver configuration table is searched for
3767          * a match.
3768          */
3769         ASSERT(un != NULL);
3770         if (sd_process_sdconf_file(un) == SD_FAILURE) {
3771                 sd_process_sdconf_table(un);
3772         }
3773 
3774         /* check for LSI device */
3775         sd_is_lsi(un);
3776 
3777 
3778 }
3779 
3780 
3781 /*
3782  *    Function: sd_process_sdconf_file
3783  *
3784  * Description: Use ddi_prop_lookup(9F) to obtain the properties from the
3785  *              driver's config file (ie, sd.conf) and update the driver
3786  *              soft state structure accordingly.
3787  *
3788  *   Arguments: un - driver soft state (unit) structure
3789  *
3790  * Return Code: SD_SUCCESS - The properties were successfully set according
3791  *                           to the driver configuration file.
3792  *              SD_FAILURE - The driver config list was not obtained or
3793  *                           there was no vid/pid match. This indicates that
3794  *                           the static config table should be used.
3795  *
3796  * The config file has a property, "sd-config-list". Currently we support
3797  * two kinds of formats. For both formats, the value of this property
3798  * is a list of duplets:
3799  *
3800  *  sd-config-list=
3801  *      <duplet>,
3802  *      [,<duplet>]*;
3803  *
3804  * For the improved format, where
3805  *
3806  *     <duplet>:= "<vid+pid>","<tunable-list>"
3807  *
3808  * and
3809  *
3810  *     <tunable-list>:=   <tunable> [, <tunable> ]*;
3811  *     <tunable> =        <name> : <value>
3812  *
3813  * The <vid+pid> is the string that is returned by the target device on a
3814  * SCSI inquiry command, the <tunable-list> contains one or more tunables
3815  * to apply to all target devices with the specified <vid+pid>.
3816  *
3817  * Each <tunable> is a "<name> : <value>" pair.
3818  *
3819  * For the old format, the structure of each duplet is as follows:
3820  *
3821  *  <duplet>:= "<vid+pid>","<data-property-name_list>"
3822  *
3823  * The first entry of the duplet is the device ID string (the concatenated
3824  * vid & pid; not to be confused with a device_id).  This is defined in
3825  * the same way as in the sd_disk_table.
3826  *
3827  * The second part of the duplet is a string that identifies a
3828  * data-property-name-list. The data-property-name-list is defined as
3829  * follows:
3830  *
3831  *  <data-property-name-list>:=<data-property-name> [<data-property-name>]
3832  *
3833  * The syntax of <data-property-name> depends on the <version> field.
3834  *
3835  * If version = SD_CONF_VERSION_1 we have the following syntax:
3836  *
3837  *      <data-property-name>:=<version>,<flags>,<prop0>,<prop1>,.....<propN>
3838  *
3839  * where the prop0 value will be used to set prop0 if bit0 set in the
3840  * flags, prop1 if bit1 set, etc. and N = SD_CONF_MAX_ITEMS -1
3841  *
3842  */
3843 
3844 static int
3845 sd_process_sdconf_file(struct sd_lun *un)
3846 {
3847         char    **config_list = NULL;
3848         uint_t  nelements;
3849         char    *vidptr;
3850         int     vidlen;
3851         char    *dnlist_ptr;
3852         char    *dataname_ptr;
3853         char    *dataname_lasts;
3854         int     *data_list = NULL;
3855         uint_t  data_list_len;
3856         int     rval = SD_FAILURE;
3857         int     i;
3858 
3859         ASSERT(un != NULL);
3860 
3861         /* Obtain the configuration list associated with the .conf file */
3862         if (ddi_prop_lookup_string_array(DDI_DEV_T_ANY, SD_DEVINFO(un),
3863             DDI_PROP_DONTPASS | DDI_PROP_NOTPROM, sd_config_list,
3864             &config_list, &nelements) != DDI_PROP_SUCCESS) {
3865                 return (SD_FAILURE);
3866         }
3867 
3868         /*
3869          * Compare vids in each duplet to the inquiry vid - if a match is
3870          * made, get the data value and update the soft state structure
3871          * accordingly.
3872          *
3873          * Each duplet should show as a pair of strings, return SD_FAILURE
3874          * otherwise.
3875          */
3876         if (nelements & 1) {
3877                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
3878                     "sd-config-list should show as pairs of strings.\n");
3879                 if (config_list)
3880                         ddi_prop_free(config_list);
3881                 return (SD_FAILURE);
3882         }
3883 
3884         for (i = 0; i < nelements; i += 2) {
3885                 /*
3886                  * Note: The assumption here is that each vid entry is on
3887                  * a unique line from its associated duplet.
3888                  */
3889                 vidptr = config_list[i];
3890                 vidlen = (int)strlen(vidptr);
3891                 if ((vidlen == 0) ||
3892                     (sd_sdconf_id_match(un, vidptr, vidlen) != SD_SUCCESS)) {
3893                         continue;
3894                 }
3895 
3896                 /*
3897                  * dnlist contains 1 or more blank separated
3898                  * data-property-name entries
3899                  */
3900                 dnlist_ptr = config_list[i + 1];
3901 
3902                 if (strchr(dnlist_ptr, ':') != NULL) {
3903                         /*
3904                          * Decode the improved format sd-config-list.
3905                          */
3906                         sd_nvpair_str_decode(un, dnlist_ptr);
3907                 } else {
3908                         /*
3909                          * The old format sd-config-list, loop through all
3910                          * data-property-name entries in the
3911                          * data-property-name-list
3912                          * setting the properties for each.
3913                          */
3914                         for (dataname_ptr = sd_strtok_r(dnlist_ptr, " \t",
3915                             &dataname_lasts); dataname_ptr != NULL;
3916                             dataname_ptr = sd_strtok_r(NULL, " \t",
3917                             &dataname_lasts)) {
3918                                 int version;
3919 
3920                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
3921                                     "sd_process_sdconf_file: disk:%s, "
3922                                     "data:%s\n", vidptr, dataname_ptr);
3923 
3924                                 /* Get the data list */
3925                                 if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY,
3926                                     SD_DEVINFO(un), 0, dataname_ptr, &data_list,
3927                                     &data_list_len) != DDI_PROP_SUCCESS) {
3928                                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
3929                                             "sd_process_sdconf_file: data "
3930                                             "property (%s) has no value\n",
3931                                             dataname_ptr);
3932                                         continue;
3933                                 }
3934 
3935                                 version = data_list[0];
3936 
3937                                 if (version == SD_CONF_VERSION_1) {
3938                                         sd_tunables values;
3939 
3940                                         /* Set the properties */
3941                                         if (sd_chk_vers1_data(un, data_list[1],
3942                                             &data_list[2], data_list_len,
3943                                             dataname_ptr) == SD_SUCCESS) {
3944                                                 sd_get_tunables_from_conf(un,
3945                                                     data_list[1], &data_list[2],
3946                                                     &values);
3947                                                 sd_set_vers1_properties(un,
3948                                                     data_list[1], &values);
3949                                                 rval = SD_SUCCESS;
3950                                         } else {
3951                                                 rval = SD_FAILURE;
3952                                         }
3953                                 } else {
3954                                         scsi_log(SD_DEVINFO(un), sd_label,
3955                                             CE_WARN, "data property %s version "
3956                                             "0x%x is invalid.",
3957                                             dataname_ptr, version);
3958                                         rval = SD_FAILURE;
3959                                 }
3960                                 if (data_list)
3961                                         ddi_prop_free(data_list);
3962                         }
3963                 }
3964         }
3965 
3966         /* free up the memory allocated by ddi_prop_lookup_string_array(). */
3967         if (config_list) {
3968                 ddi_prop_free(config_list);
3969         }
3970 
3971         return (rval);
3972 }
3973 
3974 /*
3975  *    Function: sd_nvpair_str_decode()
3976  *
3977  * Description: Parse the improved format sd-config-list to get
3978  *    each entry of tunable, which includes a name-value pair.
3979  *    Then call sd_set_properties() to set the property.
3980  *
3981  *   Arguments: un - driver soft state (unit) structure
3982  *    nvpair_str - the tunable list
3983  */
3984 static void
3985 sd_nvpair_str_decode(struct sd_lun *un, char *nvpair_str)
3986 {
3987         char    *nv, *name, *value, *token;
3988         char    *nv_lasts, *v_lasts, *x_lasts;
3989 
3990         for (nv = sd_strtok_r(nvpair_str, ",", &nv_lasts); nv != NULL;
3991             nv = sd_strtok_r(NULL, ",", &nv_lasts)) {
3992                 token = sd_strtok_r(nv, ":", &v_lasts);
3993                 name  = sd_strtok_r(token, " \t", &x_lasts);
3994                 token = sd_strtok_r(NULL, ":", &v_lasts);
3995                 value = sd_strtok_r(token, " \t", &x_lasts);
3996                 if (name == NULL || value == NULL) {
3997                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
3998                             "sd_nvpair_str_decode: "
3999                             "name or value is not valid!\n");
4000                 } else {
4001                         sd_set_properties(un, name, value);
4002                 }
4003         }
4004 }
4005 
4006 /*
4007  *    Function: sd_strtok_r()
4008  *
4009  * Description: This function uses strpbrk and strspn to break
4010  *    string into tokens on sequentially subsequent calls. Return
4011  *    NULL when no non-separator characters remain. The first
4012  *    argument is NULL for subsequent calls.
4013  */
4014 static char *
4015 sd_strtok_r(char *string, const char *sepset, char **lasts)
4016 {
4017         char    *q, *r;
4018 
4019         /* First or subsequent call */
4020         if (string == NULL)
4021                 string = *lasts;
4022 
4023         if (string == NULL)
4024                 return (NULL);
4025 
4026         /* Skip leading separators */
4027         q = string + strspn(string, sepset);
4028 
4029         if (*q == '\0')
4030                 return (NULL);
4031 
4032         if ((r = strpbrk(q, sepset)) == NULL)
4033                 *lasts = NULL;
4034         else {
4035                 *r = '\0';
4036                 *lasts = r + 1;
4037         }
4038         return (q);
4039 }
4040 
4041 /*
4042  *    Function: sd_set_properties()
4043  *
4044  * Description: Set device properties based on the improved
4045  *    format sd-config-list.
4046  *
4047  *   Arguments: un - driver soft state (unit) structure
4048  *    name  - supported tunable name
4049  *    value - tunable value
4050  */
4051 static void
4052 sd_set_properties(struct sd_lun *un, char *name, char *value)
4053 {
4054         char    *endptr = NULL;
4055         long    val = 0;
4056 
4057         if (strcasecmp(name, "cache-nonvolatile") == 0) {
4058                 if (strcasecmp(value, "true") == 0) {
4059                         un->un_f_suppress_cache_flush = TRUE;
4060                 } else if (strcasecmp(value, "false") == 0) {
4061                         un->un_f_suppress_cache_flush = FALSE;
4062                 } else {
4063                         goto value_invalid;
4064                 }
4065                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4066                     "suppress_cache_flush flag set to %d\n",
4067                     un->un_f_suppress_cache_flush);
4068                 return;
4069         }
4070 
4071         if (strcasecmp(name, "controller-type") == 0) {
4072                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4073                         un->un_ctype = val;
4074                 } else {
4075                         goto value_invalid;
4076                 }
4077                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4078                     "ctype set to %d\n", un->un_ctype);
4079                 return;
4080         }
4081 
4082         if (strcasecmp(name, "delay-busy") == 0) {
4083                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4084                         un->un_busy_timeout = drv_usectohz(val / 1000);
4085                 } else {
4086                         goto value_invalid;
4087                 }
4088                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4089                     "busy_timeout set to %d\n", un->un_busy_timeout);
4090                 return;
4091         }
4092 
4093         if (strcasecmp(name, "disksort") == 0) {
4094                 if (strcasecmp(value, "true") == 0) {
4095                         un->un_f_disksort_disabled = FALSE;
4096                 } else if (strcasecmp(value, "false") == 0) {
4097                         un->un_f_disksort_disabled = TRUE;
4098                 } else {
4099                         goto value_invalid;
4100                 }
4101                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4102                     "disksort disabled flag set to %d\n",
4103                     un->un_f_disksort_disabled);
4104                 return;
4105         }
4106 
4107         if (strcasecmp(name, "power-condition") == 0) {
4108                 if (strcasecmp(value, "true") == 0) {
4109                         un->un_f_power_condition_disabled = FALSE;
4110                 } else if (strcasecmp(value, "false") == 0) {
4111                         un->un_f_power_condition_disabled = TRUE;
4112                 } else {
4113                         goto value_invalid;
4114                 }
4115                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4116                     "power condition disabled flag set to %d\n",
4117                     un->un_f_power_condition_disabled);
4118                 return;
4119         }
4120 
4121         if (strcasecmp(name, "timeout-releasereservation") == 0) {
4122                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4123                         un->un_reserve_release_time = val;
4124                 } else {
4125                         goto value_invalid;
4126                 }
4127                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4128                     "reservation release timeout set to %d\n",
4129                     un->un_reserve_release_time);
4130                 return;
4131         }
4132 
4133         if (strcasecmp(name, "reset-lun") == 0) {
4134                 if (strcasecmp(value, "true") == 0) {
4135                         un->un_f_lun_reset_enabled = TRUE;
4136                 } else if (strcasecmp(value, "false") == 0) {
4137                         un->un_f_lun_reset_enabled = FALSE;
4138                 } else {
4139                         goto value_invalid;
4140                 }
4141                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4142                     "lun reset enabled flag set to %d\n",
4143                     un->un_f_lun_reset_enabled);
4144                 return;
4145         }
4146 
4147         if (strcasecmp(name, "retries-busy") == 0) {
4148                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4149                         un->un_busy_retry_count = val;
4150                 } else {
4151                         goto value_invalid;
4152                 }
4153                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4154                     "busy retry count set to %d\n", un->un_busy_retry_count);
4155                 return;
4156         }
4157 
4158         if (strcasecmp(name, "retries-timeout") == 0) {
4159                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4160                         un->un_retry_count = val;
4161                 } else {
4162                         goto value_invalid;
4163                 }
4164                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4165                     "timeout retry count set to %d\n", un->un_retry_count);
4166                 return;
4167         }
4168 
4169         if (strcasecmp(name, "retries-notready") == 0) {
4170                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4171                         un->un_notready_retry_count = val;
4172                 } else {
4173                         goto value_invalid;
4174                 }
4175                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4176                     "notready retry count set to %d\n",
4177                     un->un_notready_retry_count);
4178                 return;
4179         }
4180 
4181         if (strcasecmp(name, "retries-reset") == 0) {
4182                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4183                         un->un_reset_retry_count = val;
4184                 } else {
4185                         goto value_invalid;
4186                 }
4187                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4188                     "reset retry count set to %d\n",
4189                     un->un_reset_retry_count);
4190                 return;
4191         }
4192 
4193         if (strcasecmp(name, "throttle-max") == 0) {
4194                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4195                         un->un_saved_throttle = un->un_throttle = val;
4196                 } else {
4197                         goto value_invalid;
4198                 }
4199                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4200                     "throttle set to %d\n", un->un_throttle);
4201         }
4202 
4203         if (strcasecmp(name, "throttle-min") == 0) {
4204                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4205                         un->un_min_throttle = val;
4206                 } else {
4207                         goto value_invalid;
4208                 }
4209                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4210                     "min throttle set to %d\n", un->un_min_throttle);
4211         }
4212 
4213         if (strcasecmp(name, "rmw-type") == 0) {
4214                 if (ddi_strtol(value, &endptr, 0, &val) == 0) {
4215                         un->un_f_rmw_type = val;
4216                 } else {
4217                         goto value_invalid;
4218                 }
4219                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4220                     "RMW type set to %d\n", un->un_f_rmw_type);
4221         }
4222 
4223         if (strcasecmp(name, "physical-block-size") == 0) {
4224                 if (ddi_strtol(value, &endptr, 0, &val) == 0 &&
4225                     ISP2(val) && val >= un->un_tgt_blocksize &&
4226                     val >= un->un_sys_blocksize) {
4227                         un->un_phy_blocksize = val;
4228                 } else {
4229                         goto value_invalid;
4230                 }
4231                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4232                     "physical block size set to %d\n", un->un_phy_blocksize);
4233         }
4234 
4235         /*
4236          * Validate the throttle values.
4237          * If any of the numbers are invalid, set everything to defaults.
4238          */
4239         if ((un->un_throttle < SD_LOWEST_VALID_THROTTLE) ||
4240             (un->un_min_throttle < SD_LOWEST_VALID_THROTTLE) ||
4241             (un->un_min_throttle > un->un_throttle)) {
4242                 un->un_saved_throttle = un->un_throttle = sd_max_throttle;
4243                 un->un_min_throttle = sd_min_throttle;
4244         }
4245 
4246         if (strcasecmp(name, "mmc-gesn-polling") == 0) {
4247                 if (strcasecmp(value, "true") == 0) {
4248                         un->un_f_mmc_gesn_polling = TRUE;
4249                 } else if (strcasecmp(value, "false") == 0) {
4250                         un->un_f_mmc_gesn_polling = FALSE;
4251                 } else {
4252                         goto value_invalid;
4253                 }
4254                 SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4255                     "mmc-gesn-polling set to %d\n",
4256                     un->un_f_mmc_gesn_polling);
4257         }
4258 
4259         return;
4260 
4261 value_invalid:
4262         SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_set_properties: "
4263             "value of prop %s is invalid\n", name);
4264 }
4265 
4266 /*
4267  *    Function: sd_get_tunables_from_conf()
4268  *
4269  *
4270  *    This function reads the data list from the sd.conf file and pulls
4271  *    the values that can have numeric values as arguments and places
4272  *    the values in the appropriate sd_tunables member.
4273  *    Since the order of the data list members varies across platforms
4274  *    This function reads them from the data list in a platform specific
4275  *    order and places them into the correct sd_tunable member that is
4276  *    consistent across all platforms.
4277  */
4278 static void
4279 sd_get_tunables_from_conf(struct sd_lun *un, int flags, int *data_list,
4280     sd_tunables *values)
4281 {
4282         int i;
4283         int mask;
4284 
4285         bzero(values, sizeof (sd_tunables));
4286 
4287         for (i = 0; i < SD_CONF_MAX_ITEMS; i++) {
4288 
4289                 mask = 1 << i;
4290                 if (mask > flags) {
4291                         break;
4292                 }
4293 
4294                 switch (mask & flags) {
4295                 case 0: /* This mask bit not set in flags */
4296                         continue;
4297                 case SD_CONF_BSET_THROTTLE:
4298                         values->sdt_throttle = data_list[i];
4299                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4300                             "sd_get_tunables_from_conf: throttle = %d\n",
4301                             values->sdt_throttle);
4302                         break;
4303                 case SD_CONF_BSET_CTYPE:
4304                         values->sdt_ctype = data_list[i];
4305                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4306                             "sd_get_tunables_from_conf: ctype = %d\n",
4307                             values->sdt_ctype);
4308                         break;
4309                 case SD_CONF_BSET_NRR_COUNT:
4310                         values->sdt_not_rdy_retries = data_list[i];
4311                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4312                             "sd_get_tunables_from_conf: not_rdy_retries = %d\n",
4313                             values->sdt_not_rdy_retries);
4314                         break;
4315                 case SD_CONF_BSET_BSY_RETRY_COUNT:
4316                         values->sdt_busy_retries = data_list[i];
4317                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4318                             "sd_get_tunables_from_conf: busy_retries = %d\n",
4319                             values->sdt_busy_retries);
4320                         break;
4321                 case SD_CONF_BSET_RST_RETRIES:
4322                         values->sdt_reset_retries = data_list[i];
4323                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4324                             "sd_get_tunables_from_conf: reset_retries = %d\n",
4325                             values->sdt_reset_retries);
4326                         break;
4327                 case SD_CONF_BSET_RSV_REL_TIME:
4328                         values->sdt_reserv_rel_time = data_list[i];
4329                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4330                             "sd_get_tunables_from_conf: reserv_rel_time = %d\n",
4331                             values->sdt_reserv_rel_time);
4332                         break;
4333                 case SD_CONF_BSET_MIN_THROTTLE:
4334                         values->sdt_min_throttle = data_list[i];
4335                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4336                             "sd_get_tunables_from_conf: min_throttle = %d\n",
4337                             values->sdt_min_throttle);
4338                         break;
4339                 case SD_CONF_BSET_DISKSORT_DISABLED:
4340                         values->sdt_disk_sort_dis = data_list[i];
4341                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4342                             "sd_get_tunables_from_conf: disk_sort_dis = %d\n",
4343                             values->sdt_disk_sort_dis);
4344                         break;
4345                 case SD_CONF_BSET_LUN_RESET_ENABLED:
4346                         values->sdt_lun_reset_enable = data_list[i];
4347                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4348                             "sd_get_tunables_from_conf: lun_reset_enable = %d"
4349                             "\n", values->sdt_lun_reset_enable);
4350                         break;
4351                 case SD_CONF_BSET_CACHE_IS_NV:
4352                         values->sdt_suppress_cache_flush = data_list[i];
4353                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4354                             "sd_get_tunables_from_conf: \
4355                             suppress_cache_flush = %d"
4356                             "\n", values->sdt_suppress_cache_flush);
4357                         break;
4358                 case SD_CONF_BSET_PC_DISABLED:
4359                         values->sdt_disk_sort_dis = data_list[i];
4360                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4361                             "sd_get_tunables_from_conf: power_condition_dis = "
4362                             "%d\n", values->sdt_power_condition_dis);
4363                         break;
4364                 }
4365         }
4366 }
4367 
4368 /*
4369  *    Function: sd_process_sdconf_table
4370  *
4371  * Description: Search the static configuration table for a match on the
4372  *              inquiry vid/pid and update the driver soft state structure
4373  *              according to the table property values for the device.
4374  *
4375  *              The form of a configuration table entry is:
4376  *                <vid+pid>,<flags>,<property-data>
4377  *                "SEAGATE ST42400N",1,0x40000,
4378  *                0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1;
4379  *
4380  *   Arguments: un - driver soft state (unit) structure
4381  */
4382 
4383 static void
4384 sd_process_sdconf_table(struct sd_lun *un)
4385 {
4386         char    *id = NULL;
4387         int     table_index;
4388         int     idlen;
4389 
4390         ASSERT(un != NULL);
4391         for (table_index = 0; table_index < sd_disk_table_size;
4392             table_index++) {
4393                 id = sd_disk_table[table_index].device_id;
4394                 idlen = strlen(id);
4395                 if (idlen == 0) {
4396                         continue;
4397                 }
4398 
4399                 /*
4400                  * The static configuration table currently does not
4401                  * implement version 10 properties. Additionally,
4402                  * multiple data-property-name entries are not
4403                  * implemented in the static configuration table.
4404                  */
4405                 if (sd_sdconf_id_match(un, id, idlen) == SD_SUCCESS) {
4406                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4407                             "sd_process_sdconf_table: disk %s\n", id);
4408                         sd_set_vers1_properties(un,
4409                             sd_disk_table[table_index].flags,
4410                             sd_disk_table[table_index].properties);
4411                         break;
4412                 }
4413         }
4414 }
4415 
4416 
4417 /*
4418  *    Function: sd_sdconf_id_match
4419  *
4420  * Description: This local function implements a case sensitive vid/pid
4421  *              comparison as well as the boundary cases of wild card and
4422  *              multiple blanks.
4423  *
4424  *              Note: An implicit assumption made here is that the scsi
4425  *              inquiry structure will always keep the vid, pid and
4426  *              revision strings in consecutive sequence, so they can be
4427  *              read as a single string. If this assumption is not the
4428  *              case, a separate string, to be used for the check, needs
4429  *              to be built with these strings concatenated.
4430  *
4431  *   Arguments: un - driver soft state (unit) structure
4432  *              id - table or config file vid/pid
4433  *              idlen  - length of the vid/pid (bytes)
4434  *
4435  * Return Code: SD_SUCCESS - Indicates a match with the inquiry vid/pid
4436  *              SD_FAILURE - Indicates no match with the inquiry vid/pid
4437  */
4438 
4439 static int
4440 sd_sdconf_id_match(struct sd_lun *un, char *id, int idlen)
4441 {
4442         struct scsi_inquiry     *sd_inq;
4443         int                     rval = SD_SUCCESS;
4444 
4445         ASSERT(un != NULL);
4446         sd_inq = un->un_sd->sd_inq;
4447         ASSERT(id != NULL);
4448 
4449         /*
4450          * We use the inq_vid as a pointer to a buffer containing the
4451          * vid and pid and use the entire vid/pid length of the table
4452          * entry for the comparison. This works because the inq_pid
4453          * data member follows inq_vid in the scsi_inquiry structure.
4454          */
4455         if (strncasecmp(sd_inq->inq_vid, id, idlen) != 0) {
4456                 /*
4457                  * The user id string is compared to the inquiry vid/pid
4458                  * using a case insensitive comparison and ignoring
4459                  * multiple spaces.
4460                  */
4461                 rval = sd_blank_cmp(un, id, idlen);
4462                 if (rval != SD_SUCCESS) {
4463                         /*
4464                          * User id strings that start and end with a "*"
4465                          * are a special case. These do not have a
4466                          * specific vendor, and the product string can
4467                          * appear anywhere in the 16 byte PID portion of
4468                          * the inquiry data. This is a simple strstr()
4469                          * type search for the user id in the inquiry data.
4470                          */
4471                         if ((id[0] == '*') && (id[idlen - 1] == '*')) {
4472                                 char    *pidptr = &id[1];
4473                                 int     i;
4474                                 int     j;
4475                                 int     pidstrlen = idlen - 2;
4476                                 j = sizeof (SD_INQUIRY(un)->inq_pid) -
4477                                     pidstrlen;
4478 
4479                                 if (j < 0) {
4480                                         return (SD_FAILURE);
4481                                 }
4482                                 for (i = 0; i < j; i++) {
4483                                         if (bcmp(&SD_INQUIRY(un)->inq_pid[i],
4484                                             pidptr, pidstrlen) == 0) {
4485                                                 rval = SD_SUCCESS;
4486                                                 break;
4487                                         }
4488                                 }
4489                         }
4490                 }
4491         }
4492         return (rval);
4493 }
4494 
4495 
4496 /*
4497  *    Function: sd_blank_cmp
4498  *
4499  * Description: If the id string starts and ends with a space, treat
4500  *              multiple consecutive spaces as equivalent to a single
4501  *              space. For example, this causes a sd_disk_table entry
4502  *              of " NEC CDROM " to match a device's id string of
4503  *              "NEC       CDROM".
4504  *
4505  *              Note: The success exit condition for this routine is if
4506  *              the pointer to the table entry is '\0' and the cnt of
4507  *              the inquiry length is zero. This will happen if the inquiry
4508  *              string returned by the device is padded with spaces to be
4509  *              exactly 24 bytes in length (8 byte vid + 16 byte pid). The
4510  *              SCSI spec states that the inquiry string is to be padded with
4511  *              spaces.
4512  *
4513  *   Arguments: un - driver soft state (unit) structure
4514  *              id - table or config file vid/pid
4515  *              idlen  - length of the vid/pid (bytes)
4516  *
4517  * Return Code: SD_SUCCESS - Indicates a match with the inquiry vid/pid
4518  *              SD_FAILURE - Indicates no match with the inquiry vid/pid
4519  */
4520 
4521 static int
4522 sd_blank_cmp(struct sd_lun *un, char *id, int idlen)
4523 {
4524         char            *p1;
4525         char            *p2;
4526         int             cnt;
4527         cnt = sizeof (SD_INQUIRY(un)->inq_vid) +
4528             sizeof (SD_INQUIRY(un)->inq_pid);
4529 
4530         ASSERT(un != NULL);
4531         p2 = un->un_sd->sd_inq->inq_vid;
4532         ASSERT(id != NULL);
4533         p1 = id;
4534 
4535         if ((id[0] == ' ') && (id[idlen - 1] == ' ')) {
4536                 /*
4537                  * Note: string p1 is terminated by a NUL but string p2
4538                  * isn't.  The end of p2 is determined by cnt.
4539                  */
4540                 for (;;) {
4541                         /* skip over any extra blanks in both strings */
4542                         while ((*p1 != '\0') && (*p1 == ' ')) {
4543                                 p1++;
4544                         }
4545                         while ((cnt != 0) && (*p2 == ' ')) {
4546                                 p2++;
4547                                 cnt--;
4548                         }
4549 
4550                         /* compare the two strings */
4551                         if ((cnt == 0) ||
4552                             (SD_TOUPPER(*p1) != SD_TOUPPER(*p2))) {
4553                                 break;
4554                         }
4555                         while ((cnt > 0) &&
4556                             (SD_TOUPPER(*p1) == SD_TOUPPER(*p2))) {
4557                                 p1++;
4558                                 p2++;
4559                                 cnt--;
4560                         }
4561                 }
4562         }
4563 
4564         /* return SD_SUCCESS if both strings match */
4565         return (((*p1 == '\0') && (cnt == 0)) ? SD_SUCCESS : SD_FAILURE);
4566 }
4567 
4568 
4569 /*
4570  *    Function: sd_chk_vers1_data
4571  *
4572  * Description: Verify the version 1 device properties provided by the
4573  *              user via the configuration file
4574  *
4575  *   Arguments: un           - driver soft state (unit) structure
4576  *              flags        - integer mask indicating properties to be set
4577  *              prop_list    - integer list of property values
4578  *              list_len     - number of the elements
4579  *
4580  * Return Code: SD_SUCCESS - Indicates the user provided data is valid
4581  *              SD_FAILURE - Indicates the user provided data is invalid
4582  */
4583 
4584 static int
4585 sd_chk_vers1_data(struct sd_lun *un, int flags, int *prop_list,
4586     int list_len, char *dataname_ptr)
4587 {
4588         int i;
4589         int mask = 1;
4590         int index = 0;
4591 
4592         ASSERT(un != NULL);
4593 
4594         /* Check for a NULL property name and list */
4595         if (dataname_ptr == NULL) {
4596                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
4597                     "sd_chk_vers1_data: NULL data property name.");
4598                 return (SD_FAILURE);
4599         }
4600         if (prop_list == NULL) {
4601                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
4602                     "sd_chk_vers1_data: %s NULL data property list.",
4603                     dataname_ptr);
4604                 return (SD_FAILURE);
4605         }
4606 
4607         /* Display a warning if undefined bits are set in the flags */
4608         if (flags & ~SD_CONF_BIT_MASK) {
4609                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
4610                     "sd_chk_vers1_data: invalid bits 0x%x in data list %s. "
4611                     "Properties not set.",
4612                     (flags & ~SD_CONF_BIT_MASK), dataname_ptr);
4613                 return (SD_FAILURE);
4614         }
4615 
4616         /*
4617          * Verify the length of the list by identifying the highest bit set
4618          * in the flags and validating that the property list has a length
4619          * up to the index of this bit.
4620          */
4621         for (i = 0; i < SD_CONF_MAX_ITEMS; i++) {
4622                 if (flags & mask) {
4623                         index++;
4624                 }
4625                 mask = 1 << i;
4626         }
4627         if (list_len < (index + 2)) {
4628                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
4629                     "sd_chk_vers1_data: "
4630                     "Data property list %s size is incorrect. "
4631                     "Properties not set.", dataname_ptr);
4632                 scsi_log(SD_DEVINFO(un), sd_label, CE_CONT, "Size expected: "
4633                     "version + 1 flagword + %d properties", SD_CONF_MAX_ITEMS);
4634                 return (SD_FAILURE);
4635         }
4636         return (SD_SUCCESS);
4637 }
4638 
4639 
4640 /*
4641  *    Function: sd_set_vers1_properties
4642  *
4643  * Description: Set version 1 device properties based on a property list
4644  *              retrieved from the driver configuration file or static
4645  *              configuration table. Version 1 properties have the format:
4646  *
4647  *      <data-property-name>:=<version>,<flags>,<prop0>,<prop1>,.....<propN>
4648  *
4649  *              where the prop0 value will be used to set prop0 if bit0
4650  *              is set in the flags
4651  *
4652  *   Arguments: un           - driver soft state (unit) structure
4653  *              flags        - integer mask indicating properties to be set
4654  *              prop_list    - integer list of property values
4655  */
4656 
4657 static void
4658 sd_set_vers1_properties(struct sd_lun *un, int flags, sd_tunables *prop_list)
4659 {
4660         ASSERT(un != NULL);
4661 
4662         /*
4663          * Set the flag to indicate cache is to be disabled. An attempt
4664          * to disable the cache via sd_cache_control() will be made
4665          * later during attach once the basic initialization is complete.
4666          */
4667         if (flags & SD_CONF_BSET_NOCACHE) {
4668                 un->un_f_opt_disable_cache = TRUE;
4669                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4670                     "sd_set_vers1_properties: caching disabled flag set\n");
4671         }
4672 
4673         /* CD-specific configuration parameters */
4674         if (flags & SD_CONF_BSET_PLAYMSF_BCD) {
4675                 un->un_f_cfg_playmsf_bcd = TRUE;
4676                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4677                     "sd_set_vers1_properties: playmsf_bcd set\n");
4678         }
4679         if (flags & SD_CONF_BSET_READSUB_BCD) {
4680                 un->un_f_cfg_readsub_bcd = TRUE;
4681                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4682                     "sd_set_vers1_properties: readsub_bcd set\n");
4683         }
4684         if (flags & SD_CONF_BSET_READ_TOC_TRK_BCD) {
4685                 un->un_f_cfg_read_toc_trk_bcd = TRUE;
4686                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4687                     "sd_set_vers1_properties: read_toc_trk_bcd set\n");
4688         }
4689         if (flags & SD_CONF_BSET_READ_TOC_ADDR_BCD) {
4690                 un->un_f_cfg_read_toc_addr_bcd = TRUE;
4691                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4692                     "sd_set_vers1_properties: read_toc_addr_bcd set\n");
4693         }
4694         if (flags & SD_CONF_BSET_NO_READ_HEADER) {
4695                 un->un_f_cfg_no_read_header = TRUE;
4696                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4697                     "sd_set_vers1_properties: no_read_header set\n");
4698         }
4699         if (flags & SD_CONF_BSET_READ_CD_XD4) {
4700                 un->un_f_cfg_read_cd_xd4 = TRUE;
4701                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4702                     "sd_set_vers1_properties: read_cd_xd4 set\n");
4703         }
4704 
4705         /* Support for devices which do not have valid/unique serial numbers */
4706         if (flags & SD_CONF_BSET_FAB_DEVID) {
4707                 un->un_f_opt_fab_devid = TRUE;
4708                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4709                     "sd_set_vers1_properties: fab_devid bit set\n");
4710         }
4711 
4712         /* Support for user throttle configuration */
4713         if (flags & SD_CONF_BSET_THROTTLE) {
4714                 ASSERT(prop_list != NULL);
4715                 un->un_saved_throttle = un->un_throttle =
4716                     prop_list->sdt_throttle;
4717                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4718                     "sd_set_vers1_properties: throttle set to %d\n",
4719                     prop_list->sdt_throttle);
4720         }
4721 
4722         /* Set the per disk retry count according to the conf file or table. */
4723         if (flags & SD_CONF_BSET_NRR_COUNT) {
4724                 ASSERT(prop_list != NULL);
4725                 if (prop_list->sdt_not_rdy_retries) {
4726                         un->un_notready_retry_count =
4727                             prop_list->sdt_not_rdy_retries;
4728                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4729                             "sd_set_vers1_properties: not ready retry count"
4730                             " set to %d\n", un->un_notready_retry_count);
4731                 }
4732         }
4733 
4734         /* The controller type is reported for generic disk driver ioctls */
4735         if (flags & SD_CONF_BSET_CTYPE) {
4736                 ASSERT(prop_list != NULL);
4737                 switch (prop_list->sdt_ctype) {
4738                 case CTYPE_CDROM:
4739                         un->un_ctype = prop_list->sdt_ctype;
4740                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4741                             "sd_set_vers1_properties: ctype set to "
4742                             "CTYPE_CDROM\n");
4743                         break;
4744                 case CTYPE_CCS:
4745                         un->un_ctype = prop_list->sdt_ctype;
4746                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4747                             "sd_set_vers1_properties: ctype set to "
4748                             "CTYPE_CCS\n");
4749                         break;
4750                 case CTYPE_ROD:         /* RW optical */
4751                         un->un_ctype = prop_list->sdt_ctype;
4752                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
4753                             "sd_set_vers1_properties: ctype set to "
4754                             "CTYPE_ROD\n");
4755                         break;
4756                 default:
4757                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
4758                             "sd_set_vers1_properties: Could not set "
4759                             "invalid ctype value (%d)",
4760                             prop_list->sdt_ctype);
4761                 }
4762         }
4763 
4764         /* Purple failover timeout */
4765         if (flags & SD_CONF_BSET_BSY_RETRY_COUNT) {
4766                 ASSERT(prop_list != NULL);
4767                 un->un_busy_retry_count =
4768                     prop_list->sdt_busy_retries;
4769                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4770                     "sd_set_vers1_properties: "
4771                     "busy retry count set to %d\n",
4772                     un->un_busy_retry_count);
4773         }
4774 
4775         /* Purple reset retry count */
4776         if (flags & SD_CONF_BSET_RST_RETRIES) {
4777                 ASSERT(prop_list != NULL);
4778                 un->un_reset_retry_count =
4779                     prop_list->sdt_reset_retries;
4780                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4781                     "sd_set_vers1_properties: "
4782                     "reset retry count set to %d\n",
4783                     un->un_reset_retry_count);
4784         }
4785 
4786         /* Purple reservation release timeout */
4787         if (flags & SD_CONF_BSET_RSV_REL_TIME) {
4788                 ASSERT(prop_list != NULL);
4789                 un->un_reserve_release_time =
4790                     prop_list->sdt_reserv_rel_time;
4791                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4792                     "sd_set_vers1_properties: "
4793                     "reservation release timeout set to %d\n",
4794                     un->un_reserve_release_time);
4795         }
4796 
4797         /*
4798          * Driver flag telling the driver to verify that no commands are pending
4799          * for a device before issuing a Test Unit Ready. This is a workaround
4800          * for a firmware bug in some Seagate eliteI drives.
4801          */
4802         if (flags & SD_CONF_BSET_TUR_CHECK) {
4803                 un->un_f_cfg_tur_check = TRUE;
4804                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4805                     "sd_set_vers1_properties: tur queue check set\n");
4806         }
4807 
4808         if (flags & SD_CONF_BSET_MIN_THROTTLE) {
4809                 un->un_min_throttle = prop_list->sdt_min_throttle;
4810                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4811                     "sd_set_vers1_properties: min throttle set to %d\n",
4812                     un->un_min_throttle);
4813         }
4814 
4815         if (flags & SD_CONF_BSET_DISKSORT_DISABLED) {
4816                 un->un_f_disksort_disabled =
4817                     (prop_list->sdt_disk_sort_dis != 0) ?
4818                     TRUE : FALSE;
4819                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4820                     "sd_set_vers1_properties: disksort disabled "
4821                     "flag set to %d\n",
4822                     prop_list->sdt_disk_sort_dis);
4823         }
4824 
4825         if (flags & SD_CONF_BSET_LUN_RESET_ENABLED) {
4826                 un->un_f_lun_reset_enabled =
4827                     (prop_list->sdt_lun_reset_enable != 0) ?
4828                     TRUE : FALSE;
4829                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4830                     "sd_set_vers1_properties: lun reset enabled "
4831                     "flag set to %d\n",
4832                     prop_list->sdt_lun_reset_enable);
4833         }
4834 
4835         if (flags & SD_CONF_BSET_CACHE_IS_NV) {
4836                 un->un_f_suppress_cache_flush =
4837                     (prop_list->sdt_suppress_cache_flush != 0) ?
4838                     TRUE : FALSE;
4839                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4840                     "sd_set_vers1_properties: suppress_cache_flush "
4841                     "flag set to %d\n",
4842                     prop_list->sdt_suppress_cache_flush);
4843         }
4844 
4845         if (flags & SD_CONF_BSET_PC_DISABLED) {
4846                 un->un_f_power_condition_disabled =
4847                     (prop_list->sdt_power_condition_dis != 0) ?
4848                     TRUE : FALSE;
4849                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
4850                     "sd_set_vers1_properties: power_condition_disabled "
4851                     "flag set to %d\n",
4852                     prop_list->sdt_power_condition_dis);
4853         }
4854 
4855         /*
4856          * Validate the throttle values.
4857          * If any of the numbers are invalid, set everything to defaults.
4858          */
4859         if ((un->un_throttle < SD_LOWEST_VALID_THROTTLE) ||
4860             (un->un_min_throttle < SD_LOWEST_VALID_THROTTLE) ||
4861             (un->un_min_throttle > un->un_throttle)) {
4862                 un->un_saved_throttle = un->un_throttle = sd_max_throttle;
4863                 un->un_min_throttle = sd_min_throttle;
4864         }
4865 }
4866 
4867 /*
4868  *   Function: sd_is_lsi()
4869  *
4870  *   Description: Check for lsi devices, step through the static device
4871  *      table to match vid/pid.
4872  *
4873  *   Args: un - ptr to sd_lun
4874  *
4875  *   Notes:  When creating new LSI property, need to add the new LSI property
4876  *              to this function.
4877  */
4878 static void
4879 sd_is_lsi(struct sd_lun *un)
4880 {
4881         char    *id = NULL;
4882         int     table_index;
4883         int     idlen;
4884         void    *prop;
4885 
4886         ASSERT(un != NULL);
4887         for (table_index = 0; table_index < sd_disk_table_size;
4888             table_index++) {
4889                 id = sd_disk_table[table_index].device_id;
4890                 idlen = strlen(id);
4891                 if (idlen == 0) {
4892                         continue;
4893                 }
4894 
4895                 if (sd_sdconf_id_match(un, id, idlen) == SD_SUCCESS) {
4896                         prop = sd_disk_table[table_index].properties;
4897                         if (prop == &lsi_properties ||
4898                             prop == &lsi_oem_properties ||
4899                             prop == &lsi_properties_scsi ||
4900                             prop == &symbios_properties) {
4901                                 un->un_f_cfg_is_lsi = TRUE;
4902                         }
4903                         break;
4904                 }
4905         }
4906 }
4907 
4908 /*
4909  *    Function: sd_get_physical_geometry
4910  *
4911  * Description: Retrieve the MODE SENSE page 3 (Format Device Page) and
4912  *              MODE SENSE page 4 (Rigid Disk Drive Geometry Page) from the
4913  *              target, and use this information to initialize the physical
4914  *              geometry cache specified by pgeom_p.
4915  *
4916  *              MODE SENSE is an optional command, so failure in this case
4917  *              does not necessarily denote an error. We want to use the
4918  *              MODE SENSE commands to derive the physical geometry of the
4919  *              device, but if either command fails, the logical geometry is
4920  *              used as the fallback for disk label geometry in cmlb.
4921  *
4922  *              This requires that un->un_blockcount and un->un_tgt_blocksize
4923  *              have already been initialized for the current target and
4924  *              that the current values be passed as args so that we don't
4925  *              end up ever trying to use -1 as a valid value. This could
4926  *              happen if either value is reset while we're not holding
4927  *              the mutex.
4928  *
4929  *   Arguments: un - driver soft state (unit) structure
4930  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
4931  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
4932  *                      to use the USCSI "direct" chain and bypass the normal
4933  *                      command waitq.
4934  *
4935  *     Context: Kernel thread only (can sleep).
4936  */
4937 
4938 static int
4939 sd_get_physical_geometry(struct sd_lun *un, cmlb_geom_t *pgeom_p,
4940         diskaddr_t capacity, int lbasize, int path_flag)
4941 {
4942         struct  mode_format     *page3p;
4943         struct  mode_geometry   *page4p;
4944         struct  mode_header     *headerp;
4945         int     sector_size;
4946         int     nsect;
4947         int     nhead;
4948         int     ncyl;
4949         int     intrlv;
4950         int     spc;
4951         diskaddr_t      modesense_capacity;
4952         int     rpm;
4953         int     bd_len;
4954         int     mode_header_length;
4955         uchar_t *p3bufp;
4956         uchar_t *p4bufp;
4957         int     cdbsize;
4958         int     ret = EIO;
4959         sd_ssc_t *ssc;
4960         int     status;
4961 
4962         ASSERT(un != NULL);
4963 
4964         if (lbasize == 0) {
4965                 if (ISCD(un)) {
4966                         lbasize = 2048;
4967                 } else {
4968                         lbasize = un->un_sys_blocksize;
4969                 }
4970         }
4971         pgeom_p->g_secsize = (unsigned short)lbasize;
4972 
4973         /*
4974          * If the unit is a cd/dvd drive MODE SENSE page three
4975          * and MODE SENSE page four are reserved (see SBC spec
4976          * and MMC spec). To prevent soft errors just return
4977          * using the default LBA size.
4978          */
4979         if (ISCD(un))
4980                 return (ret);
4981 
4982         cdbsize = (un->un_f_cfg_is_atapi == TRUE) ? CDB_GROUP2 : CDB_GROUP0;
4983 
4984         /*
4985          * Retrieve MODE SENSE page 3 - Format Device Page
4986          */
4987         p3bufp = kmem_zalloc(SD_MODE_SENSE_PAGE3_LENGTH, KM_SLEEP);
4988         ssc = sd_ssc_init(un);
4989         status = sd_send_scsi_MODE_SENSE(ssc, cdbsize, p3bufp,
4990             SD_MODE_SENSE_PAGE3_LENGTH, SD_MODE_SENSE_PAGE3_CODE, path_flag);
4991         if (status != 0) {
4992                 SD_ERROR(SD_LOG_COMMON, un,
4993                     "sd_get_physical_geometry: mode sense page 3 failed\n");
4994                 goto page3_exit;
4995         }
4996 
4997         /*
4998          * Determine size of Block Descriptors in order to locate the mode
4999          * page data.  ATAPI devices return 0, SCSI devices should return
5000          * MODE_BLK_DESC_LENGTH.
5001          */
5002         headerp = (struct mode_header *)p3bufp;
5003         if (un->un_f_cfg_is_atapi == TRUE) {
5004                 struct mode_header_grp2 *mhp =
5005                     (struct mode_header_grp2 *)headerp;
5006                 mode_header_length = MODE_HEADER_LENGTH_GRP2;
5007                 bd_len = (mhp->bdesc_length_hi << 8) | mhp->bdesc_length_lo;
5008         } else {
5009                 mode_header_length = MODE_HEADER_LENGTH;
5010                 bd_len = ((struct mode_header *)headerp)->bdesc_length;
5011         }
5012 
5013         if (bd_len > MODE_BLK_DESC_LENGTH) {
5014                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, SD_LOG_COMMON,
5015                     "sd_get_physical_geometry: received unexpected bd_len "
5016                     "of %d, page3\n", bd_len);
5017                 status = EIO;
5018                 goto page3_exit;
5019         }
5020 
5021         page3p = (struct mode_format *)
5022             ((caddr_t)headerp + mode_header_length + bd_len);
5023 
5024         if (page3p->mode_page.code != SD_MODE_SENSE_PAGE3_CODE) {
5025                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, SD_LOG_COMMON,
5026                     "sd_get_physical_geometry: mode sense pg3 code mismatch "
5027                     "%d\n", page3p->mode_page.code);
5028                 status = EIO;
5029                 goto page3_exit;
5030         }
5031 
5032         /*
5033          * Use this physical geometry data only if BOTH MODE SENSE commands
5034          * complete successfully; otherwise, revert to the logical geometry.
5035          * So, we need to save everything in temporary variables.
5036          */
5037         sector_size = BE_16(page3p->data_bytes_sect);
5038 
5039         /*
5040          * 1243403: The NEC D38x7 drives do not support MODE SENSE sector size
5041          */
5042         if (sector_size == 0) {
5043                 sector_size = un->un_sys_blocksize;
5044         } else {
5045                 sector_size &= ~(un->un_sys_blocksize - 1);
5046         }
5047 
5048         nsect  = BE_16(page3p->sect_track);
5049         intrlv = BE_16(page3p->interleave);
5050 
5051         SD_INFO(SD_LOG_COMMON, un,
5052             "sd_get_physical_geometry: Format Parameters (page 3)\n");
5053         SD_INFO(SD_LOG_COMMON, un,
5054             "   mode page: %d; nsect: %d; sector size: %d;\n",
5055             page3p->mode_page.code, nsect, sector_size);
5056         SD_INFO(SD_LOG_COMMON, un,
5057             "   interleave: %d; track skew: %d; cylinder skew: %d;\n", intrlv,
5058             BE_16(page3p->track_skew),
5059             BE_16(page3p->cylinder_skew));
5060 
5061         sd_ssc_assessment(ssc, SD_FMT_STANDARD);
5062 
5063         /*
5064          * Retrieve MODE SENSE page 4 - Rigid Disk Drive Geometry Page
5065          */
5066         p4bufp = kmem_zalloc(SD_MODE_SENSE_PAGE4_LENGTH, KM_SLEEP);
5067         status = sd_send_scsi_MODE_SENSE(ssc, cdbsize, p4bufp,
5068             SD_MODE_SENSE_PAGE4_LENGTH, SD_MODE_SENSE_PAGE4_CODE, path_flag);
5069         if (status != 0) {
5070                 SD_ERROR(SD_LOG_COMMON, un,
5071                     "sd_get_physical_geometry: mode sense page 4 failed\n");
5072                 goto page4_exit;
5073         }
5074 
5075         /*
5076          * Determine size of Block Descriptors in order to locate the mode
5077          * page data.  ATAPI devices return 0, SCSI devices should return
5078          * MODE_BLK_DESC_LENGTH.
5079          */
5080         headerp = (struct mode_header *)p4bufp;
5081         if (un->un_f_cfg_is_atapi == TRUE) {
5082                 struct mode_header_grp2 *mhp =
5083                     (struct mode_header_grp2 *)headerp;
5084                 bd_len = (mhp->bdesc_length_hi << 8) | mhp->bdesc_length_lo;
5085         } else {
5086                 bd_len = ((struct mode_header *)headerp)->bdesc_length;
5087         }
5088 
5089         if (bd_len > MODE_BLK_DESC_LENGTH) {
5090                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, SD_LOG_COMMON,
5091                     "sd_get_physical_geometry: received unexpected bd_len of "
5092                     "%d, page4\n", bd_len);
5093                 status = EIO;
5094                 goto page4_exit;
5095         }
5096 
5097         page4p = (struct mode_geometry *)
5098             ((caddr_t)headerp + mode_header_length + bd_len);
5099 
5100         if (page4p->mode_page.code != SD_MODE_SENSE_PAGE4_CODE) {
5101                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, SD_LOG_COMMON,
5102                     "sd_get_physical_geometry: mode sense pg4 code mismatch "
5103                     "%d\n", page4p->mode_page.code);
5104                 status = EIO;
5105                 goto page4_exit;
5106         }
5107 
5108         /*
5109          * Stash the data now, after we know that both commands completed.
5110          */
5111 
5112 
5113         nhead = (int)page4p->heads;  /* uchar, so no conversion needed */
5114         spc   = nhead * nsect;
5115         ncyl  = (page4p->cyl_ub << 16) + (page4p->cyl_mb << 8) + page4p->cyl_lb;
5116         rpm   = BE_16(page4p->rpm);
5117 
5118         modesense_capacity = spc * ncyl;
5119 
5120         SD_INFO(SD_LOG_COMMON, un,
5121             "sd_get_physical_geometry: Geometry Parameters (page 4)\n");
5122         SD_INFO(SD_LOG_COMMON, un,
5123             "   cylinders: %d; heads: %d; rpm: %d;\n", ncyl, nhead, rpm);
5124         SD_INFO(SD_LOG_COMMON, un,
5125             "   computed capacity(h*s*c): %d;\n", modesense_capacity);
5126         SD_INFO(SD_LOG_COMMON, un, "   pgeom_p: %p; read cap: %d\n",
5127             (void *)pgeom_p, capacity);
5128 
5129         /*
5130          * Compensate if the drive's geometry is not rectangular, i.e.,
5131          * the product of C * H * S returned by MODE SENSE >= that returned
5132          * by read capacity. This is an idiosyncrasy of the original x86
5133          * disk subsystem.
5134          */
5135         if (modesense_capacity >= capacity) {
5136                 SD_INFO(SD_LOG_COMMON, un,
5137                     "sd_get_physical_geometry: adjusting acyl; "
5138                     "old: %d; new: %d\n", pgeom_p->g_acyl,
5139                     (modesense_capacity - capacity + spc - 1) / spc);
5140                 if (sector_size != 0) {
5141                         /* 1243403: NEC D38x7 drives don't support sec size */
5142                         pgeom_p->g_secsize = (unsigned short)sector_size;
5143                 }
5144                 pgeom_p->g_nsect    = (unsigned short)nsect;
5145                 pgeom_p->g_nhead    = (unsigned short)nhead;
5146                 pgeom_p->g_capacity = capacity;
5147                 pgeom_p->g_acyl          =
5148                     (modesense_capacity - pgeom_p->g_capacity + spc - 1) / spc;
5149                 pgeom_p->g_ncyl          = ncyl - pgeom_p->g_acyl;
5150         }
5151 
5152         pgeom_p->g_rpm    = (unsigned short)rpm;
5153         pgeom_p->g_intrlv = (unsigned short)intrlv;
5154         ret = 0;
5155 
5156         SD_INFO(SD_LOG_COMMON, un,
5157             "sd_get_physical_geometry: mode sense geometry:\n");
5158         SD_INFO(SD_LOG_COMMON, un,
5159             "   nsect: %d; sector size: %d; interlv: %d\n",
5160             nsect, sector_size, intrlv);
5161         SD_INFO(SD_LOG_COMMON, un,
5162             "   nhead: %d; ncyl: %d; rpm: %d; capacity(ms): %d\n",
5163             nhead, ncyl, rpm, modesense_capacity);
5164         SD_INFO(SD_LOG_COMMON, un,
5165             "sd_get_physical_geometry: (cached)\n");
5166         SD_INFO(SD_LOG_COMMON, un,
5167             "   ncyl: %ld; acyl: %d; nhead: %d; nsect: %d\n",
5168             pgeom_p->g_ncyl,  pgeom_p->g_acyl,
5169             pgeom_p->g_nhead, pgeom_p->g_nsect);
5170         SD_INFO(SD_LOG_COMMON, un,
5171             "   lbasize: %d; capacity: %ld; intrlv: %d; rpm: %d\n",
5172             pgeom_p->g_secsize, pgeom_p->g_capacity,
5173             pgeom_p->g_intrlv, pgeom_p->g_rpm);
5174         sd_ssc_assessment(ssc, SD_FMT_STANDARD);
5175 
5176 page4_exit:
5177         kmem_free(p4bufp, SD_MODE_SENSE_PAGE4_LENGTH);
5178 
5179 page3_exit:
5180         kmem_free(p3bufp, SD_MODE_SENSE_PAGE3_LENGTH);
5181 
5182         if (status != 0) {
5183                 if (status == EIO) {
5184                         /*
5185                          * Some disks do not support mode sense(6), we
5186                          * should ignore this kind of error(sense key is
5187                          * 0x5 - illegal request).
5188                          */
5189                         uint8_t *sensep;
5190                         int senlen;
5191 
5192                         sensep = (uint8_t *)ssc->ssc_uscsi_cmd->uscsi_rqbuf;
5193                         senlen = (int)(ssc->ssc_uscsi_cmd->uscsi_rqlen -
5194                             ssc->ssc_uscsi_cmd->uscsi_rqresid);
5195 
5196                         if (senlen > 0 &&
5197                             scsi_sense_key(sensep) == KEY_ILLEGAL_REQUEST) {
5198                                 sd_ssc_assessment(ssc,
5199                                     SD_FMT_IGNORE_COMPROMISE);
5200                         } else {
5201                                 sd_ssc_assessment(ssc, SD_FMT_STATUS_CHECK);
5202                         }
5203                 } else {
5204                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
5205                 }
5206         }
5207         sd_ssc_fini(ssc);
5208         return (ret);
5209 }
5210 
5211 /*
5212  *    Function: sd_get_virtual_geometry
5213  *
5214  * Description: Ask the controller to tell us about the target device.
5215  *
5216  *   Arguments: un - pointer to softstate
5217  *              capacity - disk capacity in #blocks
5218  *              lbasize - disk block size in bytes
5219  *
5220  *     Context: Kernel thread only
5221  */
5222 
5223 static int
5224 sd_get_virtual_geometry(struct sd_lun *un, cmlb_geom_t *lgeom_p,
5225     diskaddr_t capacity, int lbasize)
5226 {
5227         uint_t  geombuf;
5228         int     spc;
5229 
5230         ASSERT(un != NULL);
5231 
5232         /* Set sector size, and total number of sectors */
5233         (void) scsi_ifsetcap(SD_ADDRESS(un), "sector-size",   lbasize,  1);
5234         (void) scsi_ifsetcap(SD_ADDRESS(un), "total-sectors", capacity, 1);
5235 
5236         /* Let the HBA tell us its geometry */
5237         geombuf = (uint_t)scsi_ifgetcap(SD_ADDRESS(un), "geometry", 1);
5238 
5239         /* A value of -1 indicates an undefined "geometry" property */
5240         if (geombuf == (-1)) {
5241                 return (EINVAL);
5242         }
5243 
5244         /* Initialize the logical geometry cache. */
5245         lgeom_p->g_nhead   = (geombuf >> 16) & 0xffff;
5246         lgeom_p->g_nsect   = geombuf & 0xffff;
5247         lgeom_p->g_secsize = un->un_sys_blocksize;
5248 
5249         spc = lgeom_p->g_nhead * lgeom_p->g_nsect;
5250 
5251         /*
5252          * Note: The driver originally converted the capacity value from
5253          * target blocks to system blocks. However, the capacity value passed
5254          * to this routine is already in terms of system blocks (this scaling
5255          * is done when the READ CAPACITY command is issued and processed).
5256          * This 'error' may have gone undetected because the usage of g_ncyl
5257          * (which is based upon g_capacity) is very limited within the driver
5258          */
5259         lgeom_p->g_capacity = capacity;
5260 
5261         /*
5262          * Set ncyl to zero if the hba returned a zero nhead or nsect value. The
5263          * hba may return zero values if the device has been removed.
5264          */
5265         if (spc == 0) {
5266                 lgeom_p->g_ncyl = 0;
5267         } else {
5268                 lgeom_p->g_ncyl = lgeom_p->g_capacity / spc;
5269         }
5270         lgeom_p->g_acyl = 0;
5271 
5272         SD_INFO(SD_LOG_COMMON, un, "sd_get_virtual_geometry: (cached)\n");
5273         return (0);
5274 
5275 }
5276 /*
5277  *    Function: sd_update_block_info
5278  *
5279  * Description: Calculate a byte count to sector count bitshift value
5280  *              from sector size.
5281  *
5282  *   Arguments: un: unit struct.
5283  *              lbasize: new target sector size
5284  *              capacity: new target capacity, ie. block count
5285  *
5286  *     Context: Kernel thread context
5287  */
5288 
5289 static void
5290 sd_update_block_info(struct sd_lun *un, uint32_t lbasize, uint64_t capacity)
5291 {
5292         if (lbasize != 0) {
5293                 un->un_tgt_blocksize = lbasize;
5294                 un->un_f_tgt_blocksize_is_valid = TRUE;
5295                 if (!un->un_f_has_removable_media) {
5296                         un->un_sys_blocksize = lbasize;
5297                 }
5298         }
5299 
5300         if (capacity != 0) {
5301                 un->un_blockcount            = capacity;
5302                 un->un_f_blockcount_is_valid = TRUE;
5303 
5304                 /*
5305                  * The capacity has changed so update the errstats.
5306                  */
5307                 if (un->un_errstats != NULL) {
5308                         struct sd_errstats *stp;
5309 
5310                         capacity *= un->un_sys_blocksize;
5311                         stp = (struct sd_errstats *)un->un_errstats->ks_data;
5312                         if (stp->sd_capacity.value.ui64 < capacity)
5313                                 stp->sd_capacity.value.ui64 = capacity;
5314                 }
5315         }
5316 }
5317 
5318 
5319 /*
5320  *    Function: sd_register_devid
5321  *
5322  * Description: This routine will obtain the device id information from the
5323  *              target, obtain the serial number, and register the device
5324  *              id with the ddi framework.
5325  *
5326  *   Arguments: devi - the system's dev_info_t for the device.
5327  *              un - driver soft state (unit) structure
5328  *              reservation_flag - indicates if a reservation conflict
5329  *              occurred during attach
5330  *
5331  *     Context: Kernel Thread
5332  */
5333 static void
5334 sd_register_devid(sd_ssc_t *ssc, dev_info_t *devi, int reservation_flag)
5335 {
5336         int             rval            = 0;
5337         uchar_t         *inq80          = NULL;
5338         size_t          inq80_len       = MAX_INQUIRY_SIZE;
5339         size_t          inq80_resid     = 0;
5340         uchar_t         *inq83          = NULL;
5341         size_t          inq83_len       = MAX_INQUIRY_SIZE;
5342         size_t          inq83_resid     = 0;
5343         int             dlen, len;
5344         char            *sn;
5345         struct sd_lun   *un;
5346 
5347         ASSERT(ssc != NULL);
5348         un = ssc->ssc_un;
5349         ASSERT(un != NULL);
5350         ASSERT(mutex_owned(SD_MUTEX(un)));
5351         ASSERT((SD_DEVINFO(un)) == devi);
5352 
5353 
5354         /*
5355          * We check the availability of the World Wide Name (0x83) and Unit
5356          * Serial Number (0x80) pages in sd_check_vpd_page_support(), and using
5357          * un_vpd_page_mask from them, we decide which way to get the WWN.  If
5358          * 0x83 is available, that is the best choice.  Our next choice is
5359          * 0x80.  If neither are available, we munge the devid from the device
5360          * vid/pid/serial # for Sun qualified disks, or use the ddi framework
5361          * to fabricate a devid for non-Sun qualified disks.
5362          */
5363         if (sd_check_vpd_page_support(ssc) == 0) {
5364                 /* collect page 80 data if available */
5365                 if (un->un_vpd_page_mask & SD_VPD_UNIT_SERIAL_PG) {
5366 
5367                         mutex_exit(SD_MUTEX(un));
5368                         inq80 = kmem_zalloc(inq80_len, KM_SLEEP);
5369 
5370                         rval = sd_send_scsi_INQUIRY(ssc, inq80, inq80_len,
5371                             0x01, 0x80, &inq80_resid);
5372 
5373                         if (rval != 0) {
5374                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
5375                                 kmem_free(inq80, inq80_len);
5376                                 inq80 = NULL;
5377                                 inq80_len = 0;
5378                         } else if (ddi_prop_exists(
5379                             DDI_DEV_T_NONE, SD_DEVINFO(un),
5380                             DDI_PROP_NOTPROM | DDI_PROP_DONTPASS,
5381                             INQUIRY_SERIAL_NO) == 0) {
5382                                 /*
5383                                  * If we don't already have a serial number
5384                                  * property, do quick verify of data returned
5385                                  * and define property.
5386                                  */
5387                                 dlen = inq80_len - inq80_resid;
5388                                 len = (size_t)inq80[3];
5389                                 if ((dlen >= 4) && ((len + 4) <= dlen)) {
5390                                         /*
5391                                          * Ensure sn termination, skip leading
5392                                          * blanks, and create property
5393                                          * 'inquiry-serial-no'.
5394                                          */
5395                                         sn = (char *)&inq80[4];
5396                                         sn[len] = 0;
5397                                         while (*sn && (*sn == ' '))
5398                                                 sn++;
5399                                         if (*sn) {
5400                                                 (void) ddi_prop_update_string(
5401                                                     DDI_DEV_T_NONE,
5402                                                     SD_DEVINFO(un),
5403                                                     INQUIRY_SERIAL_NO, sn);
5404                                         }
5405                                 }
5406                         }
5407                         mutex_enter(SD_MUTEX(un));
5408                 }
5409 
5410                 /* collect page 83 data if available */
5411                 if (un->un_vpd_page_mask & SD_VPD_DEVID_WWN_PG) {
5412                         mutex_exit(SD_MUTEX(un));
5413                         inq83 = kmem_zalloc(inq83_len, KM_SLEEP);
5414 
5415                         rval = sd_send_scsi_INQUIRY(ssc, inq83, inq83_len,
5416                             0x01, 0x83, &inq83_resid);
5417 
5418                         if (rval != 0) {
5419                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
5420                                 kmem_free(inq83, inq83_len);
5421                                 inq83 = NULL;
5422                                 inq83_len = 0;
5423                         }
5424                         mutex_enter(SD_MUTEX(un));
5425                 }
5426         }
5427 
5428         /*
5429          * If transport has already registered a devid for this target
5430          * then that takes precedence over the driver's determination
5431          * of the devid.
5432          *
5433          * NOTE: The reason this check is done here instead of at the beginning
5434          * of the function is to allow the code above to create the
5435          * 'inquiry-serial-no' property.
5436          */
5437         if (ddi_devid_get(SD_DEVINFO(un), &un->un_devid) == DDI_SUCCESS) {
5438                 ASSERT(un->un_devid);
5439                 un->un_f_devid_transport_defined = TRUE;
5440                 goto cleanup; /* use devid registered by the transport */
5441         }
5442 
5443         /*
5444          * This is the case of antiquated Sun disk drives that have the
5445          * FAB_DEVID property set in the disk_table.  These drives
5446          * manage the devid's by storing them in last 2 available sectors
5447          * on the drive and have them fabricated by the ddi layer by calling
5448          * ddi_devid_init and passing the DEVID_FAB flag.
5449          */
5450         if (un->un_f_opt_fab_devid == TRUE) {
5451                 /*
5452                  * Depending on EINVAL isn't reliable, since a reserved disk
5453                  * may result in invalid geometry, so check to make sure a
5454                  * reservation conflict did not occur during attach.
5455                  */
5456                 if ((sd_get_devid(ssc) == EINVAL) &&
5457                     (reservation_flag != SD_TARGET_IS_RESERVED)) {
5458                         /*
5459                          * The devid is invalid AND there is no reservation
5460                          * conflict.  Fabricate a new devid.
5461                          */
5462                         (void) sd_create_devid(ssc);
5463                 }
5464 
5465                 /* Register the devid if it exists */
5466                 if (un->un_devid != NULL) {
5467                         (void) ddi_devid_register(SD_DEVINFO(un),
5468                             un->un_devid);
5469                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
5470                             "sd_register_devid: Devid Fabricated\n");
5471                 }
5472                 goto cleanup;
5473         }
5474 
5475         /* encode best devid possible based on data available */
5476         if (ddi_devid_scsi_encode(DEVID_SCSI_ENCODE_VERSION_LATEST,
5477             (char *)ddi_driver_name(SD_DEVINFO(un)),
5478             (uchar_t *)SD_INQUIRY(un), sizeof (*SD_INQUIRY(un)),
5479             inq80, inq80_len - inq80_resid, inq83, inq83_len -
5480             inq83_resid, &un->un_devid) == DDI_SUCCESS) {
5481 
5482                 /* devid successfully encoded, register devid */
5483                 (void) ddi_devid_register(SD_DEVINFO(un), un->un_devid);
5484 
5485         } else {
5486                 /*
5487                  * Unable to encode a devid based on data available.
5488                  * This is not a Sun qualified disk.  Older Sun disk
5489                  * drives that have the SD_FAB_DEVID property
5490                  * set in the disk_table and non Sun qualified
5491                  * disks are treated in the same manner.  These
5492                  * drives manage the devid's by storing them in
5493                  * last 2 available sectors on the drive and
5494                  * have them fabricated by the ddi layer by
5495                  * calling ddi_devid_init and passing the
5496                  * DEVID_FAB flag.
5497                  * Create a fabricate devid only if there's no
5498                  * fabricate devid existed.
5499                  */
5500                 if (sd_get_devid(ssc) == EINVAL) {
5501                         (void) sd_create_devid(ssc);
5502                 }
5503                 un->un_f_opt_fab_devid = TRUE;
5504 
5505                 /* Register the devid if it exists */
5506                 if (un->un_devid != NULL) {
5507                         (void) ddi_devid_register(SD_DEVINFO(un),
5508                             un->un_devid);
5509                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
5510                             "sd_register_devid: devid fabricated using "
5511                             "ddi framework\n");
5512                 }
5513         }
5514 
5515 cleanup:
5516         /* clean up resources */
5517         if (inq80 != NULL) {
5518                 kmem_free(inq80, inq80_len);
5519         }
5520         if (inq83 != NULL) {
5521                 kmem_free(inq83, inq83_len);
5522         }
5523 }
5524 
5525 
5526 
5527 /*
5528  *    Function: sd_get_devid
5529  *
5530  * Description: This routine will return 0 if a valid device id has been
5531  *              obtained from the target and stored in the soft state. If a
5532  *              valid device id has not been previously read and stored, a
5533  *              read attempt will be made.
5534  *
5535  *   Arguments: un - driver soft state (unit) structure
5536  *
5537  * Return Code: 0 if we successfully get the device id
5538  *
5539  *     Context: Kernel Thread
5540  */
5541 
5542 static int
5543 sd_get_devid(sd_ssc_t *ssc)
5544 {
5545         struct dk_devid         *dkdevid;
5546         ddi_devid_t             tmpid;
5547         uint_t                  *ip;
5548         size_t                  sz;
5549         diskaddr_t              blk;
5550         int                     status;
5551         int                     chksum;
5552         int                     i;
5553         size_t                  buffer_size;
5554         struct sd_lun           *un;
5555 
5556         ASSERT(ssc != NULL);
5557         un = ssc->ssc_un;
5558         ASSERT(un != NULL);
5559         ASSERT(mutex_owned(SD_MUTEX(un)));
5560 
5561         SD_TRACE(SD_LOG_ATTACH_DETACH, un, "sd_get_devid: entry: un: 0x%p\n",
5562             un);
5563 
5564         if (un->un_devid != NULL) {
5565                 return (0);
5566         }
5567 
5568         mutex_exit(SD_MUTEX(un));
5569         if (cmlb_get_devid_block(un->un_cmlbhandle, &blk,
5570             (void *)SD_PATH_DIRECT) != 0) {
5571                 mutex_enter(SD_MUTEX(un));
5572                 return (EINVAL);
5573         }
5574 
5575         /*
5576          * Read and verify device id, stored in the reserved cylinders at the
5577          * end of the disk. Backup label is on the odd sectors of the last
5578          * track of the last cylinder. Device id will be on track of the next
5579          * to last cylinder.
5580          */
5581         mutex_enter(SD_MUTEX(un));
5582         buffer_size = SD_REQBYTES2TGTBYTES(un, sizeof (struct dk_devid));
5583         mutex_exit(SD_MUTEX(un));
5584         dkdevid = kmem_alloc(buffer_size, KM_SLEEP);
5585         status = sd_send_scsi_READ(ssc, dkdevid, buffer_size, blk,
5586             SD_PATH_DIRECT);
5587 
5588         if (status != 0) {
5589                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
5590                 goto error;
5591         }
5592 
5593         /* Validate the revision */
5594         if ((dkdevid->dkd_rev_hi != DK_DEVID_REV_MSB) ||
5595             (dkdevid->dkd_rev_lo != DK_DEVID_REV_LSB)) {
5596                 status = EINVAL;
5597                 goto error;
5598         }
5599 
5600         /* Calculate the checksum */
5601         chksum = 0;
5602         ip = (uint_t *)dkdevid;
5603         for (i = 0; i < ((DEV_BSIZE - sizeof (int)) / sizeof (int));
5604             i++) {
5605                 chksum ^= ip[i];
5606         }
5607 
5608         /* Compare the checksums */
5609         if (DKD_GETCHKSUM(dkdevid) != chksum) {
5610                 status = EINVAL;
5611                 goto error;
5612         }
5613 
5614         /* Validate the device id */
5615         if (ddi_devid_valid((ddi_devid_t)&dkdevid->dkd_devid) != DDI_SUCCESS) {
5616                 status = EINVAL;
5617                 goto error;
5618         }
5619 
5620         /*
5621          * Store the device id in the driver soft state
5622          */
5623         sz = ddi_devid_sizeof((ddi_devid_t)&dkdevid->dkd_devid);
5624         tmpid = kmem_alloc(sz, KM_SLEEP);
5625 
5626         mutex_enter(SD_MUTEX(un));
5627 
5628         un->un_devid = tmpid;
5629         bcopy(&dkdevid->dkd_devid, un->un_devid, sz);
5630 
5631         kmem_free(dkdevid, buffer_size);
5632 
5633         SD_TRACE(SD_LOG_ATTACH_DETACH, un, "sd_get_devid: exit: un:0x%p\n", un);
5634 
5635         return (status);
5636 error:
5637         mutex_enter(SD_MUTEX(un));
5638         kmem_free(dkdevid, buffer_size);
5639         return (status);
5640 }
5641 
5642 
5643 /*
5644  *    Function: sd_create_devid
5645  *
5646  * Description: This routine will fabricate the device id and write it
5647  *              to the disk.
5648  *
5649  *   Arguments: un - driver soft state (unit) structure
5650  *
5651  * Return Code: value of the fabricated device id
5652  *
5653  *     Context: Kernel Thread
5654  */
5655 
5656 static ddi_devid_t
5657 sd_create_devid(sd_ssc_t *ssc)
5658 {
5659         struct sd_lun   *un;
5660 
5661         ASSERT(ssc != NULL);
5662         un = ssc->ssc_un;
5663         ASSERT(un != NULL);
5664 
5665         /* Fabricate the devid */
5666         if (ddi_devid_init(SD_DEVINFO(un), DEVID_FAB, 0, NULL, &un->un_devid)
5667             == DDI_FAILURE) {
5668                 return (NULL);
5669         }
5670 
5671         /* Write the devid to disk */
5672         if (sd_write_deviceid(ssc) != 0) {
5673                 ddi_devid_free(un->un_devid);
5674                 un->un_devid = NULL;
5675         }
5676 
5677         return (un->un_devid);
5678 }
5679 
5680 
5681 /*
5682  *    Function: sd_write_deviceid
5683  *
5684  * Description: This routine will write the device id to the disk
5685  *              reserved sector.
5686  *
5687  *   Arguments: un - driver soft state (unit) structure
5688  *
5689  * Return Code: EINVAL
5690  *              value returned by sd_send_scsi_cmd
5691  *
5692  *     Context: Kernel Thread
5693  */
5694 
5695 static int
5696 sd_write_deviceid(sd_ssc_t *ssc)
5697 {
5698         struct dk_devid         *dkdevid;
5699         uchar_t                 *buf;
5700         diskaddr_t              blk;
5701         uint_t                  *ip, chksum;
5702         int                     status;
5703         int                     i;
5704         struct sd_lun           *un;
5705 
5706         ASSERT(ssc != NULL);
5707         un = ssc->ssc_un;
5708         ASSERT(un != NULL);
5709         ASSERT(mutex_owned(SD_MUTEX(un)));
5710 
5711         mutex_exit(SD_MUTEX(un));
5712         if (cmlb_get_devid_block(un->un_cmlbhandle, &blk,
5713             (void *)SD_PATH_DIRECT) != 0) {
5714                 mutex_enter(SD_MUTEX(un));
5715                 return (-1);
5716         }
5717 
5718 
5719         /* Allocate the buffer */
5720         buf = kmem_zalloc(un->un_sys_blocksize, KM_SLEEP);
5721         dkdevid = (struct dk_devid *)buf;
5722 
5723         /* Fill in the revision */
5724         dkdevid->dkd_rev_hi = DK_DEVID_REV_MSB;
5725         dkdevid->dkd_rev_lo = DK_DEVID_REV_LSB;
5726 
5727         /* Copy in the device id */
5728         mutex_enter(SD_MUTEX(un));
5729         bcopy(un->un_devid, &dkdevid->dkd_devid,
5730             ddi_devid_sizeof(un->un_devid));
5731         mutex_exit(SD_MUTEX(un));
5732 
5733         /* Calculate the checksum */
5734         chksum = 0;
5735         ip = (uint_t *)dkdevid;
5736         for (i = 0; i < ((DEV_BSIZE - sizeof (int)) / sizeof (int));
5737             i++) {
5738                 chksum ^= ip[i];
5739         }
5740 
5741         /* Fill-in checksum */
5742         DKD_FORMCHKSUM(chksum, dkdevid);
5743 
5744         /* Write the reserved sector */
5745         status = sd_send_scsi_WRITE(ssc, buf, un->un_sys_blocksize, blk,
5746             SD_PATH_DIRECT);
5747         if (status != 0)
5748                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
5749 
5750         kmem_free(buf, un->un_sys_blocksize);
5751 
5752         mutex_enter(SD_MUTEX(un));
5753         return (status);
5754 }
5755 
5756 
5757 /*
5758  *    Function: sd_check_vpd_page_support
5759  *
5760  * Description: This routine sends an inquiry command with the EVPD bit set and
5761  *              a page code of 0x00 to the device. It is used to determine which
5762  *              vital product pages are available to find the devid. We are
5763  *              looking for pages 0x83 0x80 or 0xB1.  If we return a negative 1,
5764  *              the device does not support that command.
5765  *
5766  *   Arguments: un  - driver soft state (unit) structure
5767  *
5768  * Return Code: 0 - success
5769  *              1 - check condition
5770  *
5771  *     Context: This routine can sleep.
5772  */
5773 
5774 static int
5775 sd_check_vpd_page_support(sd_ssc_t *ssc)
5776 {
5777         uchar_t *page_list      = NULL;
5778         uchar_t page_length     = 0xff; /* Use max possible length */
5779         uchar_t evpd            = 0x01; /* Set the EVPD bit */
5780         uchar_t page_code       = 0x00; /* Supported VPD Pages */
5781         int     rval            = 0;
5782         int     counter;
5783         struct sd_lun           *un;
5784 
5785         ASSERT(ssc != NULL);
5786         un = ssc->ssc_un;
5787         ASSERT(un != NULL);
5788         ASSERT(mutex_owned(SD_MUTEX(un)));
5789 
5790         mutex_exit(SD_MUTEX(un));
5791 
5792         /*
5793          * We'll set the page length to the maximum to save figuring it out
5794          * with an additional call.
5795          */
5796         page_list =  kmem_zalloc(page_length, KM_SLEEP);
5797 
5798         rval = sd_send_scsi_INQUIRY(ssc, page_list, page_length, evpd,
5799             page_code, NULL);
5800 
5801         if (rval != 0)
5802                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
5803 
5804         mutex_enter(SD_MUTEX(un));
5805 
5806         /*
5807          * Now we must validate that the device accepted the command, as some
5808          * drives do not support it.  If the drive does support it, we will
5809          * return 0, and the supported pages will be in un_vpd_page_mask.  If
5810          * not, we return -1.
5811          */
5812         if ((rval == 0) && (page_list[VPD_MODE_PAGE] == 0x00)) {
5813                 /* Loop to find one of the 2 pages we need */
5814                 counter = 4;  /* Supported pages start at byte 4, with 0x00 */
5815 
5816                 /*
5817                  * Pages are returned in ascending order, and 0x83 is what we
5818                  * are hoping for.
5819                  */
5820                 while ((page_list[counter] <= 0xB1) &&
5821                     (counter <= (page_list[VPD_PAGE_LENGTH] +
5822                     VPD_HEAD_OFFSET))) {
5823                         /*
5824                          * Add 3 because page_list[3] is the number of
5825                          * pages minus 3
5826                          */
5827 
5828                         switch (page_list[counter]) {
5829                         case 0x00:
5830                                 un->un_vpd_page_mask |= SD_VPD_SUPPORTED_PG;
5831                                 break;
5832                         case 0x80:
5833                                 un->un_vpd_page_mask |= SD_VPD_UNIT_SERIAL_PG;
5834                                 break;
5835                         case 0x81:
5836                                 un->un_vpd_page_mask |= SD_VPD_OPERATING_PG;
5837                                 break;
5838                         case 0x82:
5839                                 un->un_vpd_page_mask |= SD_VPD_ASCII_OP_PG;
5840                                 break;
5841                         case 0x83:
5842                                 un->un_vpd_page_mask |= SD_VPD_DEVID_WWN_PG;
5843                                 break;
5844                         case 0x86:
5845                                 un->un_vpd_page_mask |= SD_VPD_EXTENDED_DATA_PG;
5846                                 break;
5847                         case 0xB1:
5848                                 un->un_vpd_page_mask |= SD_VPD_DEV_CHARACTER_PG;
5849                                 break;
5850                         }
5851                         counter++;
5852                 }
5853 
5854         } else {
5855                 rval = -1;
5856 
5857                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
5858                     "sd_check_vpd_page_support: This drive does not implement "
5859                     "VPD pages.\n");
5860         }
5861 
5862         kmem_free(page_list, page_length);
5863 
5864         return (rval);
5865 }
5866 
5867 
5868 /*
5869  *    Function: sd_setup_pm
5870  *
5871  * Description: Initialize Power Management on the device
5872  *
5873  *     Context: Kernel Thread
5874  */
5875 
5876 static void
5877 sd_setup_pm(sd_ssc_t *ssc, dev_info_t *devi)
5878 {
5879         uint_t          log_page_size;
5880         uchar_t         *log_page_data;
5881         int             rval = 0;
5882         struct sd_lun   *un;
5883 
5884         ASSERT(ssc != NULL);
5885         un = ssc->ssc_un;
5886         ASSERT(un != NULL);
5887 
5888         /*
5889          * Since we are called from attach, holding a mutex for
5890          * un is unnecessary. Because some of the routines called
5891          * from here require SD_MUTEX to not be held, assert this
5892          * right up front.
5893          */
5894         ASSERT(!mutex_owned(SD_MUTEX(un)));
5895         /*
5896          * Since the sd device does not have the 'reg' property,
5897          * cpr will not call its DDI_SUSPEND/DDI_RESUME entries.
5898          * The following code is to tell cpr that this device
5899          * DOES need to be suspended and resumed.
5900          */
5901         (void) ddi_prop_update_string(DDI_DEV_T_NONE, devi,
5902             "pm-hardware-state", "needs-suspend-resume");
5903 
5904         /*
5905          * This complies with the new power management framework
5906          * for certain desktop machines. Create the pm_components
5907          * property as a string array property.
5908          * If un_f_pm_supported is TRUE, that means the disk
5909          * attached HBA has set the "pm-capable" property and
5910          * the value of this property is bigger than 0.
5911          */
5912         if (un->un_f_pm_supported) {
5913                 /*
5914                  * not all devices have a motor, try it first.
5915                  * some devices may return ILLEGAL REQUEST, some
5916                  * will hang
5917                  * The following START_STOP_UNIT is used to check if target
5918                  * device has a motor.
5919                  */
5920                 un->un_f_start_stop_supported = TRUE;
5921 
5922                 if (un->un_f_power_condition_supported) {
5923                         rval = sd_send_scsi_START_STOP_UNIT(ssc,
5924                             SD_POWER_CONDITION, SD_TARGET_ACTIVE,
5925                             SD_PATH_DIRECT);
5926                         if (rval != 0) {
5927                                 un->un_f_power_condition_supported = FALSE;
5928                         }
5929                 }
5930                 if (!un->un_f_power_condition_supported) {
5931                         rval = sd_send_scsi_START_STOP_UNIT(ssc,
5932                             SD_START_STOP, SD_TARGET_START, SD_PATH_DIRECT);
5933                 }
5934                 if (rval != 0) {
5935                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
5936                         un->un_f_start_stop_supported = FALSE;
5937                 }
5938 
5939                 /*
5940                  * create pm properties anyways otherwise the parent can't
5941                  * go to sleep
5942                  */
5943                 un->un_f_pm_is_enabled = TRUE;
5944                 (void) sd_create_pm_components(devi, un);
5945 
5946                 /*
5947                  * If it claims that log sense is supported, check it out.
5948                  */
5949                 if (un->un_f_log_sense_supported) {
5950                         rval = sd_log_page_supported(ssc,
5951                             START_STOP_CYCLE_PAGE);
5952                         if (rval == 1) {
5953                                 /* Page found, use it. */
5954                                 un->un_start_stop_cycle_page =
5955                                     START_STOP_CYCLE_PAGE;
5956                         } else {
5957                                 /*
5958                                  * Page not found or log sense is not
5959                                  * supported.
5960                                  * Notice we do not check the old style
5961                                  * START_STOP_CYCLE_VU_PAGE because this
5962                                  * code path does not apply to old disks.
5963                                  */
5964                                 un->un_f_log_sense_supported = FALSE;
5965                                 un->un_f_pm_log_sense_smart = FALSE;
5966                         }
5967                 }
5968 
5969                 return;
5970         }
5971 
5972         /*
5973          * For the disk whose attached HBA has not set the "pm-capable"
5974          * property, check if it supports the power management.
5975          */
5976         if (!un->un_f_log_sense_supported) {
5977                 un->un_power_level = SD_SPINDLE_ON;
5978                 un->un_f_pm_is_enabled = FALSE;
5979                 return;
5980         }
5981 
5982         rval = sd_log_page_supported(ssc, START_STOP_CYCLE_PAGE);
5983 
5984 #ifdef  SDDEBUG
5985         if (sd_force_pm_supported) {
5986                 /* Force a successful result */
5987                 rval = 1;
5988         }
5989 #endif
5990 
5991         /*
5992          * If the start-stop cycle counter log page is not supported
5993          * or if the pm-capable property is set to be false (0),
5994          * then we should not create the pm_components property.
5995          */
5996         if (rval == -1) {
5997                 /*
5998                  * Error.
5999                  * Reading log sense failed, most likely this is
6000                  * an older drive that does not support log sense.
6001                  * If this fails auto-pm is not supported.
6002                  */
6003                 un->un_power_level = SD_SPINDLE_ON;
6004                 un->un_f_pm_is_enabled = FALSE;
6005 
6006         } else if (rval == 0) {
6007                 /*
6008                  * Page not found.
6009                  * The start stop cycle counter is implemented as page
6010                  * START_STOP_CYCLE_PAGE_VU_PAGE (0x31) in older disks. For
6011                  * newer disks it is implemented as START_STOP_CYCLE_PAGE (0xE).
6012                  */
6013                 if (sd_log_page_supported(ssc, START_STOP_CYCLE_VU_PAGE) == 1) {
6014                         /*
6015                          * Page found, use this one.
6016                          */
6017                         un->un_start_stop_cycle_page = START_STOP_CYCLE_VU_PAGE;
6018                         un->un_f_pm_is_enabled = TRUE;
6019                 } else {
6020                         /*
6021                          * Error or page not found.
6022                          * auto-pm is not supported for this device.
6023                          */
6024                         un->un_power_level = SD_SPINDLE_ON;
6025                         un->un_f_pm_is_enabled = FALSE;
6026                 }
6027         } else {
6028                 /*
6029                  * Page found, use it.
6030                  */
6031                 un->un_start_stop_cycle_page = START_STOP_CYCLE_PAGE;
6032                 un->un_f_pm_is_enabled = TRUE;
6033         }
6034 
6035 
6036         if (un->un_f_pm_is_enabled == TRUE) {
6037                 log_page_size = START_STOP_CYCLE_COUNTER_PAGE_SIZE;
6038                 log_page_data = kmem_zalloc(log_page_size, KM_SLEEP);
6039 
6040                 rval = sd_send_scsi_LOG_SENSE(ssc, log_page_data,
6041                     log_page_size, un->un_start_stop_cycle_page,
6042                     0x01, 0, SD_PATH_DIRECT);
6043 
6044                 if (rval != 0) {
6045                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
6046                 }
6047 
6048 #ifdef  SDDEBUG
6049                 if (sd_force_pm_supported) {
6050                         /* Force a successful result */
6051                         rval = 0;
6052                 }
6053 #endif
6054 
6055                 /*
6056                  * If the Log sense for Page( Start/stop cycle counter page)
6057                  * succeeds, then power management is supported and we can
6058                  * enable auto-pm.
6059                  */
6060                 if (rval == 0)  {
6061                         (void) sd_create_pm_components(devi, un);
6062                 } else {
6063                         un->un_power_level = SD_SPINDLE_ON;
6064                         un->un_f_pm_is_enabled = FALSE;
6065                 }
6066 
6067                 kmem_free(log_page_data, log_page_size);
6068         }
6069 }
6070 
6071 
6072 /*
6073  *    Function: sd_create_pm_components
6074  *
6075  * Description: Initialize PM property.
6076  *
6077  *     Context: Kernel thread context
6078  */
6079 
6080 static void
6081 sd_create_pm_components(dev_info_t *devi, struct sd_lun *un)
6082 {
6083         ASSERT(!mutex_owned(SD_MUTEX(un)));
6084 
6085         if (un->un_f_power_condition_supported) {
6086                 if (ddi_prop_update_string_array(DDI_DEV_T_NONE, devi,
6087                     "pm-components", sd_pwr_pc.pm_comp, 5)
6088                     != DDI_PROP_SUCCESS) {
6089                         un->un_power_level = SD_SPINDLE_ACTIVE;
6090                         un->un_f_pm_is_enabled = FALSE;
6091                         return;
6092                 }
6093         } else {
6094                 if (ddi_prop_update_string_array(DDI_DEV_T_NONE, devi,
6095                     "pm-components", sd_pwr_ss.pm_comp, 3)
6096                     != DDI_PROP_SUCCESS) {
6097                         un->un_power_level = SD_SPINDLE_ON;
6098                         un->un_f_pm_is_enabled = FALSE;
6099                         return;
6100                 }
6101         }
6102         /*
6103          * When components are initially created they are idle,
6104          * power up any non-removables.
6105          * Note: the return value of pm_raise_power can't be used
6106          * for determining if PM should be enabled for this device.
6107          * Even if you check the return values and remove this
6108          * property created above, the PM framework will not honor the
6109          * change after the first call to pm_raise_power. Hence,
6110          * removal of that property does not help if pm_raise_power
6111          * fails. In the case of removable media, the start/stop
6112          * will fail if the media is not present.
6113          */
6114         if (un->un_f_attach_spinup && (pm_raise_power(SD_DEVINFO(un), 0,
6115             SD_PM_STATE_ACTIVE(un)) == DDI_SUCCESS)) {
6116                 mutex_enter(SD_MUTEX(un));
6117                 un->un_power_level = SD_PM_STATE_ACTIVE(un);
6118                 mutex_enter(&un->un_pm_mutex);
6119                 /* Set to on and not busy. */
6120                 un->un_pm_count = 0;
6121         } else {
6122                 mutex_enter(SD_MUTEX(un));
6123                 un->un_power_level = SD_PM_STATE_STOPPED(un);
6124                 mutex_enter(&un->un_pm_mutex);
6125                 /* Set to off. */
6126                 un->un_pm_count = -1;
6127         }
6128         mutex_exit(&un->un_pm_mutex);
6129         mutex_exit(SD_MUTEX(un));
6130 }
6131 
6132 
6133 /*
6134  *    Function: sd_ddi_suspend
6135  *
6136  * Description: Performs system power-down operations. This includes
6137  *              setting the drive state to indicate its suspended so
6138  *              that no new commands will be accepted. Also, wait for
6139  *              all commands that are in transport or queued to a timer
6140  *              for retry to complete. All timeout threads are cancelled.
6141  *
6142  * Return Code: DDI_FAILURE or DDI_SUCCESS
6143  *
6144  *     Context: Kernel thread context
6145  */
6146 
6147 static int
6148 sd_ddi_suspend(dev_info_t *devi)
6149 {
6150         struct  sd_lun  *un;
6151         clock_t         wait_cmds_complete;
6152 
6153         un = ddi_get_soft_state(sd_state, ddi_get_instance(devi));
6154         if (un == NULL) {
6155                 return (DDI_FAILURE);
6156         }
6157 
6158         SD_TRACE(SD_LOG_IO_PM, un, "sd_ddi_suspend: entry\n");
6159 
6160         mutex_enter(SD_MUTEX(un));
6161 
6162         /* Return success if the device is already suspended. */
6163         if (un->un_state == SD_STATE_SUSPENDED) {
6164                 mutex_exit(SD_MUTEX(un));
6165                 SD_TRACE(SD_LOG_IO_PM, un, "sd_ddi_suspend: "
6166                     "device already suspended, exiting\n");
6167                 return (DDI_SUCCESS);
6168         }
6169 
6170         /* Return failure if the device is being used by HA */
6171         if (un->un_resvd_status &
6172             (SD_RESERVE | SD_WANT_RESERVE | SD_LOST_RESERVE)) {
6173                 mutex_exit(SD_MUTEX(un));
6174                 SD_TRACE(SD_LOG_IO_PM, un, "sd_ddi_suspend: "
6175                     "device in use by HA, exiting\n");
6176                 return (DDI_FAILURE);
6177         }
6178 
6179         /*
6180          * Return failure if the device is in a resource wait
6181          * or power changing state.
6182          */
6183         if ((un->un_state == SD_STATE_RWAIT) ||
6184             (un->un_state == SD_STATE_PM_CHANGING)) {
6185                 mutex_exit(SD_MUTEX(un));
6186                 SD_TRACE(SD_LOG_IO_PM, un, "sd_ddi_suspend: "
6187                     "device in resource wait state, exiting\n");
6188                 return (DDI_FAILURE);
6189         }
6190 
6191 
6192         un->un_save_state = un->un_last_state;
6193         New_state(un, SD_STATE_SUSPENDED);
6194 
6195         /*
6196          * Wait for all commands that are in transport or queued to a timer
6197          * for retry to complete.
6198          *
6199          * While waiting, no new commands will be accepted or sent because of
6200          * the new state we set above.
6201          *
6202          * Wait till current operation has completed. If we are in the resource
6203          * wait state (with an intr outstanding) then we need to wait till the
6204          * intr completes and starts the next cmd. We want to wait for
6205          * SD_WAIT_CMDS_COMPLETE seconds before failing the DDI_SUSPEND.
6206          */
6207         wait_cmds_complete = ddi_get_lbolt() +
6208             (sd_wait_cmds_complete * drv_usectohz(1000000));
6209 
6210         while (un->un_ncmds_in_transport != 0) {
6211                 /*
6212                  * Fail if commands do not finish in the specified time.
6213                  */
6214                 if (cv_timedwait(&un->un_disk_busy_cv, SD_MUTEX(un),
6215                     wait_cmds_complete) == -1) {
6216                         /*
6217                          * Undo the state changes made above. Everything
6218                          * must go back to it's original value.
6219                          */
6220                         Restore_state(un);
6221                         un->un_last_state = un->un_save_state;
6222                         /* Wake up any threads that might be waiting. */
6223                         cv_broadcast(&un->un_suspend_cv);
6224                         mutex_exit(SD_MUTEX(un));
6225                         SD_ERROR(SD_LOG_IO_PM, un,
6226                             "sd_ddi_suspend: failed due to outstanding cmds\n");
6227                         SD_TRACE(SD_LOG_IO_PM, un, "sd_ddi_suspend: exiting\n");
6228                         return (DDI_FAILURE);
6229                 }
6230         }
6231 
6232         /*
6233          * Cancel SCSI watch thread and timeouts, if any are active
6234          */
6235 
6236         if (SD_OK_TO_SUSPEND_SCSI_WATCHER(un)) {
6237                 opaque_t temp_token = un->un_swr_token;
6238                 mutex_exit(SD_MUTEX(un));
6239                 scsi_watch_suspend(temp_token);
6240                 mutex_enter(SD_MUTEX(un));
6241         }
6242 
6243         if (un->un_reset_throttle_timeid != NULL) {
6244                 timeout_id_t temp_id = un->un_reset_throttle_timeid;
6245                 un->un_reset_throttle_timeid = NULL;
6246                 mutex_exit(SD_MUTEX(un));
6247                 (void) untimeout(temp_id);
6248                 mutex_enter(SD_MUTEX(un));
6249         }
6250 
6251         if (un->un_dcvb_timeid != NULL) {
6252                 timeout_id_t temp_id = un->un_dcvb_timeid;
6253                 un->un_dcvb_timeid = NULL;
6254                 mutex_exit(SD_MUTEX(un));
6255                 (void) untimeout(temp_id);
6256                 mutex_enter(SD_MUTEX(un));
6257         }
6258 
6259         mutex_enter(&un->un_pm_mutex);
6260         if (un->un_pm_timeid != NULL) {
6261                 timeout_id_t temp_id = un->un_pm_timeid;
6262                 un->un_pm_timeid = NULL;
6263                 mutex_exit(&un->un_pm_mutex);
6264                 mutex_exit(SD_MUTEX(un));
6265                 (void) untimeout(temp_id);
6266                 mutex_enter(SD_MUTEX(un));
6267         } else {
6268                 mutex_exit(&un->un_pm_mutex);
6269         }
6270 
6271         if (un->un_rmw_msg_timeid != NULL) {
6272                 timeout_id_t temp_id = un->un_rmw_msg_timeid;
6273                 un->un_rmw_msg_timeid = NULL;
6274                 mutex_exit(SD_MUTEX(un));
6275                 (void) untimeout(temp_id);
6276                 mutex_enter(SD_MUTEX(un));
6277         }
6278 
6279         if (un->un_retry_timeid != NULL) {
6280                 timeout_id_t temp_id = un->un_retry_timeid;
6281                 un->un_retry_timeid = NULL;
6282                 mutex_exit(SD_MUTEX(un));
6283                 (void) untimeout(temp_id);
6284                 mutex_enter(SD_MUTEX(un));
6285 
6286                 if (un->un_retry_bp != NULL) {
6287                         un->un_retry_bp->av_forw = un->un_waitq_headp;
6288                         un->un_waitq_headp = un->un_retry_bp;
6289                         if (un->un_waitq_tailp == NULL) {
6290                                 un->un_waitq_tailp = un->un_retry_bp;
6291                         }
6292                         un->un_retry_bp = NULL;
6293                         un->un_retry_statp = NULL;
6294                 }
6295         }
6296 
6297         if (un->un_direct_priority_timeid != NULL) {
6298                 timeout_id_t temp_id = un->un_direct_priority_timeid;
6299                 un->un_direct_priority_timeid = NULL;
6300                 mutex_exit(SD_MUTEX(un));
6301                 (void) untimeout(temp_id);
6302                 mutex_enter(SD_MUTEX(un));
6303         }
6304 
6305         if (un->un_f_is_fibre == TRUE) {
6306                 /*
6307                  * Remove callbacks for insert and remove events
6308                  */
6309                 if (un->un_insert_event != NULL) {
6310                         mutex_exit(SD_MUTEX(un));
6311                         (void) ddi_remove_event_handler(un->un_insert_cb_id);
6312                         mutex_enter(SD_MUTEX(un));
6313                         un->un_insert_event = NULL;
6314                 }
6315 
6316                 if (un->un_remove_event != NULL) {
6317                         mutex_exit(SD_MUTEX(un));
6318                         (void) ddi_remove_event_handler(un->un_remove_cb_id);
6319                         mutex_enter(SD_MUTEX(un));
6320                         un->un_remove_event = NULL;
6321                 }
6322         }
6323 
6324         mutex_exit(SD_MUTEX(un));
6325 
6326         SD_TRACE(SD_LOG_IO_PM, un, "sd_ddi_suspend: exit\n");
6327 
6328         return (DDI_SUCCESS);
6329 }
6330 
6331 
6332 /*
6333  *    Function: sd_ddi_resume
6334  *
6335  * Description: Performs system power-up operations..
6336  *
6337  * Return Code: DDI_SUCCESS
6338  *              DDI_FAILURE
6339  *
6340  *     Context: Kernel thread context
6341  */
6342 
6343 static int
6344 sd_ddi_resume(dev_info_t *devi)
6345 {
6346         struct  sd_lun  *un;
6347 
6348         un = ddi_get_soft_state(sd_state, ddi_get_instance(devi));
6349         if (un == NULL) {
6350                 return (DDI_FAILURE);
6351         }
6352 
6353         SD_TRACE(SD_LOG_IO_PM, un, "sd_ddi_resume: entry\n");
6354 
6355         mutex_enter(SD_MUTEX(un));
6356         Restore_state(un);
6357 
6358         /*
6359          * Restore the state which was saved to give the
6360          * the right state in un_last_state
6361          */
6362         un->un_last_state = un->un_save_state;
6363         /*
6364          * Note: throttle comes back at full.
6365          * Also note: this MUST be done before calling pm_raise_power
6366          * otherwise the system can get hung in biowait. The scenario where
6367          * this'll happen is under cpr suspend. Writing of the system
6368          * state goes through sddump, which writes 0 to un_throttle. If
6369          * writing the system state then fails, example if the partition is
6370          * too small, then cpr attempts a resume. If throttle isn't restored
6371          * from the saved value until after calling pm_raise_power then
6372          * cmds sent in sdpower are not transported and sd_send_scsi_cmd hangs
6373          * in biowait.
6374          */
6375         un->un_throttle = un->un_saved_throttle;
6376 
6377         /*
6378          * The chance of failure is very rare as the only command done in power
6379          * entry point is START command when you transition from 0->1 or
6380          * unknown->1. Put it to SPINDLE ON state irrespective of the state at
6381          * which suspend was done. Ignore the return value as the resume should
6382          * not be failed. In the case of removable media the media need not be
6383          * inserted and hence there is a chance that raise power will fail with
6384          * media not present.
6385          */
6386         if (un->un_f_attach_spinup) {
6387                 mutex_exit(SD_MUTEX(un));
6388                 (void) pm_raise_power(SD_DEVINFO(un), 0,
6389                     SD_PM_STATE_ACTIVE(un));
6390                 mutex_enter(SD_MUTEX(un));
6391         }
6392 
6393         /*
6394          * Don't broadcast to the suspend cv and therefore possibly
6395          * start I/O until after power has been restored.
6396          */
6397         cv_broadcast(&un->un_suspend_cv);
6398         cv_broadcast(&un->un_state_cv);
6399 
6400         /* restart thread */
6401         if (SD_OK_TO_RESUME_SCSI_WATCHER(un)) {
6402                 scsi_watch_resume(un->un_swr_token);
6403         }
6404 
6405 #if (defined(__fibre))
6406         if (un->un_f_is_fibre == TRUE) {
6407                 /*
6408                  * Add callbacks for insert and remove events
6409                  */
6410                 if (strcmp(un->un_node_type, DDI_NT_BLOCK_CHAN)) {
6411                         sd_init_event_callbacks(un);
6412                 }
6413         }
6414 #endif
6415 
6416         /*
6417          * Transport any pending commands to the target.
6418          *
6419          * If this is a low-activity device commands in queue will have to wait
6420          * until new commands come in, which may take awhile. Also, we
6421          * specifically don't check un_ncmds_in_transport because we know that
6422          * there really are no commands in progress after the unit was
6423          * suspended and we could have reached the throttle level, been
6424          * suspended, and have no new commands coming in for awhile. Highly
6425          * unlikely, but so is the low-activity disk scenario.
6426          */
6427         ddi_xbuf_dispatch(un->un_xbuf_attr);
6428 
6429         sd_start_cmds(un, NULL);
6430         mutex_exit(SD_MUTEX(un));
6431 
6432         SD_TRACE(SD_LOG_IO_PM, un, "sd_ddi_resume: exit\n");
6433 
6434         return (DDI_SUCCESS);
6435 }
6436 
6437 
6438 /*
6439  *    Function: sd_pm_state_change
6440  *
6441  * Description: Change the driver power state.
6442  *              Someone else is required to actually change the driver
6443  *              power level.
6444  *
6445  *   Arguments: un - driver soft state (unit) structure
6446  *              level - the power level that is changed to
6447  *              flag - to decide how to change the power state
6448  *
6449  * Return Code: DDI_SUCCESS
6450  *
6451  *     Context: Kernel thread context
6452  */
6453 static int
6454 sd_pm_state_change(struct sd_lun *un, int level, int flag)
6455 {
6456         ASSERT(un != NULL);
6457         SD_TRACE(SD_LOG_POWER, un, "sd_pm_state_change: entry\n");
6458 
6459         ASSERT(!mutex_owned(SD_MUTEX(un)));
6460         mutex_enter(SD_MUTEX(un));
6461 
6462         if (flag == SD_PM_STATE_ROLLBACK || SD_PM_IS_IO_CAPABLE(un, level)) {
6463                 un->un_power_level = level;
6464                 ASSERT(!mutex_owned(&un->un_pm_mutex));
6465                 mutex_enter(&un->un_pm_mutex);
6466                 if (SD_DEVICE_IS_IN_LOW_POWER(un)) {
6467                         un->un_pm_count++;
6468                         ASSERT(un->un_pm_count == 0);
6469                 }
6470                 mutex_exit(&un->un_pm_mutex);
6471         } else {
6472                 /*
6473                  * Exit if power management is not enabled for this device,
6474                  * or if the device is being used by HA.
6475                  */
6476                 if ((un->un_f_pm_is_enabled == FALSE) || (un->un_resvd_status &
6477                     (SD_RESERVE | SD_WANT_RESERVE | SD_LOST_RESERVE))) {
6478                         mutex_exit(SD_MUTEX(un));
6479                         SD_TRACE(SD_LOG_POWER, un,
6480                             "sd_pm_state_change: exiting\n");
6481                         return (DDI_FAILURE);
6482                 }
6483 
6484                 SD_INFO(SD_LOG_POWER, un, "sd_pm_state_change: "
6485                     "un_ncmds_in_driver=%ld\n", un->un_ncmds_in_driver);
6486 
6487                 /*
6488                  * See if the device is not busy, ie.:
6489                  *    - we have no commands in the driver for this device
6490                  *    - not waiting for resources
6491                  */
6492                 if ((un->un_ncmds_in_driver == 0) &&
6493                     (un->un_state != SD_STATE_RWAIT)) {
6494                         /*
6495                          * The device is not busy, so it is OK to go to low
6496                          * power state. Indicate low power, but rely on someone
6497                          * else to actually change it.
6498                          */
6499                         mutex_enter(&un->un_pm_mutex);
6500                         un->un_pm_count = -1;
6501                         mutex_exit(&un->un_pm_mutex);
6502                         un->un_power_level = level;
6503                 }
6504         }
6505 
6506         mutex_exit(SD_MUTEX(un));
6507 
6508         SD_TRACE(SD_LOG_POWER, un, "sd_pm_state_change: exit\n");
6509 
6510         return (DDI_SUCCESS);
6511 }
6512 
6513 
6514 /*
6515  *    Function: sd_pm_idletimeout_handler
6516  *
6517  * Description: A timer routine that's active only while a device is busy.
6518  *              The purpose is to extend slightly the pm framework's busy
6519  *              view of the device to prevent busy/idle thrashing for
6520  *              back-to-back commands. Do this by comparing the current time
6521  *              to the time at which the last command completed and when the
6522  *              difference is greater than sd_pm_idletime, call
6523  *              pm_idle_component. In addition to indicating idle to the pm
6524  *              framework, update the chain type to again use the internal pm
6525  *              layers of the driver.
6526  *
6527  *   Arguments: arg - driver soft state (unit) structure
6528  *
6529  *     Context: Executes in a timeout(9F) thread context
6530  */
6531 
6532 static void
6533 sd_pm_idletimeout_handler(void *arg)
6534 {
6535         struct sd_lun *un = arg;
6536 
6537         time_t  now;
6538 
6539         mutex_enter(&sd_detach_mutex);
6540         if (un->un_detach_count != 0) {
6541                 /* Abort if the instance is detaching */
6542                 mutex_exit(&sd_detach_mutex);
6543                 return;
6544         }
6545         mutex_exit(&sd_detach_mutex);
6546 
6547         now = ddi_get_time();
6548         /*
6549          * Grab both mutexes, in the proper order, since we're accessing
6550          * both PM and softstate variables.
6551          */
6552         mutex_enter(SD_MUTEX(un));
6553         mutex_enter(&un->un_pm_mutex);
6554         if (((now - un->un_pm_idle_time) > sd_pm_idletime) &&
6555             (un->un_ncmds_in_driver == 0) && (un->un_pm_count == 0)) {
6556                 /*
6557                  * Update the chain types.
6558                  * This takes affect on the next new command received.
6559                  */
6560                 if (un->un_f_non_devbsize_supported) {
6561                         un->un_buf_chain_type = SD_CHAIN_INFO_RMMEDIA;
6562                 } else {
6563                         un->un_buf_chain_type = SD_CHAIN_INFO_DISK;
6564                 }
6565                 un->un_uscsi_chain_type = SD_CHAIN_INFO_USCSI_CMD;
6566 
6567                 SD_TRACE(SD_LOG_IO_PM, un,
6568                     "sd_pm_idletimeout_handler: idling device\n");
6569                 (void) pm_idle_component(SD_DEVINFO(un), 0);
6570                 un->un_pm_idle_timeid = NULL;
6571         } else {
6572                 un->un_pm_idle_timeid =
6573                     timeout(sd_pm_idletimeout_handler, un,
6574                     (drv_usectohz((clock_t)300000))); /* 300 ms. */
6575         }
6576         mutex_exit(&un->un_pm_mutex);
6577         mutex_exit(SD_MUTEX(un));
6578 }
6579 
6580 
6581 /*
6582  *    Function: sd_pm_timeout_handler
6583  *
6584  * Description: Callback to tell framework we are idle.
6585  *
6586  *     Context: timeout(9f) thread context.
6587  */
6588 
6589 static void
6590 sd_pm_timeout_handler(void *arg)
6591 {
6592         struct sd_lun *un = arg;
6593 
6594         (void) pm_idle_component(SD_DEVINFO(un), 0);
6595         mutex_enter(&un->un_pm_mutex);
6596         un->un_pm_timeid = NULL;
6597         mutex_exit(&un->un_pm_mutex);
6598 }
6599 
6600 
6601 /*
6602  *    Function: sdpower
6603  *
6604  * Description: PM entry point.
6605  *
6606  * Return Code: DDI_SUCCESS
6607  *              DDI_FAILURE
6608  *
6609  *     Context: Kernel thread context
6610  */
6611 
6612 static int
6613 sdpower(dev_info_t *devi, int component, int level)
6614 {
6615         struct sd_lun   *un;
6616         int             instance;
6617         int             rval = DDI_SUCCESS;
6618         uint_t          i, log_page_size, maxcycles, ncycles;
6619         uchar_t         *log_page_data;
6620         int             log_sense_page;
6621         int             medium_present;
6622         time_t          intvlp;
6623         struct pm_trans_data    sd_pm_tran_data;
6624         uchar_t         save_state;
6625         int             sval;
6626         uchar_t         state_before_pm;
6627         int             got_semaphore_here;
6628         sd_ssc_t        *ssc;
6629         int     last_power_level;
6630 
6631         instance = ddi_get_instance(devi);
6632 
6633         if (((un = ddi_get_soft_state(sd_state, instance)) == NULL) ||
6634             !SD_PM_IS_LEVEL_VALID(un, level) || component != 0) {
6635                 return (DDI_FAILURE);
6636         }
6637 
6638         ssc = sd_ssc_init(un);
6639 
6640         SD_TRACE(SD_LOG_IO_PM, un, "sdpower: entry, level = %d\n", level);
6641 
6642         /*
6643          * Must synchronize power down with close.
6644          * Attempt to decrement/acquire the open/close semaphore,
6645          * but do NOT wait on it. If it's not greater than zero,
6646          * ie. it can't be decremented without waiting, then
6647          * someone else, either open or close, already has it
6648          * and the try returns 0. Use that knowledge here to determine
6649          * if it's OK to change the device power level.
6650          * Also, only increment it on exit if it was decremented, ie. gotten,
6651          * here.
6652          */
6653         got_semaphore_here = sema_tryp(&un->un_semoclose);
6654 
6655         mutex_enter(SD_MUTEX(un));
6656 
6657         SD_INFO(SD_LOG_POWER, un, "sdpower: un_ncmds_in_driver = %ld\n",
6658             un->un_ncmds_in_driver);
6659 
6660         /*
6661          * If un_ncmds_in_driver is non-zero it indicates commands are
6662          * already being processed in the driver, or if the semaphore was
6663          * not gotten here it indicates an open or close is being processed.
6664          * At the same time somebody is requesting to go to a lower power
6665          * that can't perform I/O, which can't happen, therefore we need to
6666          * return failure.
6667          */
6668         if ((!SD_PM_IS_IO_CAPABLE(un, level)) &&
6669             ((un->un_ncmds_in_driver != 0) || (got_semaphore_here == 0))) {
6670                 mutex_exit(SD_MUTEX(un));
6671 
6672                 if (got_semaphore_here != 0) {
6673                         sema_v(&un->un_semoclose);
6674                 }
6675                 SD_TRACE(SD_LOG_IO_PM, un,
6676                     "sdpower: exit, device has queued cmds.\n");
6677 
6678                 goto sdpower_failed;
6679         }
6680 
6681         /*
6682          * if it is OFFLINE that means the disk is completely dead
6683          * in our case we have to put the disk in on or off by sending commands
6684          * Of course that will fail anyway so return back here.
6685          *
6686          * Power changes to a device that's OFFLINE or SUSPENDED
6687          * are not allowed.
6688          */
6689         if ((un->un_state == SD_STATE_OFFLINE) ||
6690             (un->un_state == SD_STATE_SUSPENDED)) {
6691                 mutex_exit(SD_MUTEX(un));
6692 
6693                 if (got_semaphore_here != 0) {
6694                         sema_v(&un->un_semoclose);
6695                 }
6696                 SD_TRACE(SD_LOG_IO_PM, un,
6697                     "sdpower: exit, device is off-line.\n");
6698 
6699                 goto sdpower_failed;
6700         }
6701 
6702         /*
6703          * Change the device's state to indicate it's power level
6704          * is being changed. Do this to prevent a power off in the
6705          * middle of commands, which is especially bad on devices
6706          * that are really powered off instead of just spun down.
6707          */
6708         state_before_pm = un->un_state;
6709         un->un_state = SD_STATE_PM_CHANGING;
6710 
6711         mutex_exit(SD_MUTEX(un));
6712 
6713         /*
6714          * If log sense command is not supported, bypass the
6715          * following checking, otherwise, check the log sense
6716          * information for this device.
6717          */
6718         if (SD_PM_STOP_MOTOR_NEEDED(un, level) &&
6719             un->un_f_log_sense_supported) {
6720                 /*
6721                  * Get the log sense information to understand whether the
6722                  * the powercycle counts have gone beyond the threshhold.
6723                  */
6724                 log_page_size = START_STOP_CYCLE_COUNTER_PAGE_SIZE;
6725                 log_page_data = kmem_zalloc(log_page_size, KM_SLEEP);
6726 
6727                 mutex_enter(SD_MUTEX(un));
6728                 log_sense_page = un->un_start_stop_cycle_page;
6729                 mutex_exit(SD_MUTEX(un));
6730 
6731                 rval = sd_send_scsi_LOG_SENSE(ssc, log_page_data,
6732                     log_page_size, log_sense_page, 0x01, 0, SD_PATH_DIRECT);
6733 
6734                 if (rval != 0) {
6735                         if (rval == EIO)
6736                                 sd_ssc_assessment(ssc, SD_FMT_STATUS_CHECK);
6737                         else
6738                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
6739                 }
6740 
6741 #ifdef  SDDEBUG
6742                 if (sd_force_pm_supported) {
6743                         /* Force a successful result */
6744                         rval = 0;
6745                 }
6746 #endif
6747                 if (rval != 0) {
6748                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
6749                             "Log Sense Failed\n");
6750 
6751                         kmem_free(log_page_data, log_page_size);
6752                         /* Cannot support power management on those drives */
6753 
6754                         if (got_semaphore_here != 0) {
6755                                 sema_v(&un->un_semoclose);
6756                         }
6757                         /*
6758                          * On exit put the state back to it's original value
6759                          * and broadcast to anyone waiting for the power
6760                          * change completion.
6761                          */
6762                         mutex_enter(SD_MUTEX(un));
6763                         un->un_state = state_before_pm;
6764                         cv_broadcast(&un->un_suspend_cv);
6765                         mutex_exit(SD_MUTEX(un));
6766                         SD_TRACE(SD_LOG_IO_PM, un,
6767                             "sdpower: exit, Log Sense Failed.\n");
6768 
6769                         goto sdpower_failed;
6770                 }
6771 
6772                 /*
6773                  * From the page data - Convert the essential information to
6774                  * pm_trans_data
6775                  */
6776                 maxcycles =
6777                     (log_page_data[0x1c] << 24) | (log_page_data[0x1d] << 16) |
6778                     (log_page_data[0x1E] << 8)  | log_page_data[0x1F];
6779 
6780                 ncycles =
6781                     (log_page_data[0x24] << 24) | (log_page_data[0x25] << 16) |
6782                     (log_page_data[0x26] << 8)  | log_page_data[0x27];
6783 
6784                 if (un->un_f_pm_log_sense_smart) {
6785                         sd_pm_tran_data.un.smart_count.allowed = maxcycles;
6786                         sd_pm_tran_data.un.smart_count.consumed = ncycles;
6787                         sd_pm_tran_data.un.smart_count.flag = 0;
6788                         sd_pm_tran_data.format = DC_SMART_FORMAT;
6789                 } else {
6790                         sd_pm_tran_data.un.scsi_cycles.lifemax = maxcycles;
6791                         sd_pm_tran_data.un.scsi_cycles.ncycles = ncycles;
6792                         for (i = 0; i < DC_SCSI_MFR_LEN; i++) {
6793                                 sd_pm_tran_data.un.scsi_cycles.svc_date[i] =
6794                                     log_page_data[8+i];
6795                         }
6796                         sd_pm_tran_data.un.scsi_cycles.flag = 0;
6797                         sd_pm_tran_data.format = DC_SCSI_FORMAT;
6798                 }
6799 
6800                 kmem_free(log_page_data, log_page_size);
6801 
6802                 /*
6803                  * Call pm_trans_check routine to get the Ok from
6804                  * the global policy
6805                  */
6806                 rval = pm_trans_check(&sd_pm_tran_data, &intvlp);
6807 #ifdef  SDDEBUG
6808                 if (sd_force_pm_supported) {
6809                         /* Force a successful result */
6810                         rval = 1;
6811                 }
6812 #endif
6813                 switch (rval) {
6814                 case 0:
6815                         /*
6816                          * Not Ok to Power cycle or error in parameters passed
6817                          * Would have given the advised time to consider power
6818                          * cycle. Based on the new intvlp parameter we are
6819                          * supposed to pretend we are busy so that pm framework
6820                          * will never call our power entry point. Because of
6821                          * that install a timeout handler and wait for the
6822                          * recommended time to elapse so that power management
6823                          * can be effective again.
6824                          *
6825                          * To effect this behavior, call pm_busy_component to
6826                          * indicate to the framework this device is busy.
6827                          * By not adjusting un_pm_count the rest of PM in
6828                          * the driver will function normally, and independent
6829                          * of this but because the framework is told the device
6830                          * is busy it won't attempt powering down until it gets
6831                          * a matching idle. The timeout handler sends this.
6832                          * Note: sd_pm_entry can't be called here to do this
6833                          * because sdpower may have been called as a result
6834                          * of a call to pm_raise_power from within sd_pm_entry.
6835                          *
6836                          * If a timeout handler is already active then
6837                          * don't install another.
6838                          */
6839                         mutex_enter(&un->un_pm_mutex);
6840                         if (un->un_pm_timeid == NULL) {
6841                                 un->un_pm_timeid =
6842                                     timeout(sd_pm_timeout_handler,
6843                                     un, intvlp * drv_usectohz(1000000));
6844                                 mutex_exit(&un->un_pm_mutex);
6845                                 (void) pm_busy_component(SD_DEVINFO(un), 0);
6846                         } else {
6847                                 mutex_exit(&un->un_pm_mutex);
6848                         }
6849                         if (got_semaphore_here != 0) {
6850                                 sema_v(&un->un_semoclose);
6851                         }
6852                         /*
6853                          * On exit put the state back to it's original value
6854                          * and broadcast to anyone waiting for the power
6855                          * change completion.
6856                          */
6857                         mutex_enter(SD_MUTEX(un));
6858                         un->un_state = state_before_pm;
6859                         cv_broadcast(&un->un_suspend_cv);
6860                         mutex_exit(SD_MUTEX(un));
6861 
6862                         SD_TRACE(SD_LOG_IO_PM, un, "sdpower: exit, "
6863                             "trans check Failed, not ok to power cycle.\n");
6864 
6865                         goto sdpower_failed;
6866                 case -1:
6867                         if (got_semaphore_here != 0) {
6868                                 sema_v(&un->un_semoclose);
6869                         }
6870                         /*
6871                          * On exit put the state back to it's original value
6872                          * and broadcast to anyone waiting for the power
6873                          * change completion.
6874                          */
6875                         mutex_enter(SD_MUTEX(un));
6876                         un->un_state = state_before_pm;
6877                         cv_broadcast(&un->un_suspend_cv);
6878                         mutex_exit(SD_MUTEX(un));
6879                         SD_TRACE(SD_LOG_IO_PM, un,
6880                             "sdpower: exit, trans check command Failed.\n");
6881 
6882                         goto sdpower_failed;
6883                 }
6884         }
6885 
6886         if (!SD_PM_IS_IO_CAPABLE(un, level)) {
6887                 /*
6888                  * Save the last state... if the STOP FAILS we need it
6889                  * for restoring
6890                  */
6891                 mutex_enter(SD_MUTEX(un));
6892                 save_state = un->un_last_state;
6893                 last_power_level = un->un_power_level;
6894                 /*
6895                  * There must not be any cmds. getting processed
6896                  * in the driver when we get here. Power to the
6897                  * device is potentially going off.
6898                  */
6899                 ASSERT(un->un_ncmds_in_driver == 0);
6900                 mutex_exit(SD_MUTEX(un));
6901 
6902                 /*
6903                  * For now PM suspend the device completely before spindle is
6904                  * turned off
6905                  */
6906                 if ((rval = sd_pm_state_change(un, level, SD_PM_STATE_CHANGE))
6907                     == DDI_FAILURE) {
6908                         if (got_semaphore_here != 0) {
6909                                 sema_v(&un->un_semoclose);
6910                         }
6911                         /*
6912                          * On exit put the state back to it's original value
6913                          * and broadcast to anyone waiting for the power
6914                          * change completion.
6915                          */
6916                         mutex_enter(SD_MUTEX(un));
6917                         un->un_state = state_before_pm;
6918                         un->un_power_level = last_power_level;
6919                         cv_broadcast(&un->un_suspend_cv);
6920                         mutex_exit(SD_MUTEX(un));
6921                         SD_TRACE(SD_LOG_IO_PM, un,
6922                             "sdpower: exit, PM suspend Failed.\n");
6923 
6924                         goto sdpower_failed;
6925                 }
6926         }
6927 
6928         /*
6929          * The transition from SPINDLE_OFF to SPINDLE_ON can happen in open,
6930          * close, or strategy. Dump no long uses this routine, it uses it's
6931          * own code so it can be done in polled mode.
6932          */
6933 
6934         medium_present = TRUE;
6935 
6936         /*
6937          * When powering up, issue a TUR in case the device is at unit
6938          * attention.  Don't do retries. Bypass the PM layer, otherwise
6939          * a deadlock on un_pm_busy_cv will occur.
6940          */
6941         if (SD_PM_IS_IO_CAPABLE(un, level)) {
6942                 sval = sd_send_scsi_TEST_UNIT_READY(ssc,
6943                     SD_DONT_RETRY_TUR | SD_BYPASS_PM);
6944                 if (sval != 0)
6945                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
6946         }
6947 
6948         if (un->un_f_power_condition_supported) {
6949                 char *pm_condition_name[] = {"STOPPED", "STANDBY",
6950                     "IDLE", "ACTIVE"};
6951                 SD_TRACE(SD_LOG_IO_PM, un,
6952                     "sdpower: sending \'%s\' power condition",
6953                     pm_condition_name[level]);
6954                 sval = sd_send_scsi_START_STOP_UNIT(ssc, SD_POWER_CONDITION,
6955                     sd_pl2pc[level], SD_PATH_DIRECT);
6956         } else {
6957                 SD_TRACE(SD_LOG_IO_PM, un, "sdpower: sending \'%s\' unit\n",
6958                     ((level == SD_SPINDLE_ON) ? "START" : "STOP"));
6959                 sval = sd_send_scsi_START_STOP_UNIT(ssc, SD_START_STOP,
6960                     ((level == SD_SPINDLE_ON) ? SD_TARGET_START :
6961                     SD_TARGET_STOP), SD_PATH_DIRECT);
6962         }
6963         if (sval != 0) {
6964                 if (sval == EIO)
6965                         sd_ssc_assessment(ssc, SD_FMT_STATUS_CHECK);
6966                 else
6967                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
6968         }
6969 
6970         /* Command failed, check for media present. */
6971         if ((sval == ENXIO) && un->un_f_has_removable_media) {
6972                 medium_present = FALSE;
6973         }
6974 
6975         /*
6976          * The conditions of interest here are:
6977          *   if a spindle off with media present fails,
6978          *      then restore the state and return an error.
6979          *   else if a spindle on fails,
6980          *      then return an error (there's no state to restore).
6981          * In all other cases we setup for the new state
6982          * and return success.
6983          */
6984         if (!SD_PM_IS_IO_CAPABLE(un, level)) {
6985                 if ((medium_present == TRUE) && (sval != 0)) {
6986                         /* The stop command from above failed */
6987                         rval = DDI_FAILURE;
6988                         /*
6989                          * The stop command failed, and we have media
6990                          * present. Put the level back by calling the
6991                          * sd_pm_resume() and set the state back to
6992                          * it's previous value.
6993                          */
6994                         (void) sd_pm_state_change(un, last_power_level,
6995                             SD_PM_STATE_ROLLBACK);
6996                         mutex_enter(SD_MUTEX(un));
6997                         un->un_last_state = save_state;
6998                         mutex_exit(SD_MUTEX(un));
6999                 } else if (un->un_f_monitor_media_state) {
7000                         /*
7001                          * The stop command from above succeeded.
7002                          * Terminate watch thread in case of removable media
7003                          * devices going into low power state. This is as per
7004                          * the requirements of pm framework, otherwise commands
7005                          * will be generated for the device (through watch
7006                          * thread), even when the device is in low power state.
7007                          */
7008                         mutex_enter(SD_MUTEX(un));
7009                         un->un_f_watcht_stopped = FALSE;
7010                         if (un->un_swr_token != NULL) {
7011                                 opaque_t temp_token = un->un_swr_token;
7012                                 un->un_f_watcht_stopped = TRUE;
7013                                 un->un_swr_token = NULL;
7014                                 mutex_exit(SD_MUTEX(un));
7015                                 (void) scsi_watch_request_terminate(temp_token,
7016                                     SCSI_WATCH_TERMINATE_ALL_WAIT);
7017                         } else {
7018                                 mutex_exit(SD_MUTEX(un));
7019                         }
7020                 }
7021         } else {
7022                 /*
7023                  * The level requested is I/O capable.
7024                  * Legacy behavior: return success on a failed spinup
7025                  * if there is no media in the drive.
7026                  * Do this by looking at medium_present here.
7027                  */
7028                 if ((sval != 0) && medium_present) {
7029                         /* The start command from above failed */
7030                         rval = DDI_FAILURE;
7031                 } else {
7032                         /*
7033                          * The start command from above succeeded
7034                          * PM resume the devices now that we have
7035                          * started the disks
7036                          */
7037                         (void) sd_pm_state_change(un, level,
7038                             SD_PM_STATE_CHANGE);
7039 
7040                         /*
7041                          * Resume the watch thread since it was suspended
7042                          * when the device went into low power mode.
7043                          */
7044                         if (un->un_f_monitor_media_state) {
7045                                 mutex_enter(SD_MUTEX(un));
7046                                 if (un->un_f_watcht_stopped == TRUE) {
7047                                         opaque_t temp_token;
7048 
7049                                         un->un_f_watcht_stopped = FALSE;
7050                                         mutex_exit(SD_MUTEX(un));
7051                                         temp_token =
7052                                             sd_watch_request_submit(un);
7053                                         mutex_enter(SD_MUTEX(un));
7054                                         un->un_swr_token = temp_token;
7055                                 }
7056                                 mutex_exit(SD_MUTEX(un));
7057                         }
7058                 }
7059         }
7060 
7061         if (got_semaphore_here != 0) {
7062                 sema_v(&un->un_semoclose);
7063         }
7064         /*
7065          * On exit put the state back to it's original value
7066          * and broadcast to anyone waiting for the power
7067          * change completion.
7068          */
7069         mutex_enter(SD_MUTEX(un));
7070         un->un_state = state_before_pm;
7071         cv_broadcast(&un->un_suspend_cv);
7072         mutex_exit(SD_MUTEX(un));
7073 
7074         SD_TRACE(SD_LOG_IO_PM, un, "sdpower: exit, status = 0x%x\n", rval);
7075 
7076         sd_ssc_fini(ssc);
7077         return (rval);
7078 
7079 sdpower_failed:
7080 
7081         sd_ssc_fini(ssc);
7082         return (DDI_FAILURE);
7083 }
7084 
7085 
7086 
7087 /*
7088  *    Function: sdattach
7089  *
7090  * Description: Driver's attach(9e) entry point function.
7091  *
7092  *   Arguments: devi - opaque device info handle
7093  *              cmd  - attach  type
7094  *
7095  * Return Code: DDI_SUCCESS
7096  *              DDI_FAILURE
7097  *
7098  *     Context: Kernel thread context
7099  */
7100 
7101 static int
7102 sdattach(dev_info_t *devi, ddi_attach_cmd_t cmd)
7103 {
7104         switch (cmd) {
7105         case DDI_ATTACH:
7106                 return (sd_unit_attach(devi));
7107         case DDI_RESUME:
7108                 return (sd_ddi_resume(devi));
7109         default:
7110                 break;
7111         }
7112         return (DDI_FAILURE);
7113 }
7114 
7115 
7116 /*
7117  *    Function: sddetach
7118  *
7119  * Description: Driver's detach(9E) entry point function.
7120  *
7121  *   Arguments: devi - opaque device info handle
7122  *              cmd  - detach  type
7123  *
7124  * Return Code: DDI_SUCCESS
7125  *              DDI_FAILURE
7126  *
7127  *     Context: Kernel thread context
7128  */
7129 
7130 static int
7131 sddetach(dev_info_t *devi, ddi_detach_cmd_t cmd)
7132 {
7133         switch (cmd) {
7134         case DDI_DETACH:
7135                 return (sd_unit_detach(devi));
7136         case DDI_SUSPEND:
7137                 return (sd_ddi_suspend(devi));
7138         default:
7139                 break;
7140         }
7141         return (DDI_FAILURE);
7142 }
7143 
7144 
7145 /*
7146  *     Function: sd_sync_with_callback
7147  *
7148  *  Description: Prevents sd_unit_attach or sd_unit_detach from freeing the soft
7149  *               state while the callback routine is active.
7150  *
7151  *    Arguments: un: softstate structure for the instance
7152  *
7153  *      Context: Kernel thread context
7154  */
7155 
7156 static void
7157 sd_sync_with_callback(struct sd_lun *un)
7158 {
7159         ASSERT(un != NULL);
7160 
7161         mutex_enter(SD_MUTEX(un));
7162 
7163         ASSERT(un->un_in_callback >= 0);
7164 
7165         while (un->un_in_callback > 0) {
7166                 mutex_exit(SD_MUTEX(un));
7167                 delay(2);
7168                 mutex_enter(SD_MUTEX(un));
7169         }
7170 
7171         mutex_exit(SD_MUTEX(un));
7172 }
7173 
7174 /*
7175  *    Function: sd_unit_attach
7176  *
7177  * Description: Performs DDI_ATTACH processing for sdattach(). Allocates
7178  *              the soft state structure for the device and performs
7179  *              all necessary structure and device initializations.
7180  *
7181  *   Arguments: devi: the system's dev_info_t for the device.
7182  *
7183  * Return Code: DDI_SUCCESS if attach is successful.
7184  *              DDI_FAILURE if any part of the attach fails.
7185  *
7186  *     Context: Called at attach(9e) time for the DDI_ATTACH flag.
7187  *              Kernel thread context only.  Can sleep.
7188  */
7189 
7190 static int
7191 sd_unit_attach(dev_info_t *devi)
7192 {
7193         struct  scsi_device     *devp;
7194         struct  sd_lun          *un;
7195         char                    *variantp;
7196         char                    name_str[48];
7197         int     reservation_flag = SD_TARGET_IS_UNRESERVED;
7198         int     instance;
7199         int     rval;
7200         int     wc_enabled;
7201         int     tgt;
7202         uint64_t        capacity;
7203         uint_t          lbasize = 0;
7204         dev_info_t      *pdip = ddi_get_parent(devi);
7205         int             offbyone = 0;
7206         int             geom_label_valid = 0;
7207         sd_ssc_t        *ssc;
7208         int             status;
7209         struct sd_fm_internal   *sfip = NULL;
7210         int             max_xfer_size;
7211 
7212         /*
7213          * Retrieve the target driver's private data area. This was set
7214          * up by the HBA.
7215          */
7216         devp = ddi_get_driver_private(devi);
7217 
7218         /*
7219          * Retrieve the target ID of the device.
7220          */
7221         tgt = ddi_prop_get_int(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS,
7222             SCSI_ADDR_PROP_TARGET, -1);
7223 
7224         /*
7225          * Since we have no idea what state things were left in by the last
7226          * user of the device, set up some 'default' settings, ie. turn 'em
7227          * off. The scsi_ifsetcap calls force re-negotiations with the drive.
7228          * Do this before the scsi_probe, which sends an inquiry.
7229          * This is a fix for bug (4430280).
7230          * Of special importance is wide-xfer. The drive could have been left
7231          * in wide transfer mode by the last driver to communicate with it,
7232          * this includes us. If that's the case, and if the following is not
7233          * setup properly or we don't re-negotiate with the drive prior to
7234          * transferring data to/from the drive, it causes bus parity errors,
7235          * data overruns, and unexpected interrupts. This first occurred when
7236          * the fix for bug (4378686) was made.
7237          */
7238         (void) scsi_ifsetcap(&devp->sd_address, "lun-reset", 0, 1);
7239         (void) scsi_ifsetcap(&devp->sd_address, "wide-xfer", 0, 1);
7240         (void) scsi_ifsetcap(&devp->sd_address, "auto-rqsense", 0, 1);
7241 
7242         /*
7243          * Currently, scsi_ifsetcap sets tagged-qing capability for all LUNs
7244          * on a target. Setting it per lun instance actually sets the
7245          * capability of this target, which affects those luns already
7246          * attached on the same target. So during attach, we can only disable
7247          * this capability only when no other lun has been attached on this
7248          * target. By doing this, we assume a target has the same tagged-qing
7249          * capability for every lun. The condition can be removed when HBA
7250          * is changed to support per lun based tagged-qing capability.
7251          */
7252         if (sd_scsi_get_target_lun_count(pdip, tgt) < 1) {
7253                 (void) scsi_ifsetcap(&devp->sd_address, "tagged-qing", 0, 1);
7254         }
7255 
7256         /*
7257          * Use scsi_probe() to issue an INQUIRY command to the device.
7258          * This call will allocate and fill in the scsi_inquiry structure
7259          * and point the sd_inq member of the scsi_device structure to it.
7260          * If the attach succeeds, then this memory will not be de-allocated
7261          * (via scsi_unprobe()) until the instance is detached.
7262          */
7263         if (scsi_probe(devp, SLEEP_FUNC) != SCSIPROBE_EXISTS) {
7264                 goto probe_failed;
7265         }
7266 
7267         /*
7268          * Check the device type as specified in the inquiry data and
7269          * claim it if it is of a type that we support.
7270          */
7271         switch (devp->sd_inq->inq_dtype) {
7272         case DTYPE_DIRECT:
7273                 break;
7274         case DTYPE_RODIRECT:
7275                 break;
7276         case DTYPE_OPTICAL:
7277                 break;
7278         case DTYPE_NOTPRESENT:
7279         default:
7280                 /* Unsupported device type; fail the attach. */
7281                 goto probe_failed;
7282         }
7283 
7284         /*
7285          * Allocate the soft state structure for this unit.
7286          *
7287          * We rely upon this memory being set to all zeroes by
7288          * ddi_soft_state_zalloc().  We assume that any member of the
7289          * soft state structure that is not explicitly initialized by
7290          * this routine will have a value of zero.
7291          */
7292         instance = ddi_get_instance(devp->sd_dev);
7293 #ifndef XPV_HVM_DRIVER
7294         if (ddi_soft_state_zalloc(sd_state, instance) != DDI_SUCCESS) {
7295                 goto probe_failed;
7296         }
7297 #endif /* !XPV_HVM_DRIVER */
7298 
7299         /*
7300          * Retrieve a pointer to the newly-allocated soft state.
7301          *
7302          * This should NEVER fail if the ddi_soft_state_zalloc() call above
7303          * was successful, unless something has gone horribly wrong and the
7304          * ddi's soft state internals are corrupt (in which case it is
7305          * probably better to halt here than just fail the attach....)
7306          */
7307         if ((un = ddi_get_soft_state(sd_state, instance)) == NULL) {
7308                 panic("sd_unit_attach: NULL soft state on instance:0x%x",
7309                     instance);
7310                 /*NOTREACHED*/
7311         }
7312 
7313         /*
7314          * Link the back ptr of the driver soft state to the scsi_device
7315          * struct for this lun.
7316          * Save a pointer to the softstate in the driver-private area of
7317          * the scsi_device struct.
7318          * Note: We cannot call SD_INFO, SD_TRACE, SD_ERROR, or SD_DIAG until
7319          * we first set un->un_sd below.
7320          */
7321         un->un_sd = devp;
7322         devp->sd_private = (opaque_t)un;
7323 
7324         /*
7325          * The following must be after devp is stored in the soft state struct.
7326          */
7327 #ifdef SDDEBUG
7328         SD_TRACE(SD_LOG_ATTACH_DETACH, un,
7329             "%s_unit_attach: un:0x%p instance:%d\n",
7330             ddi_driver_name(devi), un, instance);
7331 #endif
7332 
7333         /*
7334          * Set up the device type and node type (for the minor nodes).
7335          * By default we assume that the device can at least support the
7336          * Common Command Set. Call it a CD-ROM if it reports itself
7337          * as a RODIRECT device.
7338          */
7339         switch (devp->sd_inq->inq_dtype) {
7340         case DTYPE_RODIRECT:
7341                 un->un_node_type = DDI_NT_CD_CHAN;
7342                 un->un_ctype  = CTYPE_CDROM;
7343                 break;
7344         case DTYPE_OPTICAL:
7345                 un->un_node_type = DDI_NT_BLOCK_CHAN;
7346                 un->un_ctype  = CTYPE_ROD;
7347                 break;
7348         default:
7349                 un->un_node_type = DDI_NT_BLOCK_CHAN;
7350                 un->un_ctype  = CTYPE_CCS;
7351                 break;
7352         }
7353 
7354         /*
7355          * Try to read the interconnect type from the HBA.
7356          *
7357          * Note: This driver is currently compiled as two binaries, a parallel
7358          * scsi version (sd) and a fibre channel version (ssd). All functional
7359          * differences are determined at compile time. In the future a single
7360          * binary will be provided and the interconnect type will be used to
7361          * differentiate between fibre and parallel scsi behaviors. At that time
7362          * it will be necessary for all fibre channel HBAs to support this
7363          * property.
7364          *
7365          * set un_f_is_fiber to TRUE ( default fiber )
7366          */
7367         un->un_f_is_fibre = TRUE;
7368         switch (scsi_ifgetcap(SD_ADDRESS(un), "interconnect-type", -1)) {
7369         case INTERCONNECT_SSA:
7370                 un->un_interconnect_type = SD_INTERCONNECT_SSA;
7371                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7372                     "sd_unit_attach: un:0x%p SD_INTERCONNECT_SSA\n", un);
7373                 break;
7374         case INTERCONNECT_PARALLEL:
7375                 un->un_f_is_fibre = FALSE;
7376                 un->un_interconnect_type = SD_INTERCONNECT_PARALLEL;
7377                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7378                     "sd_unit_attach: un:0x%p SD_INTERCONNECT_PARALLEL\n", un);
7379                 break;
7380         case INTERCONNECT_SAS:
7381                 un->un_f_is_fibre = FALSE;
7382                 un->un_interconnect_type = SD_INTERCONNECT_SAS;
7383                 un->un_node_type = DDI_NT_BLOCK_SAS;
7384                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7385                     "sd_unit_attach: un:0x%p SD_INTERCONNECT_SAS\n", un);
7386                 break;
7387         case INTERCONNECT_SATA:
7388                 un->un_f_is_fibre = FALSE;
7389                 un->un_interconnect_type = SD_INTERCONNECT_SATA;
7390                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7391                     "sd_unit_attach: un:0x%p SD_INTERCONNECT_SATA\n", un);
7392                 break;
7393         case INTERCONNECT_FIBRE:
7394                 un->un_interconnect_type = SD_INTERCONNECT_FIBRE;
7395                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7396                     "sd_unit_attach: un:0x%p SD_INTERCONNECT_FIBRE\n", un);
7397                 break;
7398         case INTERCONNECT_FABRIC:
7399                 un->un_interconnect_type = SD_INTERCONNECT_FABRIC;
7400                 un->un_node_type = DDI_NT_BLOCK_FABRIC;
7401                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7402                     "sd_unit_attach: un:0x%p SD_INTERCONNECT_FABRIC\n", un);
7403                 break;
7404         default:
7405 #ifdef SD_DEFAULT_INTERCONNECT_TYPE
7406                 /*
7407                  * The HBA does not support the "interconnect-type" property
7408                  * (or did not provide a recognized type).
7409                  *
7410                  * Note: This will be obsoleted when a single fibre channel
7411                  * and parallel scsi driver is delivered. In the meantime the
7412                  * interconnect type will be set to the platform default.If that
7413                  * type is not parallel SCSI, it means that we should be
7414                  * assuming "ssd" semantics. However, here this also means that
7415                  * the FC HBA is not supporting the "interconnect-type" property
7416                  * like we expect it to, so log this occurrence.
7417                  */
7418                 un->un_interconnect_type = SD_DEFAULT_INTERCONNECT_TYPE;
7419                 if (!SD_IS_PARALLEL_SCSI(un)) {
7420                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
7421                             "sd_unit_attach: un:0x%p Assuming "
7422                             "INTERCONNECT_FIBRE\n", un);
7423                 } else {
7424                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
7425                             "sd_unit_attach: un:0x%p Assuming "
7426                             "INTERCONNECT_PARALLEL\n", un);
7427                         un->un_f_is_fibre = FALSE;
7428                 }
7429 #else
7430                 /*
7431                  * Note: This source will be implemented when a single fibre
7432                  * channel and parallel scsi driver is delivered. The default
7433                  * will be to assume that if a device does not support the
7434                  * "interconnect-type" property it is a parallel SCSI HBA and
7435                  * we will set the interconnect type for parallel scsi.
7436                  */
7437                 un->un_interconnect_type = SD_INTERCONNECT_PARALLEL;
7438                 un->un_f_is_fibre = FALSE;
7439 #endif
7440                 break;
7441         }
7442 
7443         if (un->un_f_is_fibre == TRUE) {
7444                 if (scsi_ifgetcap(SD_ADDRESS(un), "scsi-version", 1) ==
7445                     SCSI_VERSION_3) {
7446                         switch (un->un_interconnect_type) {
7447                         case SD_INTERCONNECT_FIBRE:
7448                         case SD_INTERCONNECT_SSA:
7449                                 un->un_node_type = DDI_NT_BLOCK_WWN;
7450                                 break;
7451                         default:
7452                                 break;
7453                         }
7454                 }
7455         }
7456 
7457         /*
7458          * Initialize the Request Sense command for the target
7459          */
7460         if (sd_alloc_rqs(devp, un) != DDI_SUCCESS) {
7461                 goto alloc_rqs_failed;
7462         }
7463 
7464         /*
7465          * Set un_retry_count with SD_RETRY_COUNT, this is ok for Sparc
7466          * with separate binary for sd and ssd.
7467          *
7468          * x86 has 1 binary, un_retry_count is set base on connection type.
7469          * The hardcoded values will go away when Sparc uses 1 binary
7470          * for sd and ssd.  This hardcoded values need to match
7471          * SD_RETRY_COUNT in sddef.h
7472          * The value used is base on interconnect type.
7473          * fibre = 3, parallel = 5
7474          */
7475 #if defined(__i386) || defined(__amd64)
7476         un->un_retry_count = un->un_f_is_fibre ? 3 : 5;
7477 #else
7478         un->un_retry_count = SD_RETRY_COUNT;
7479 #endif
7480 
7481         /*
7482          * Set the per disk retry count to the default number of retries
7483          * for disks and CDROMs. This value can be overridden by the
7484          * disk property list or an entry in sd.conf.
7485          */
7486         un->un_notready_retry_count =
7487             ISCD(un) ? CD_NOT_READY_RETRY_COUNT(un)
7488             : DISK_NOT_READY_RETRY_COUNT(un);
7489 
7490         /*
7491          * Set the busy retry count to the default value of un_retry_count.
7492          * This can be overridden by entries in sd.conf or the device
7493          * config table.
7494          */
7495         un->un_busy_retry_count = un->un_retry_count;
7496 
7497         /*
7498          * Init the reset threshold for retries.  This number determines
7499          * how many retries must be performed before a reset can be issued
7500          * (for certain error conditions). This can be overridden by entries
7501          * in sd.conf or the device config table.
7502          */
7503         un->un_reset_retry_count = (un->un_retry_count / 2);
7504 
7505         /*
7506          * Set the victim_retry_count to the default un_retry_count
7507          */
7508         un->un_victim_retry_count = (2 * un->un_retry_count);
7509 
7510         /*
7511          * Set the reservation release timeout to the default value of
7512          * 5 seconds. This can be overridden by entries in ssd.conf or the
7513          * device config table.
7514          */
7515         un->un_reserve_release_time = 5;
7516 
7517         /*
7518          * Set up the default maximum transfer size. Note that this may
7519          * get updated later in the attach, when setting up default wide
7520          * operations for disks.
7521          */
7522 #if defined(__i386) || defined(__amd64)
7523         un->un_max_xfer_size = (uint_t)SD_DEFAULT_MAX_XFER_SIZE;
7524         un->un_partial_dma_supported = 1;
7525 #else
7526         un->un_max_xfer_size = (uint_t)maxphys;
7527 #endif
7528 
7529         /*
7530          * Get "allow bus device reset" property (defaults to "enabled" if
7531          * the property was not defined). This is to disable bus resets for
7532          * certain kinds of error recovery. Note: In the future when a run-time
7533          * fibre check is available the soft state flag should default to
7534          * enabled.
7535          */
7536         if (un->un_f_is_fibre == TRUE) {
7537                 un->un_f_allow_bus_device_reset = TRUE;
7538         } else {
7539                 if (ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS,
7540                     "allow-bus-device-reset", 1) != 0) {
7541                         un->un_f_allow_bus_device_reset = TRUE;
7542                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
7543                             "sd_unit_attach: un:0x%p Bus device reset "
7544                             "enabled\n", un);
7545                 } else {
7546                         un->un_f_allow_bus_device_reset = FALSE;
7547                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
7548                             "sd_unit_attach: un:0x%p Bus device reset "
7549                             "disabled\n", un);
7550                 }
7551         }
7552 
7553         /*
7554          * Check if this is an ATAPI device. ATAPI devices use Group 1
7555          * Read/Write commands and Group 2 Mode Sense/Select commands.
7556          *
7557          * Note: The "obsolete" way of doing this is to check for the "atapi"
7558          * property. The new "variant" property with a value of "atapi" has been
7559          * introduced so that future 'variants' of standard SCSI behavior (like
7560          * atapi) could be specified by the underlying HBA drivers by supplying
7561          * a new value for the "variant" property, instead of having to define a
7562          * new property.
7563          */
7564         if (ddi_prop_get_int(DDI_DEV_T_ANY, devi, 0, "atapi", -1) != -1) {
7565                 un->un_f_cfg_is_atapi = TRUE;
7566                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7567                     "sd_unit_attach: un:0x%p Atapi device\n", un);
7568         }
7569         if (ddi_prop_lookup_string(DDI_DEV_T_ANY, devi, 0, "variant",
7570             &variantp) == DDI_PROP_SUCCESS) {
7571                 if (strcmp(variantp, "atapi") == 0) {
7572                         un->un_f_cfg_is_atapi = TRUE;
7573                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
7574                             "sd_unit_attach: un:0x%p Atapi device\n", un);
7575                 }
7576                 ddi_prop_free(variantp);
7577         }
7578 
7579         un->un_cmd_timeout   = SD_IO_TIME;
7580 
7581         un->un_busy_timeout  = SD_BSY_TIMEOUT;
7582 
7583         /* Info on current states, statuses, etc. (Updated frequently) */
7584         un->un_state         = SD_STATE_NORMAL;
7585         un->un_last_state    = SD_STATE_NORMAL;
7586 
7587         /* Control & status info for command throttling */
7588         un->un_throttle              = sd_max_throttle;
7589         un->un_saved_throttle        = sd_max_throttle;
7590         un->un_min_throttle  = sd_min_throttle;
7591 
7592         if (un->un_f_is_fibre == TRUE) {
7593                 un->un_f_use_adaptive_throttle = TRUE;
7594         } else {
7595                 un->un_f_use_adaptive_throttle = FALSE;
7596         }
7597 
7598         /* Removable media support. */
7599         cv_init(&un->un_state_cv, NULL, CV_DRIVER, NULL);
7600         un->un_mediastate            = DKIO_NONE;
7601         un->un_specified_mediastate  = DKIO_NONE;
7602 
7603         /* CVs for suspend/resume (PM or DR) */
7604         cv_init(&un->un_suspend_cv,   NULL, CV_DRIVER, NULL);
7605         cv_init(&un->un_disk_busy_cv, NULL, CV_DRIVER, NULL);
7606 
7607         /* Power management support. */
7608         un->un_power_level = SD_SPINDLE_UNINIT;
7609 
7610         cv_init(&un->un_wcc_cv,   NULL, CV_DRIVER, NULL);
7611         un->un_f_wcc_inprog = 0;
7612 
7613         /*
7614          * The open/close semaphore is used to serialize threads executing
7615          * in the driver's open & close entry point routines for a given
7616          * instance.
7617          */
7618         (void) sema_init(&un->un_semoclose, 1, NULL, SEMA_DRIVER, NULL);
7619 
7620         /*
7621          * The conf file entry and softstate variable is a forceful override,
7622          * meaning a non-zero value must be entered to change the default.
7623          */
7624         un->un_f_disksort_disabled = FALSE;
7625         un->un_f_rmw_type = SD_RMW_TYPE_DEFAULT;
7626         un->un_f_enable_rmw = FALSE;
7627 
7628         /*
7629          * GET EVENT STATUS NOTIFICATION media polling enabled by default, but
7630          * can be overridden via [s]sd-config-list "mmc-gesn-polling" property.
7631          */
7632         un->un_f_mmc_gesn_polling = TRUE;
7633 
7634         /*
7635          * physical sector size defaults to DEV_BSIZE currently. We can
7636          * override this value via the driver configuration file so we must
7637          * set it before calling sd_read_unit_properties().
7638          */
7639         un->un_phy_blocksize = DEV_BSIZE;
7640 
7641         /*
7642          * Retrieve the properties from the static driver table or the driver
7643          * configuration file (.conf) for this unit and update the soft state
7644          * for the device as needed for the indicated properties.
7645          * Note: the property configuration needs to occur here as some of the
7646          * following routines may have dependencies on soft state flags set
7647          * as part of the driver property configuration.
7648          */
7649         sd_read_unit_properties(un);
7650         SD_TRACE(SD_LOG_ATTACH_DETACH, un,
7651             "sd_unit_attach: un:0x%p property configuration complete.\n", un);
7652 
7653         /*
7654          * Only if a device has "hotpluggable" property, it is
7655          * treated as hotpluggable device. Otherwise, it is
7656          * regarded as non-hotpluggable one.
7657          */
7658         if (ddi_prop_get_int(DDI_DEV_T_ANY, devi, 0, "hotpluggable",
7659             -1) != -1) {
7660                 un->un_f_is_hotpluggable = TRUE;
7661         }
7662 
7663         /*
7664          * set unit's attributes(flags) according to "hotpluggable" and
7665          * RMB bit in INQUIRY data.
7666          */
7667         sd_set_unit_attributes(un, devi);
7668 
7669         /*
7670          * By default, we mark the capacity, lbasize, and geometry
7671          * as invalid. Only if we successfully read a valid capacity
7672          * will we update the un_blockcount and un_tgt_blocksize with the
7673          * valid values (the geometry will be validated later).
7674          */
7675         un->un_f_blockcount_is_valid = FALSE;
7676         un->un_f_tgt_blocksize_is_valid      = FALSE;
7677 
7678         /*
7679          * Use DEV_BSIZE and DEV_BSHIFT as defaults, until we can determine
7680          * otherwise.
7681          */
7682         un->un_tgt_blocksize  = un->un_sys_blocksize  = DEV_BSIZE;
7683         un->un_blockcount = 0;
7684 
7685         /*
7686          * Set up the per-instance info needed to determine the correct
7687          * CDBs and other info for issuing commands to the target.
7688          */
7689         sd_init_cdb_limits(un);
7690 
7691         /*
7692          * Set up the IO chains to use, based upon the target type.
7693          */
7694         if (un->un_f_non_devbsize_supported) {
7695                 un->un_buf_chain_type = SD_CHAIN_INFO_RMMEDIA;
7696         } else {
7697                 un->un_buf_chain_type = SD_CHAIN_INFO_DISK;
7698         }
7699         un->un_uscsi_chain_type  = SD_CHAIN_INFO_USCSI_CMD;
7700         un->un_direct_chain_type = SD_CHAIN_INFO_DIRECT_CMD;
7701         un->un_priority_chain_type = SD_CHAIN_INFO_PRIORITY_CMD;
7702 
7703         un->un_xbuf_attr = ddi_xbuf_attr_create(sizeof (struct sd_xbuf),
7704             sd_xbuf_strategy, un, sd_xbuf_active_limit,  sd_xbuf_reserve_limit,
7705             ddi_driver_major(devi), DDI_XBUF_QTHREAD_DRIVER);
7706         ddi_xbuf_attr_register_devinfo(un->un_xbuf_attr, devi);
7707 
7708 
7709         if (ISCD(un)) {
7710                 un->un_additional_codes = sd_additional_codes;
7711         } else {
7712                 un->un_additional_codes = NULL;
7713         }
7714 
7715         /*
7716          * Create the kstats here so they can be available for attach-time
7717          * routines that send commands to the unit (either polled or via
7718          * sd_send_scsi_cmd).
7719          *
7720          * Note: This is a critical sequence that needs to be maintained:
7721          *      1) Instantiate the kstats here, before any routines using the
7722          *         iopath (i.e. sd_send_scsi_cmd).
7723          *      2) Instantiate and initialize the partition stats
7724          *         (sd_set_pstats).
7725          *      3) Initialize the error stats (sd_set_errstats), following
7726          *         sd_validate_geometry(),sd_register_devid(),
7727          *         and sd_cache_control().
7728          */
7729 
7730         un->un_stats = kstat_create(sd_label, instance,
7731             NULL, "disk", KSTAT_TYPE_IO, 1, KSTAT_FLAG_PERSISTENT);
7732         if (un->un_stats != NULL) {
7733                 un->un_stats->ks_lock = SD_MUTEX(un);
7734                 kstat_install(un->un_stats);
7735         }
7736         SD_TRACE(SD_LOG_ATTACH_DETACH, un,
7737             "sd_unit_attach: un:0x%p un_stats created\n", un);
7738 
7739         sd_create_errstats(un, instance);
7740         if (un->un_errstats == NULL) {
7741                 goto create_errstats_failed;
7742         }
7743         SD_TRACE(SD_LOG_ATTACH_DETACH, un,
7744             "sd_unit_attach: un:0x%p errstats created\n", un);
7745 
7746         /*
7747          * The following if/else code was relocated here from below as part
7748          * of the fix for bug (4430280). However with the default setup added
7749          * on entry to this routine, it's no longer absolutely necessary for
7750          * this to be before the call to sd_spin_up_unit.
7751          */
7752         if (SD_IS_PARALLEL_SCSI(un) || SD_IS_SERIAL(un)) {
7753                 int tq_trigger_flag = (((devp->sd_inq->inq_ansi == 4) ||
7754                     (devp->sd_inq->inq_ansi == 5)) &&
7755                     devp->sd_inq->inq_bque) || devp->sd_inq->inq_cmdque;
7756 
7757                 /*
7758                  * If tagged queueing is supported by the target
7759                  * and by the host adapter then we will enable it
7760                  */
7761                 un->un_tagflags = 0;
7762                 if ((devp->sd_inq->inq_rdf == RDF_SCSI2) && tq_trigger_flag &&
7763                     (un->un_f_arq_enabled == TRUE)) {
7764                         if (scsi_ifsetcap(SD_ADDRESS(un), "tagged-qing",
7765                             1, 1) == 1) {
7766                                 un->un_tagflags = FLAG_STAG;
7767                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7768                                     "sd_unit_attach: un:0x%p tag queueing "
7769                                     "enabled\n", un);
7770                         } else if (scsi_ifgetcap(SD_ADDRESS(un),
7771                             "untagged-qing", 0) == 1) {
7772                                 un->un_f_opt_queueing = TRUE;
7773                                 un->un_saved_throttle = un->un_throttle =
7774                                     min(un->un_throttle, 3);
7775                         } else {
7776                                 un->un_f_opt_queueing = FALSE;
7777                                 un->un_saved_throttle = un->un_throttle = 1;
7778                         }
7779                 } else if ((scsi_ifgetcap(SD_ADDRESS(un), "untagged-qing", 0)
7780                     == 1) && (un->un_f_arq_enabled == TRUE)) {
7781                         /* The Host Adapter supports internal queueing. */
7782                         un->un_f_opt_queueing = TRUE;
7783                         un->un_saved_throttle = un->un_throttle =
7784                             min(un->un_throttle, 3);
7785                 } else {
7786                         un->un_f_opt_queueing = FALSE;
7787                         un->un_saved_throttle = un->un_throttle = 1;
7788                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
7789                             "sd_unit_attach: un:0x%p no tag queueing\n", un);
7790                 }
7791 
7792                 /*
7793                  * Enable large transfers for SATA/SAS drives
7794                  */
7795                 if (SD_IS_SERIAL(un)) {
7796                         un->un_max_xfer_size =
7797                             ddi_getprop(DDI_DEV_T_ANY, devi, 0,
7798                             sd_max_xfer_size, SD_MAX_XFER_SIZE);
7799                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
7800                             "sd_unit_attach: un:0x%p max transfer "
7801                             "size=0x%x\n", un, un->un_max_xfer_size);
7802 
7803                 }
7804 
7805                 /* Setup or tear down default wide operations for disks */
7806 
7807                 /*
7808                  * Note: Legacy: it may be possible for both "sd_max_xfer_size"
7809                  * and "ssd_max_xfer_size" to exist simultaneously on the same
7810                  * system and be set to different values. In the future this
7811                  * code may need to be updated when the ssd module is
7812                  * obsoleted and removed from the system. (4299588)
7813                  */
7814                 if (SD_IS_PARALLEL_SCSI(un) &&
7815                     (devp->sd_inq->inq_rdf == RDF_SCSI2) &&
7816                     (devp->sd_inq->inq_wbus16 || devp->sd_inq->inq_wbus32)) {
7817                         if (scsi_ifsetcap(SD_ADDRESS(un), "wide-xfer",
7818                             1, 1) == 1) {
7819                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7820                                     "sd_unit_attach: un:0x%p Wide Transfer "
7821                                     "enabled\n", un);
7822                         }
7823 
7824                         /*
7825                          * If tagged queuing has also been enabled, then
7826                          * enable large xfers
7827                          */
7828                         if (un->un_saved_throttle == sd_max_throttle) {
7829                                 un->un_max_xfer_size =
7830                                     ddi_getprop(DDI_DEV_T_ANY, devi, 0,
7831                                     sd_max_xfer_size, SD_MAX_XFER_SIZE);
7832                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7833                                     "sd_unit_attach: un:0x%p max transfer "
7834                                     "size=0x%x\n", un, un->un_max_xfer_size);
7835                         }
7836                 } else {
7837                         if (scsi_ifsetcap(SD_ADDRESS(un), "wide-xfer",
7838                             0, 1) == 1) {
7839                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
7840                                     "sd_unit_attach: un:0x%p "
7841                                     "Wide Transfer disabled\n", un);
7842                         }
7843                 }
7844         } else {
7845                 un->un_tagflags = FLAG_STAG;
7846                 un->un_max_xfer_size = ddi_getprop(DDI_DEV_T_ANY,
7847                     devi, 0, sd_max_xfer_size, SD_MAX_XFER_SIZE);
7848         }
7849 
7850         /*
7851          * If this target supports LUN reset, try to enable it.
7852          */
7853         if (un->un_f_lun_reset_enabled) {
7854                 if (scsi_ifsetcap(SD_ADDRESS(un), "lun-reset", 1, 1) == 1) {
7855                         SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_unit_attach: "
7856                             "un:0x%p lun_reset capability set\n", un);
7857                 } else {
7858                         SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_unit_attach: "
7859                             "un:0x%p lun-reset capability not set\n", un);
7860                 }
7861         }
7862 
7863         /*
7864          * Adjust the maximum transfer size. This is to fix
7865          * the problem of partial DMA support on SPARC. Some
7866          * HBA driver, like aac, has very small dma_attr_maxxfer
7867          * size, which requires partial DMA support on SPARC.
7868          * In the future the SPARC pci nexus driver may solve
7869          * the problem instead of this fix.
7870          */
7871         max_xfer_size = scsi_ifgetcap(SD_ADDRESS(un), "dma-max", 1);
7872         if ((max_xfer_size > 0) && (max_xfer_size < un->un_max_xfer_size)) {
7873                 /* We need DMA partial even on sparc to ensure sddump() works */
7874                 un->un_max_xfer_size = max_xfer_size;
7875                 if (un->un_partial_dma_supported == 0)
7876                         un->un_partial_dma_supported = 1;
7877         }
7878         if (ddi_prop_get_int(DDI_DEV_T_ANY, SD_DEVINFO(un),
7879             DDI_PROP_DONTPASS, "buf_break", 0) == 1) {
7880                 if (ddi_xbuf_attr_setup_brk(un->un_xbuf_attr,
7881                     un->un_max_xfer_size) == 1) {
7882                         un->un_buf_breakup_supported = 1;
7883                         SD_INFO(SD_LOG_ATTACH_DETACH, un, "sd_unit_attach: "
7884                             "un:0x%p Buf breakup enabled\n", un);
7885                 }
7886         }
7887 
7888         /*
7889          * Set PKT_DMA_PARTIAL flag.
7890          */
7891         if (un->un_partial_dma_supported == 1) {
7892                 un->un_pkt_flags = PKT_DMA_PARTIAL;
7893         } else {
7894                 un->un_pkt_flags = 0;
7895         }
7896 
7897         /* Initialize sd_ssc_t for internal uscsi commands */
7898         ssc = sd_ssc_init(un);
7899         scsi_fm_init(devp);
7900 
7901         /*
7902          * Allocate memory for SCSI FMA stuffs.
7903          */
7904         un->un_fm_private =
7905             kmem_zalloc(sizeof (struct sd_fm_internal), KM_SLEEP);
7906         sfip = (struct sd_fm_internal *)un->un_fm_private;
7907         sfip->fm_ssc.ssc_uscsi_cmd = &sfip->fm_ucmd;
7908         sfip->fm_ssc.ssc_uscsi_info = &sfip->fm_uinfo;
7909         sfip->fm_ssc.ssc_un = un;
7910 
7911         if (ISCD(un) ||
7912             un->un_f_has_removable_media ||
7913             devp->sd_fm_capable == DDI_FM_NOT_CAPABLE) {
7914                 /*
7915                  * We don't touch CDROM or the DDI_FM_NOT_CAPABLE device.
7916                  * Their log are unchanged.
7917                  */
7918                 sfip->fm_log_level = SD_FM_LOG_NSUP;
7919         } else {
7920                 /*
7921                  * If enter here, it should be non-CDROM and FM-capable
7922                  * device, and it will not keep the old scsi_log as before
7923                  * in /var/adm/messages. However, the property
7924                  * "fm-scsi-log" will control whether the FM telemetry will
7925                  * be logged in /var/adm/messages.
7926                  */
7927                 int fm_scsi_log;
7928                 fm_scsi_log = ddi_prop_get_int(DDI_DEV_T_ANY, SD_DEVINFO(un),
7929                     DDI_PROP_DONTPASS | DDI_PROP_NOTPROM, "fm-scsi-log", 0);
7930 
7931                 if (fm_scsi_log)
7932                         sfip->fm_log_level = SD_FM_LOG_EREPORT;
7933                 else
7934                         sfip->fm_log_level = SD_FM_LOG_SILENT;
7935         }
7936 
7937         /*
7938          * At this point in the attach, we have enough info in the
7939          * soft state to be able to issue commands to the target.
7940          *
7941          * All command paths used below MUST issue their commands as
7942          * SD_PATH_DIRECT. This is important as intermediate layers
7943          * are not all initialized yet (such as PM).
7944          */
7945 
7946         /*
7947          * Send a TEST UNIT READY command to the device. This should clear
7948          * any outstanding UNIT ATTENTION that may be present.
7949          *
7950          * Note: Don't check for success, just track if there is a reservation,
7951          * this is a throw away command to clear any unit attentions.
7952          *
7953          * Note: This MUST be the first command issued to the target during
7954          * attach to ensure power on UNIT ATTENTIONS are cleared.
7955          * Pass in flag SD_DONT_RETRY_TUR to prevent the long delays associated
7956          * with attempts at spinning up a device with no media.
7957          */
7958         status = sd_send_scsi_TEST_UNIT_READY(ssc, SD_DONT_RETRY_TUR);
7959         if (status != 0) {
7960                 if (status == EACCES)
7961                         reservation_flag = SD_TARGET_IS_RESERVED;
7962                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
7963         }
7964 
7965         /*
7966          * If the device is NOT a removable media device, attempt to spin
7967          * it up (using the START_STOP_UNIT command) and read its capacity
7968          * (using the READ CAPACITY command).  Note, however, that either
7969          * of these could fail and in some cases we would continue with
7970          * the attach despite the failure (see below).
7971          */
7972         if (un->un_f_descr_format_supported) {
7973 
7974                 switch (sd_spin_up_unit(ssc)) {
7975                 case 0:
7976                         /*
7977                          * Spin-up was successful; now try to read the
7978                          * capacity.  If successful then save the results
7979                          * and mark the capacity & lbasize as valid.
7980                          */
7981                         SD_TRACE(SD_LOG_ATTACH_DETACH, un,
7982                             "sd_unit_attach: un:0x%p spin-up successful\n", un);
7983 
7984                         status = sd_send_scsi_READ_CAPACITY(ssc, &capacity,
7985                             &lbasize, SD_PATH_DIRECT);
7986 
7987                         switch (status) {
7988                         case 0: {
7989                                 if (capacity > DK_MAX_BLOCKS) {
7990 #ifdef _LP64
7991                                         if ((capacity + 1) >
7992                                             SD_GROUP1_MAX_ADDRESS) {
7993                                                 /*
7994                                                  * Enable descriptor format
7995                                                  * sense data so that we can
7996                                                  * get 64 bit sense data
7997                                                  * fields.
7998                                                  */
7999                                                 sd_enable_descr_sense(ssc);
8000                                         }
8001 #else
8002                                         /* 32-bit kernels can't handle this */
8003                                         scsi_log(SD_DEVINFO(un),
8004                                             sd_label, CE_WARN,
8005                                             "disk has %llu blocks, which "
8006                                             "is too large for a 32-bit "
8007                                             "kernel", capacity);
8008 
8009 #if defined(__i386) || defined(__amd64)
8010                                         /*
8011                                          * 1TB disk was treated as (1T - 512)B
8012                                          * in the past, so that it might have
8013                                          * valid VTOC and solaris partitions,
8014                                          * we have to allow it to continue to
8015                                          * work.
8016                                          */
8017                                         if (capacity -1 > DK_MAX_BLOCKS)
8018 #endif
8019                                         goto spinup_failed;
8020 #endif
8021                                 }
8022 
8023                                 /*
8024                                  * Here it's not necessary to check the case:
8025                                  * the capacity of the device is bigger than
8026                                  * what the max hba cdb can support. Because
8027                                  * sd_send_scsi_READ_CAPACITY will retrieve
8028                                  * the capacity by sending USCSI command, which
8029                                  * is constrained by the max hba cdb. Actually,
8030                                  * sd_send_scsi_READ_CAPACITY will return
8031                                  * EINVAL when using bigger cdb than required
8032                                  * cdb length. Will handle this case in
8033                                  * "case EINVAL".
8034                                  */
8035 
8036                                 /*
8037                                  * The following relies on
8038                                  * sd_send_scsi_READ_CAPACITY never
8039                                  * returning 0 for capacity and/or lbasize.
8040                                  */
8041                                 sd_update_block_info(un, lbasize, capacity);
8042 
8043                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
8044                                     "sd_unit_attach: un:0x%p capacity = %ld "
8045                                     "blocks; lbasize= %ld.\n", un,
8046                                     un->un_blockcount, un->un_tgt_blocksize);
8047 
8048                                 break;
8049                         }
8050                         case EINVAL:
8051                                 /*
8052                                  * In the case where the max-cdb-length property
8053                                  * is smaller than the required CDB length for
8054                                  * a SCSI device, a target driver can fail to
8055                                  * attach to that device.
8056                                  */
8057                                 scsi_log(SD_DEVINFO(un),
8058                                     sd_label, CE_WARN,
8059                                     "disk capacity is too large "
8060                                     "for current cdb length");
8061                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
8062 
8063                                 goto spinup_failed;
8064                         case EACCES:
8065                                 /*
8066                                  * Should never get here if the spin-up
8067                                  * succeeded, but code it in anyway.
8068                                  * From here, just continue with the attach...
8069                                  */
8070                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
8071                                     "sd_unit_attach: un:0x%p "
8072                                     "sd_send_scsi_READ_CAPACITY "
8073                                     "returned reservation conflict\n", un);
8074                                 reservation_flag = SD_TARGET_IS_RESERVED;
8075                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
8076                                 break;
8077                         default:
8078                                 /*
8079                                  * Likewise, should never get here if the
8080                                  * spin-up succeeded. Just continue with
8081                                  * the attach...
8082                                  */
8083                                 if (status == EIO)
8084                                         sd_ssc_assessment(ssc,
8085                                             SD_FMT_STATUS_CHECK);
8086                                 else
8087                                         sd_ssc_assessment(ssc,
8088                                             SD_FMT_IGNORE);
8089                                 break;
8090                         }
8091                         break;
8092                 case EACCES:
8093                         /*
8094                          * Device is reserved by another host.  In this case
8095                          * we could not spin it up or read the capacity, but
8096                          * we continue with the attach anyway.
8097                          */
8098                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
8099                             "sd_unit_attach: un:0x%p spin-up reservation "
8100                             "conflict.\n", un);
8101                         reservation_flag = SD_TARGET_IS_RESERVED;
8102                         break;
8103                 default:
8104                         /* Fail the attach if the spin-up failed. */
8105                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
8106                             "sd_unit_attach: un:0x%p spin-up failed.", un);
8107                         goto spinup_failed;
8108                 }
8109 
8110         }
8111 
8112         /*
8113          * Check to see if this is a MMC drive
8114          */
8115         if (ISCD(un)) {
8116                 sd_set_mmc_caps(ssc);
8117         }
8118 
8119         /*
8120          * Add a zero-length attribute to tell the world we support
8121          * kernel ioctls (for layered drivers)
8122          */
8123         (void) ddi_prop_create(DDI_DEV_T_NONE, devi, DDI_PROP_CANSLEEP,
8124             DDI_KERNEL_IOCTL, NULL, 0);
8125 
8126         /*
8127          * Add a boolean property to tell the world we support
8128          * the B_FAILFAST flag (for layered drivers)
8129          */
8130         (void) ddi_prop_create(DDI_DEV_T_NONE, devi, DDI_PROP_CANSLEEP,
8131             "ddi-failfast-supported", NULL, 0);
8132 
8133         /*
8134          * Initialize power management
8135          */
8136         mutex_init(&un->un_pm_mutex, NULL, MUTEX_DRIVER, NULL);
8137         cv_init(&un->un_pm_busy_cv, NULL, CV_DRIVER, NULL);
8138         sd_setup_pm(ssc, devi);
8139         if (un->un_f_pm_is_enabled == FALSE) {
8140                 /*
8141                  * For performance, point to a jump table that does
8142                  * not include pm.
8143                  * The direct and priority chains don't change with PM.
8144                  *
8145                  * Note: this is currently done based on individual device
8146                  * capabilities. When an interface for determining system
8147                  * power enabled state becomes available, or when additional
8148                  * layers are added to the command chain, these values will
8149                  * have to be re-evaluated for correctness.
8150                  */
8151                 if (un->un_f_non_devbsize_supported) {
8152                         un->un_buf_chain_type = SD_CHAIN_INFO_RMMEDIA_NO_PM;
8153                 } else {
8154                         un->un_buf_chain_type = SD_CHAIN_INFO_DISK_NO_PM;
8155                 }
8156                 un->un_uscsi_chain_type  = SD_CHAIN_INFO_USCSI_CMD_NO_PM;
8157         }
8158 
8159         /*
8160          * This property is set to 0 by HA software to avoid retries
8161          * on a reserved disk. (The preferred property name is
8162          * "retry-on-reservation-conflict") (1189689)
8163          *
8164          * Note: The use of a global here can have unintended consequences. A
8165          * per instance variable is preferable to match the capabilities of
8166          * different underlying hba's (4402600)
8167          */
8168         sd_retry_on_reservation_conflict = ddi_getprop(DDI_DEV_T_ANY, devi,
8169             DDI_PROP_DONTPASS, "retry-on-reservation-conflict",
8170             sd_retry_on_reservation_conflict);
8171         if (sd_retry_on_reservation_conflict != 0) {
8172                 sd_retry_on_reservation_conflict = ddi_getprop(DDI_DEV_T_ANY,
8173                     devi, DDI_PROP_DONTPASS, sd_resv_conflict_name,
8174                     sd_retry_on_reservation_conflict);
8175         }
8176 
8177         /* Set up options for QFULL handling. */
8178         if ((rval = ddi_getprop(DDI_DEV_T_ANY, devi, 0,
8179             "qfull-retries", -1)) != -1) {
8180                 (void) scsi_ifsetcap(SD_ADDRESS(un), "qfull-retries",
8181                     rval, 1);
8182         }
8183         if ((rval = ddi_getprop(DDI_DEV_T_ANY, devi, 0,
8184             "qfull-retry-interval", -1)) != -1) {
8185                 (void) scsi_ifsetcap(SD_ADDRESS(un), "qfull-retry-interval",
8186                     rval, 1);
8187         }
8188 
8189         /*
8190          * This just prints a message that announces the existence of the
8191          * device. The message is always printed in the system logfile, but
8192          * only appears on the console if the system is booted with the
8193          * -v (verbose) argument.
8194          */
8195         ddi_report_dev(devi);
8196 
8197         un->un_mediastate = DKIO_NONE;
8198 
8199         /*
8200          * Check if this is a SSD(Solid State Drive).
8201          */
8202         sd_check_solid_state(ssc);
8203 
8204         /*
8205          * Check whether the drive is in emulation mode.
8206          */
8207         sd_check_emulation_mode(ssc);
8208 
8209         cmlb_alloc_handle(&un->un_cmlbhandle);
8210 
8211 #if defined(__i386) || defined(__amd64)
8212         /*
8213          * On x86, compensate for off-by-1 legacy error
8214          */
8215         if (!un->un_f_has_removable_media && !un->un_f_is_hotpluggable &&
8216             (lbasize == un->un_sys_blocksize))
8217                 offbyone = CMLB_OFF_BY_ONE;
8218 #endif
8219 
8220         if (cmlb_attach(devi, &sd_tgops, (int)devp->sd_inq->inq_dtype,
8221             VOID2BOOLEAN(un->un_f_has_removable_media != 0),
8222             VOID2BOOLEAN(un->un_f_is_hotpluggable != 0),
8223             un->un_node_type, offbyone, un->un_cmlbhandle,
8224             (void *)SD_PATH_DIRECT) != 0) {
8225                 goto cmlb_attach_failed;
8226         }
8227 
8228 
8229         /*
8230          * Read and validate the device's geometry (ie, disk label)
8231          * A new unformatted drive will not have a valid geometry, but
8232          * the driver needs to successfully attach to this device so
8233          * the drive can be formatted via ioctls.
8234          */
8235         geom_label_valid = (cmlb_validate(un->un_cmlbhandle, 0,
8236             (void *)SD_PATH_DIRECT) == 0) ? 1: 0;
8237 
8238         mutex_enter(SD_MUTEX(un));
8239 
8240         /*
8241          * Read and initialize the devid for the unit.
8242          */
8243         if (un->un_f_devid_supported) {
8244                 sd_register_devid(ssc, devi, reservation_flag);
8245         }
8246         mutex_exit(SD_MUTEX(un));
8247 
8248 #if (defined(__fibre))
8249         /*
8250          * Register callbacks for fibre only.  You can't do this solely
8251          * on the basis of the devid_type because this is hba specific.
8252          * We need to query our hba capabilities to find out whether to
8253          * register or not.
8254          */
8255         if (un->un_f_is_fibre) {
8256                 if (strcmp(un->un_node_type, DDI_NT_BLOCK_CHAN)) {
8257                         sd_init_event_callbacks(un);
8258                         SD_TRACE(SD_LOG_ATTACH_DETACH, un,
8259                             "sd_unit_attach: un:0x%p event callbacks inserted",
8260                             un);
8261                 }
8262         }
8263 #endif
8264 
8265         if (un->un_f_opt_disable_cache == TRUE) {
8266                 /*
8267                  * Disable both read cache and write cache.  This is
8268                  * the historic behavior of the keywords in the config file.
8269                  */
8270                 if (sd_cache_control(ssc, SD_CACHE_DISABLE, SD_CACHE_DISABLE) !=
8271                     0) {
8272                         SD_ERROR(SD_LOG_ATTACH_DETACH, un,
8273                             "sd_unit_attach: un:0x%p Could not disable "
8274                             "caching", un);
8275                         goto devid_failed;
8276                 }
8277         }
8278 
8279         /*
8280          * Check the value of the WCE bit now and
8281          * set un_f_write_cache_enabled accordingly.
8282          */
8283         (void) sd_get_write_cache_enabled(ssc, &wc_enabled);
8284         mutex_enter(SD_MUTEX(un));
8285         un->un_f_write_cache_enabled = (wc_enabled != 0);
8286         mutex_exit(SD_MUTEX(un));
8287 
8288         if ((un->un_f_rmw_type != SD_RMW_TYPE_RETURN_ERROR &&
8289             un->un_tgt_blocksize != DEV_BSIZE) ||
8290             un->un_f_enable_rmw) {
8291                 if (!(un->un_wm_cache)) {
8292                         (void) snprintf(name_str, sizeof (name_str),
8293                             "%s%d_cache",
8294                             ddi_driver_name(SD_DEVINFO(un)),
8295                             ddi_get_instance(SD_DEVINFO(un)));
8296                         un->un_wm_cache = kmem_cache_create(
8297                             name_str, sizeof (struct sd_w_map),
8298                             8, sd_wm_cache_constructor,
8299                             sd_wm_cache_destructor, NULL,
8300                             (void *)un, NULL, 0);
8301                         if (!(un->un_wm_cache)) {
8302                                 goto wm_cache_failed;
8303                         }
8304                 }
8305         }
8306 
8307         /*
8308          * Check the value of the NV_SUP bit and set
8309          * un_f_suppress_cache_flush accordingly.
8310          */
8311         sd_get_nv_sup(ssc);
8312 
8313         /*
8314          * Find out what type of reservation this disk supports.
8315          */
8316         status = sd_send_scsi_PERSISTENT_RESERVE_IN(ssc, SD_READ_KEYS, 0, NULL);
8317 
8318         switch (status) {
8319         case 0:
8320                 /*
8321                  * SCSI-3 reservations are supported.
8322                  */
8323                 un->un_reservation_type = SD_SCSI3_RESERVATION;
8324                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
8325                     "sd_unit_attach: un:0x%p SCSI-3 reservations\n", un);
8326                 break;
8327         case ENOTSUP:
8328                 /*
8329                  * The PERSISTENT RESERVE IN command would not be recognized by
8330                  * a SCSI-2 device, so assume the reservation type is SCSI-2.
8331                  */
8332                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
8333                     "sd_unit_attach: un:0x%p SCSI-2 reservations\n", un);
8334                 un->un_reservation_type = SD_SCSI2_RESERVATION;
8335 
8336                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
8337                 break;
8338         default:
8339                 /*
8340                  * default to SCSI-3 reservations
8341                  */
8342                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
8343                     "sd_unit_attach: un:0x%p default SCSI3 reservations\n", un);
8344                 un->un_reservation_type = SD_SCSI3_RESERVATION;
8345 
8346                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
8347                 break;
8348         }
8349 
8350         /*
8351          * Set the pstat and error stat values here, so data obtained during the
8352          * previous attach-time routines is available.
8353          *
8354          * Note: This is a critical sequence that needs to be maintained:
8355          *      1) Instantiate the kstats before any routines using the iopath
8356          *         (i.e. sd_send_scsi_cmd).
8357          *      2) Initialize the error stats (sd_set_errstats) and partition
8358          *         stats (sd_set_pstats)here, following
8359          *         cmlb_validate_geometry(), sd_register_devid(), and
8360          *         sd_cache_control().
8361          */
8362 
8363         if (un->un_f_pkstats_enabled && geom_label_valid) {
8364                 sd_set_pstats(un);
8365                 SD_TRACE(SD_LOG_IO_PARTITION, un,
8366                     "sd_unit_attach: un:0x%p pstats created and set\n", un);
8367         }
8368 
8369         sd_set_errstats(un);
8370         SD_TRACE(SD_LOG_ATTACH_DETACH, un,
8371             "sd_unit_attach: un:0x%p errstats set\n", un);
8372 
8373 
8374         /*
8375          * After successfully attaching an instance, we record the information
8376          * of how many luns have been attached on the relative target and
8377          * controller for parallel SCSI. This information is used when sd tries
8378          * to set the tagged queuing capability in HBA.
8379          */
8380         if (SD_IS_PARALLEL_SCSI(un) && (tgt >= 0) && (tgt < NTARGETS_WIDE)) {
8381                 sd_scsi_update_lun_on_target(pdip, tgt, SD_SCSI_LUN_ATTACH);
8382         }
8383 
8384         SD_TRACE(SD_LOG_ATTACH_DETACH, un,
8385             "sd_unit_attach: un:0x%p exit success\n", un);
8386 
8387         /* Uninitialize sd_ssc_t pointer */
8388         sd_ssc_fini(ssc);
8389 
8390         return (DDI_SUCCESS);
8391 
8392         /*
8393          * An error occurred during the attach; clean up & return failure.
8394          */
8395 wm_cache_failed:
8396 devid_failed:
8397 
8398 setup_pm_failed:
8399         ddi_remove_minor_node(devi, NULL);
8400 
8401 cmlb_attach_failed:
8402         /*
8403          * Cleanup from the scsi_ifsetcap() calls (437868)
8404          */
8405         (void) scsi_ifsetcap(SD_ADDRESS(un), "lun-reset", 0, 1);
8406         (void) scsi_ifsetcap(SD_ADDRESS(un), "wide-xfer", 0, 1);
8407 
8408         /*
8409          * Refer to the comments of setting tagged-qing in the beginning of
8410          * sd_unit_attach. We can only disable tagged queuing when there is
8411          * no lun attached on the target.
8412          */
8413         if (sd_scsi_get_target_lun_count(pdip, tgt) < 1) {
8414                 (void) scsi_ifsetcap(SD_ADDRESS(un), "tagged-qing", 0, 1);
8415         }
8416 
8417         if (un->un_f_is_fibre == FALSE) {
8418                 (void) scsi_ifsetcap(SD_ADDRESS(un), "auto-rqsense", 0, 1);
8419         }
8420 
8421 spinup_failed:
8422 
8423         /* Uninitialize sd_ssc_t pointer */
8424         sd_ssc_fini(ssc);
8425 
8426         mutex_enter(SD_MUTEX(un));
8427 
8428         /* Deallocate SCSI FMA memory spaces */
8429         kmem_free(un->un_fm_private, sizeof (struct sd_fm_internal));
8430 
8431         /* Cancel callback for SD_PATH_DIRECT_PRIORITY cmd. restart */
8432         if (un->un_direct_priority_timeid != NULL) {
8433                 timeout_id_t temp_id = un->un_direct_priority_timeid;
8434                 un->un_direct_priority_timeid = NULL;
8435                 mutex_exit(SD_MUTEX(un));
8436                 (void) untimeout(temp_id);
8437                 mutex_enter(SD_MUTEX(un));
8438         }
8439 
8440         /* Cancel any pending start/stop timeouts */
8441         if (un->un_startstop_timeid != NULL) {
8442                 timeout_id_t temp_id = un->un_startstop_timeid;
8443                 un->un_startstop_timeid = NULL;
8444                 mutex_exit(SD_MUTEX(un));
8445                 (void) untimeout(temp_id);
8446                 mutex_enter(SD_MUTEX(un));
8447         }
8448 
8449         /* Cancel any pending reset-throttle timeouts */
8450         if (un->un_reset_throttle_timeid != NULL) {
8451                 timeout_id_t temp_id = un->un_reset_throttle_timeid;
8452                 un->un_reset_throttle_timeid = NULL;
8453                 mutex_exit(SD_MUTEX(un));
8454                 (void) untimeout(temp_id);
8455                 mutex_enter(SD_MUTEX(un));
8456         }
8457 
8458         /* Cancel rmw warning message timeouts */
8459         if (un->un_rmw_msg_timeid != NULL) {
8460                 timeout_id_t temp_id = un->un_rmw_msg_timeid;
8461                 un->un_rmw_msg_timeid = NULL;
8462                 mutex_exit(SD_MUTEX(un));
8463                 (void) untimeout(temp_id);
8464                 mutex_enter(SD_MUTEX(un));
8465         }
8466 
8467         /* Cancel any pending retry timeouts */
8468         if (un->un_retry_timeid != NULL) {
8469                 timeout_id_t temp_id = un->un_retry_timeid;
8470                 un->un_retry_timeid = NULL;
8471                 mutex_exit(SD_MUTEX(un));
8472                 (void) untimeout(temp_id);
8473                 mutex_enter(SD_MUTEX(un));
8474         }
8475 
8476         /* Cancel any pending delayed cv broadcast timeouts */
8477         if (un->un_dcvb_timeid != NULL) {
8478                 timeout_id_t temp_id = un->un_dcvb_timeid;
8479                 un->un_dcvb_timeid = NULL;
8480                 mutex_exit(SD_MUTEX(un));
8481                 (void) untimeout(temp_id);
8482                 mutex_enter(SD_MUTEX(un));
8483         }
8484 
8485         mutex_exit(SD_MUTEX(un));
8486 
8487         /* There should not be any in-progress I/O so ASSERT this check */
8488         ASSERT(un->un_ncmds_in_transport == 0);
8489         ASSERT(un->un_ncmds_in_driver == 0);
8490 
8491         /* Do not free the softstate if the callback routine is active */
8492         sd_sync_with_callback(un);
8493 
8494         /*
8495          * Partition stats apparently are not used with removables. These would
8496          * not have been created during attach, so no need to clean them up...
8497          */
8498         if (un->un_errstats != NULL) {
8499                 kstat_delete(un->un_errstats);
8500                 un->un_errstats = NULL;
8501         }
8502 
8503 create_errstats_failed:
8504 
8505         if (un->un_stats != NULL) {
8506                 kstat_delete(un->un_stats);
8507                 un->un_stats = NULL;
8508         }
8509 
8510         ddi_xbuf_attr_unregister_devinfo(un->un_xbuf_attr, devi);
8511         ddi_xbuf_attr_destroy(un->un_xbuf_attr);
8512 
8513         ddi_prop_remove_all(devi);
8514         sema_destroy(&un->un_semoclose);
8515         cv_destroy(&un->un_state_cv);
8516 
8517 getrbuf_failed:
8518 
8519         sd_free_rqs(un);
8520 
8521 alloc_rqs_failed:
8522 
8523         devp->sd_private = NULL;
8524         bzero(un, sizeof (struct sd_lun));      /* Clear any stale data! */
8525 
8526 get_softstate_failed:
8527         /*
8528          * Note: the man pages are unclear as to whether or not doing a
8529          * ddi_soft_state_free(sd_state, instance) is the right way to
8530          * clean up after the ddi_soft_state_zalloc() if the subsequent
8531          * ddi_get_soft_state() fails.  The implication seems to be
8532          * that the get_soft_state cannot fail if the zalloc succeeds.
8533          */
8534 #ifndef XPV_HVM_DRIVER
8535         ddi_soft_state_free(sd_state, instance);
8536 #endif /* !XPV_HVM_DRIVER */
8537 
8538 probe_failed:
8539         scsi_unprobe(devp);
8540 
8541         return (DDI_FAILURE);
8542 }
8543 
8544 
8545 /*
8546  *    Function: sd_unit_detach
8547  *
8548  * Description: Performs DDI_DETACH processing for sddetach().
8549  *
8550  * Return Code: DDI_SUCCESS
8551  *              DDI_FAILURE
8552  *
8553  *     Context: Kernel thread context
8554  */
8555 
8556 static int
8557 sd_unit_detach(dev_info_t *devi)
8558 {
8559         struct scsi_device      *devp;
8560         struct sd_lun           *un;
8561         int                     i;
8562         int                     tgt;
8563         dev_t                   dev;
8564         dev_info_t              *pdip = ddi_get_parent(devi);
8565 #ifndef XPV_HVM_DRIVER
8566         int                     instance = ddi_get_instance(devi);
8567 #endif /* !XPV_HVM_DRIVER */
8568 
8569         mutex_enter(&sd_detach_mutex);
8570 
8571         /*
8572          * Fail the detach for any of the following:
8573          *  - Unable to get the sd_lun struct for the instance
8574          *  - A layered driver has an outstanding open on the instance
8575          *  - Another thread is already detaching this instance
8576          *  - Another thread is currently performing an open
8577          */
8578         devp = ddi_get_driver_private(devi);
8579         if ((devp == NULL) ||
8580             ((un = (struct sd_lun *)devp->sd_private) == NULL) ||
8581             (un->un_ncmds_in_driver != 0) || (un->un_layer_count != 0) ||
8582             (un->un_detach_count != 0) || (un->un_opens_in_progress != 0)) {
8583                 mutex_exit(&sd_detach_mutex);
8584                 return (DDI_FAILURE);
8585         }
8586 
8587         SD_TRACE(SD_LOG_ATTACH_DETACH, un, "sd_unit_detach: entry 0x%p\n", un);
8588 
8589         /*
8590          * Mark this instance as currently in a detach, to inhibit any
8591          * opens from a layered driver.
8592          */
8593         un->un_detach_count++;
8594         mutex_exit(&sd_detach_mutex);
8595 
8596         tgt = ddi_prop_get_int(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS,
8597             SCSI_ADDR_PROP_TARGET, -1);
8598 
8599         dev = sd_make_device(SD_DEVINFO(un));
8600 
8601 #ifndef lint
8602         _NOTE(COMPETING_THREADS_NOW);
8603 #endif
8604 
8605         mutex_enter(SD_MUTEX(un));
8606 
8607         /*
8608          * Fail the detach if there are any outstanding layered
8609          * opens on this device.
8610          */
8611         for (i = 0; i < NDKMAP; i++) {
8612                 if (un->un_ocmap.lyropen[i] != 0) {
8613                         goto err_notclosed;
8614                 }
8615         }
8616 
8617         /*
8618          * Verify there are NO outstanding commands issued to this device.
8619          * ie, un_ncmds_in_transport == 0.
8620          * It's possible to have outstanding commands through the physio
8621          * code path, even though everything's closed.
8622          */
8623         if ((un->un_ncmds_in_transport != 0) || (un->un_retry_timeid != NULL) ||
8624             (un->un_direct_priority_timeid != NULL) ||
8625             (un->un_state == SD_STATE_RWAIT)) {
8626                 mutex_exit(SD_MUTEX(un));
8627                 SD_ERROR(SD_LOG_ATTACH_DETACH, un,
8628                     "sd_dr_detach: Detach failure due to outstanding cmds\n");
8629                 goto err_stillbusy;
8630         }
8631 
8632         /*
8633          * If we have the device reserved, release the reservation.
8634          */
8635         if ((un->un_resvd_status & SD_RESERVE) &&
8636             !(un->un_resvd_status & SD_LOST_RESERVE)) {
8637                 mutex_exit(SD_MUTEX(un));
8638                 /*
8639                  * Note: sd_reserve_release sends a command to the device
8640                  * via the sd_ioctlcmd() path, and can sleep.
8641                  */
8642                 if (sd_reserve_release(dev, SD_RELEASE) != 0) {
8643                         SD_ERROR(SD_LOG_ATTACH_DETACH, un,
8644                             "sd_dr_detach: Cannot release reservation \n");
8645                 }
8646         } else {
8647                 mutex_exit(SD_MUTEX(un));
8648         }
8649 
8650         /*
8651          * Untimeout any reserve recover, throttle reset, restart unit
8652          * and delayed broadcast timeout threads. Protect the timeout pointer
8653          * from getting nulled by their callback functions.
8654          */
8655         mutex_enter(SD_MUTEX(un));
8656         if (un->un_resvd_timeid != NULL) {
8657                 timeout_id_t temp_id = un->un_resvd_timeid;
8658                 un->un_resvd_timeid = NULL;
8659                 mutex_exit(SD_MUTEX(un));
8660                 (void) untimeout(temp_id);
8661                 mutex_enter(SD_MUTEX(un));
8662         }
8663 
8664         if (un->un_reset_throttle_timeid != NULL) {
8665                 timeout_id_t temp_id = un->un_reset_throttle_timeid;
8666                 un->un_reset_throttle_timeid = NULL;
8667                 mutex_exit(SD_MUTEX(un));
8668                 (void) untimeout(temp_id);
8669                 mutex_enter(SD_MUTEX(un));
8670         }
8671 
8672         if (un->un_startstop_timeid != NULL) {
8673                 timeout_id_t temp_id = un->un_startstop_timeid;
8674                 un->un_startstop_timeid = NULL;
8675                 mutex_exit(SD_MUTEX(un));
8676                 (void) untimeout(temp_id);
8677                 mutex_enter(SD_MUTEX(un));
8678         }
8679 
8680         if (un->un_rmw_msg_timeid != NULL) {
8681                 timeout_id_t temp_id = un->un_rmw_msg_timeid;
8682                 un->un_rmw_msg_timeid = NULL;
8683                 mutex_exit(SD_MUTEX(un));
8684                 (void) untimeout(temp_id);
8685                 mutex_enter(SD_MUTEX(un));
8686         }
8687 
8688         if (un->un_dcvb_timeid != NULL) {
8689                 timeout_id_t temp_id = un->un_dcvb_timeid;
8690                 un->un_dcvb_timeid = NULL;
8691                 mutex_exit(SD_MUTEX(un));
8692                 (void) untimeout(temp_id);
8693         } else {
8694                 mutex_exit(SD_MUTEX(un));
8695         }
8696 
8697         /* Remove any pending reservation reclaim requests for this device */
8698         sd_rmv_resv_reclaim_req(dev);
8699 
8700         mutex_enter(SD_MUTEX(un));
8701 
8702         /* Cancel any pending callbacks for SD_PATH_DIRECT_PRIORITY cmd. */
8703         if (un->un_direct_priority_timeid != NULL) {
8704                 timeout_id_t temp_id = un->un_direct_priority_timeid;
8705                 un->un_direct_priority_timeid = NULL;
8706                 mutex_exit(SD_MUTEX(un));
8707                 (void) untimeout(temp_id);
8708                 mutex_enter(SD_MUTEX(un));
8709         }
8710 
8711         /* Cancel any active multi-host disk watch thread requests */
8712         if (un->un_mhd_token != NULL) {
8713                 mutex_exit(SD_MUTEX(un));
8714                  _NOTE(DATA_READABLE_WITHOUT_LOCK(sd_lun::un_mhd_token));
8715                 if (scsi_watch_request_terminate(un->un_mhd_token,
8716                     SCSI_WATCH_TERMINATE_NOWAIT)) {
8717                         SD_ERROR(SD_LOG_ATTACH_DETACH, un,
8718                             "sd_dr_detach: Cannot cancel mhd watch request\n");
8719                         /*
8720                          * Note: We are returning here after having removed
8721                          * some driver timeouts above. This is consistent with
8722                          * the legacy implementation but perhaps the watch
8723                          * terminate call should be made with the wait flag set.
8724                          */
8725                         goto err_stillbusy;
8726                 }
8727                 mutex_enter(SD_MUTEX(un));
8728                 un->un_mhd_token = NULL;
8729         }
8730 
8731         if (un->un_swr_token != NULL) {
8732                 mutex_exit(SD_MUTEX(un));
8733                 _NOTE(DATA_READABLE_WITHOUT_LOCK(sd_lun::un_swr_token));
8734                 if (scsi_watch_request_terminate(un->un_swr_token,
8735                     SCSI_WATCH_TERMINATE_NOWAIT)) {
8736                         SD_ERROR(SD_LOG_ATTACH_DETACH, un,
8737                             "sd_dr_detach: Cannot cancel swr watch request\n");
8738                         /*
8739                          * Note: We are returning here after having removed
8740                          * some driver timeouts above. This is consistent with
8741                          * the legacy implementation but perhaps the watch
8742                          * terminate call should be made with the wait flag set.
8743                          */
8744                         goto err_stillbusy;
8745                 }
8746                 mutex_enter(SD_MUTEX(un));
8747                 un->un_swr_token = NULL;
8748         }
8749 
8750         mutex_exit(SD_MUTEX(un));
8751 
8752         /*
8753          * Clear any scsi_reset_notifies. We clear the reset notifies
8754          * if we have not registered one.
8755          * Note: The sd_mhd_reset_notify_cb() fn tries to acquire SD_MUTEX!
8756          */
8757         (void) scsi_reset_notify(SD_ADDRESS(un), SCSI_RESET_CANCEL,
8758             sd_mhd_reset_notify_cb, (caddr_t)un);
8759 
8760         /*
8761          * protect the timeout pointers from getting nulled by
8762          * their callback functions during the cancellation process.
8763          * In such a scenario untimeout can be invoked with a null value.
8764          */
8765         _NOTE(NO_COMPETING_THREADS_NOW);
8766 
8767         mutex_enter(&un->un_pm_mutex);
8768         if (un->un_pm_idle_timeid != NULL) {
8769                 timeout_id_t temp_id = un->un_pm_idle_timeid;
8770                 un->un_pm_idle_timeid = NULL;
8771                 mutex_exit(&un->un_pm_mutex);
8772 
8773                 /*
8774                  * Timeout is active; cancel it.
8775                  * Note that it'll never be active on a device
8776                  * that does not support PM therefore we don't
8777                  * have to check before calling pm_idle_component.
8778                  */
8779                 (void) untimeout(temp_id);
8780                 (void) pm_idle_component(SD_DEVINFO(un), 0);
8781                 mutex_enter(&un->un_pm_mutex);
8782         }
8783 
8784         /*
8785          * Check whether there is already a timeout scheduled for power
8786          * management. If yes then don't lower the power here, that's.
8787          * the timeout handler's job.
8788          */
8789         if (un->un_pm_timeid != NULL) {
8790                 timeout_id_t temp_id = un->un_pm_timeid;
8791                 un->un_pm_timeid = NULL;
8792                 mutex_exit(&un->un_pm_mutex);
8793                 /*
8794                  * Timeout is active; cancel it.
8795                  * Note that it'll never be active on a device
8796                  * that does not support PM therefore we don't
8797                  * have to check before calling pm_idle_component.
8798                  */
8799                 (void) untimeout(temp_id);
8800                 (void) pm_idle_component(SD_DEVINFO(un), 0);
8801 
8802         } else {
8803                 mutex_exit(&un->un_pm_mutex);
8804                 if ((un->un_f_pm_is_enabled == TRUE) &&
8805                     (pm_lower_power(SD_DEVINFO(un), 0, SD_PM_STATE_STOPPED(un))
8806                     != DDI_SUCCESS)) {
8807                         SD_ERROR(SD_LOG_ATTACH_DETACH, un,
8808                     "sd_dr_detach: Lower power request failed, ignoring.\n");
8809                         /*
8810                          * Fix for bug: 4297749, item # 13
8811                          * The above test now includes a check to see if PM is
8812                          * supported by this device before call
8813                          * pm_lower_power().
8814                          * Note, the following is not dead code. The call to
8815                          * pm_lower_power above will generate a call back into
8816                          * our sdpower routine which might result in a timeout
8817                          * handler getting activated. Therefore the following
8818                          * code is valid and necessary.
8819                          */
8820                         mutex_enter(&un->un_pm_mutex);
8821                         if (un->un_pm_timeid != NULL) {
8822                                 timeout_id_t temp_id = un->un_pm_timeid;
8823                                 un->un_pm_timeid = NULL;
8824                                 mutex_exit(&un->un_pm_mutex);
8825                                 (void) untimeout(temp_id);
8826                                 (void) pm_idle_component(SD_DEVINFO(un), 0);
8827                         } else {
8828                                 mutex_exit(&un->un_pm_mutex);
8829                         }
8830                 }
8831         }
8832 
8833         /*
8834          * Cleanup from the scsi_ifsetcap() calls (437868)
8835          * Relocated here from above to be after the call to
8836          * pm_lower_power, which was getting errors.
8837          */
8838         (void) scsi_ifsetcap(SD_ADDRESS(un), "lun-reset", 0, 1);
8839         (void) scsi_ifsetcap(SD_ADDRESS(un), "wide-xfer", 0, 1);
8840 
8841         /*
8842          * Currently, tagged queuing is supported per target based by HBA.
8843          * Setting this per lun instance actually sets the capability of this
8844          * target in HBA, which affects those luns already attached on the
8845          * same target. So during detach, we can only disable this capability
8846          * only when this is the only lun left on this target. By doing
8847          * this, we assume a target has the same tagged queuing capability
8848          * for every lun. The condition can be removed when HBA is changed to
8849          * support per lun based tagged queuing capability.
8850          */
8851         if (sd_scsi_get_target_lun_count(pdip, tgt) <= 1) {
8852                 (void) scsi_ifsetcap(SD_ADDRESS(un), "tagged-qing", 0, 1);
8853         }
8854 
8855         if (un->un_f_is_fibre == FALSE) {
8856                 (void) scsi_ifsetcap(SD_ADDRESS(un), "auto-rqsense", 0, 1);
8857         }
8858 
8859         /*
8860          * Remove any event callbacks, fibre only
8861          */
8862         if (un->un_f_is_fibre == TRUE) {
8863                 if ((un->un_insert_event != NULL) &&
8864                     (ddi_remove_event_handler(un->un_insert_cb_id) !=
8865                     DDI_SUCCESS)) {
8866                         /*
8867                          * Note: We are returning here after having done
8868                          * substantial cleanup above. This is consistent
8869                          * with the legacy implementation but this may not
8870                          * be the right thing to do.
8871                          */
8872                         SD_ERROR(SD_LOG_ATTACH_DETACH, un,
8873                             "sd_dr_detach: Cannot cancel insert event\n");
8874                         goto err_remove_event;
8875                 }
8876                 un->un_insert_event = NULL;
8877 
8878                 if ((un->un_remove_event != NULL) &&
8879                     (ddi_remove_event_handler(un->un_remove_cb_id) !=
8880                     DDI_SUCCESS)) {
8881                         /*
8882                          * Note: We are returning here after having done
8883                          * substantial cleanup above. This is consistent
8884                          * with the legacy implementation but this may not
8885                          * be the right thing to do.
8886                          */
8887                         SD_ERROR(SD_LOG_ATTACH_DETACH, un,
8888                             "sd_dr_detach: Cannot cancel remove event\n");
8889                         goto err_remove_event;
8890                 }
8891                 un->un_remove_event = NULL;
8892         }
8893 
8894         /* Do not free the softstate if the callback routine is active */
8895         sd_sync_with_callback(un);
8896 
8897         cmlb_detach(un->un_cmlbhandle, (void *)SD_PATH_DIRECT);
8898         cmlb_free_handle(&un->un_cmlbhandle);
8899 
8900         /*
8901          * Hold the detach mutex here, to make sure that no other threads ever
8902          * can access a (partially) freed soft state structure.
8903          */
8904         mutex_enter(&sd_detach_mutex);
8905 
8906         /*
8907          * Clean up the soft state struct.
8908          * Cleanup is done in reverse order of allocs/inits.
8909          * At this point there should be no competing threads anymore.
8910          */
8911 
8912         scsi_fm_fini(devp);
8913 
8914         /*
8915          * Deallocate memory for SCSI FMA.
8916          */
8917         kmem_free(un->un_fm_private, sizeof (struct sd_fm_internal));
8918 
8919         /*
8920          * Unregister and free device id if it was not registered
8921          * by the transport.
8922          */
8923         if (un->un_f_devid_transport_defined == FALSE)
8924                 ddi_devid_unregister(devi);
8925 
8926         /*
8927          * free the devid structure if allocated before (by ddi_devid_init()
8928          * or ddi_devid_get()).
8929          */
8930         if (un->un_devid) {
8931                 ddi_devid_free(un->un_devid);
8932                 un->un_devid = NULL;
8933         }
8934 
8935         /*
8936          * Destroy wmap cache if it exists.
8937          */
8938         if (un->un_wm_cache != NULL) {
8939                 kmem_cache_destroy(un->un_wm_cache);
8940                 un->un_wm_cache = NULL;
8941         }
8942 
8943         /*
8944          * kstat cleanup is done in detach for all device types (4363169).
8945          * We do not want to fail detach if the device kstats are not deleted
8946          * since there is a confusion about the devo_refcnt for the device.
8947          * We just delete the kstats and let detach complete successfully.
8948          */
8949         if (un->un_stats != NULL) {
8950                 kstat_delete(un->un_stats);
8951                 un->un_stats = NULL;
8952         }
8953         if (un->un_errstats != NULL) {
8954                 kstat_delete(un->un_errstats);
8955                 un->un_errstats = NULL;
8956         }
8957 
8958         /* Remove partition stats */
8959         if (un->un_f_pkstats_enabled) {
8960                 for (i = 0; i < NSDMAP; i++) {
8961                         if (un->un_pstats[i] != NULL) {
8962                                 kstat_delete(un->un_pstats[i]);
8963                                 un->un_pstats[i] = NULL;
8964                         }
8965                 }
8966         }
8967 
8968         /* Remove xbuf registration */
8969         ddi_xbuf_attr_unregister_devinfo(un->un_xbuf_attr, devi);
8970         ddi_xbuf_attr_destroy(un->un_xbuf_attr);
8971 
8972         /* Remove driver properties */
8973         ddi_prop_remove_all(devi);
8974 
8975         mutex_destroy(&un->un_pm_mutex);
8976         cv_destroy(&un->un_pm_busy_cv);
8977 
8978         cv_destroy(&un->un_wcc_cv);
8979 
8980         /* Open/close semaphore */
8981         sema_destroy(&un->un_semoclose);
8982 
8983         /* Removable media condvar. */
8984         cv_destroy(&un->un_state_cv);
8985 
8986         /* Suspend/resume condvar. */
8987         cv_destroy(&un->un_suspend_cv);
8988         cv_destroy(&un->un_disk_busy_cv);
8989 
8990         sd_free_rqs(un);
8991 
8992         /* Free up soft state */
8993         devp->sd_private = NULL;
8994 
8995         bzero(un, sizeof (struct sd_lun));
8996 #ifndef XPV_HVM_DRIVER
8997         ddi_soft_state_free(sd_state, instance);
8998 #endif /* !XPV_HVM_DRIVER */
8999 
9000         mutex_exit(&sd_detach_mutex);
9001 
9002         /* This frees up the INQUIRY data associated with the device. */
9003         scsi_unprobe(devp);
9004 
9005         /*
9006          * After successfully detaching an instance, we update the information
9007          * of how many luns have been attached in the relative target and
9008          * controller for parallel SCSI. This information is used when sd tries
9009          * to set the tagged queuing capability in HBA.
9010          * Since un has been released, we can't use SD_IS_PARALLEL_SCSI(un) to
9011          * check if the device is parallel SCSI. However, we don't need to
9012          * check here because we've already checked during attach. No device
9013          * that is not parallel SCSI is in the chain.
9014          */
9015         if ((tgt >= 0) && (tgt < NTARGETS_WIDE)) {
9016                 sd_scsi_update_lun_on_target(pdip, tgt, SD_SCSI_LUN_DETACH);
9017         }
9018 
9019         return (DDI_SUCCESS);
9020 
9021 err_notclosed:
9022         mutex_exit(SD_MUTEX(un));
9023 
9024 err_stillbusy:
9025         _NOTE(NO_COMPETING_THREADS_NOW);
9026 
9027 err_remove_event:
9028         mutex_enter(&sd_detach_mutex);
9029         un->un_detach_count--;
9030         mutex_exit(&sd_detach_mutex);
9031 
9032         SD_TRACE(SD_LOG_ATTACH_DETACH, un, "sd_unit_detach: exit failure\n");
9033         return (DDI_FAILURE);
9034 }
9035 
9036 
9037 /*
9038  *    Function: sd_create_errstats
9039  *
9040  * Description: This routine instantiates the device error stats.
9041  *
9042  *              Note: During attach the stats are instantiated first so they are
9043  *              available for attach-time routines that utilize the driver
9044  *              iopath to send commands to the device. The stats are initialized
9045  *              separately so data obtained during some attach-time routines is
9046  *              available. (4362483)
9047  *
9048  *   Arguments: un - driver soft state (unit) structure
9049  *              instance - driver instance
9050  *
9051  *     Context: Kernel thread context
9052  */
9053 
9054 static void
9055 sd_create_errstats(struct sd_lun *un, int instance)
9056 {
9057         struct  sd_errstats     *stp;
9058         char    kstatmodule_err[KSTAT_STRLEN];
9059         char    kstatname[KSTAT_STRLEN];
9060         int     ndata = (sizeof (struct sd_errstats) / sizeof (kstat_named_t));
9061 
9062         ASSERT(un != NULL);
9063 
9064         if (un->un_errstats != NULL) {
9065                 return;
9066         }
9067 
9068         (void) snprintf(kstatmodule_err, sizeof (kstatmodule_err),
9069             "%serr", sd_label);
9070         (void) snprintf(kstatname, sizeof (kstatname),
9071             "%s%d,err", sd_label, instance);
9072 
9073         un->un_errstats = kstat_create(kstatmodule_err, instance, kstatname,
9074             "device_error", KSTAT_TYPE_NAMED, ndata, KSTAT_FLAG_PERSISTENT);
9075 
9076         if (un->un_errstats == NULL) {
9077                 SD_ERROR(SD_LOG_ATTACH_DETACH, un,
9078                     "sd_create_errstats: Failed kstat_create\n");
9079                 return;
9080         }
9081 
9082         stp = (struct sd_errstats *)un->un_errstats->ks_data;
9083         kstat_named_init(&stp->sd_softerrs,      "Soft Errors",
9084             KSTAT_DATA_UINT32);
9085         kstat_named_init(&stp->sd_harderrs,      "Hard Errors",
9086             KSTAT_DATA_UINT32);
9087         kstat_named_init(&stp->sd_transerrs,     "Transport Errors",
9088             KSTAT_DATA_UINT32);
9089         kstat_named_init(&stp->sd_vid,           "Vendor",
9090             KSTAT_DATA_CHAR);
9091         kstat_named_init(&stp->sd_pid,           "Product",
9092             KSTAT_DATA_CHAR);
9093         kstat_named_init(&stp->sd_revision,      "Revision",
9094             KSTAT_DATA_CHAR);
9095         kstat_named_init(&stp->sd_serial,        "Serial No",
9096             KSTAT_DATA_CHAR);
9097         kstat_named_init(&stp->sd_capacity,      "Size",
9098             KSTAT_DATA_ULONGLONG);
9099         kstat_named_init(&stp->sd_rq_media_err,  "Media Error",
9100             KSTAT_DATA_UINT32);
9101         kstat_named_init(&stp->sd_rq_ntrdy_err,  "Device Not Ready",
9102             KSTAT_DATA_UINT32);
9103         kstat_named_init(&stp->sd_rq_nodev_err,  "No Device",
9104             KSTAT_DATA_UINT32);
9105         kstat_named_init(&stp->sd_rq_recov_err,  "Recoverable",
9106             KSTAT_DATA_UINT32);
9107         kstat_named_init(&stp->sd_rq_illrq_err,  "Illegal Request",
9108             KSTAT_DATA_UINT32);
9109         kstat_named_init(&stp->sd_rq_pfa_err,    "Predictive Failure Analysis",
9110             KSTAT_DATA_UINT32);
9111 
9112         un->un_errstats->ks_private = un;
9113         un->un_errstats->ks_update  = nulldev;
9114 
9115         kstat_install(un->un_errstats);
9116 }
9117 
9118 
9119 /*
9120  *    Function: sd_set_errstats
9121  *
9122  * Description: This routine sets the value of the vendor id, product id,
9123  *              revision, serial number, and capacity device error stats.
9124  *
9125  *              Note: During attach the stats are instantiated first so they are
9126  *              available for attach-time routines that utilize the driver
9127  *              iopath to send commands to the device. The stats are initialized
9128  *              separately so data obtained during some attach-time routines is
9129  *              available. (4362483)
9130  *
9131  *   Arguments: un - driver soft state (unit) structure
9132  *
9133  *     Context: Kernel thread context
9134  */
9135 
9136 static void
9137 sd_set_errstats(struct sd_lun *un)
9138 {
9139         struct  sd_errstats     *stp;
9140         char                    *sn;
9141 
9142         ASSERT(un != NULL);
9143         ASSERT(un->un_errstats != NULL);
9144         stp = (struct sd_errstats *)un->un_errstats->ks_data;
9145         ASSERT(stp != NULL);
9146         (void) strncpy(stp->sd_vid.value.c, un->un_sd->sd_inq->inq_vid, 8);
9147         (void) strncpy(stp->sd_pid.value.c, un->un_sd->sd_inq->inq_pid, 16);
9148         (void) strncpy(stp->sd_revision.value.c,
9149             un->un_sd->sd_inq->inq_revision, 4);
9150 
9151         /*
9152          * All the errstats are persistent across detach/attach,
9153          * so reset all the errstats here in case of the hot
9154          * replacement of disk drives, except for not changed
9155          * Sun qualified drives.
9156          */
9157         if ((bcmp(&SD_INQUIRY(un)->inq_pid[9], "SUN", 3) != 0) ||
9158             (bcmp(&SD_INQUIRY(un)->inq_serial, stp->sd_serial.value.c,
9159             sizeof (SD_INQUIRY(un)->inq_serial)) != 0)) {
9160                 stp->sd_softerrs.value.ui32 = 0;
9161                 stp->sd_harderrs.value.ui32 = 0;
9162                 stp->sd_transerrs.value.ui32 = 0;
9163                 stp->sd_rq_media_err.value.ui32 = 0;
9164                 stp->sd_rq_ntrdy_err.value.ui32 = 0;
9165                 stp->sd_rq_nodev_err.value.ui32 = 0;
9166                 stp->sd_rq_recov_err.value.ui32 = 0;
9167                 stp->sd_rq_illrq_err.value.ui32 = 0;
9168                 stp->sd_rq_pfa_err.value.ui32 = 0;
9169         }
9170 
9171         /*
9172          * Set the "Serial No" kstat for Sun qualified drives (indicated by
9173          * "SUN" in bytes 25-27 of the inquiry data (bytes 9-11 of the pid)
9174          * (4376302))
9175          */
9176         if (bcmp(&SD_INQUIRY(un)->inq_pid[9], "SUN", 3) == 0) {
9177                 bcopy(&SD_INQUIRY(un)->inq_serial, stp->sd_serial.value.c,
9178                     sizeof (SD_INQUIRY(un)->inq_serial));
9179         } else {
9180                 /*
9181                  * Set the "Serial No" kstat for non-Sun qualified drives
9182                  */
9183                 if (ddi_prop_lookup_string(DDI_DEV_T_ANY, SD_DEVINFO(un),
9184                     DDI_PROP_NOTPROM | DDI_PROP_DONTPASS,
9185                     INQUIRY_SERIAL_NO, &sn) == DDI_SUCCESS) {
9186                         (void) strlcpy(stp->sd_serial.value.c, sn,
9187                             sizeof (stp->sd_serial.value.c));
9188                         ddi_prop_free(sn);
9189                 }
9190         }
9191 
9192         if (un->un_f_blockcount_is_valid != TRUE) {
9193                 /*
9194                  * Set capacity error stat to 0 for no media. This ensures
9195                  * a valid capacity is displayed in response to 'iostat -E'
9196                  * when no media is present in the device.
9197                  */
9198                 stp->sd_capacity.value.ui64 = 0;
9199         } else {
9200                 /*
9201                  * Multiply un_blockcount by un->un_sys_blocksize to get
9202                  * capacity.
9203                  *
9204                  * Note: for non-512 blocksize devices "un_blockcount" has been
9205                  * "scaled" in sd_send_scsi_READ_CAPACITY by multiplying by
9206                  * (un_tgt_blocksize / un->un_sys_blocksize).
9207                  */
9208                 stp->sd_capacity.value.ui64 = (uint64_t)
9209                     ((uint64_t)un->un_blockcount * un->un_sys_blocksize);
9210         }
9211 }
9212 
9213 
9214 /*
9215  *    Function: sd_set_pstats
9216  *
9217  * Description: This routine instantiates and initializes the partition
9218  *              stats for each partition with more than zero blocks.
9219  *              (4363169)
9220  *
9221  *   Arguments: un - driver soft state (unit) structure
9222  *
9223  *     Context: Kernel thread context
9224  */
9225 
9226 static void
9227 sd_set_pstats(struct sd_lun *un)
9228 {
9229         char    kstatname[KSTAT_STRLEN];
9230         int     instance;
9231         int     i;
9232         diskaddr_t      nblks = 0;
9233         char    *partname = NULL;
9234 
9235         ASSERT(un != NULL);
9236 
9237         instance = ddi_get_instance(SD_DEVINFO(un));
9238 
9239         /* Note:x86: is this a VTOC8/VTOC16 difference? */
9240         for (i = 0; i < NSDMAP; i++) {
9241 
9242                 if (cmlb_partinfo(un->un_cmlbhandle, i,
9243                     &nblks, NULL, &partname, NULL, (void *)SD_PATH_DIRECT) != 0)
9244                         continue;
9245                 mutex_enter(SD_MUTEX(un));
9246 
9247                 if ((un->un_pstats[i] == NULL) &&
9248                     (nblks != 0)) {
9249 
9250                         (void) snprintf(kstatname, sizeof (kstatname),
9251                             "%s%d,%s", sd_label, instance,
9252                             partname);
9253 
9254                         un->un_pstats[i] = kstat_create(sd_label,
9255                             instance, kstatname, "partition", KSTAT_TYPE_IO,
9256                             1, KSTAT_FLAG_PERSISTENT);
9257                         if (un->un_pstats[i] != NULL) {
9258                                 un->un_pstats[i]->ks_lock = SD_MUTEX(un);
9259                                 kstat_install(un->un_pstats[i]);
9260                         }
9261                 }
9262                 mutex_exit(SD_MUTEX(un));
9263         }
9264 }
9265 
9266 
9267 #if (defined(__fibre))
9268 /*
9269  *    Function: sd_init_event_callbacks
9270  *
9271  * Description: This routine initializes the insertion and removal event
9272  *              callbacks. (fibre only)
9273  *
9274  *   Arguments: un - driver soft state (unit) structure
9275  *
9276  *     Context: Kernel thread context
9277  */
9278 
9279 static void
9280 sd_init_event_callbacks(struct sd_lun *un)
9281 {
9282         ASSERT(un != NULL);
9283 
9284         if ((un->un_insert_event == NULL) &&
9285             (ddi_get_eventcookie(SD_DEVINFO(un), FCAL_INSERT_EVENT,
9286             &un->un_insert_event) == DDI_SUCCESS)) {
9287                 /*
9288                  * Add the callback for an insertion event
9289                  */
9290                 (void) ddi_add_event_handler(SD_DEVINFO(un),
9291                     un->un_insert_event, sd_event_callback, (void *)un,
9292                     &(un->un_insert_cb_id));
9293         }
9294 
9295         if ((un->un_remove_event == NULL) &&
9296             (ddi_get_eventcookie(SD_DEVINFO(un), FCAL_REMOVE_EVENT,
9297             &un->un_remove_event) == DDI_SUCCESS)) {
9298                 /*
9299                  * Add the callback for a removal event
9300                  */
9301                 (void) ddi_add_event_handler(SD_DEVINFO(un),
9302                     un->un_remove_event, sd_event_callback, (void *)un,
9303                     &(un->un_remove_cb_id));
9304         }
9305 }
9306 
9307 
9308 /*
9309  *    Function: sd_event_callback
9310  *
9311  * Description: This routine handles insert/remove events (photon). The
9312  *              state is changed to OFFLINE which can be used to supress
9313  *              error msgs. (fibre only)
9314  *
9315  *   Arguments: un - driver soft state (unit) structure
9316  *
9317  *     Context: Callout thread context
9318  */
9319 /* ARGSUSED */
9320 static void
9321 sd_event_callback(dev_info_t *dip, ddi_eventcookie_t event, void *arg,
9322     void *bus_impldata)
9323 {
9324         struct sd_lun *un = (struct sd_lun *)arg;
9325 
9326         _NOTE(DATA_READABLE_WITHOUT_LOCK(sd_lun::un_insert_event));
9327         if (event == un->un_insert_event) {
9328                 SD_TRACE(SD_LOG_COMMON, un, "sd_event_callback: insert event");
9329                 mutex_enter(SD_MUTEX(un));
9330                 if (un->un_state == SD_STATE_OFFLINE) {
9331                         if (un->un_last_state != SD_STATE_SUSPENDED) {
9332                                 un->un_state = un->un_last_state;
9333                         } else {
9334                                 /*
9335                                  * We have gone through SUSPEND/RESUME while
9336                                  * we were offline. Restore the last state
9337                                  */
9338                                 un->un_state = un->un_save_state;
9339                         }
9340                 }
9341                 mutex_exit(SD_MUTEX(un));
9342 
9343         _NOTE(DATA_READABLE_WITHOUT_LOCK(sd_lun::un_remove_event));
9344         } else if (event == un->un_remove_event) {
9345                 SD_TRACE(SD_LOG_COMMON, un, "sd_event_callback: remove event");
9346                 mutex_enter(SD_MUTEX(un));
9347                 /*
9348                  * We need to handle an event callback that occurs during
9349                  * the suspend operation, since we don't prevent it.
9350                  */
9351                 if (un->un_state != SD_STATE_OFFLINE) {
9352                         if (un->un_state != SD_STATE_SUSPENDED) {
9353                                 New_state(un, SD_STATE_OFFLINE);
9354                         } else {
9355                                 un->un_last_state = SD_STATE_OFFLINE;
9356                         }
9357                 }
9358                 mutex_exit(SD_MUTEX(un));
9359         } else {
9360                 scsi_log(SD_DEVINFO(un), sd_label, CE_NOTE,
9361                     "!Unknown event\n");
9362         }
9363 
9364 }
9365 #endif
9366 
9367 /*
9368  *    Function: sd_cache_control()
9369  *
9370  * Description: This routine is the driver entry point for setting
9371  *              read and write caching by modifying the WCE (write cache
9372  *              enable) and RCD (read cache disable) bits of mode
9373  *              page 8 (MODEPAGE_CACHING).
9374  *
9375  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
9376  *                      structure for this target.
9377  *              rcd_flag - flag for controlling the read cache
9378  *              wce_flag - flag for controlling the write cache
9379  *
9380  * Return Code: EIO
9381  *              code returned by sd_send_scsi_MODE_SENSE and
9382  *              sd_send_scsi_MODE_SELECT
9383  *
9384  *     Context: Kernel Thread
9385  */
9386 
9387 static int
9388 sd_cache_control(sd_ssc_t *ssc, int rcd_flag, int wce_flag)
9389 {
9390         struct mode_caching     *mode_caching_page;
9391         uchar_t                 *header;
9392         size_t                  buflen;
9393         int                     hdrlen;
9394         int                     bd_len;
9395         int                     rval = 0;
9396         struct mode_header_grp2 *mhp;
9397         struct sd_lun           *un;
9398         int                     status;
9399 
9400         ASSERT(ssc != NULL);
9401         un = ssc->ssc_un;
9402         ASSERT(un != NULL);
9403 
9404         /*
9405          * Do a test unit ready, otherwise a mode sense may not work if this
9406          * is the first command sent to the device after boot.
9407          */
9408         status = sd_send_scsi_TEST_UNIT_READY(ssc, 0);
9409         if (status != 0)
9410                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
9411 
9412         if (un->un_f_cfg_is_atapi == TRUE) {
9413                 hdrlen = MODE_HEADER_LENGTH_GRP2;
9414         } else {
9415                 hdrlen = MODE_HEADER_LENGTH;
9416         }
9417 
9418         /*
9419          * Allocate memory for the retrieved mode page and its headers.  Set
9420          * a pointer to the page itself.  Use mode_cache_scsi3 to insure
9421          * we get all of the mode sense data otherwise, the mode select
9422          * will fail.  mode_cache_scsi3 is a superset of mode_caching.
9423          */
9424         buflen = hdrlen + MODE_BLK_DESC_LENGTH +
9425             sizeof (struct mode_cache_scsi3);
9426 
9427         header = kmem_zalloc(buflen, KM_SLEEP);
9428 
9429         /* Get the information from the device. */
9430         if (un->un_f_cfg_is_atapi == TRUE) {
9431                 rval = sd_send_scsi_MODE_SENSE(ssc, CDB_GROUP1, header, buflen,
9432                     MODEPAGE_CACHING, SD_PATH_DIRECT);
9433         } else {
9434                 rval = sd_send_scsi_MODE_SENSE(ssc, CDB_GROUP0, header, buflen,
9435                     MODEPAGE_CACHING, SD_PATH_DIRECT);
9436         }
9437 
9438         if (rval != 0) {
9439                 SD_ERROR(SD_LOG_IOCTL_RMMEDIA, un,
9440                     "sd_cache_control: Mode Sense Failed\n");
9441                 goto mode_sense_failed;
9442         }
9443 
9444         /*
9445          * Determine size of Block Descriptors in order to locate
9446          * the mode page data. ATAPI devices return 0, SCSI devices
9447          * should return MODE_BLK_DESC_LENGTH.
9448          */
9449         if (un->un_f_cfg_is_atapi == TRUE) {
9450                 mhp     = (struct mode_header_grp2 *)header;
9451                 bd_len  = (mhp->bdesc_length_hi << 8) | mhp->bdesc_length_lo;
9452         } else {
9453                 bd_len  = ((struct mode_header *)header)->bdesc_length;
9454         }
9455 
9456         if (bd_len > MODE_BLK_DESC_LENGTH) {
9457                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, 0,
9458                     "sd_cache_control: Mode Sense returned invalid block "
9459                     "descriptor length\n");
9460                 rval = EIO;
9461                 goto mode_sense_failed;
9462         }
9463 
9464         mode_caching_page = (struct mode_caching *)(header + hdrlen + bd_len);
9465         if (mode_caching_page->mode_page.code != MODEPAGE_CACHING) {
9466                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, SD_LOG_COMMON,
9467                     "sd_cache_control: Mode Sense caching page code mismatch "
9468                     "%d\n", mode_caching_page->mode_page.code);
9469                 rval = EIO;
9470                 goto mode_sense_failed;
9471         }
9472 
9473         /* Check the relevant bits on successful mode sense. */
9474         if ((mode_caching_page->rcd && rcd_flag == SD_CACHE_ENABLE) ||
9475             (!mode_caching_page->rcd && rcd_flag == SD_CACHE_DISABLE) ||
9476             (mode_caching_page->wce && wce_flag == SD_CACHE_DISABLE) ||
9477             (!mode_caching_page->wce && wce_flag == SD_CACHE_ENABLE)) {
9478 
9479                 size_t sbuflen;
9480                 uchar_t save_pg;
9481 
9482                 /*
9483                  * Construct select buffer length based on the
9484                  * length of the sense data returned.
9485                  */
9486                 sbuflen =  hdrlen + bd_len +
9487                     sizeof (struct mode_page) +
9488                     (int)mode_caching_page->mode_page.length;
9489 
9490                 /*
9491                  * Set the caching bits as requested.
9492                  */
9493                 if (rcd_flag == SD_CACHE_ENABLE)
9494                         mode_caching_page->rcd = 0;
9495                 else if (rcd_flag == SD_CACHE_DISABLE)
9496                         mode_caching_page->rcd = 1;
9497 
9498                 if (wce_flag == SD_CACHE_ENABLE)
9499                         mode_caching_page->wce = 1;
9500                 else if (wce_flag == SD_CACHE_DISABLE)
9501                         mode_caching_page->wce = 0;
9502 
9503                 /*
9504                  * Save the page if the mode sense says the
9505                  * drive supports it.
9506                  */
9507                 save_pg = mode_caching_page->mode_page.ps ?
9508                     SD_SAVE_PAGE : SD_DONTSAVE_PAGE;
9509 
9510                 /* Clear reserved bits before mode select. */
9511                 mode_caching_page->mode_page.ps = 0;
9512 
9513                 /*
9514                  * Clear out mode header for mode select.
9515                  * The rest of the retrieved page will be reused.
9516                  */
9517                 bzero(header, hdrlen);
9518 
9519                 if (un->un_f_cfg_is_atapi == TRUE) {
9520                         mhp = (struct mode_header_grp2 *)header;
9521                         mhp->bdesc_length_hi = bd_len >> 8;
9522                         mhp->bdesc_length_lo = (uchar_t)bd_len & 0xff;
9523                 } else {
9524                         ((struct mode_header *)header)->bdesc_length = bd_len;
9525                 }
9526 
9527                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
9528 
9529                 /* Issue mode select to change the cache settings */
9530                 if (un->un_f_cfg_is_atapi == TRUE) {
9531                         rval = sd_send_scsi_MODE_SELECT(ssc, CDB_GROUP1, header,
9532                             sbuflen, save_pg, SD_PATH_DIRECT);
9533                 } else {
9534                         rval = sd_send_scsi_MODE_SELECT(ssc, CDB_GROUP0, header,
9535                             sbuflen, save_pg, SD_PATH_DIRECT);
9536                 }
9537 
9538         }
9539 
9540 
9541 mode_sense_failed:
9542 
9543         kmem_free(header, buflen);
9544 
9545         if (rval != 0) {
9546                 if (rval == EIO)
9547                         sd_ssc_assessment(ssc, SD_FMT_STATUS_CHECK);
9548                 else
9549                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
9550         }
9551         return (rval);
9552 }
9553 
9554 
9555 /*
9556  *    Function: sd_get_write_cache_enabled()
9557  *
9558  * Description: This routine is the driver entry point for determining if
9559  *              write caching is enabled.  It examines the WCE (write cache
9560  *              enable) bits of mode page 8 (MODEPAGE_CACHING).
9561  *
9562  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
9563  *                      structure for this target.
9564  *              is_enabled - pointer to int where write cache enabled state
9565  *              is returned (non-zero -> write cache enabled)
9566  *
9567  *
9568  * Return Code: EIO
9569  *              code returned by sd_send_scsi_MODE_SENSE
9570  *
9571  *     Context: Kernel Thread
9572  *
9573  * NOTE: If ioctl is added to disable write cache, this sequence should
9574  * be followed so that no locking is required for accesses to
9575  * un->un_f_write_cache_enabled:
9576  *      do mode select to clear wce
9577  *      do synchronize cache to flush cache
9578  *      set un->un_f_write_cache_enabled = FALSE
9579  *
9580  * Conversely, an ioctl to enable the write cache should be done
9581  * in this order:
9582  *      set un->un_f_write_cache_enabled = TRUE
9583  *      do mode select to set wce
9584  */
9585 
9586 static int
9587 sd_get_write_cache_enabled(sd_ssc_t *ssc, int *is_enabled)
9588 {
9589         struct mode_caching     *mode_caching_page;
9590         uchar_t                 *header;
9591         size_t                  buflen;
9592         int                     hdrlen;
9593         int                     bd_len;
9594         int                     rval = 0;
9595         struct sd_lun           *un;
9596         int                     status;
9597 
9598         ASSERT(ssc != NULL);
9599         un = ssc->ssc_un;
9600         ASSERT(un != NULL);
9601         ASSERT(is_enabled != NULL);
9602 
9603         /* in case of error, flag as enabled */
9604         *is_enabled = TRUE;
9605 
9606         /*
9607          * Do a test unit ready, otherwise a mode sense may not work if this
9608          * is the first command sent to the device after boot.
9609          */
9610         status = sd_send_scsi_TEST_UNIT_READY(ssc, 0);
9611 
9612         if (status != 0)
9613                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
9614 
9615         if (un->un_f_cfg_is_atapi == TRUE) {
9616                 hdrlen = MODE_HEADER_LENGTH_GRP2;
9617         } else {
9618                 hdrlen = MODE_HEADER_LENGTH;
9619         }
9620 
9621         /*
9622          * Allocate memory for the retrieved mode page and its headers.  Set
9623          * a pointer to the page itself.
9624          */
9625         buflen = hdrlen + MODE_BLK_DESC_LENGTH + sizeof (struct mode_caching);
9626         header = kmem_zalloc(buflen, KM_SLEEP);
9627 
9628         /* Get the information from the device. */
9629         if (un->un_f_cfg_is_atapi == TRUE) {
9630                 rval = sd_send_scsi_MODE_SENSE(ssc, CDB_GROUP1, header, buflen,
9631                     MODEPAGE_CACHING, SD_PATH_DIRECT);
9632         } else {
9633                 rval = sd_send_scsi_MODE_SENSE(ssc, CDB_GROUP0, header, buflen,
9634                     MODEPAGE_CACHING, SD_PATH_DIRECT);
9635         }
9636 
9637         if (rval != 0) {
9638                 SD_ERROR(SD_LOG_IOCTL_RMMEDIA, un,
9639                     "sd_get_write_cache_enabled: Mode Sense Failed\n");
9640                 goto mode_sense_failed;
9641         }
9642 
9643         /*
9644          * Determine size of Block Descriptors in order to locate
9645          * the mode page data. ATAPI devices return 0, SCSI devices
9646          * should return MODE_BLK_DESC_LENGTH.
9647          */
9648         if (un->un_f_cfg_is_atapi == TRUE) {
9649                 struct mode_header_grp2 *mhp;
9650                 mhp     = (struct mode_header_grp2 *)header;
9651                 bd_len  = (mhp->bdesc_length_hi << 8) | mhp->bdesc_length_lo;
9652         } else {
9653                 bd_len  = ((struct mode_header *)header)->bdesc_length;
9654         }
9655 
9656         if (bd_len > MODE_BLK_DESC_LENGTH) {
9657                 /* FMA should make upset complain here */
9658                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, 0,
9659                     "sd_get_write_cache_enabled: Mode Sense returned invalid "
9660                     "block descriptor length\n");
9661                 rval = EIO;
9662                 goto mode_sense_failed;
9663         }
9664 
9665         mode_caching_page = (struct mode_caching *)(header + hdrlen + bd_len);
9666         if (mode_caching_page->mode_page.code != MODEPAGE_CACHING) {
9667                 /* FMA could make upset complain here */
9668                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, SD_LOG_COMMON,
9669                     "sd_get_write_cache_enabled: Mode Sense caching page "
9670                     "code mismatch %d\n", mode_caching_page->mode_page.code);
9671                 rval = EIO;
9672                 goto mode_sense_failed;
9673         }
9674         *is_enabled = mode_caching_page->wce;
9675 
9676 mode_sense_failed:
9677         if (rval == 0) {
9678                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
9679         } else if (rval == EIO) {
9680                 /*
9681                  * Some disks do not support mode sense(6), we
9682                  * should ignore this kind of error(sense key is
9683                  * 0x5 - illegal request).
9684                  */
9685                 uint8_t *sensep;
9686                 int senlen;
9687 
9688                 sensep = (uint8_t *)ssc->ssc_uscsi_cmd->uscsi_rqbuf;
9689                 senlen = (int)(ssc->ssc_uscsi_cmd->uscsi_rqlen -
9690                     ssc->ssc_uscsi_cmd->uscsi_rqresid);
9691 
9692                 if (senlen > 0 &&
9693                     scsi_sense_key(sensep) == KEY_ILLEGAL_REQUEST) {
9694                         sd_ssc_assessment(ssc, SD_FMT_IGNORE_COMPROMISE);
9695                 } else {
9696                         sd_ssc_assessment(ssc, SD_FMT_STATUS_CHECK);
9697                 }
9698         } else {
9699                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
9700         }
9701         kmem_free(header, buflen);
9702         return (rval);
9703 }
9704 
9705 /*
9706  *    Function: sd_get_nv_sup()
9707  *
9708  * Description: This routine is the driver entry point for
9709  * determining whether non-volatile cache is supported. This
9710  * determination process works as follows:
9711  *
9712  * 1. sd first queries sd.conf on whether
9713  * suppress_cache_flush bit is set for this device.
9714  *
9715  * 2. if not there, then queries the internal disk table.
9716  *
9717  * 3. if either sd.conf or internal disk table specifies
9718  * cache flush be suppressed, we don't bother checking
9719  * NV_SUP bit.
9720  *
9721  * If SUPPRESS_CACHE_FLUSH bit is not set to 1, sd queries
9722  * the optional INQUIRY VPD page 0x86. If the device
9723  * supports VPD page 0x86, sd examines the NV_SUP
9724  * (non-volatile cache support) bit in the INQUIRY VPD page
9725  * 0x86:
9726  *   o If NV_SUP bit is set, sd assumes the device has a
9727  *   non-volatile cache and set the
9728  *   un_f_sync_nv_supported to TRUE.
9729  *   o Otherwise cache is not non-volatile,
9730  *   un_f_sync_nv_supported is set to FALSE.
9731  *
9732  * Arguments: un - driver soft state (unit) structure
9733  *
9734  * Return Code:
9735  *
9736  *     Context: Kernel Thread
9737  */
9738 
9739 static void
9740 sd_get_nv_sup(sd_ssc_t *ssc)
9741 {
9742         int             rval            = 0;
9743         uchar_t         *inq86          = NULL;
9744         size_t          inq86_len       = MAX_INQUIRY_SIZE;
9745         size_t          inq86_resid     = 0;
9746         struct          dk_callback *dkc;
9747         struct sd_lun   *un;
9748 
9749         ASSERT(ssc != NULL);
9750         un = ssc->ssc_un;
9751         ASSERT(un != NULL);
9752 
9753         mutex_enter(SD_MUTEX(un));
9754 
9755         /*
9756          * Be conservative on the device's support of
9757          * SYNC_NV bit: un_f_sync_nv_supported is
9758          * initialized to be false.
9759          */
9760         un->un_f_sync_nv_supported = FALSE;
9761 
9762         /*
9763          * If either sd.conf or internal disk table
9764          * specifies cache flush be suppressed, then
9765          * we don't bother checking NV_SUP bit.
9766          */
9767         if (un->un_f_suppress_cache_flush == TRUE) {
9768                 mutex_exit(SD_MUTEX(un));
9769                 return;
9770         }
9771 
9772         if (sd_check_vpd_page_support(ssc) == 0 &&
9773             un->un_vpd_page_mask & SD_VPD_EXTENDED_DATA_PG) {
9774                 mutex_exit(SD_MUTEX(un));
9775                 /* collect page 86 data if available */
9776                 inq86 = kmem_zalloc(inq86_len, KM_SLEEP);
9777 
9778                 rval = sd_send_scsi_INQUIRY(ssc, inq86, inq86_len,
9779                     0x01, 0x86, &inq86_resid);
9780 
9781                 if (rval == 0 && (inq86_len - inq86_resid > 6)) {
9782                         SD_TRACE(SD_LOG_COMMON, un,
9783                             "sd_get_nv_sup: \
9784                             successfully get VPD page: %x \
9785                             PAGE LENGTH: %x BYTE 6: %x\n",
9786                             inq86[1], inq86[3], inq86[6]);
9787 
9788                         mutex_enter(SD_MUTEX(un));
9789                         /*
9790                          * check the value of NV_SUP bit: only if the device
9791                          * reports NV_SUP bit to be 1, the
9792                          * un_f_sync_nv_supported bit will be set to true.
9793                          */
9794                         if (inq86[6] & SD_VPD_NV_SUP) {
9795                                 un->un_f_sync_nv_supported = TRUE;
9796                         }
9797                         mutex_exit(SD_MUTEX(un));
9798                 } else if (rval != 0) {
9799                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
9800                 }
9801 
9802                 kmem_free(inq86, inq86_len);
9803         } else {
9804                 mutex_exit(SD_MUTEX(un));
9805         }
9806 
9807         /*
9808          * Send a SYNC CACHE command to check whether
9809          * SYNC_NV bit is supported. This command should have
9810          * un_f_sync_nv_supported set to correct value.
9811          */
9812         mutex_enter(SD_MUTEX(un));
9813         if (un->un_f_sync_nv_supported) {
9814                 mutex_exit(SD_MUTEX(un));
9815                 dkc = kmem_zalloc(sizeof (struct dk_callback), KM_SLEEP);
9816                 dkc->dkc_flag = FLUSH_VOLATILE;
9817                 (void) sd_send_scsi_SYNCHRONIZE_CACHE(un, dkc);
9818 
9819                 /*
9820                  * Send a TEST UNIT READY command to the device. This should
9821                  * clear any outstanding UNIT ATTENTION that may be present.
9822                  */
9823                 rval = sd_send_scsi_TEST_UNIT_READY(ssc, SD_DONT_RETRY_TUR);
9824                 if (rval != 0)
9825                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
9826 
9827                 kmem_free(dkc, sizeof (struct dk_callback));
9828         } else {
9829                 mutex_exit(SD_MUTEX(un));
9830         }
9831 
9832         SD_TRACE(SD_LOG_COMMON, un, "sd_get_nv_sup: \
9833             un_f_suppress_cache_flush is set to %d\n",
9834             un->un_f_suppress_cache_flush);
9835 }
9836 
9837 /*
9838  *    Function: sd_make_device
9839  *
9840  * Description: Utility routine to return the Solaris device number from
9841  *              the data in the device's dev_info structure.
9842  *
9843  * Return Code: The Solaris device number
9844  *
9845  *     Context: Any
9846  */
9847 
9848 static dev_t
9849 sd_make_device(dev_info_t *devi)
9850 {
9851         return (makedevice(ddi_driver_major(devi),
9852             ddi_get_instance(devi) << SDUNIT_SHIFT));
9853 }
9854 
9855 
9856 /*
9857  *    Function: sd_pm_entry
9858  *
9859  * Description: Called at the start of a new command to manage power
9860  *              and busy status of a device. This includes determining whether
9861  *              the current power state of the device is sufficient for
9862  *              performing the command or whether it must be changed.
9863  *              The PM framework is notified appropriately.
9864  *              Only with a return status of DDI_SUCCESS will the
9865  *              component be busy to the framework.
9866  *
9867  *              All callers of sd_pm_entry must check the return status
9868  *              and only call sd_pm_exit it it was DDI_SUCCESS. A status
9869  *              of DDI_FAILURE indicates the device failed to power up.
9870  *              In this case un_pm_count has been adjusted so the result
9871  *              on exit is still powered down, ie. count is less than 0.
9872  *              Calling sd_pm_exit with this count value hits an ASSERT.
9873  *
9874  * Return Code: DDI_SUCCESS or DDI_FAILURE
9875  *
9876  *     Context: Kernel thread context.
9877  */
9878 
9879 static int
9880 sd_pm_entry(struct sd_lun *un)
9881 {
9882         int return_status = DDI_SUCCESS;
9883 
9884         ASSERT(!mutex_owned(SD_MUTEX(un)));
9885         ASSERT(!mutex_owned(&un->un_pm_mutex));
9886 
9887         SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_entry: entry\n");
9888 
9889         if (un->un_f_pm_is_enabled == FALSE) {
9890                 SD_TRACE(SD_LOG_IO_PM, un,
9891                     "sd_pm_entry: exiting, PM not enabled\n");
9892                 return (return_status);
9893         }
9894 
9895         /*
9896          * Just increment a counter if PM is enabled. On the transition from
9897          * 0 ==> 1, mark the device as busy.  The iodone side will decrement
9898          * the count with each IO and mark the device as idle when the count
9899          * hits 0.
9900          *
9901          * If the count is less than 0 the device is powered down. If a powered
9902          * down device is successfully powered up then the count must be
9903          * incremented to reflect the power up. Note that it'll get incremented
9904          * a second time to become busy.
9905          *
9906          * Because the following has the potential to change the device state
9907          * and must release the un_pm_mutex to do so, only one thread can be
9908          * allowed through at a time.
9909          */
9910 
9911         mutex_enter(&un->un_pm_mutex);
9912         while (un->un_pm_busy == TRUE) {
9913                 cv_wait(&un->un_pm_busy_cv, &un->un_pm_mutex);
9914         }
9915         un->un_pm_busy = TRUE;
9916 
9917         if (un->un_pm_count < 1) {
9918 
9919                 SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_entry: busy component\n");
9920 
9921                 /*
9922                  * Indicate we are now busy so the framework won't attempt to
9923                  * power down the device. This call will only fail if either
9924                  * we passed a bad component number or the device has no
9925                  * components. Neither of these should ever happen.
9926                  */
9927                 mutex_exit(&un->un_pm_mutex);
9928                 return_status = pm_busy_component(SD_DEVINFO(un), 0);
9929                 ASSERT(return_status == DDI_SUCCESS);
9930 
9931                 mutex_enter(&un->un_pm_mutex);
9932 
9933                 if (un->un_pm_count < 0) {
9934                         mutex_exit(&un->un_pm_mutex);
9935 
9936                         SD_TRACE(SD_LOG_IO_PM, un,
9937                             "sd_pm_entry: power up component\n");
9938 
9939                         /*
9940                          * pm_raise_power will cause sdpower to be called
9941                          * which brings the device power level to the
9942                          * desired state, If successful, un_pm_count and
9943                          * un_power_level will be updated appropriately.
9944                          */
9945                         return_status = pm_raise_power(SD_DEVINFO(un), 0,
9946                             SD_PM_STATE_ACTIVE(un));
9947 
9948                         mutex_enter(&un->un_pm_mutex);
9949 
9950                         if (return_status != DDI_SUCCESS) {
9951                                 /*
9952                                  * Power up failed.
9953                                  * Idle the device and adjust the count
9954                                  * so the result on exit is that we're
9955                                  * still powered down, ie. count is less than 0.
9956                                  */
9957                                 SD_TRACE(SD_LOG_IO_PM, un,
9958                                     "sd_pm_entry: power up failed,"
9959                                     " idle the component\n");
9960 
9961                                 (void) pm_idle_component(SD_DEVINFO(un), 0);
9962                                 un->un_pm_count--;
9963                         } else {
9964                                 /*
9965                                  * Device is powered up, verify the
9966                                  * count is non-negative.
9967                                  * This is debug only.
9968                                  */
9969                                 ASSERT(un->un_pm_count == 0);
9970                         }
9971                 }
9972 
9973                 if (return_status == DDI_SUCCESS) {
9974                         /*
9975                          * For performance, now that the device has been tagged
9976                          * as busy, and it's known to be powered up, update the
9977                          * chain types to use jump tables that do not include
9978                          * pm. This significantly lowers the overhead and
9979                          * therefore improves performance.
9980                          */
9981 
9982                         mutex_exit(&un->un_pm_mutex);
9983                         mutex_enter(SD_MUTEX(un));
9984                         SD_TRACE(SD_LOG_IO_PM, un,
9985                             "sd_pm_entry: changing uscsi_chain_type from %d\n",
9986                             un->un_uscsi_chain_type);
9987 
9988                         if (un->un_f_non_devbsize_supported) {
9989                                 un->un_buf_chain_type =
9990                                     SD_CHAIN_INFO_RMMEDIA_NO_PM;
9991                         } else {
9992                                 un->un_buf_chain_type =
9993                                     SD_CHAIN_INFO_DISK_NO_PM;
9994                         }
9995                         un->un_uscsi_chain_type = SD_CHAIN_INFO_USCSI_CMD_NO_PM;
9996 
9997                         SD_TRACE(SD_LOG_IO_PM, un,
9998                             "             changed  uscsi_chain_type to   %d\n",
9999                             un->un_uscsi_chain_type);
10000                         mutex_exit(SD_MUTEX(un));
10001                         mutex_enter(&un->un_pm_mutex);
10002 
10003                         if (un->un_pm_idle_timeid == NULL) {
10004                                 /* 300 ms. */
10005                                 un->un_pm_idle_timeid =
10006                                     timeout(sd_pm_idletimeout_handler, un,
10007                                     (drv_usectohz((clock_t)300000)));
10008                                 /*
10009                                  * Include an extra call to busy which keeps the
10010                                  * device busy with-respect-to the PM layer
10011                                  * until the timer fires, at which time it'll
10012                                  * get the extra idle call.
10013                                  */
10014                                 (void) pm_busy_component(SD_DEVINFO(un), 0);
10015                         }
10016                 }
10017         }
10018         un->un_pm_busy = FALSE;
10019         /* Next... */
10020         cv_signal(&un->un_pm_busy_cv);
10021 
10022         un->un_pm_count++;
10023 
10024         SD_TRACE(SD_LOG_IO_PM, un,
10025             "sd_pm_entry: exiting, un_pm_count = %d\n", un->un_pm_count);
10026 
10027         mutex_exit(&un->un_pm_mutex);
10028 
10029         return (return_status);
10030 }
10031 
10032 
10033 /*
10034  *    Function: sd_pm_exit
10035  *
10036  * Description: Called at the completion of a command to manage busy
10037  *              status for the device. If the device becomes idle the
10038  *              PM framework is notified.
10039  *
10040  *     Context: Kernel thread context
10041  */
10042 
10043 static void
10044 sd_pm_exit(struct sd_lun *un)
10045 {
10046         ASSERT(!mutex_owned(SD_MUTEX(un)));
10047         ASSERT(!mutex_owned(&un->un_pm_mutex));
10048 
10049         SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_exit: entry\n");
10050 
10051         /*
10052          * After attach the following flag is only read, so don't
10053          * take the penalty of acquiring a mutex for it.
10054          */
10055         if (un->un_f_pm_is_enabled == TRUE) {
10056 
10057                 mutex_enter(&un->un_pm_mutex);
10058                 un->un_pm_count--;
10059 
10060                 SD_TRACE(SD_LOG_IO_PM, un,
10061                     "sd_pm_exit: un_pm_count = %d\n", un->un_pm_count);
10062 
10063                 ASSERT(un->un_pm_count >= 0);
10064                 if (un->un_pm_count == 0) {
10065                         mutex_exit(&un->un_pm_mutex);
10066 
10067                         SD_TRACE(SD_LOG_IO_PM, un,
10068                             "sd_pm_exit: idle component\n");
10069 
10070                         (void) pm_idle_component(SD_DEVINFO(un), 0);
10071 
10072                 } else {
10073                         mutex_exit(&un->un_pm_mutex);
10074                 }
10075         }
10076 
10077         SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_exit: exiting\n");
10078 }
10079 
10080 
10081 /*
10082  *    Function: sdopen
10083  *
10084  * Description: Driver's open(9e) entry point function.
10085  *
10086  *   Arguments: dev_i   - pointer to device number
10087  *              flag    - how to open file (FEXCL, FNDELAY, FREAD, FWRITE)
10088  *              otyp    - open type (OTYP_BLK, OTYP_CHR, OTYP_LYR)
10089  *              cred_p  - user credential pointer
10090  *
10091  * Return Code: EINVAL
10092  *              ENXIO
10093  *              EIO
10094  *              EROFS
10095  *              EBUSY
10096  *
10097  *     Context: Kernel thread context
10098  */
10099 /* ARGSUSED */
10100 static int
10101 sdopen(dev_t *dev_p, int flag, int otyp, cred_t *cred_p)
10102 {
10103         struct sd_lun   *un;
10104         int             nodelay;
10105         int             part;
10106         uint64_t        partmask;
10107         int             instance;
10108         dev_t           dev;
10109         int             rval = EIO;
10110         diskaddr_t      nblks = 0;
10111         diskaddr_t      label_cap;
10112 
10113         /* Validate the open type */
10114         if (otyp >= OTYPCNT) {
10115                 return (EINVAL);
10116         }
10117 
10118         dev = *dev_p;
10119         instance = SDUNIT(dev);
10120         mutex_enter(&sd_detach_mutex);
10121 
10122         /*
10123          * Fail the open if there is no softstate for the instance, or
10124          * if another thread somewhere is trying to detach the instance.
10125          */
10126         if (((un = ddi_get_soft_state(sd_state, instance)) == NULL) ||
10127             (un->un_detach_count != 0)) {
10128                 mutex_exit(&sd_detach_mutex);
10129                 /*
10130                  * The probe cache only needs to be cleared when open (9e) fails
10131                  * with ENXIO (4238046).
10132                  */
10133                 /*
10134                  * un-conditionally clearing probe cache is ok with
10135                  * separate sd/ssd binaries
10136                  * x86 platform can be an issue with both parallel
10137                  * and fibre in 1 binary
10138                  */
10139                 sd_scsi_clear_probe_cache();
10140                 return (ENXIO);
10141         }
10142 
10143         /*
10144          * The un_layer_count is to prevent another thread in specfs from
10145          * trying to detach the instance, which can happen when we are
10146          * called from a higher-layer driver instead of thru specfs.
10147          * This will not be needed when DDI provides a layered driver
10148          * interface that allows specfs to know that an instance is in
10149          * use by a layered driver & should not be detached.
10150          *
10151          * Note: the semantics for layered driver opens are exactly one
10152          * close for every open.
10153          */
10154         if (otyp == OTYP_LYR) {
10155                 un->un_layer_count++;
10156         }
10157 
10158         /*
10159          * Keep a count of the current # of opens in progress. This is because
10160          * some layered drivers try to call us as a regular open. This can
10161          * cause problems that we cannot prevent, however by keeping this count
10162          * we can at least keep our open and detach routines from racing against
10163          * each other under such conditions.
10164          */
10165         un->un_opens_in_progress++;
10166         mutex_exit(&sd_detach_mutex);
10167 
10168         nodelay  = (flag & (FNDELAY | FNONBLOCK));
10169         part     = SDPART(dev);
10170         partmask = 1 << part;
10171 
10172         /*
10173          * We use a semaphore here in order to serialize
10174          * open and close requests on the device.
10175          */
10176         sema_p(&un->un_semoclose);
10177 
10178         mutex_enter(SD_MUTEX(un));
10179 
10180         /*
10181          * All device accesses go thru sdstrategy() where we check
10182          * on suspend status but there could be a scsi_poll command,
10183          * which bypasses sdstrategy(), so we need to check pm
10184          * status.
10185          */
10186 
10187         if (!nodelay) {
10188                 while ((un->un_state == SD_STATE_SUSPENDED) ||
10189                     (un->un_state == SD_STATE_PM_CHANGING)) {
10190                         cv_wait(&un->un_suspend_cv, SD_MUTEX(un));
10191                 }
10192 
10193                 mutex_exit(SD_MUTEX(un));
10194                 if (sd_pm_entry(un) != DDI_SUCCESS) {
10195                         rval = EIO;
10196                         SD_ERROR(SD_LOG_OPEN_CLOSE, un,
10197                             "sdopen: sd_pm_entry failed\n");
10198                         goto open_failed_with_pm;
10199                 }
10200                 mutex_enter(SD_MUTEX(un));
10201         }
10202 
10203         /* check for previous exclusive open */
10204         SD_TRACE(SD_LOG_OPEN_CLOSE, un, "sdopen: un=%p\n", (void *)un);
10205         SD_TRACE(SD_LOG_OPEN_CLOSE, un,
10206             "sdopen: exclopen=%x, flag=%x, regopen=%x\n",
10207             un->un_exclopen, flag, un->un_ocmap.regopen[otyp]);
10208 
10209         if (un->un_exclopen & (partmask)) {
10210                 goto excl_open_fail;
10211         }
10212 
10213         if (flag & FEXCL) {
10214                 int i;
10215                 if (un->un_ocmap.lyropen[part]) {
10216                         goto excl_open_fail;
10217                 }
10218                 for (i = 0; i < (OTYPCNT - 1); i++) {
10219                         if (un->un_ocmap.regopen[i] & (partmask)) {
10220                                 goto excl_open_fail;
10221                         }
10222                 }
10223         }
10224 
10225         /*
10226          * Check the write permission if this is a removable media device,
10227          * NDELAY has not been set, and writable permission is requested.
10228          *
10229          * Note: If NDELAY was set and this is write-protected media the WRITE
10230          * attempt will fail with EIO as part of the I/O processing. This is a
10231          * more permissive implementation that allows the open to succeed and
10232          * WRITE attempts to fail when appropriate.
10233          */
10234         if (un->un_f_chk_wp_open) {
10235                 if ((flag & FWRITE) && (!nodelay)) {
10236                         mutex_exit(SD_MUTEX(un));
10237                         /*
10238                          * Defer the check for write permission on writable
10239                          * DVD drive till sdstrategy and will not fail open even
10240                          * if FWRITE is set as the device can be writable
10241                          * depending upon the media and the media can change
10242                          * after the call to open().
10243                          */
10244                         if (un->un_f_dvdram_writable_device == FALSE) {
10245                                 if (ISCD(un) || sr_check_wp(dev)) {
10246                                 rval = EROFS;
10247                                 mutex_enter(SD_MUTEX(un));
10248                                 SD_ERROR(SD_LOG_OPEN_CLOSE, un, "sdopen: "
10249                                     "write to cd or write protected media\n");
10250                                 goto open_fail;
10251                                 }
10252                         }
10253                         mutex_enter(SD_MUTEX(un));
10254                 }
10255         }
10256 
10257         /*
10258          * If opening in NDELAY/NONBLOCK mode, just return.
10259          * Check if disk is ready and has a valid geometry later.
10260          */
10261         if (!nodelay) {
10262                 sd_ssc_t        *ssc;
10263 
10264                 mutex_exit(SD_MUTEX(un));
10265                 ssc = sd_ssc_init(un);
10266                 rval = sd_ready_and_valid(ssc, part);
10267                 sd_ssc_fini(ssc);
10268                 mutex_enter(SD_MUTEX(un));
10269                 /*
10270                  * Fail if device is not ready or if the number of disk
10271                  * blocks is zero or negative for non CD devices.
10272                  */
10273 
10274                 nblks = 0;
10275 
10276                 if (rval == SD_READY_VALID && (!ISCD(un))) {
10277                         /* if cmlb_partinfo fails, nblks remains 0 */
10278                         mutex_exit(SD_MUTEX(un));
10279                         (void) cmlb_partinfo(un->un_cmlbhandle, part, &nblks,
10280                             NULL, NULL, NULL, (void *)SD_PATH_DIRECT);
10281                         mutex_enter(SD_MUTEX(un));
10282                 }
10283 
10284                 if ((rval != SD_READY_VALID) ||
10285                     (!ISCD(un) && nblks <= 0)) {
10286                         rval = un->un_f_has_removable_media ? ENXIO : EIO;
10287                         SD_ERROR(SD_LOG_OPEN_CLOSE, un, "sdopen: "
10288                             "device not ready or invalid disk block value\n");
10289                         goto open_fail;
10290                 }
10291 #if defined(__i386) || defined(__amd64)
10292         } else {
10293                 uchar_t *cp;
10294                 /*
10295                  * x86 requires special nodelay handling, so that p0 is
10296                  * always defined and accessible.
10297                  * Invalidate geometry only if device is not already open.
10298                  */
10299                 cp = &un->un_ocmap.chkd[0];
10300                 while (cp < &un->un_ocmap.chkd[OCSIZE]) {
10301                         if (*cp != (uchar_t)0) {
10302                                 break;
10303                         }
10304                         cp++;
10305                 }
10306                 if (cp == &un->un_ocmap.chkd[OCSIZE]) {
10307                         mutex_exit(SD_MUTEX(un));
10308                         cmlb_invalidate(un->un_cmlbhandle,
10309                             (void *)SD_PATH_DIRECT);
10310                         mutex_enter(SD_MUTEX(un));
10311                 }
10312 
10313 #endif
10314         }
10315 
10316         if (otyp == OTYP_LYR) {
10317                 un->un_ocmap.lyropen[part]++;
10318         } else {
10319                 un->un_ocmap.regopen[otyp] |= partmask;
10320         }
10321 
10322         /* Set up open and exclusive open flags */
10323         if (flag & FEXCL) {
10324                 un->un_exclopen |= (partmask);
10325         }
10326 
10327         /*
10328          * If the lun is EFI labeled and lun capacity is greater than the
10329          * capacity contained in the label, log a sys-event to notify the
10330          * interested module.
10331          * To avoid an infinite loop of logging sys-event, we only log the
10332          * event when the lun is not opened in NDELAY mode. The event handler
10333          * should open the lun in NDELAY mode.
10334          */
10335         if (!nodelay) {
10336                 mutex_exit(SD_MUTEX(un));
10337                 if (cmlb_efi_label_capacity(un->un_cmlbhandle, &label_cap,
10338                     (void*)SD_PATH_DIRECT) == 0) {
10339                         mutex_enter(SD_MUTEX(un));
10340                         if (un->un_f_blockcount_is_valid &&
10341                             un->un_blockcount > label_cap &&
10342                             un->un_f_expnevent == B_FALSE) {
10343                                 un->un_f_expnevent = B_TRUE;
10344                                 mutex_exit(SD_MUTEX(un));
10345                                 sd_log_lun_expansion_event(un,
10346                                     (nodelay ? KM_NOSLEEP : KM_SLEEP));
10347                                 mutex_enter(SD_MUTEX(un));
10348                         }
10349                 } else {
10350                         mutex_enter(SD_MUTEX(un));
10351                 }
10352         }
10353 
10354         SD_TRACE(SD_LOG_OPEN_CLOSE, un, "sdopen: "
10355             "open of part %d type %d\n", part, otyp);
10356 
10357         mutex_exit(SD_MUTEX(un));
10358         if (!nodelay) {
10359                 sd_pm_exit(un);
10360         }
10361 
10362         sema_v(&un->un_semoclose);
10363 
10364         mutex_enter(&sd_detach_mutex);
10365         un->un_opens_in_progress--;
10366         mutex_exit(&sd_detach_mutex);
10367 
10368         SD_TRACE(SD_LOG_OPEN_CLOSE, un, "sdopen: exit success\n");
10369         return (DDI_SUCCESS);
10370 
10371 excl_open_fail:
10372         SD_ERROR(SD_LOG_OPEN_CLOSE, un, "sdopen: fail exclusive open\n");
10373         rval = EBUSY;
10374 
10375 open_fail:
10376         mutex_exit(SD_MUTEX(un));
10377 
10378         /*
10379          * On a failed open we must exit the pm management.
10380          */
10381         if (!nodelay) {
10382                 sd_pm_exit(un);
10383         }
10384 open_failed_with_pm:
10385         sema_v(&un->un_semoclose);
10386 
10387         mutex_enter(&sd_detach_mutex);
10388         un->un_opens_in_progress--;
10389         if (otyp == OTYP_LYR) {
10390                 un->un_layer_count--;
10391         }
10392         mutex_exit(&sd_detach_mutex);
10393 
10394         return (rval);
10395 }
10396 
10397 
10398 /*
10399  *    Function: sdclose
10400  *
10401  * Description: Driver's close(9e) entry point function.
10402  *
10403  *   Arguments: dev    - device number
10404  *              flag   - file status flag, informational only
10405  *              otyp   - close type (OTYP_BLK, OTYP_CHR, OTYP_LYR)
10406  *              cred_p - user credential pointer
10407  *
10408  * Return Code: ENXIO
10409  *
10410  *     Context: Kernel thread context
10411  */
10412 /* ARGSUSED */
10413 static int
10414 sdclose(dev_t dev, int flag, int otyp, cred_t *cred_p)
10415 {
10416         struct sd_lun   *un;
10417         uchar_t         *cp;
10418         int             part;
10419         int             nodelay;
10420         int             rval = 0;
10421 
10422         /* Validate the open type */
10423         if (otyp >= OTYPCNT) {
10424                 return (ENXIO);
10425         }
10426 
10427         if ((un = ddi_get_soft_state(sd_state, SDUNIT(dev))) == NULL) {
10428                 return (ENXIO);
10429         }
10430 
10431         part = SDPART(dev);
10432         nodelay = flag & (FNDELAY | FNONBLOCK);
10433 
10434         SD_TRACE(SD_LOG_OPEN_CLOSE, un,
10435             "sdclose: close of part %d type %d\n", part, otyp);
10436 
10437         /*
10438          * We use a semaphore here in order to serialize
10439          * open and close requests on the device.
10440          */
10441         sema_p(&un->un_semoclose);
10442 
10443         mutex_enter(SD_MUTEX(un));
10444 
10445         /* Don't proceed if power is being changed. */
10446         while (un->un_state == SD_STATE_PM_CHANGING) {
10447                 cv_wait(&un->un_suspend_cv, SD_MUTEX(un));
10448         }
10449 
10450         if (un->un_exclopen & (1 << part)) {
10451                 un->un_exclopen &= ~(1 << part);
10452         }
10453 
10454         /* Update the open partition map */
10455         if (otyp == OTYP_LYR) {
10456                 un->un_ocmap.lyropen[part] -= 1;
10457         } else {
10458                 un->un_ocmap.regopen[otyp] &= ~(1 << part);
10459         }
10460 
10461         cp = &un->un_ocmap.chkd[0];
10462         while (cp < &un->un_ocmap.chkd[OCSIZE]) {
10463                 if (*cp != NULL) {
10464                         break;
10465                 }
10466                 cp++;
10467         }
10468 
10469         if (cp == &un->un_ocmap.chkd[OCSIZE]) {
10470                 SD_TRACE(SD_LOG_OPEN_CLOSE, un, "sdclose: last close\n");
10471 
10472                 /*
10473                  * We avoid persistance upon the last close, and set
10474                  * the throttle back to the maximum.
10475                  */
10476                 un->un_throttle = un->un_saved_throttle;
10477 
10478                 if (un->un_state == SD_STATE_OFFLINE) {
10479                         if (un->un_f_is_fibre == FALSE) {
10480                                 scsi_log(SD_DEVINFO(un), sd_label,
10481                                     CE_WARN, "offline\n");
10482                         }
10483                         mutex_exit(SD_MUTEX(un));
10484                         cmlb_invalidate(un->un_cmlbhandle,
10485                             (void *)SD_PATH_DIRECT);
10486                         mutex_enter(SD_MUTEX(un));
10487 
10488                 } else {
10489                         /*
10490                          * Flush any outstanding writes in NVRAM cache.
10491                          * Note: SYNCHRONIZE CACHE is an optional SCSI-2
10492                          * cmd, it may not work for non-Pluto devices.
10493                          * SYNCHRONIZE CACHE is not required for removables,
10494                          * except DVD-RAM drives.
10495                          *
10496                          * Also note: because SYNCHRONIZE CACHE is currently
10497                          * the only command issued here that requires the
10498                          * drive be powered up, only do the power up before
10499                          * sending the Sync Cache command. If additional
10500                          * commands are added which require a powered up
10501                          * drive, the following sequence may have to change.
10502                          *
10503                          * And finally, note that parallel SCSI on SPARC
10504                          * only issues a Sync Cache to DVD-RAM, a newly
10505                          * supported device.
10506                          */
10507 #if defined(__i386) || defined(__amd64)
10508                         if ((un->un_f_sync_cache_supported &&
10509                             un->un_f_sync_cache_required) ||
10510                             un->un_f_dvdram_writable_device == TRUE) {
10511 #else
10512                         if (un->un_f_dvdram_writable_device == TRUE) {
10513 #endif
10514                                 mutex_exit(SD_MUTEX(un));
10515                                 if (sd_pm_entry(un) == DDI_SUCCESS) {
10516                                         rval =
10517                                             sd_send_scsi_SYNCHRONIZE_CACHE(un,
10518                                             NULL);
10519                                         /* ignore error if not supported */
10520                                         if (rval == ENOTSUP) {
10521                                                 rval = 0;
10522                                         } else if (rval != 0) {
10523                                                 rval = EIO;
10524                                         }
10525                                         sd_pm_exit(un);
10526                                 } else {
10527                                         rval = EIO;
10528                                 }
10529                                 mutex_enter(SD_MUTEX(un));
10530                         }
10531 
10532                         /*
10533                          * For devices which supports DOOR_LOCK, send an ALLOW
10534                          * MEDIA REMOVAL command, but don't get upset if it
10535                          * fails. We need to raise the power of the drive before
10536                          * we can call sd_send_scsi_DOORLOCK()
10537                          */
10538                         if (un->un_f_doorlock_supported) {
10539                                 mutex_exit(SD_MUTEX(un));
10540                                 if (sd_pm_entry(un) == DDI_SUCCESS) {
10541                                         sd_ssc_t        *ssc;
10542 
10543                                         ssc = sd_ssc_init(un);
10544                                         rval = sd_send_scsi_DOORLOCK(ssc,
10545                                             SD_REMOVAL_ALLOW, SD_PATH_DIRECT);
10546                                         if (rval != 0)
10547                                                 sd_ssc_assessment(ssc,
10548                                                     SD_FMT_IGNORE);
10549                                         sd_ssc_fini(ssc);
10550 
10551                                         sd_pm_exit(un);
10552                                         if (ISCD(un) && (rval != 0) &&
10553                                             (nodelay != 0)) {
10554                                                 rval = ENXIO;
10555                                         }
10556                                 } else {
10557                                         rval = EIO;
10558                                 }
10559                                 mutex_enter(SD_MUTEX(un));
10560                         }
10561 
10562                         /*
10563                          * If a device has removable media, invalidate all
10564                          * parameters related to media, such as geometry,
10565                          * blocksize, and blockcount.
10566                          */
10567                         if (un->un_f_has_removable_media) {
10568                                 sr_ejected(un);
10569                         }
10570 
10571                         /*
10572                          * Destroy the cache (if it exists) which was
10573                          * allocated for the write maps since this is
10574                          * the last close for this media.
10575                          */
10576                         if (un->un_wm_cache) {
10577                                 /*
10578                                  * Check if there are pending commands.
10579                                  * and if there are give a warning and
10580                                  * do not destroy the cache.
10581                                  */
10582                                 if (un->un_ncmds_in_driver > 0) {
10583                                         scsi_log(SD_DEVINFO(un),
10584                                             sd_label, CE_WARN,
10585                                             "Unable to clean up memory "
10586                                             "because of pending I/O\n");
10587                                 } else {
10588                                         kmem_cache_destroy(
10589                                             un->un_wm_cache);
10590                                         un->un_wm_cache = NULL;
10591                                 }
10592                         }
10593                 }
10594         }
10595 
10596         mutex_exit(SD_MUTEX(un));
10597         sema_v(&un->un_semoclose);
10598 
10599         if (otyp == OTYP_LYR) {
10600                 mutex_enter(&sd_detach_mutex);
10601                 /*
10602                  * The detach routine may run when the layer count
10603                  * drops to zero.
10604                  */
10605                 un->un_layer_count--;
10606                 mutex_exit(&sd_detach_mutex);
10607         }
10608 
10609         return (rval);
10610 }
10611 
10612 
10613 /*
10614  *    Function: sd_ready_and_valid
10615  *
10616  * Description: Test if device is ready and has a valid geometry.
10617  *
10618  *   Arguments: ssc - sd_ssc_t will contain un
10619  *              un  - driver soft state (unit) structure
10620  *
10621  * Return Code: SD_READY_VALID          ready and valid label
10622  *              SD_NOT_READY_VALID      not ready, no label
10623  *              SD_RESERVED_BY_OTHERS   reservation conflict
10624  *
10625  *     Context: Never called at interrupt context.
10626  */
10627 
10628 static int
10629 sd_ready_and_valid(sd_ssc_t *ssc, int part)
10630 {
10631         struct sd_errstats      *stp;
10632         uint64_t                capacity;
10633         uint_t                  lbasize;
10634         int                     rval = SD_READY_VALID;
10635         char                    name_str[48];
10636         boolean_t               is_valid;
10637         struct sd_lun           *un;
10638         int                     status;
10639 
10640         ASSERT(ssc != NULL);
10641         un = ssc->ssc_un;
10642         ASSERT(un != NULL);
10643         ASSERT(!mutex_owned(SD_MUTEX(un)));
10644 
10645         mutex_enter(SD_MUTEX(un));
10646         /*
10647          * If a device has removable media, we must check if media is
10648          * ready when checking if this device is ready and valid.
10649          */
10650         if (un->un_f_has_removable_media) {
10651                 mutex_exit(SD_MUTEX(un));
10652                 status = sd_send_scsi_TEST_UNIT_READY(ssc, 0);
10653 
10654                 if (status != 0) {
10655                         rval = SD_NOT_READY_VALID;
10656                         mutex_enter(SD_MUTEX(un));
10657 
10658                         /* Ignore all failed status for removalbe media */
10659                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
10660 
10661                         goto done;
10662                 }
10663 
10664                 is_valid = SD_IS_VALID_LABEL(un);
10665                 mutex_enter(SD_MUTEX(un));
10666                 if (!is_valid ||
10667                     (un->un_f_blockcount_is_valid == FALSE) ||
10668                     (un->un_f_tgt_blocksize_is_valid == FALSE)) {
10669 
10670                         /* capacity has to be read every open. */
10671                         mutex_exit(SD_MUTEX(un));
10672                         status = sd_send_scsi_READ_CAPACITY(ssc, &capacity,
10673                             &lbasize, SD_PATH_DIRECT);
10674 
10675                         if (status != 0) {
10676                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
10677 
10678                                 cmlb_invalidate(un->un_cmlbhandle,
10679                                     (void *)SD_PATH_DIRECT);
10680                                 mutex_enter(SD_MUTEX(un));
10681                                 rval = SD_NOT_READY_VALID;
10682 
10683                                 goto done;
10684                         } else {
10685                                 mutex_enter(SD_MUTEX(un));
10686                                 sd_update_block_info(un, lbasize, capacity);
10687                         }
10688                 }
10689 
10690                 /*
10691                  * Check if the media in the device is writable or not.
10692                  */
10693                 if (!is_valid && ISCD(un)) {
10694                         sd_check_for_writable_cd(ssc, SD_PATH_DIRECT);
10695                 }
10696 
10697         } else {
10698                 /*
10699                  * Do a test unit ready to clear any unit attention from non-cd
10700                  * devices.
10701                  */
10702                 mutex_exit(SD_MUTEX(un));
10703 
10704                 status = sd_send_scsi_TEST_UNIT_READY(ssc, 0);
10705                 if (status != 0) {
10706                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
10707                 }
10708 
10709                 mutex_enter(SD_MUTEX(un));
10710         }
10711 
10712 
10713         /*
10714          * If this is a non 512 block device, allocate space for
10715          * the wmap cache. This is being done here since every time
10716          * a media is changed this routine will be called and the
10717          * block size is a function of media rather than device.
10718          */
10719         if (((un->un_f_rmw_type != SD_RMW_TYPE_RETURN_ERROR ||
10720             un->un_f_non_devbsize_supported) &&
10721             un->un_tgt_blocksize != DEV_BSIZE) ||
10722             un->un_f_enable_rmw) {
10723                 if (!(un->un_wm_cache)) {
10724                         (void) snprintf(name_str, sizeof (name_str),
10725                             "%s%d_cache",
10726                             ddi_driver_name(SD_DEVINFO(un)),
10727                             ddi_get_instance(SD_DEVINFO(un)));
10728                         un->un_wm_cache = kmem_cache_create(
10729                             name_str, sizeof (struct sd_w_map),
10730                             8, sd_wm_cache_constructor,
10731                             sd_wm_cache_destructor, NULL,
10732                             (void *)un, NULL, 0);
10733                         if (!(un->un_wm_cache)) {
10734                                 rval = ENOMEM;
10735                                 goto done;
10736                         }
10737                 }
10738         }
10739 
10740         if (un->un_state == SD_STATE_NORMAL) {
10741                 /*
10742                  * If the target is not yet ready here (defined by a TUR
10743                  * failure), invalidate the geometry and print an 'offline'
10744                  * message. This is a legacy message, as the state of the
10745                  * target is not actually changed to SD_STATE_OFFLINE.
10746                  *
10747                  * If the TUR fails for EACCES (Reservation Conflict),
10748                  * SD_RESERVED_BY_OTHERS will be returned to indicate
10749                  * reservation conflict. If the TUR fails for other
10750                  * reasons, SD_NOT_READY_VALID will be returned.
10751                  */
10752                 int err;
10753 
10754                 mutex_exit(SD_MUTEX(un));
10755                 err = sd_send_scsi_TEST_UNIT_READY(ssc, 0);
10756                 mutex_enter(SD_MUTEX(un));
10757 
10758                 if (err != 0) {
10759                         mutex_exit(SD_MUTEX(un));
10760                         cmlb_invalidate(un->un_cmlbhandle,
10761                             (void *)SD_PATH_DIRECT);
10762                         mutex_enter(SD_MUTEX(un));
10763                         if (err == EACCES) {
10764                                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
10765                                     "reservation conflict\n");
10766                                 rval = SD_RESERVED_BY_OTHERS;
10767                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
10768                         } else {
10769                                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
10770                                     "drive offline\n");
10771                                 rval = SD_NOT_READY_VALID;
10772                                 sd_ssc_assessment(ssc, SD_FMT_STATUS_CHECK);
10773                         }
10774                         goto done;
10775                 }
10776         }
10777 
10778         if (un->un_f_format_in_progress == FALSE) {
10779                 mutex_exit(SD_MUTEX(un));
10780 
10781                 (void) cmlb_validate(un->un_cmlbhandle, 0,
10782                     (void *)SD_PATH_DIRECT);
10783                 if (cmlb_partinfo(un->un_cmlbhandle, part, NULL, NULL, NULL,
10784                     NULL, (void *) SD_PATH_DIRECT) != 0) {
10785                         rval = SD_NOT_READY_VALID;
10786                         mutex_enter(SD_MUTEX(un));
10787 
10788                         goto done;
10789                 }
10790                 if (un->un_f_pkstats_enabled) {
10791                         sd_set_pstats(un);
10792                         SD_TRACE(SD_LOG_IO_PARTITION, un,
10793                             "sd_ready_and_valid: un:0x%p pstats created and "
10794                             "set\n", un);
10795                 }
10796                 mutex_enter(SD_MUTEX(un));
10797         }
10798 
10799         /*
10800          * If this device supports DOOR_LOCK command, try and send
10801          * this command to PREVENT MEDIA REMOVAL, but don't get upset
10802          * if it fails. For a CD, however, it is an error
10803          */
10804         if (un->un_f_doorlock_supported) {
10805                 mutex_exit(SD_MUTEX(un));
10806                 status = sd_send_scsi_DOORLOCK(ssc, SD_REMOVAL_PREVENT,
10807                     SD_PATH_DIRECT);
10808 
10809                 if ((status != 0) && ISCD(un)) {
10810                         rval = SD_NOT_READY_VALID;
10811                         mutex_enter(SD_MUTEX(un));
10812 
10813                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
10814 
10815                         goto done;
10816                 } else if (status != 0)
10817                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
10818                 mutex_enter(SD_MUTEX(un));
10819         }
10820 
10821         /* The state has changed, inform the media watch routines */
10822         un->un_mediastate = DKIO_INSERTED;
10823         cv_broadcast(&un->un_state_cv);
10824         rval = SD_READY_VALID;
10825 
10826 done:
10827 
10828         /*
10829          * Initialize the capacity kstat value, if no media previously
10830          * (capacity kstat is 0) and a media has been inserted
10831          * (un_blockcount > 0).
10832          */
10833         if (un->un_errstats != NULL) {
10834                 stp = (struct sd_errstats *)un->un_errstats->ks_data;
10835                 if ((stp->sd_capacity.value.ui64 == 0) &&
10836                     (un->un_f_blockcount_is_valid == TRUE)) {
10837                         stp->sd_capacity.value.ui64 =
10838                             (uint64_t)((uint64_t)un->un_blockcount *
10839                             un->un_sys_blocksize);
10840                 }
10841         }
10842 
10843         mutex_exit(SD_MUTEX(un));
10844         return (rval);
10845 }
10846 
10847 
10848 /*
10849  *    Function: sdmin
10850  *
10851  * Description: Routine to limit the size of a data transfer. Used in
10852  *              conjunction with physio(9F).
10853  *
10854  *   Arguments: bp - pointer to the indicated buf(9S) struct.
10855  *
10856  *     Context: Kernel thread context.
10857  */
10858 
10859 static void
10860 sdmin(struct buf *bp)
10861 {
10862         struct sd_lun   *un;
10863         int             instance;
10864 
10865         instance = SDUNIT(bp->b_edev);
10866 
10867         un = ddi_get_soft_state(sd_state, instance);
10868         ASSERT(un != NULL);
10869 
10870         /*
10871          * We depend on buf breakup to restrict
10872          * IO size if it is enabled.
10873          */
10874         if (un->un_buf_breakup_supported) {
10875                 return;
10876         }
10877 
10878         if (bp->b_bcount > un->un_max_xfer_size) {
10879                 bp->b_bcount = un->un_max_xfer_size;
10880         }
10881 }
10882 
10883 
10884 /*
10885  *    Function: sdread
10886  *
10887  * Description: Driver's read(9e) entry point function.
10888  *
10889  *   Arguments: dev   - device number
10890  *              uio   - structure pointer describing where data is to be stored
10891  *                      in user's space
10892  *              cred_p  - user credential pointer
10893  *
10894  * Return Code: ENXIO
10895  *              EIO
10896  *              EINVAL
10897  *              value returned by physio
10898  *
10899  *     Context: Kernel thread context.
10900  */
10901 /* ARGSUSED */
10902 static int
10903 sdread(dev_t dev, struct uio *uio, cred_t *cred_p)
10904 {
10905         struct sd_lun   *un = NULL;
10906         int             secmask;
10907         int             err = 0;
10908         sd_ssc_t        *ssc;
10909 
10910         if ((un = ddi_get_soft_state(sd_state, SDUNIT(dev))) == NULL) {
10911                 return (ENXIO);
10912         }
10913 
10914         ASSERT(!mutex_owned(SD_MUTEX(un)));
10915 
10916 
10917         if (!SD_IS_VALID_LABEL(un) && !ISCD(un)) {
10918                 mutex_enter(SD_MUTEX(un));
10919                 /*
10920                  * Because the call to sd_ready_and_valid will issue I/O we
10921                  * must wait here if either the device is suspended or
10922                  * if it's power level is changing.
10923                  */
10924                 while ((un->un_state == SD_STATE_SUSPENDED) ||
10925                     (un->un_state == SD_STATE_PM_CHANGING)) {
10926                         cv_wait(&un->un_suspend_cv, SD_MUTEX(un));
10927                 }
10928                 un->un_ncmds_in_driver++;
10929                 mutex_exit(SD_MUTEX(un));
10930 
10931                 /* Initialize sd_ssc_t for internal uscsi commands */
10932                 ssc = sd_ssc_init(un);
10933                 if ((sd_ready_and_valid(ssc, SDPART(dev))) != SD_READY_VALID) {
10934                         err = EIO;
10935                 } else {
10936                         err = 0;
10937                 }
10938                 sd_ssc_fini(ssc);
10939 
10940                 mutex_enter(SD_MUTEX(un));
10941                 un->un_ncmds_in_driver--;
10942                 ASSERT(un->un_ncmds_in_driver >= 0);
10943                 mutex_exit(SD_MUTEX(un));
10944                 if (err != 0)
10945                         return (err);
10946         }
10947 
10948         /*
10949          * Read requests are restricted to multiples of the system block size.
10950          */
10951         if (un->un_f_rmw_type == SD_RMW_TYPE_RETURN_ERROR &&
10952             !un->un_f_enable_rmw)
10953                 secmask = un->un_tgt_blocksize - 1;
10954         else
10955                 secmask = DEV_BSIZE - 1;
10956 
10957         if (uio->uio_loffset & ((offset_t)(secmask))) {
10958                 SD_ERROR(SD_LOG_READ_WRITE, un,
10959                     "sdread: file offset not modulo %d\n",
10960                     secmask + 1);
10961                 err = EINVAL;
10962         } else if (uio->uio_iov->iov_len & (secmask)) {
10963                 SD_ERROR(SD_LOG_READ_WRITE, un,
10964                     "sdread: transfer length not modulo %d\n",
10965                     secmask + 1);
10966                 err = EINVAL;
10967         } else {
10968                 err = physio(sdstrategy, NULL, dev, B_READ, sdmin, uio);
10969         }
10970 
10971         return (err);
10972 }
10973 
10974 
10975 /*
10976  *    Function: sdwrite
10977  *
10978  * Description: Driver's write(9e) entry point function.
10979  *
10980  *   Arguments: dev   - device number
10981  *              uio   - structure pointer describing where data is stored in
10982  *                      user's space
10983  *              cred_p  - user credential pointer
10984  *
10985  * Return Code: ENXIO
10986  *              EIO
10987  *              EINVAL
10988  *              value returned by physio
10989  *
10990  *     Context: Kernel thread context.
10991  */
10992 /* ARGSUSED */
10993 static int
10994 sdwrite(dev_t dev, struct uio *uio, cred_t *cred_p)
10995 {
10996         struct sd_lun   *un = NULL;
10997         int             secmask;
10998         int             err = 0;
10999         sd_ssc_t        *ssc;
11000 
11001         if ((un = ddi_get_soft_state(sd_state, SDUNIT(dev))) == NULL) {
11002                 return (ENXIO);
11003         }
11004 
11005         ASSERT(!mutex_owned(SD_MUTEX(un)));
11006 
11007         if (!SD_IS_VALID_LABEL(un) && !ISCD(un)) {
11008                 mutex_enter(SD_MUTEX(un));
11009                 /*
11010                  * Because the call to sd_ready_and_valid will issue I/O we
11011                  * must wait here if either the device is suspended or
11012                  * if it's power level is changing.
11013                  */
11014                 while ((un->un_state == SD_STATE_SUSPENDED) ||
11015                     (un->un_state == SD_STATE_PM_CHANGING)) {
11016                         cv_wait(&un->un_suspend_cv, SD_MUTEX(un));
11017                 }
11018                 un->un_ncmds_in_driver++;
11019                 mutex_exit(SD_MUTEX(un));
11020 
11021                 /* Initialize sd_ssc_t for internal uscsi commands */
11022                 ssc = sd_ssc_init(un);
11023                 if ((sd_ready_and_valid(ssc, SDPART(dev))) != SD_READY_VALID) {
11024                         err = EIO;
11025                 } else {
11026                         err = 0;
11027                 }
11028                 sd_ssc_fini(ssc);
11029 
11030                 mutex_enter(SD_MUTEX(un));
11031                 un->un_ncmds_in_driver--;
11032                 ASSERT(un->un_ncmds_in_driver >= 0);
11033                 mutex_exit(SD_MUTEX(un));
11034                 if (err != 0)
11035                         return (err);
11036         }
11037 
11038         /*
11039          * Write requests are restricted to multiples of the system block size.
11040          */
11041         if (un->un_f_rmw_type == SD_RMW_TYPE_RETURN_ERROR &&
11042             !un->un_f_enable_rmw)
11043                 secmask = un->un_tgt_blocksize - 1;
11044         else
11045                 secmask = DEV_BSIZE - 1;
11046 
11047         if (uio->uio_loffset & ((offset_t)(secmask))) {
11048                 SD_ERROR(SD_LOG_READ_WRITE, un,
11049                     "sdwrite: file offset not modulo %d\n",
11050                     secmask + 1);
11051                 err = EINVAL;
11052         } else if (uio->uio_iov->iov_len & (secmask)) {
11053                 SD_ERROR(SD_LOG_READ_WRITE, un,
11054                     "sdwrite: transfer length not modulo %d\n",
11055                     secmask + 1);
11056                 err = EINVAL;
11057         } else {
11058                 err = physio(sdstrategy, NULL, dev, B_WRITE, sdmin, uio);
11059         }
11060 
11061         return (err);
11062 }
11063 
11064 
11065 /*
11066  *    Function: sdaread
11067  *
11068  * Description: Driver's aread(9e) entry point function.
11069  *
11070  *   Arguments: dev   - device number
11071  *              aio   - structure pointer describing where data is to be stored
11072  *              cred_p  - user credential pointer
11073  *
11074  * Return Code: ENXIO
11075  *              EIO
11076  *              EINVAL
11077  *              value returned by aphysio
11078  *
11079  *     Context: Kernel thread context.
11080  */
11081 /* ARGSUSED */
11082 static int
11083 sdaread(dev_t dev, struct aio_req *aio, cred_t *cred_p)
11084 {
11085         struct sd_lun   *un = NULL;
11086         struct uio      *uio = aio->aio_uio;
11087         int             secmask;
11088         int             err = 0;
11089         sd_ssc_t        *ssc;
11090 
11091         if ((un = ddi_get_soft_state(sd_state, SDUNIT(dev))) == NULL) {
11092                 return (ENXIO);
11093         }
11094 
11095         ASSERT(!mutex_owned(SD_MUTEX(un)));
11096 
11097         if (!SD_IS_VALID_LABEL(un) && !ISCD(un)) {
11098                 mutex_enter(SD_MUTEX(un));
11099                 /*
11100                  * Because the call to sd_ready_and_valid will issue I/O we
11101                  * must wait here if either the device is suspended or
11102                  * if it's power level is changing.
11103                  */
11104                 while ((un->un_state == SD_STATE_SUSPENDED) ||
11105                     (un->un_state == SD_STATE_PM_CHANGING)) {
11106                         cv_wait(&un->un_suspend_cv, SD_MUTEX(un));
11107                 }
11108                 un->un_ncmds_in_driver++;
11109                 mutex_exit(SD_MUTEX(un));
11110 
11111                 /* Initialize sd_ssc_t for internal uscsi commands */
11112                 ssc = sd_ssc_init(un);
11113                 if ((sd_ready_and_valid(ssc, SDPART(dev))) != SD_READY_VALID) {
11114                         err = EIO;
11115                 } else {
11116                         err = 0;
11117                 }
11118                 sd_ssc_fini(ssc);
11119 
11120                 mutex_enter(SD_MUTEX(un));
11121                 un->un_ncmds_in_driver--;
11122                 ASSERT(un->un_ncmds_in_driver >= 0);
11123                 mutex_exit(SD_MUTEX(un));
11124                 if (err != 0)
11125                         return (err);
11126         }
11127 
11128         /*
11129          * Read requests are restricted to multiples of the system block size.
11130          */
11131         if (un->un_f_rmw_type == SD_RMW_TYPE_RETURN_ERROR &&
11132             !un->un_f_enable_rmw)
11133                 secmask = un->un_tgt_blocksize - 1;
11134         else
11135                 secmask = DEV_BSIZE - 1;
11136 
11137         if (uio->uio_loffset & ((offset_t)(secmask))) {
11138                 SD_ERROR(SD_LOG_READ_WRITE, un,
11139                     "sdaread: file offset not modulo %d\n",
11140                     secmask + 1);
11141                 err = EINVAL;
11142         } else if (uio->uio_iov->iov_len & (secmask)) {
11143                 SD_ERROR(SD_LOG_READ_WRITE, un,
11144                     "sdaread: transfer length not modulo %d\n",
11145                     secmask + 1);
11146                 err = EINVAL;
11147         } else {
11148                 err = aphysio(sdstrategy, anocancel, dev, B_READ, sdmin, aio);
11149         }
11150 
11151         return (err);
11152 }
11153 
11154 
11155 /*
11156  *    Function: sdawrite
11157  *
11158  * Description: Driver's awrite(9e) entry point function.
11159  *
11160  *   Arguments: dev   - device number
11161  *              aio   - structure pointer describing where data is stored
11162  *              cred_p  - user credential pointer
11163  *
11164  * Return Code: ENXIO
11165  *              EIO
11166  *              EINVAL
11167  *              value returned by aphysio
11168  *
11169  *     Context: Kernel thread context.
11170  */
11171 /* ARGSUSED */
11172 static int
11173 sdawrite(dev_t dev, struct aio_req *aio, cred_t *cred_p)
11174 {
11175         struct sd_lun   *un = NULL;
11176         struct uio      *uio = aio->aio_uio;
11177         int             secmask;
11178         int             err = 0;
11179         sd_ssc_t        *ssc;
11180 
11181         if ((un = ddi_get_soft_state(sd_state, SDUNIT(dev))) == NULL) {
11182                 return (ENXIO);
11183         }
11184 
11185         ASSERT(!mutex_owned(SD_MUTEX(un)));
11186 
11187         if (!SD_IS_VALID_LABEL(un) && !ISCD(un)) {
11188                 mutex_enter(SD_MUTEX(un));
11189                 /*
11190                  * Because the call to sd_ready_and_valid will issue I/O we
11191                  * must wait here if either the device is suspended or
11192                  * if it's power level is changing.
11193                  */
11194                 while ((un->un_state == SD_STATE_SUSPENDED) ||
11195                     (un->un_state == SD_STATE_PM_CHANGING)) {
11196                         cv_wait(&un->un_suspend_cv, SD_MUTEX(un));
11197                 }
11198                 un->un_ncmds_in_driver++;
11199                 mutex_exit(SD_MUTEX(un));
11200 
11201                 /* Initialize sd_ssc_t for internal uscsi commands */
11202                 ssc = sd_ssc_init(un);
11203                 if ((sd_ready_and_valid(ssc, SDPART(dev))) != SD_READY_VALID) {
11204                         err = EIO;
11205                 } else {
11206                         err = 0;
11207                 }
11208                 sd_ssc_fini(ssc);
11209 
11210                 mutex_enter(SD_MUTEX(un));
11211                 un->un_ncmds_in_driver--;
11212                 ASSERT(un->un_ncmds_in_driver >= 0);
11213                 mutex_exit(SD_MUTEX(un));
11214                 if (err != 0)
11215                         return (err);
11216         }
11217 
11218         /*
11219          * Write requests are restricted to multiples of the system block size.
11220          */
11221         if (un->un_f_rmw_type == SD_RMW_TYPE_RETURN_ERROR &&
11222             !un->un_f_enable_rmw)
11223                 secmask = un->un_tgt_blocksize - 1;
11224         else
11225                 secmask = DEV_BSIZE - 1;
11226 
11227         if (uio->uio_loffset & ((offset_t)(secmask))) {
11228                 SD_ERROR(SD_LOG_READ_WRITE, un,
11229                     "sdawrite: file offset not modulo %d\n",
11230                     secmask + 1);
11231                 err = EINVAL;
11232         } else if (uio->uio_iov->iov_len & (secmask)) {
11233                 SD_ERROR(SD_LOG_READ_WRITE, un,
11234                     "sdawrite: transfer length not modulo %d\n",
11235                     secmask + 1);
11236                 err = EINVAL;
11237         } else {
11238                 err = aphysio(sdstrategy, anocancel, dev, B_WRITE, sdmin, aio);
11239         }
11240 
11241         return (err);
11242 }
11243 
11244 
11245 
11246 
11247 
11248 /*
11249  * Driver IO processing follows the following sequence:
11250  *
11251  *     sdioctl(9E)     sdstrategy(9E)         biodone(9F)
11252  *         |                |                     ^
11253  *         v                v                     |
11254  * sd_send_scsi_cmd()  ddi_xbuf_qstrategy()       +-------------------+
11255  *         |                |                     |                   |
11256  *         v                |                     |                   |
11257  * sd_uscsi_strategy() sd_xbuf_strategy()   sd_buf_iodone()   sd_uscsi_iodone()
11258  *         |                |                     ^                   ^
11259  *         v                v                     |                   |
11260  * SD_BEGIN_IOSTART()  SD_BEGIN_IOSTART()         |                   |
11261  *         |                |                     |                   |
11262  *     +---+                |                     +------------+      +-------+
11263  *     |                    |                                  |              |
11264  *     |   SD_NEXT_IOSTART()|                  SD_NEXT_IODONE()|              |
11265  *     |                    v                                  |              |
11266  *     |         sd_mapblockaddr_iostart()           sd_mapblockaddr_iodone() |
11267  *     |                    |                                  ^              |
11268  *     |   SD_NEXT_IOSTART()|                  SD_NEXT_IODONE()|              |
11269  *     |                    v                                  |              |
11270  *     |         sd_mapblocksize_iostart()           sd_mapblocksize_iodone() |
11271  *     |                    |                                  ^              |
11272  *     |   SD_NEXT_IOSTART()|                  SD_NEXT_IODONE()|              |
11273  *     |                    v                                  |              |
11274  *     |           sd_checksum_iostart()               sd_checksum_iodone()   |
11275  *     |                    |                                  ^              |
11276  *     +-> SD_NEXT_IOSTART()|                  SD_NEXT_IODONE()+------------->+
11277  *     |                    v                                  |              |
11278  *     |              sd_pm_iostart()                     sd_pm_iodone()      |
11279  *     |                    |                                  ^              |
11280  *     |                    |                                  |              |
11281  *     +-> SD_NEXT_IOSTART()|               SD_BEGIN_IODONE()--+--------------+
11282  *                          |                           ^
11283  *                          v                           |
11284  *                   sd_core_iostart()                  |
11285  *                          |                           |
11286  *                          |                           +------>(*destroypkt)()
11287  *                          +-> sd_start_cmds() <-+     |           |
11288  *                          |                     |     |           v
11289  *                          |                     |     |  scsi_destroy_pkt(9F)
11290  *                          |                     |     |
11291  *                          +->(*initpkt)()       +- sdintr()
11292  *                          |  |                        |  |
11293  *                          |  +-> scsi_init_pkt(9F)    |  +-> sd_handle_xxx()
11294  *                          |  +-> scsi_setup_cdb(9F)   |
11295  *                          |                           |
11296  *                          +--> scsi_transport(9F)     |
11297  *                                     |                |
11298  *                                     +----> SCSA ---->+
11299  *
11300  *
11301  * This code is based upon the following presumptions:
11302  *
11303  *   - iostart and iodone functions operate on buf(9S) structures. These
11304  *     functions perform the necessary operations on the buf(9S) and pass
11305  *     them along to the next function in the chain by using the macros
11306  *     SD_NEXT_IOSTART() (for iostart side functions) and SD_NEXT_IODONE()
11307  *     (for iodone side functions).
11308  *
11309  *   - The iostart side functions may sleep. The iodone side functions
11310  *     are called under interrupt context and may NOT sleep. Therefore
11311  *     iodone side functions also may not call iostart side functions.
11312  *     (NOTE: iostart side functions should NOT sleep for memory, as
11313  *     this could result in deadlock.)
11314  *
11315  *   - An iostart side function may call its corresponding iodone side
11316  *     function directly (if necessary).
11317  *
11318  *   - In the event of an error, an iostart side function can return a buf(9S)
11319  *     to its caller by calling SD_BEGIN_IODONE() (after setting B_ERROR and
11320  *     b_error in the usual way of course).
11321  *
11322  *   - The taskq mechanism may be used by the iodone side functions to dispatch
11323  *     requests to the iostart side functions.  The iostart side functions in
11324  *     this case would be called under the context of a taskq thread, so it's
11325  *     OK for them to block/sleep/spin in this case.
11326  *
11327  *   - iostart side functions may allocate "shadow" buf(9S) structs and
11328  *     pass them along to the next function in the chain.  The corresponding
11329  *     iodone side functions must coalesce the "shadow" bufs and return
11330  *     the "original" buf to the next higher layer.
11331  *
11332  *   - The b_private field of the buf(9S) struct holds a pointer to
11333  *     an sd_xbuf struct, which contains information needed to
11334  *     construct the scsi_pkt for the command.
11335  *
11336  *   - The SD_MUTEX(un) is NOT held across calls to the next layer. Each
11337  *     layer must acquire & release the SD_MUTEX(un) as needed.
11338  */
11339 
11340 
11341 /*
11342  * Create taskq for all targets in the system. This is created at
11343  * _init(9E) and destroyed at _fini(9E).
11344  *
11345  * Note: here we set the minalloc to a reasonably high number to ensure that
11346  * we will have an adequate supply of task entries available at interrupt time.
11347  * This is used in conjunction with the TASKQ_PREPOPULATE flag in
11348  * sd_create_taskq().  Since we do not want to sleep for allocations at
11349  * interrupt time, set maxalloc equal to minalloc. That way we will just fail
11350  * the command if we ever try to dispatch more than SD_TASKQ_MAXALLOC taskq
11351  * requests any one instant in time.
11352  */
11353 #define SD_TASKQ_NUMTHREADS     8
11354 #define SD_TASKQ_MINALLOC       256
11355 #define SD_TASKQ_MAXALLOC       256
11356 
11357 static taskq_t  *sd_tq = NULL;
11358 _NOTE(SCHEME_PROTECTS_DATA("stable data", sd_tq))
11359 
11360 static int      sd_taskq_minalloc = SD_TASKQ_MINALLOC;
11361 static int      sd_taskq_maxalloc = SD_TASKQ_MAXALLOC;
11362 
11363 /*
11364  * The following task queue is being created for the write part of
11365  * read-modify-write of non-512 block size devices.
11366  * Limit the number of threads to 1 for now. This number has been chosen
11367  * considering the fact that it applies only to dvd ram drives/MO drives
11368  * currently. Performance for which is not main criteria at this stage.
11369  * Note: It needs to be explored if we can use a single taskq in future
11370  */
11371 #define SD_WMR_TASKQ_NUMTHREADS 1
11372 static taskq_t  *sd_wmr_tq = NULL;
11373 _NOTE(SCHEME_PROTECTS_DATA("stable data", sd_wmr_tq))
11374 
11375 /*
11376  *    Function: sd_taskq_create
11377  *
11378  * Description: Create taskq thread(s) and preallocate task entries
11379  *
11380  * Return Code: Returns a pointer to the allocated taskq_t.
11381  *
11382  *     Context: Can sleep. Requires blockable context.
11383  *
11384  *       Notes: - The taskq() facility currently is NOT part of the DDI.
11385  *                (definitely NOT recommeded for 3rd-party drivers!) :-)
11386  *              - taskq_create() will block for memory, also it will panic
11387  *                if it cannot create the requested number of threads.
11388  *              - Currently taskq_create() creates threads that cannot be
11389  *                swapped.
11390  *              - We use TASKQ_PREPOPULATE to ensure we have an adequate
11391  *                supply of taskq entries at interrupt time (ie, so that we
11392  *                do not have to sleep for memory)
11393  */
11394 
11395 static void
11396 sd_taskq_create(void)
11397 {
11398         char    taskq_name[TASKQ_NAMELEN];
11399 
11400         ASSERT(sd_tq == NULL);
11401         ASSERT(sd_wmr_tq == NULL);
11402 
11403         (void) snprintf(taskq_name, sizeof (taskq_name),
11404             "%s_drv_taskq", sd_label);
11405         sd_tq = (taskq_create(taskq_name, SD_TASKQ_NUMTHREADS,
11406             (v.v_maxsyspri - 2), sd_taskq_minalloc, sd_taskq_maxalloc,
11407             TASKQ_PREPOPULATE));
11408 
11409         (void) snprintf(taskq_name, sizeof (taskq_name),
11410             "%s_rmw_taskq", sd_label);
11411         sd_wmr_tq = (taskq_create(taskq_name, SD_WMR_TASKQ_NUMTHREADS,
11412             (v.v_maxsyspri - 2), sd_taskq_minalloc, sd_taskq_maxalloc,
11413             TASKQ_PREPOPULATE));
11414 }
11415 
11416 
11417 /*
11418  *    Function: sd_taskq_delete
11419  *
11420  * Description: Complementary cleanup routine for sd_taskq_create().
11421  *
11422  *     Context: Kernel thread context.
11423  */
11424 
11425 static void
11426 sd_taskq_delete(void)
11427 {
11428         ASSERT(sd_tq != NULL);
11429         ASSERT(sd_wmr_tq != NULL);
11430         taskq_destroy(sd_tq);
11431         taskq_destroy(sd_wmr_tq);
11432         sd_tq = NULL;
11433         sd_wmr_tq = NULL;
11434 }
11435 
11436 
11437 /*
11438  *    Function: sdstrategy
11439  *
11440  * Description: Driver's strategy (9E) entry point function.
11441  *
11442  *   Arguments: bp - pointer to buf(9S)
11443  *
11444  * Return Code: Always returns zero
11445  *
11446  *     Context: Kernel thread context.
11447  */
11448 
11449 static int
11450 sdstrategy(struct buf *bp)
11451 {
11452         struct sd_lun *un;
11453 
11454         un = ddi_get_soft_state(sd_state, SD_GET_INSTANCE_FROM_BUF(bp));
11455         if (un == NULL) {
11456                 bioerror(bp, EIO);
11457                 bp->b_resid = bp->b_bcount;
11458                 biodone(bp);
11459                 return (0);
11460         }
11461 
11462         /* As was done in the past, fail new cmds. if state is dumping. */
11463         if (un->un_state == SD_STATE_DUMPING) {
11464                 bioerror(bp, ENXIO);
11465                 bp->b_resid = bp->b_bcount;
11466                 biodone(bp);
11467                 return (0);
11468         }
11469 
11470         ASSERT(!mutex_owned(SD_MUTEX(un)));
11471 
11472         /*
11473          * Commands may sneak in while we released the mutex in
11474          * DDI_SUSPEND, we should block new commands. However, old
11475          * commands that are still in the driver at this point should
11476          * still be allowed to drain.
11477          */
11478         mutex_enter(SD_MUTEX(un));
11479         /*
11480          * Must wait here if either the device is suspended or
11481          * if it's power level is changing.
11482          */
11483         while ((un->un_state == SD_STATE_SUSPENDED) ||
11484             (un->un_state == SD_STATE_PM_CHANGING)) {
11485                 cv_wait(&un->un_suspend_cv, SD_MUTEX(un));
11486         }
11487 
11488         un->un_ncmds_in_driver++;
11489 
11490         /*
11491          * atapi: Since we are running the CD for now in PIO mode we need to
11492          * call bp_mapin here to avoid bp_mapin called interrupt context under
11493          * the HBA's init_pkt routine.
11494          */
11495         if (un->un_f_cfg_is_atapi == TRUE) {
11496                 mutex_exit(SD_MUTEX(un));
11497                 bp_mapin(bp);
11498                 mutex_enter(SD_MUTEX(un));
11499         }
11500         SD_INFO(SD_LOG_IO, un, "sdstrategy: un_ncmds_in_driver = %ld\n",
11501             un->un_ncmds_in_driver);
11502 
11503         if (bp->b_flags & B_WRITE)
11504                 un->un_f_sync_cache_required = TRUE;
11505 
11506         mutex_exit(SD_MUTEX(un));
11507 
11508         /*
11509          * This will (eventually) allocate the sd_xbuf area and
11510          * call sd_xbuf_strategy().  We just want to return the
11511          * result of ddi_xbuf_qstrategy so that we have an opt-
11512          * imized tail call which saves us a stack frame.
11513          */
11514         return (ddi_xbuf_qstrategy(bp, un->un_xbuf_attr));
11515 }
11516 
11517 
11518 /*
11519  *    Function: sd_xbuf_strategy
11520  *
11521  * Description: Function for initiating IO operations via the
11522  *              ddi_xbuf_qstrategy() mechanism.
11523  *
11524  *     Context: Kernel thread context.
11525  */
11526 
11527 static void
11528 sd_xbuf_strategy(struct buf *bp, ddi_xbuf_t xp, void *arg)
11529 {
11530         struct sd_lun *un = arg;
11531 
11532         ASSERT(bp != NULL);
11533         ASSERT(xp != NULL);
11534         ASSERT(un != NULL);
11535         ASSERT(!mutex_owned(SD_MUTEX(un)));
11536 
11537         /*
11538          * Initialize the fields in the xbuf and save a pointer to the
11539          * xbuf in bp->b_private.
11540          */
11541         sd_xbuf_init(un, bp, xp, SD_CHAIN_BUFIO, NULL);
11542 
11543         /* Send the buf down the iostart chain */
11544         SD_BEGIN_IOSTART(((struct sd_xbuf *)xp)->xb_chain_iostart, un, bp);
11545 }
11546 
11547 
11548 /*
11549  *    Function: sd_xbuf_init
11550  *
11551  * Description: Prepare the given sd_xbuf struct for use.
11552  *
11553  *   Arguments: un - ptr to softstate
11554  *              bp - ptr to associated buf(9S)
11555  *              xp - ptr to associated sd_xbuf
11556  *              chain_type - IO chain type to use:
11557  *                      SD_CHAIN_NULL
11558  *                      SD_CHAIN_BUFIO
11559  *                      SD_CHAIN_USCSI
11560  *                      SD_CHAIN_DIRECT
11561  *                      SD_CHAIN_DIRECT_PRIORITY
11562  *              pktinfop - ptr to private data struct for scsi_pkt(9S)
11563  *                      initialization; may be NULL if none.
11564  *
11565  *     Context: Kernel thread context
11566  */
11567 
11568 static void
11569 sd_xbuf_init(struct sd_lun *un, struct buf *bp, struct sd_xbuf *xp,
11570         uchar_t chain_type, void *pktinfop)
11571 {
11572         int index;
11573 
11574         ASSERT(un != NULL);
11575         ASSERT(bp != NULL);
11576         ASSERT(xp != NULL);
11577 
11578         SD_INFO(SD_LOG_IO, un, "sd_xbuf_init: buf:0x%p chain type:0x%x\n",
11579             bp, chain_type);
11580 
11581         xp->xb_un    = un;
11582         xp->xb_pktp  = NULL;
11583         xp->xb_pktinfo       = pktinfop;
11584         xp->xb_private       = bp->b_private;
11585         xp->xb_blkno = (daddr_t)bp->b_blkno;
11586 
11587         /*
11588          * Set up the iostart and iodone chain indexes in the xbuf, based
11589          * upon the specified chain type to use.
11590          */
11591         switch (chain_type) {
11592         case SD_CHAIN_NULL:
11593                 /*
11594                  * Fall thru to just use the values for the buf type, even
11595                  * tho for the NULL chain these values will never be used.
11596                  */
11597                 /* FALLTHRU */
11598         case SD_CHAIN_BUFIO:
11599                 index = un->un_buf_chain_type;
11600                 if ((!un->un_f_has_removable_media) &&
11601                     (un->un_tgt_blocksize != 0) &&
11602                     (un->un_tgt_blocksize != DEV_BSIZE ||
11603                     un->un_f_enable_rmw)) {
11604                         int secmask = 0, blknomask = 0;
11605                         if (un->un_f_enable_rmw) {
11606                                 blknomask =
11607                                     (un->un_phy_blocksize / DEV_BSIZE) - 1;
11608                                 secmask = un->un_phy_blocksize - 1;
11609                         } else {
11610                                 blknomask =
11611                                     (un->un_tgt_blocksize / DEV_BSIZE) - 1;
11612                                 secmask = un->un_tgt_blocksize - 1;
11613                         }
11614 
11615                         if ((bp->b_lblkno & (blknomask)) ||
11616                             (bp->b_bcount & (secmask))) {
11617                                 if ((un->un_f_rmw_type !=
11618                                     SD_RMW_TYPE_RETURN_ERROR) ||
11619                                     un->un_f_enable_rmw) {
11620                                         if (un->un_f_pm_is_enabled == FALSE)
11621                                                 index =
11622                                                     SD_CHAIN_INFO_MSS_DSK_NO_PM;
11623                                         else
11624                                                 index =
11625                                                     SD_CHAIN_INFO_MSS_DISK;
11626                                 }
11627                         }
11628                 }
11629                 break;
11630         case SD_CHAIN_USCSI:
11631                 index = un->un_uscsi_chain_type;
11632                 break;
11633         case SD_CHAIN_DIRECT:
11634                 index = un->un_direct_chain_type;
11635                 break;
11636         case SD_CHAIN_DIRECT_PRIORITY:
11637                 index = un->un_priority_chain_type;
11638                 break;
11639         default:
11640                 /* We're really broken if we ever get here... */
11641                 panic("sd_xbuf_init: illegal chain type!");
11642                 /*NOTREACHED*/
11643         }
11644 
11645         xp->xb_chain_iostart = sd_chain_index_map[index].sci_iostart_index;
11646         xp->xb_chain_iodone = sd_chain_index_map[index].sci_iodone_index;
11647 
11648         /*
11649          * It might be a bit easier to simply bzero the entire xbuf above,
11650          * but it turns out that since we init a fair number of members anyway,
11651          * we save a fair number cycles by doing explicit assignment of zero.
11652          */
11653         xp->xb_pkt_flags     = 0;
11654         xp->xb_dma_resid     = 0;
11655         xp->xb_retry_count   = 0;
11656         xp->xb_victim_retry_count = 0;
11657         xp->xb_ua_retry_count        = 0;
11658         xp->xb_nr_retry_count        = 0;
11659         xp->xb_sense_bp              = NULL;
11660         xp->xb_sense_status  = 0;
11661         xp->xb_sense_state   = 0;
11662         xp->xb_sense_resid   = 0;
11663         xp->xb_ena           = 0;
11664 
11665         bp->b_private        = xp;
11666         bp->b_flags  &= ~(B_DONE | B_ERROR);
11667         bp->b_resid  = 0;
11668         bp->av_forw  = NULL;
11669         bp->av_back  = NULL;
11670         bioerror(bp, 0);
11671 
11672         SD_INFO(SD_LOG_IO, un, "sd_xbuf_init: done.\n");
11673 }
11674 
11675 
11676 /*
11677  *    Function: sd_uscsi_strategy
11678  *
11679  * Description: Wrapper for calling into the USCSI chain via physio(9F)
11680  *
11681  *   Arguments: bp - buf struct ptr
11682  *
11683  * Return Code: Always returns 0
11684  *
11685  *     Context: Kernel thread context
11686  */
11687 
11688 static int
11689 sd_uscsi_strategy(struct buf *bp)
11690 {
11691         struct sd_lun           *un;
11692         struct sd_uscsi_info    *uip;
11693         struct sd_xbuf          *xp;
11694         uchar_t                 chain_type;
11695         uchar_t                 cmd;
11696 
11697         ASSERT(bp != NULL);
11698 
11699         un = ddi_get_soft_state(sd_state, SD_GET_INSTANCE_FROM_BUF(bp));
11700         if (un == NULL) {
11701                 bioerror(bp, EIO);
11702                 bp->b_resid = bp->b_bcount;
11703                 biodone(bp);
11704                 return (0);
11705         }
11706 
11707         ASSERT(!mutex_owned(SD_MUTEX(un)));
11708 
11709         SD_TRACE(SD_LOG_IO, un, "sd_uscsi_strategy: entry: buf:0x%p\n", bp);
11710 
11711         /*
11712          * A pointer to a struct sd_uscsi_info is expected in bp->b_private
11713          */
11714         ASSERT(bp->b_private != NULL);
11715         uip = (struct sd_uscsi_info *)bp->b_private;
11716         cmd = ((struct uscsi_cmd *)(uip->ui_cmdp))->uscsi_cdb[0];
11717 
11718         mutex_enter(SD_MUTEX(un));
11719         /*
11720          * atapi: Since we are running the CD for now in PIO mode we need to
11721          * call bp_mapin here to avoid bp_mapin called interrupt context under
11722          * the HBA's init_pkt routine.
11723          */
11724         if (un->un_f_cfg_is_atapi == TRUE) {
11725                 mutex_exit(SD_MUTEX(un));
11726                 bp_mapin(bp);
11727                 mutex_enter(SD_MUTEX(un));
11728         }
11729         un->un_ncmds_in_driver++;
11730         SD_INFO(SD_LOG_IO, un, "sd_uscsi_strategy: un_ncmds_in_driver = %ld\n",
11731             un->un_ncmds_in_driver);
11732 
11733         if ((bp->b_flags & B_WRITE) && (bp->b_bcount != 0) &&
11734             (cmd != SCMD_MODE_SELECT) && (cmd != SCMD_MODE_SELECT_G1))
11735                 un->un_f_sync_cache_required = TRUE;
11736 
11737         mutex_exit(SD_MUTEX(un));
11738 
11739         switch (uip->ui_flags) {
11740         case SD_PATH_DIRECT:
11741                 chain_type = SD_CHAIN_DIRECT;
11742                 break;
11743         case SD_PATH_DIRECT_PRIORITY:
11744                 chain_type = SD_CHAIN_DIRECT_PRIORITY;
11745                 break;
11746         default:
11747                 chain_type = SD_CHAIN_USCSI;
11748                 break;
11749         }
11750 
11751         /*
11752          * We may allocate extra buf for external USCSI commands. If the
11753          * application asks for bigger than 20-byte sense data via USCSI,
11754          * SCSA layer will allocate 252 bytes sense buf for that command.
11755          */
11756         if (((struct uscsi_cmd *)(uip->ui_cmdp))->uscsi_rqlen >
11757             SENSE_LENGTH) {
11758                 xp = kmem_zalloc(sizeof (struct sd_xbuf) - SENSE_LENGTH +
11759                     MAX_SENSE_LENGTH, KM_SLEEP);
11760         } else {
11761                 xp = kmem_zalloc(sizeof (struct sd_xbuf), KM_SLEEP);
11762         }
11763 
11764         sd_xbuf_init(un, bp, xp, chain_type, uip->ui_cmdp);
11765 
11766         /* Use the index obtained within xbuf_init */
11767         SD_BEGIN_IOSTART(xp->xb_chain_iostart, un, bp);
11768 
11769         SD_TRACE(SD_LOG_IO, un, "sd_uscsi_strategy: exit: buf:0x%p\n", bp);
11770 
11771         return (0);
11772 }
11773 
11774 /*
11775  *    Function: sd_send_scsi_cmd
11776  *
11777  * Description: Runs a USCSI command for user (when called thru sdioctl),
11778  *              or for the driver
11779  *
11780  *   Arguments: dev - the dev_t for the device
11781  *              incmd - ptr to a valid uscsi_cmd struct
11782  *              flag - bit flag, indicating open settings, 32/64 bit type
11783  *              dataspace - UIO_USERSPACE or UIO_SYSSPACE
11784  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
11785  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
11786  *                      to use the USCSI "direct" chain and bypass the normal
11787  *                      command waitq.
11788  *
11789  * Return Code: 0 -  successful completion of the given command
11790  *              EIO - scsi_uscsi_handle_command() failed
11791  *              ENXIO  - soft state not found for specified dev
11792  *              EINVAL
11793  *              EFAULT - copyin/copyout error
11794  *              return code of scsi_uscsi_handle_command():
11795  *                      EIO
11796  *                      ENXIO
11797  *                      EACCES
11798  *
11799  *     Context: Waits for command to complete. Can sleep.
11800  */
11801 
11802 static int
11803 sd_send_scsi_cmd(dev_t dev, struct uscsi_cmd *incmd, int flag,
11804         enum uio_seg dataspace, int path_flag)
11805 {
11806         struct sd_lun   *un;
11807         sd_ssc_t        *ssc;
11808         int             rval;
11809 
11810         un = ddi_get_soft_state(sd_state, SDUNIT(dev));
11811         if (un == NULL) {
11812                 return (ENXIO);
11813         }
11814 
11815         /*
11816          * Using sd_ssc_send to handle uscsi cmd
11817          */
11818         ssc = sd_ssc_init(un);
11819         rval = sd_ssc_send(ssc, incmd, flag, dataspace, path_flag);
11820         sd_ssc_fini(ssc);
11821 
11822         return (rval);
11823 }
11824 
11825 /*
11826  *    Function: sd_ssc_init
11827  *
11828  * Description: Uscsi end-user call this function to initialize necessary
11829  *              fields, such as uscsi_cmd and sd_uscsi_info struct.
11830  *
11831  *              The return value of sd_send_scsi_cmd will be treated as a
11832  *              fault in various conditions. Even it is not Zero, some
11833  *              callers may ignore the return value. That is to say, we can
11834  *              not make an accurate assessment in sdintr, since if a
11835  *              command is failed in sdintr it does not mean the caller of
11836  *              sd_send_scsi_cmd will treat it as a real failure.
11837  *
11838  *              To avoid printing too many error logs for a failed uscsi
11839  *              packet that the caller may not treat it as a failure, the
11840  *              sd will keep silent for handling all uscsi commands.
11841  *
11842  *              During detach->attach and attach-open, for some types of
11843  *              problems, the driver should be providing information about
11844  *              the problem encountered. Device use USCSI_SILENT, which
11845  *              suppresses all driver information. The result is that no
11846  *              information about the problem is available. Being
11847  *              completely silent during this time is inappropriate. The
11848  *              driver needs a more selective filter than USCSI_SILENT, so
11849  *              that information related to faults is provided.
11850  *
11851  *              To make the accurate accessment, the caller  of
11852  *              sd_send_scsi_USCSI_CMD should take the ownership and
11853  *              get necessary information to print error messages.
11854  *
11855  *              If we want to print necessary info of uscsi command, we need to
11856  *              keep the uscsi_cmd and sd_uscsi_info till we can make the
11857  *              assessment. We use sd_ssc_init to alloc necessary
11858  *              structs for sending an uscsi command and we are also
11859  *              responsible for free the memory by calling
11860  *              sd_ssc_fini.
11861  *
11862  *              The calling secquences will look like:
11863  *              sd_ssc_init->
11864  *
11865  *                  ...
11866  *
11867  *                  sd_send_scsi_USCSI_CMD->
11868  *                      sd_ssc_send-> - - - sdintr
11869  *                  ...
11870  *
11871  *                  if we think the return value should be treated as a
11872  *                  failure, we make the accessment here and print out
11873  *                  necessary by retrieving uscsi_cmd and sd_uscsi_info'
11874  *
11875  *                  ...
11876  *
11877  *              sd_ssc_fini
11878  *
11879  *
11880  *   Arguments: un - pointer to driver soft state (unit) structure for this
11881  *                   target.
11882  *
11883  * Return code: sd_ssc_t - pointer to allocated sd_ssc_t struct, it contains
11884  *                         uscsi_cmd and sd_uscsi_info.
11885  *                  NULL - if can not alloc memory for sd_ssc_t struct
11886  *
11887  *     Context: Kernel Thread.
11888  */
11889 static sd_ssc_t *
11890 sd_ssc_init(struct sd_lun *un)
11891 {
11892         sd_ssc_t                *ssc;
11893         struct uscsi_cmd        *ucmdp;
11894         struct sd_uscsi_info    *uip;
11895 
11896         ASSERT(un != NULL);
11897         ASSERT(!mutex_owned(SD_MUTEX(un)));
11898 
11899         /*
11900          * Allocate sd_ssc_t structure
11901          */
11902         ssc = kmem_zalloc(sizeof (sd_ssc_t), KM_SLEEP);
11903 
11904         /*
11905          * Allocate uscsi_cmd by calling scsi_uscsi_alloc common routine
11906          */
11907         ucmdp = scsi_uscsi_alloc();
11908 
11909         /*
11910          * Allocate sd_uscsi_info structure
11911          */
11912         uip = kmem_zalloc(sizeof (struct sd_uscsi_info), KM_SLEEP);
11913 
11914         ssc->ssc_uscsi_cmd = ucmdp;
11915         ssc->ssc_uscsi_info = uip;
11916         ssc->ssc_un = un;
11917 
11918         return (ssc);
11919 }
11920 
11921 /*
11922  * Function: sd_ssc_fini
11923  *
11924  * Description: To free sd_ssc_t and it's hanging off
11925  *
11926  * Arguments: ssc - struct pointer of sd_ssc_t.
11927  */
11928 static void
11929 sd_ssc_fini(sd_ssc_t *ssc)
11930 {
11931         scsi_uscsi_free(ssc->ssc_uscsi_cmd);
11932 
11933         if (ssc->ssc_uscsi_info != NULL) {
11934                 kmem_free(ssc->ssc_uscsi_info, sizeof (struct sd_uscsi_info));
11935                 ssc->ssc_uscsi_info = NULL;
11936         }
11937 
11938         kmem_free(ssc, sizeof (sd_ssc_t));
11939         ssc = NULL;
11940 }
11941 
11942 /*
11943  * Function: sd_ssc_send
11944  *
11945  * Description: Runs a USCSI command for user when called through sdioctl,
11946  *              or for the driver.
11947  *
11948  *   Arguments: ssc - the struct of sd_ssc_t will bring uscsi_cmd and
11949  *                    sd_uscsi_info in.
11950  *              incmd - ptr to a valid uscsi_cmd struct
11951  *              flag - bit flag, indicating open settings, 32/64 bit type
11952  *              dataspace - UIO_USERSPACE or UIO_SYSSPACE
11953  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
11954  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
11955  *                      to use the USCSI "direct" chain and bypass the normal
11956  *                      command waitq.
11957  *
11958  * Return Code: 0 -  successful completion of the given command
11959  *              EIO - scsi_uscsi_handle_command() failed
11960  *              ENXIO  - soft state not found for specified dev
11961  *              ECANCELED - command cancelled due to low power
11962  *              EINVAL
11963  *              EFAULT - copyin/copyout error
11964  *              return code of scsi_uscsi_handle_command():
11965  *                      EIO
11966  *                      ENXIO
11967  *                      EACCES
11968  *
11969  *     Context: Kernel Thread;
11970  *              Waits for command to complete. Can sleep.
11971  */
11972 static int
11973 sd_ssc_send(sd_ssc_t *ssc, struct uscsi_cmd *incmd, int flag,
11974         enum uio_seg dataspace, int path_flag)
11975 {
11976         struct sd_uscsi_info    *uip;
11977         struct uscsi_cmd        *uscmd;
11978         struct sd_lun           *un;
11979         dev_t                   dev;
11980 
11981         int     format = 0;
11982         int     rval;
11983 
11984         ASSERT(ssc != NULL);
11985         un = ssc->ssc_un;
11986         ASSERT(un != NULL);
11987         uscmd = ssc->ssc_uscsi_cmd;
11988         ASSERT(uscmd != NULL);
11989         ASSERT(!mutex_owned(SD_MUTEX(un)));
11990         if (ssc->ssc_flags & SSC_FLAGS_NEED_ASSESSMENT) {
11991                 /*
11992                  * If enter here, it indicates that the previous uscsi
11993                  * command has not been processed by sd_ssc_assessment.
11994                  * This is violating our rules of FMA telemetry processing.
11995                  * We should print out this message and the last undisposed
11996                  * uscsi command.
11997                  */
11998                 if (uscmd->uscsi_cdb != NULL) {
11999                         SD_INFO(SD_LOG_SDTEST, un,
12000                             "sd_ssc_send is missing the alternative "
12001                             "sd_ssc_assessment when running command 0x%x.\n",
12002                             uscmd->uscsi_cdb[0]);
12003                 }
12004                 /*
12005                  * Set the ssc_flags to SSC_FLAGS_UNKNOWN, which should be
12006                  * the initial status.
12007                  */
12008                 ssc->ssc_flags = SSC_FLAGS_UNKNOWN;
12009         }
12010 
12011         /*
12012          * We need to make sure sd_ssc_send will have sd_ssc_assessment
12013          * followed to avoid missing FMA telemetries.
12014          */
12015         ssc->ssc_flags |= SSC_FLAGS_NEED_ASSESSMENT;
12016 
12017         /*
12018          * if USCSI_PMFAILFAST is set and un is in low power, fail the
12019          * command immediately.
12020          */
12021         mutex_enter(SD_MUTEX(un));
12022         mutex_enter(&un->un_pm_mutex);
12023         if ((uscmd->uscsi_flags & USCSI_PMFAILFAST) &&
12024             SD_DEVICE_IS_IN_LOW_POWER(un)) {
12025                 SD_TRACE(SD_LOG_IO, un, "sd_ssc_send:"
12026                     "un:0x%p is in low power\n", un);
12027                 mutex_exit(&un->un_pm_mutex);
12028                 mutex_exit(SD_MUTEX(un));
12029                 return (ECANCELED);
12030         }
12031         mutex_exit(&un->un_pm_mutex);
12032         mutex_exit(SD_MUTEX(un));
12033 
12034 #ifdef SDDEBUG
12035         switch (dataspace) {
12036         case UIO_USERSPACE:
12037                 SD_TRACE(SD_LOG_IO, un,
12038                     "sd_ssc_send: entry: un:0x%p UIO_USERSPACE\n", un);
12039                 break;
12040         case UIO_SYSSPACE:
12041                 SD_TRACE(SD_LOG_IO, un,
12042                     "sd_ssc_send: entry: un:0x%p UIO_SYSSPACE\n", un);
12043                 break;
12044         default:
12045                 SD_TRACE(SD_LOG_IO, un,
12046                     "sd_ssc_send: entry: un:0x%p UNEXPECTED SPACE\n", un);
12047                 break;
12048         }
12049 #endif
12050 
12051         rval = scsi_uscsi_copyin((intptr_t)incmd, flag,
12052             SD_ADDRESS(un), &uscmd);
12053         if (rval != 0) {
12054                 SD_TRACE(SD_LOG_IO, un, "sd_sense_scsi_cmd: "
12055                     "scsi_uscsi_alloc_and_copyin failed\n", un);
12056                 return (rval);
12057         }
12058 
12059         if ((uscmd->uscsi_cdb != NULL) &&
12060             (uscmd->uscsi_cdb[0] == SCMD_FORMAT)) {
12061                 mutex_enter(SD_MUTEX(un));
12062                 un->un_f_format_in_progress = TRUE;
12063                 mutex_exit(SD_MUTEX(un));
12064                 format = 1;
12065         }
12066 
12067         /*
12068          * Allocate an sd_uscsi_info struct and fill it with the info
12069          * needed by sd_initpkt_for_uscsi().  Then put the pointer into
12070          * b_private in the buf for sd_initpkt_for_uscsi().  Note that
12071          * since we allocate the buf here in this function, we do not
12072          * need to preserve the prior contents of b_private.
12073          * The sd_uscsi_info struct is also used by sd_uscsi_strategy()
12074          */
12075         uip = ssc->ssc_uscsi_info;
12076         uip->ui_flags = path_flag;
12077         uip->ui_cmdp = uscmd;
12078 
12079         /*
12080          * Commands sent with priority are intended for error recovery
12081          * situations, and do not have retries performed.
12082          */
12083         if (path_flag == SD_PATH_DIRECT_PRIORITY) {
12084                 uscmd->uscsi_flags |= USCSI_DIAGNOSE;
12085         }
12086         uscmd->uscsi_flags &= ~USCSI_NOINTR;
12087 
12088         dev = SD_GET_DEV(un);
12089         rval = scsi_uscsi_handle_cmd(dev, dataspace, uscmd,
12090             sd_uscsi_strategy, NULL, uip);
12091 
12092         /*
12093          * mark ssc_flags right after handle_cmd to make sure
12094          * the uscsi has been sent
12095          */
12096         ssc->ssc_flags |= SSC_FLAGS_CMD_ISSUED;
12097 
12098 #ifdef SDDEBUG
12099         SD_INFO(SD_LOG_IO, un, "sd_ssc_send: "
12100             "uscsi_status: 0x%02x  uscsi_resid:0x%x\n",
12101             uscmd->uscsi_status, uscmd->uscsi_resid);
12102         if (uscmd->uscsi_bufaddr != NULL) {
12103                 SD_INFO(SD_LOG_IO, un, "sd_ssc_send: "
12104                     "uscmd->uscsi_bufaddr: 0x%p  uscmd->uscsi_buflen:%d\n",
12105                     uscmd->uscsi_bufaddr, uscmd->uscsi_buflen);
12106                 if (dataspace == UIO_SYSSPACE) {
12107                         SD_DUMP_MEMORY(un, SD_LOG_IO,
12108                             "data", (uchar_t *)uscmd->uscsi_bufaddr,
12109                             uscmd->uscsi_buflen, SD_LOG_HEX);
12110                 }
12111         }
12112 #endif
12113 
12114         if (format == 1) {
12115                 mutex_enter(SD_MUTEX(un));
12116                 un->un_f_format_in_progress = FALSE;
12117                 mutex_exit(SD_MUTEX(un));
12118         }
12119 
12120         (void) scsi_uscsi_copyout((intptr_t)incmd, uscmd);
12121 
12122         return (rval);
12123 }
12124 
12125 /*
12126  *     Function: sd_ssc_print
12127  *
12128  * Description: Print information available to the console.
12129  *
12130  * Arguments: ssc - the struct of sd_ssc_t will bring uscsi_cmd and
12131  *                    sd_uscsi_info in.
12132  *            sd_severity - log level.
12133  *     Context: Kernel thread or interrupt context.
12134  */
12135 static void
12136 sd_ssc_print(sd_ssc_t *ssc, int sd_severity)
12137 {
12138         struct uscsi_cmd        *ucmdp;
12139         struct scsi_device      *devp;
12140         dev_info_t              *devinfo;
12141         uchar_t                 *sensep;
12142         int                     senlen;
12143         union scsi_cdb          *cdbp;
12144         uchar_t                 com;
12145         extern struct scsi_key_strings scsi_cmds[];
12146 
12147         ASSERT(ssc != NULL);
12148         ASSERT(ssc->ssc_un != NULL);
12149 
12150         if (SD_FM_LOG(ssc->ssc_un) != SD_FM_LOG_EREPORT)
12151                 return;
12152         ucmdp = ssc->ssc_uscsi_cmd;
12153         devp = SD_SCSI_DEVP(ssc->ssc_un);
12154         devinfo = SD_DEVINFO(ssc->ssc_un);
12155         ASSERT(ucmdp != NULL);
12156         ASSERT(devp != NULL);
12157         ASSERT(devinfo != NULL);
12158         sensep = (uint8_t *)ucmdp->uscsi_rqbuf;
12159         senlen = ucmdp->uscsi_rqlen - ucmdp->uscsi_rqresid;
12160         cdbp = (union scsi_cdb *)ucmdp->uscsi_cdb;
12161 
12162         /* In certain case (like DOORLOCK), the cdb could be NULL. */
12163         if (cdbp == NULL)
12164                 return;
12165         /* We don't print log if no sense data available. */
12166         if (senlen == 0)
12167                 sensep = NULL;
12168         com = cdbp->scc_cmd;
12169         scsi_generic_errmsg(devp, sd_label, sd_severity, 0, 0, com,
12170             scsi_cmds, sensep, ssc->ssc_un->un_additional_codes, NULL);
12171 }
12172 
12173 /*
12174  *     Function: sd_ssc_assessment
12175  *
12176  * Description: We use this function to make an assessment at the point
12177  *              where SD driver may encounter a potential error.
12178  *
12179  * Arguments: ssc - the struct of sd_ssc_t will bring uscsi_cmd and
12180  *                  sd_uscsi_info in.
12181  *            tp_assess - a hint of strategy for ereport posting.
12182  *            Possible values of tp_assess include:
12183  *                SD_FMT_IGNORE - we don't post any ereport because we're
12184  *                sure that it is ok to ignore the underlying problems.
12185  *                SD_FMT_IGNORE_COMPROMISE - we don't post any ereport for now
12186  *                but it might be not correct to ignore the underlying hardware
12187  *                error.
12188  *                SD_FMT_STATUS_CHECK - we will post an ereport with the
12189  *                payload driver-assessment of value "fail" or
12190  *                "fatal"(depending on what information we have here). This
12191  *                assessment value is usually set when SD driver think there
12192  *                is a potential error occurred(Typically, when return value
12193  *                of the SCSI command is EIO).
12194  *                SD_FMT_STANDARD - we will post an ereport with the payload
12195  *                driver-assessment of value "info". This assessment value is
12196  *                set when the SCSI command returned successfully and with
12197  *                sense data sent back.
12198  *
12199  *     Context: Kernel thread.
12200  */
12201 static void
12202 sd_ssc_assessment(sd_ssc_t *ssc, enum sd_type_assessment tp_assess)
12203 {
12204         int senlen = 0;
12205         struct uscsi_cmd *ucmdp = NULL;
12206         struct sd_lun *un;
12207 
12208         ASSERT(ssc != NULL);
12209         un = ssc->ssc_un;
12210         ASSERT(un != NULL);
12211         ucmdp = ssc->ssc_uscsi_cmd;
12212         ASSERT(ucmdp != NULL);
12213 
12214         if (ssc->ssc_flags & SSC_FLAGS_NEED_ASSESSMENT) {
12215                 ssc->ssc_flags &= ~SSC_FLAGS_NEED_ASSESSMENT;
12216         } else {
12217                 /*
12218                  * If enter here, it indicates that we have a wrong
12219                  * calling sequence of sd_ssc_send and sd_ssc_assessment,
12220                  * both of which should be called in a pair in case of
12221                  * loss of FMA telemetries.
12222                  */
12223                 if (ucmdp->uscsi_cdb != NULL) {
12224                         SD_INFO(SD_LOG_SDTEST, un,
12225                             "sd_ssc_assessment is missing the "
12226                             "alternative sd_ssc_send when running 0x%x, "
12227                             "or there are superfluous sd_ssc_assessment for "
12228                             "the same sd_ssc_send.\n",
12229                             ucmdp->uscsi_cdb[0]);
12230                 }
12231                 /*
12232                  * Set the ssc_flags to the initial value to avoid passing
12233                  * down dirty flags to the following sd_ssc_send function.
12234                  */
12235                 ssc->ssc_flags = SSC_FLAGS_UNKNOWN;
12236                 return;
12237         }
12238 
12239         /*
12240          * Only handle an issued command which is waiting for assessment.
12241          * A command which is not issued will not have
12242          * SSC_FLAGS_INVALID_DATA set, so it'ok we just return here.
12243          */
12244         if (!(ssc->ssc_flags & SSC_FLAGS_CMD_ISSUED)) {
12245                 sd_ssc_print(ssc, SCSI_ERR_INFO);
12246                 return;
12247         } else {
12248                 /*
12249                  * For an issued command, we should clear this flag in
12250                  * order to make the sd_ssc_t structure be used off
12251                  * multiple uscsi commands.
12252                  */
12253                 ssc->ssc_flags &= ~SSC_FLAGS_CMD_ISSUED;
12254         }
12255 
12256         /*
12257          * We will not deal with non-retryable(flag USCSI_DIAGNOSE set)
12258          * commands here. And we should clear the ssc_flags before return.
12259          */
12260         if (ucmdp->uscsi_flags & USCSI_DIAGNOSE) {
12261                 ssc->ssc_flags = SSC_FLAGS_UNKNOWN;
12262                 return;
12263         }
12264 
12265         switch (tp_assess) {
12266         case SD_FMT_IGNORE:
12267         case SD_FMT_IGNORE_COMPROMISE:
12268                 break;
12269         case SD_FMT_STATUS_CHECK:
12270                 /*
12271                  * For a failed command(including the succeeded command
12272                  * with invalid data sent back).
12273                  */
12274                 sd_ssc_post(ssc, SD_FM_DRV_FATAL);
12275                 break;
12276         case SD_FMT_STANDARD:
12277                 /*
12278                  * Always for the succeeded commands probably with sense
12279                  * data sent back.
12280                  * Limitation:
12281                  *      We can only handle a succeeded command with sense
12282                  *      data sent back when auto-request-sense is enabled.
12283                  */
12284                 senlen = ssc->ssc_uscsi_cmd->uscsi_rqlen -
12285                     ssc->ssc_uscsi_cmd->uscsi_rqresid;
12286                 if ((ssc->ssc_uscsi_info->ui_pkt_state & STATE_ARQ_DONE) &&
12287                     (un->un_f_arq_enabled == TRUE) &&
12288                     senlen > 0 &&
12289                     ssc->ssc_uscsi_cmd->uscsi_rqbuf != NULL) {
12290                         sd_ssc_post(ssc, SD_FM_DRV_NOTICE);
12291                 }
12292                 break;
12293         default:
12294                 /*
12295                  * Should not have other type of assessment.
12296                  */
12297                 scsi_log(SD_DEVINFO(un), sd_label, CE_CONT,
12298                     "sd_ssc_assessment got wrong "
12299                     "sd_type_assessment %d.\n", tp_assess);
12300                 break;
12301         }
12302         /*
12303          * Clear up the ssc_flags before return.
12304          */
12305         ssc->ssc_flags = SSC_FLAGS_UNKNOWN;
12306 }
12307 
12308 /*
12309  *    Function: sd_ssc_post
12310  *
12311  * Description: 1. read the driver property to get fm-scsi-log flag.
12312  *              2. print log if fm_log_capable is non-zero.
12313  *              3. call sd_ssc_ereport_post to post ereport if possible.
12314  *
12315  *    Context: May be called from kernel thread or interrupt context.
12316  */
12317 static void
12318 sd_ssc_post(sd_ssc_t *ssc, enum sd_driver_assessment sd_assess)
12319 {
12320         struct sd_lun   *un;
12321         int             sd_severity;
12322 
12323         ASSERT(ssc != NULL);
12324         un = ssc->ssc_un;
12325         ASSERT(un != NULL);
12326 
12327         /*
12328          * We may enter here from sd_ssc_assessment(for USCSI command) or
12329          * by directly called from sdintr context.
12330          * We don't handle a non-disk drive(CD-ROM, removable media).
12331          * Clear the ssc_flags before return in case we've set
12332          * SSC_FLAGS_INVALID_XXX which should be skipped for a non-disk
12333          * driver.
12334          */
12335         if (ISCD(un) || un->un_f_has_removable_media) {
12336                 ssc->ssc_flags = SSC_FLAGS_UNKNOWN;
12337                 return;
12338         }
12339 
12340         switch (sd_assess) {
12341                 case SD_FM_DRV_FATAL:
12342                         sd_severity = SCSI_ERR_FATAL;
12343                         break;
12344                 case SD_FM_DRV_RECOVERY:
12345                         sd_severity = SCSI_ERR_RECOVERED;
12346                         break;
12347                 case SD_FM_DRV_RETRY:
12348                         sd_severity = SCSI_ERR_RETRYABLE;
12349                         break;
12350                 case SD_FM_DRV_NOTICE:
12351                         sd_severity = SCSI_ERR_INFO;
12352                         break;
12353                 default:
12354                         sd_severity = SCSI_ERR_UNKNOWN;
12355         }
12356         /* print log */
12357         sd_ssc_print(ssc, sd_severity);
12358 
12359         /* always post ereport */
12360         sd_ssc_ereport_post(ssc, sd_assess);
12361 }
12362 
12363 /*
12364  *    Function: sd_ssc_set_info
12365  *
12366  * Description: Mark ssc_flags and set ssc_info which would be the
12367  *              payload of uderr ereport. This function will cause
12368  *              sd_ssc_ereport_post to post uderr ereport only.
12369  *              Besides, when ssc_flags == SSC_FLAGS_INVALID_DATA(USCSI),
12370  *              the function will also call SD_ERROR or scsi_log for a
12371  *              CDROM/removable-media/DDI_FM_NOT_CAPABLE device.
12372  *
12373  * Arguments: ssc - the struct of sd_ssc_t will bring uscsi_cmd and
12374  *                  sd_uscsi_info in.
12375  *            ssc_flags - indicate the sub-category of a uderr.
12376  *            comp - this argument is meaningful only when
12377  *                   ssc_flags == SSC_FLAGS_INVALID_DATA, and its possible
12378  *                   values include:
12379  *                   > 0, SD_ERROR is used with comp as the driver logging
12380  *                   component;
12381  *                   = 0, scsi-log is used to log error telemetries;
12382  *                   < 0, no log available for this telemetry.
12383  *
12384  *    Context: Kernel thread or interrupt context
12385  */
12386 static void
12387 sd_ssc_set_info(sd_ssc_t *ssc, int ssc_flags, uint_t comp, const char *fmt, ...)
12388 {
12389         va_list ap;
12390 
12391         ASSERT(ssc != NULL);
12392         ASSERT(ssc->ssc_un != NULL);
12393 
12394         ssc->ssc_flags |= ssc_flags;
12395         va_start(ap, fmt);
12396         (void) vsnprintf(ssc->ssc_info, sizeof (ssc->ssc_info), fmt, ap);
12397         va_end(ap);
12398 
12399         /*
12400          * If SSC_FLAGS_INVALID_DATA is set, it should be a uscsi command
12401          * with invalid data sent back. For non-uscsi command, the
12402          * following code will be bypassed.
12403          */
12404         if (ssc_flags & SSC_FLAGS_INVALID_DATA) {
12405                 if (SD_FM_LOG(ssc->ssc_un) == SD_FM_LOG_NSUP) {
12406                         /*
12407                          * If the error belong to certain component and we
12408                          * do not want it to show up on the console, we
12409                          * will use SD_ERROR, otherwise scsi_log is
12410                          * preferred.
12411                          */
12412                         if (comp > 0) {
12413                                 SD_ERROR(comp, ssc->ssc_un, ssc->ssc_info);
12414                         } else if (comp == 0) {
12415                                 scsi_log(SD_DEVINFO(ssc->ssc_un), sd_label,
12416                                     CE_WARN, ssc->ssc_info);
12417                         }
12418                 }
12419         }
12420 }
12421 
12422 /*
12423  *    Function: sd_buf_iodone
12424  *
12425  * Description: Frees the sd_xbuf & returns the buf to its originator.
12426  *
12427  *     Context: May be called from interrupt context.
12428  */
12429 /* ARGSUSED */
12430 static void
12431 sd_buf_iodone(int index, struct sd_lun *un, struct buf *bp)
12432 {
12433         struct sd_xbuf *xp;
12434 
12435         ASSERT(un != NULL);
12436         ASSERT(bp != NULL);
12437         ASSERT(!mutex_owned(SD_MUTEX(un)));
12438 
12439         SD_TRACE(SD_LOG_IO_CORE, un, "sd_buf_iodone: entry.\n");
12440 
12441         xp = SD_GET_XBUF(bp);
12442         ASSERT(xp != NULL);
12443 
12444         /* xbuf is gone after this */
12445         if (ddi_xbuf_done(bp, un->un_xbuf_attr)) {
12446                 mutex_enter(SD_MUTEX(un));
12447 
12448                 /*
12449                  * Grab time when the cmd completed.
12450                  * This is used for determining if the system has been
12451                  * idle long enough to make it idle to the PM framework.
12452                  * This is for lowering the overhead, and therefore improving
12453                  * performance per I/O operation.
12454                  */
12455                 un->un_pm_idle_time = ddi_get_time();
12456 
12457                 un->un_ncmds_in_driver--;
12458                 ASSERT(un->un_ncmds_in_driver >= 0);
12459                 SD_INFO(SD_LOG_IO, un,
12460                     "sd_buf_iodone: un_ncmds_in_driver = %ld\n",
12461                     un->un_ncmds_in_driver);
12462 
12463                 mutex_exit(SD_MUTEX(un));
12464         }
12465 
12466         biodone(bp);                            /* bp is gone after this */
12467 
12468         SD_TRACE(SD_LOG_IO_CORE, un, "sd_buf_iodone: exit.\n");
12469 }
12470 
12471 
12472 /*
12473  *    Function: sd_uscsi_iodone
12474  *
12475  * Description: Frees the sd_xbuf & returns the buf to its originator.
12476  *
12477  *     Context: May be called from interrupt context.
12478  */
12479 /* ARGSUSED */
12480 static void
12481 sd_uscsi_iodone(int index, struct sd_lun *un, struct buf *bp)
12482 {
12483         struct sd_xbuf *xp;
12484 
12485         ASSERT(un != NULL);
12486         ASSERT(bp != NULL);
12487 
12488         xp = SD_GET_XBUF(bp);
12489         ASSERT(xp != NULL);
12490         ASSERT(!mutex_owned(SD_MUTEX(un)));
12491 
12492         SD_INFO(SD_LOG_IO, un, "sd_uscsi_iodone: entry.\n");
12493 
12494         bp->b_private = xp->xb_private;
12495 
12496         mutex_enter(SD_MUTEX(un));
12497 
12498         /*
12499          * Grab time when the cmd completed.
12500          * This is used for determining if the system has been
12501          * idle long enough to make it idle to the PM framework.
12502          * This is for lowering the overhead, and therefore improving
12503          * performance per I/O operation.
12504          */
12505         un->un_pm_idle_time = ddi_get_time();
12506 
12507         un->un_ncmds_in_driver--;
12508         ASSERT(un->un_ncmds_in_driver >= 0);
12509         SD_INFO(SD_LOG_IO, un, "sd_uscsi_iodone: un_ncmds_in_driver = %ld\n",
12510             un->un_ncmds_in_driver);
12511 
12512         mutex_exit(SD_MUTEX(un));
12513 
12514         if (((struct uscsi_cmd *)(xp->xb_pktinfo))->uscsi_rqlen >
12515             SENSE_LENGTH) {
12516                 kmem_free(xp, sizeof (struct sd_xbuf) - SENSE_LENGTH +
12517                     MAX_SENSE_LENGTH);
12518         } else {
12519                 kmem_free(xp, sizeof (struct sd_xbuf));
12520         }
12521 
12522         biodone(bp);
12523 
12524         SD_INFO(SD_LOG_IO, un, "sd_uscsi_iodone: exit.\n");
12525 }
12526 
12527 
12528 /*
12529  *    Function: sd_mapblockaddr_iostart
12530  *
12531  * Description: Verify request lies within the partition limits for
12532  *              the indicated minor device.  Issue "overrun" buf if
12533  *              request would exceed partition range.  Converts
12534  *              partition-relative block address to absolute.
12535  *
12536  *              Upon exit of this function:
12537  *              1.I/O is aligned
12538  *                 xp->xb_blkno represents the absolute sector address
12539  *              2.I/O is misaligned
12540  *                 xp->xb_blkno represents the absolute logical block address
12541  *                 based on DEV_BSIZE. The logical block address will be
12542  *                 converted to physical sector address in sd_mapblocksize_\
12543  *                 iostart.
12544  *              3.I/O is misaligned but is aligned in "overrun" buf
12545  *                 xp->xb_blkno represents the absolute logical block address
12546  *                 based on DEV_BSIZE. The logical block address will be
12547  *                 converted to physical sector address in sd_mapblocksize_\
12548  *                 iostart. But no RMW will be issued in this case.
12549  *
12550  *     Context: Can sleep
12551  *
12552  *      Issues: This follows what the old code did, in terms of accessing
12553  *              some of the partition info in the unit struct without holding
12554  *              the mutext.  This is a general issue, if the partition info
12555  *              can be altered while IO is in progress... as soon as we send
12556  *              a buf, its partitioning can be invalid before it gets to the
12557  *              device.  Probably the right fix is to move partitioning out
12558  *              of the driver entirely.
12559  */
12560 
12561 static void
12562 sd_mapblockaddr_iostart(int index, struct sd_lun *un, struct buf *bp)
12563 {
12564         diskaddr_t      nblocks;        /* #blocks in the given partition */
12565         daddr_t blocknum;       /* Block number specified by the buf */
12566         size_t  requested_nblocks;
12567         size_t  available_nblocks;
12568         int     partition;
12569         diskaddr_t      partition_offset;
12570         struct sd_xbuf *xp;
12571         int secmask = 0, blknomask = 0;
12572         ushort_t is_aligned = TRUE;
12573 
12574         ASSERT(un != NULL);
12575         ASSERT(bp != NULL);
12576         ASSERT(!mutex_owned(SD_MUTEX(un)));
12577 
12578         SD_TRACE(SD_LOG_IO_PARTITION, un,
12579             "sd_mapblockaddr_iostart: entry: buf:0x%p\n", bp);
12580 
12581         xp = SD_GET_XBUF(bp);
12582         ASSERT(xp != NULL);
12583 
12584         /*
12585          * If the geometry is not indicated as valid, attempt to access
12586          * the unit & verify the geometry/label. This can be the case for
12587          * removable-media devices, of if the device was opened in
12588          * NDELAY/NONBLOCK mode.
12589          */
12590         partition = SDPART(bp->b_edev);
12591 
12592         if (!SD_IS_VALID_LABEL(un)) {
12593                 sd_ssc_t *ssc;
12594                 /*
12595                  * Initialize sd_ssc_t for internal uscsi commands
12596                  * In case of potential porformance issue, we need
12597                  * to alloc memory only if there is invalid label
12598                  */
12599                 ssc = sd_ssc_init(un);
12600 
12601                 if (sd_ready_and_valid(ssc, partition) != SD_READY_VALID) {
12602                         /*
12603                          * For removable devices it is possible to start an
12604                          * I/O without a media by opening the device in nodelay
12605                          * mode. Also for writable CDs there can be many
12606                          * scenarios where there is no geometry yet but volume
12607                          * manager is trying to issue a read() just because
12608                          * it can see TOC on the CD. So do not print a message
12609                          * for removables.
12610                          */
12611                         if (!un->un_f_has_removable_media) {
12612                                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
12613                                     "i/o to invalid geometry\n");
12614                         }
12615                         bioerror(bp, EIO);
12616                         bp->b_resid = bp->b_bcount;
12617                         SD_BEGIN_IODONE(index, un, bp);
12618 
12619                         sd_ssc_fini(ssc);
12620                         return;
12621                 }
12622                 sd_ssc_fini(ssc);
12623         }
12624 
12625         nblocks = 0;
12626         (void) cmlb_partinfo(un->un_cmlbhandle, partition,
12627             &nblocks, &partition_offset, NULL, NULL, (void *)SD_PATH_DIRECT);
12628 
12629         if (un->un_f_enable_rmw) {
12630                 blknomask = (un->un_phy_blocksize / DEV_BSIZE) - 1;
12631                 secmask = un->un_phy_blocksize - 1;
12632         } else {
12633                 blknomask = (un->un_tgt_blocksize / DEV_BSIZE) - 1;
12634                 secmask = un->un_tgt_blocksize - 1;
12635         }
12636 
12637         if ((bp->b_lblkno & (blknomask)) || (bp->b_bcount & (secmask))) {
12638                 is_aligned = FALSE;
12639         }
12640 
12641         if (!(NOT_DEVBSIZE(un)) || un->un_f_enable_rmw) {
12642                 /*
12643                  * If I/O is aligned, no need to involve RMW(Read Modify Write)
12644                  * Convert the logical block number to target's physical sector
12645                  * number.
12646                  */
12647                 if (is_aligned) {
12648                         xp->xb_blkno = SD_SYS2TGTBLOCK(un, xp->xb_blkno);
12649                 } else {
12650                         /*
12651                          * There is no RMW if we're just reading, so don't
12652                          * warn or error out because of it.
12653                          */
12654                         if (bp->b_flags & B_READ) {
12655                                 /*EMPTY*/
12656                         } else if (!un->un_f_enable_rmw &&
12657                             un->un_f_rmw_type == SD_RMW_TYPE_RETURN_ERROR) {
12658                                 bp->b_flags |= B_ERROR;
12659                                 goto error_exit;
12660                         } else if (un->un_f_rmw_type == SD_RMW_TYPE_DEFAULT) {
12661                                 mutex_enter(SD_MUTEX(un));
12662                                 if (!un->un_f_enable_rmw &&
12663                                     un->un_rmw_msg_timeid == NULL) {
12664                                         scsi_log(SD_DEVINFO(un), sd_label,
12665                                             CE_WARN, "I/O request is not "
12666                                             "aligned with %d disk sector size. "
12667                                             "It is handled through Read Modify "
12668                                             "Write but the performance is "
12669                                             "very low.\n",
12670                                             un->un_tgt_blocksize);
12671                                         un->un_rmw_msg_timeid =
12672                                             timeout(sd_rmw_msg_print_handler,
12673                                             un, SD_RMW_MSG_PRINT_TIMEOUT);
12674                                 } else {
12675                                         un->un_rmw_incre_count ++;
12676                                 }
12677                                 mutex_exit(SD_MUTEX(un));
12678                         }
12679 
12680                         nblocks = SD_TGT2SYSBLOCK(un, nblocks);
12681                         partition_offset = SD_TGT2SYSBLOCK(un,
12682                             partition_offset);
12683                 }
12684         }
12685 
12686         /*
12687          * blocknum is the starting block number of the request. At this
12688          * point it is still relative to the start of the minor device.
12689          */
12690         blocknum = xp->xb_blkno;
12691 
12692         /*
12693          * Legacy: If the starting block number is one past the last block
12694          * in the partition, do not set B_ERROR in the buf.
12695          */
12696         if (blocknum == nblocks)  {
12697                 goto error_exit;
12698         }
12699 
12700         /*
12701          * Confirm that the first block of the request lies within the
12702          * partition limits. Also the requested number of bytes must be
12703          * a multiple of the system block size.
12704          */
12705         if ((blocknum < 0) || (blocknum >= nblocks) ||
12706             ((bp->b_bcount & (DEV_BSIZE - 1)) != 0)) {
12707                 bp->b_flags |= B_ERROR;
12708                 goto error_exit;
12709         }
12710 
12711         /*
12712          * If the requsted # blocks exceeds the available # blocks, that
12713          * is an overrun of the partition.
12714          */
12715         if ((!NOT_DEVBSIZE(un)) && is_aligned) {
12716                 requested_nblocks = SD_BYTES2TGTBLOCKS(un, bp->b_bcount);
12717         } else {
12718                 requested_nblocks = SD_BYTES2SYSBLOCKS(bp->b_bcount);
12719         }
12720 
12721         available_nblocks = (size_t)(nblocks - blocknum);
12722         ASSERT(nblocks >= blocknum);
12723 
12724         if (requested_nblocks > available_nblocks) {
12725                 size_t resid;
12726 
12727                 /*
12728                  * Allocate an "overrun" buf to allow the request to proceed
12729                  * for the amount of space available in the partition. The
12730                  * amount not transferred will be added into the b_resid
12731                  * when the operation is complete. The overrun buf
12732                  * replaces the original buf here, and the original buf
12733                  * is saved inside the overrun buf, for later use.
12734                  */
12735                 if ((!NOT_DEVBSIZE(un)) && is_aligned) {
12736                         resid = SD_TGTBLOCKS2BYTES(un,
12737                             (offset_t)(requested_nblocks - available_nblocks));
12738                 } else {
12739                         resid = SD_SYSBLOCKS2BYTES(
12740                             (offset_t)(requested_nblocks - available_nblocks));
12741                 }
12742 
12743                 size_t count = bp->b_bcount - resid;
12744                 /*
12745                  * Note: count is an unsigned entity thus it'll NEVER
12746                  * be less than 0 so ASSERT the original values are
12747                  * correct.
12748                  */
12749                 ASSERT(bp->b_bcount >= resid);
12750 
12751                 bp = sd_bioclone_alloc(bp, count, blocknum,
12752                     (int (*)(struct buf *)) sd_mapblockaddr_iodone);
12753                 xp = SD_GET_XBUF(bp); /* Update for 'new' bp! */
12754                 ASSERT(xp != NULL);
12755         }
12756 
12757         /* At this point there should be no residual for this buf. */
12758         ASSERT(bp->b_resid == 0);
12759 
12760         /* Convert the block number to an absolute address. */
12761         xp->xb_blkno += partition_offset;
12762 
12763         SD_NEXT_IOSTART(index, un, bp);
12764 
12765         SD_TRACE(SD_LOG_IO_PARTITION, un,
12766             "sd_mapblockaddr_iostart: exit 0: buf:0x%p\n", bp);
12767 
12768         return;
12769 
12770 error_exit:
12771         bp->b_resid = bp->b_bcount;
12772         SD_BEGIN_IODONE(index, un, bp);
12773         SD_TRACE(SD_LOG_IO_PARTITION, un,
12774             "sd_mapblockaddr_iostart: exit 1: buf:0x%p\n", bp);
12775 }
12776 
12777 
12778 /*
12779  *    Function: sd_mapblockaddr_iodone
12780  *
12781  * Description: Completion-side processing for partition management.
12782  *
12783  *     Context: May be called under interrupt context
12784  */
12785 
12786 static void
12787 sd_mapblockaddr_iodone(int index, struct sd_lun *un, struct buf *bp)
12788 {
12789         /* int  partition; */   /* Not used, see below. */
12790         ASSERT(un != NULL);
12791         ASSERT(bp != NULL);
12792         ASSERT(!mutex_owned(SD_MUTEX(un)));
12793 
12794         SD_TRACE(SD_LOG_IO_PARTITION, un,
12795             "sd_mapblockaddr_iodone: entry: buf:0x%p\n", bp);
12796 
12797         if (bp->b_iodone == (int (*)(struct buf *)) sd_mapblockaddr_iodone) {
12798                 /*
12799                  * We have an "overrun" buf to deal with...
12800                  */
12801                 struct sd_xbuf  *xp;
12802                 struct buf      *obp;   /* ptr to the original buf */
12803 
12804                 xp = SD_GET_XBUF(bp);
12805                 ASSERT(xp != NULL);
12806 
12807                 /* Retrieve the pointer to the original buf */
12808                 obp = (struct buf *)xp->xb_private;
12809                 ASSERT(obp != NULL);
12810 
12811                 obp->b_resid = obp->b_bcount - (bp->b_bcount - bp->b_resid);
12812                 bioerror(obp, bp->b_error);
12813 
12814                 sd_bioclone_free(bp);
12815 
12816                 /*
12817                  * Get back the original buf.
12818                  * Note that since the restoration of xb_blkno below
12819                  * was removed, the sd_xbuf is not needed.
12820                  */
12821                 bp = obp;
12822                 /*
12823                  * xp = SD_GET_XBUF(bp);
12824                  * ASSERT(xp != NULL);
12825                  */
12826         }
12827 
12828         /*
12829          * Convert sd->xb_blkno back to a minor-device relative value.
12830          * Note: this has been commented out, as it is not needed in the
12831          * current implementation of the driver (ie, since this function
12832          * is at the top of the layering chains, so the info will be
12833          * discarded) and it is in the "hot" IO path.
12834          *
12835          * partition = getminor(bp->b_edev) & SDPART_MASK;
12836          * xp->xb_blkno -= un->un_offset[partition];
12837          */
12838 
12839         SD_NEXT_IODONE(index, un, bp);
12840 
12841         SD_TRACE(SD_LOG_IO_PARTITION, un,
12842             "sd_mapblockaddr_iodone: exit: buf:0x%p\n", bp);
12843 }
12844 
12845 
12846 /*
12847  *    Function: sd_mapblocksize_iostart
12848  *
12849  * Description: Convert between system block size (un->un_sys_blocksize)
12850  *              and target block size (un->un_tgt_blocksize).
12851  *
12852  *     Context: Can sleep to allocate resources.
12853  *
12854  * Assumptions: A higher layer has already performed any partition validation,
12855  *              and converted the xp->xb_blkno to an absolute value relative
12856  *              to the start of the device.
12857  *
12858  *              It is also assumed that the higher layer has implemented
12859  *              an "overrun" mechanism for the case where the request would
12860  *              read/write beyond the end of a partition.  In this case we
12861  *              assume (and ASSERT) that bp->b_resid == 0.
12862  *
12863  *              Note: The implementation for this routine assumes the target
12864  *              block size remains constant between allocation and transport.
12865  */
12866 
12867 static void
12868 sd_mapblocksize_iostart(int index, struct sd_lun *un, struct buf *bp)
12869 {
12870         struct sd_mapblocksize_info     *bsp;
12871         struct sd_xbuf                  *xp;
12872         offset_t first_byte;
12873         daddr_t start_block, end_block;
12874         daddr_t request_bytes;
12875         ushort_t is_aligned = FALSE;
12876 
12877         ASSERT(un != NULL);
12878         ASSERT(bp != NULL);
12879         ASSERT(!mutex_owned(SD_MUTEX(un)));
12880         ASSERT(bp->b_resid == 0);
12881 
12882         SD_TRACE(SD_LOG_IO_RMMEDIA, un,
12883             "sd_mapblocksize_iostart: entry: buf:0x%p\n", bp);
12884 
12885         /*
12886          * For a non-writable CD, a write request is an error
12887          */
12888         if (ISCD(un) && ((bp->b_flags & B_READ) == 0) &&
12889             (un->un_f_mmc_writable_media == FALSE)) {
12890                 bioerror(bp, EIO);
12891                 bp->b_resid = bp->b_bcount;
12892                 SD_BEGIN_IODONE(index, un, bp);
12893                 return;
12894         }
12895 
12896         /*
12897          * We do not need a shadow buf if the device is using
12898          * un->un_sys_blocksize as its block size or if bcount == 0.
12899          * In this case there is no layer-private data block allocated.
12900          */
12901         if ((un->un_tgt_blocksize == DEV_BSIZE && !un->un_f_enable_rmw) ||
12902             (bp->b_bcount == 0)) {
12903                 goto done;
12904         }
12905 
12906 #if defined(__i386) || defined(__amd64)
12907         /* We do not support non-block-aligned transfers for ROD devices */
12908         ASSERT(!ISROD(un));
12909 #endif
12910 
12911         xp = SD_GET_XBUF(bp);
12912         ASSERT(xp != NULL);
12913 
12914         SD_INFO(SD_LOG_IO_RMMEDIA, un, "sd_mapblocksize_iostart: "
12915             "tgt_blocksize:0x%x sys_blocksize: 0x%x\n",
12916             un->un_tgt_blocksize, DEV_BSIZE);
12917         SD_INFO(SD_LOG_IO_RMMEDIA, un, "sd_mapblocksize_iostart: "
12918             "request start block:0x%x\n", xp->xb_blkno);
12919         SD_INFO(SD_LOG_IO_RMMEDIA, un, "sd_mapblocksize_iostart: "
12920             "request len:0x%x\n", bp->b_bcount);
12921 
12922         /*
12923          * Allocate the layer-private data area for the mapblocksize layer.
12924          * Layers are allowed to use the xp_private member of the sd_xbuf
12925          * struct to store the pointer to their layer-private data block, but
12926          * each layer also has the responsibility of restoring the prior
12927          * contents of xb_private before returning the buf/xbuf to the
12928          * higher layer that sent it.
12929          *
12930          * Here we save the prior contents of xp->xb_private into the
12931          * bsp->mbs_oprivate field of our layer-private data area. This value
12932          * is restored by sd_mapblocksize_iodone() just prior to freeing up
12933          * the layer-private area and returning the buf/xbuf to the layer
12934          * that sent it.
12935          *
12936          * Note that here we use kmem_zalloc for the allocation as there are
12937          * parts of the mapblocksize code that expect certain fields to be
12938          * zero unless explicitly set to a required value.
12939          */
12940         bsp = kmem_zalloc(sizeof (struct sd_mapblocksize_info), KM_SLEEP);
12941         bsp->mbs_oprivate = xp->xb_private;
12942         xp->xb_private = bsp;
12943 
12944         /*
12945          * This treats the data on the disk (target) as an array of bytes.
12946          * first_byte is the byte offset, from the beginning of the device,
12947          * to the location of the request. This is converted from a
12948          * un->un_sys_blocksize block address to a byte offset, and then back
12949          * to a block address based upon a un->un_tgt_blocksize block size.
12950          *
12951          * xp->xb_blkno should be absolute upon entry into this function,
12952          * but, but it is based upon partitions that use the "system"
12953          * block size. It must be adjusted to reflect the block size of
12954          * the target.
12955          *
12956          * Note that end_block is actually the block that follows the last
12957          * block of the request, but that's what is needed for the computation.
12958          */
12959         first_byte  = SD_SYSBLOCKS2BYTES((offset_t)xp->xb_blkno);
12960         if (un->un_f_enable_rmw) {
12961                 start_block = xp->xb_blkno =
12962                     (first_byte / un->un_phy_blocksize) *
12963                     (un->un_phy_blocksize / DEV_BSIZE);
12964                 end_block   = ((first_byte + bp->b_bcount +
12965                     un->un_phy_blocksize - 1) / un->un_phy_blocksize) *
12966                     (un->un_phy_blocksize / DEV_BSIZE);
12967         } else {
12968                 start_block = xp->xb_blkno = first_byte / un->un_tgt_blocksize;
12969                 end_block   = (first_byte + bp->b_bcount +
12970                     un->un_tgt_blocksize - 1) / un->un_tgt_blocksize;
12971         }
12972 
12973         /* request_bytes is rounded up to a multiple of the target block size */
12974         request_bytes = (end_block - start_block) * un->un_tgt_blocksize;
12975 
12976         /*
12977          * See if the starting address of the request and the request
12978          * length are aligned on a un->un_tgt_blocksize boundary. If aligned
12979          * then we do not need to allocate a shadow buf to handle the request.
12980          */
12981         if (un->un_f_enable_rmw) {
12982                 if (((first_byte % un->un_phy_blocksize) == 0) &&
12983                     ((bp->b_bcount % un->un_phy_blocksize) == 0)) {
12984                         is_aligned = TRUE;
12985                 }
12986         } else {
12987                 if (((first_byte % un->un_tgt_blocksize) == 0) &&
12988                     ((bp->b_bcount % un->un_tgt_blocksize) == 0)) {
12989                         is_aligned = TRUE;
12990                 }
12991         }
12992 
12993         if ((bp->b_flags & B_READ) == 0) {
12994                 /*
12995                  * Lock the range for a write operation. An aligned request is
12996                  * considered a simple write; otherwise the request must be a
12997                  * read-modify-write.
12998                  */
12999                 bsp->mbs_wmp = sd_range_lock(un, start_block, end_block - 1,
13000                     (is_aligned == TRUE) ? SD_WTYPE_SIMPLE : SD_WTYPE_RMW);
13001         }
13002 
13003         /*
13004          * Alloc a shadow buf if the request is not aligned. Also, this is
13005          * where the READ command is generated for a read-modify-write. (The
13006          * write phase is deferred until after the read completes.)
13007          */
13008         if (is_aligned == FALSE) {
13009 
13010                 struct sd_mapblocksize_info     *shadow_bsp;
13011                 struct sd_xbuf  *shadow_xp;
13012                 struct buf      *shadow_bp;
13013 
13014                 /*
13015                  * Allocate the shadow buf and it associated xbuf. Note that
13016                  * after this call the xb_blkno value in both the original
13017                  * buf's sd_xbuf _and_ the shadow buf's sd_xbuf will be the
13018                  * same: absolute relative to the start of the device, and
13019                  * adjusted for the target block size. The b_blkno in the
13020                  * shadow buf will also be set to this value. We should never
13021                  * change b_blkno in the original bp however.
13022                  *
13023                  * Note also that the shadow buf will always need to be a
13024                  * READ command, regardless of whether the incoming command
13025                  * is a READ or a WRITE.
13026                  */
13027                 shadow_bp = sd_shadow_buf_alloc(bp, request_bytes, B_READ,
13028                     xp->xb_blkno,
13029                     (int (*)(struct buf *)) sd_mapblocksize_iodone);
13030 
13031                 shadow_xp = SD_GET_XBUF(shadow_bp);
13032 
13033                 /*
13034                  * Allocate the layer-private data for the shadow buf.
13035                  * (No need to preserve xb_private in the shadow xbuf.)
13036                  */
13037                 shadow_xp->xb_private = shadow_bsp =
13038                     kmem_zalloc(sizeof (struct sd_mapblocksize_info), KM_SLEEP);
13039 
13040                 /*
13041                  * bsp->mbs_copy_offset is used later by sd_mapblocksize_iodone
13042                  * to figure out where the start of the user data is (based upon
13043                  * the system block size) in the data returned by the READ
13044                  * command (which will be based upon the target blocksize). Note
13045                  * that this is only really used if the request is unaligned.
13046                  */
13047                 if (un->un_f_enable_rmw) {
13048                         bsp->mbs_copy_offset = (ssize_t)(first_byte -
13049                             ((offset_t)xp->xb_blkno * un->un_sys_blocksize));
13050                         ASSERT((bsp->mbs_copy_offset >= 0) &&
13051                             (bsp->mbs_copy_offset < un->un_phy_blocksize));
13052                 } else {
13053                         bsp->mbs_copy_offset = (ssize_t)(first_byte -
13054                             ((offset_t)xp->xb_blkno * un->un_tgt_blocksize));
13055                         ASSERT((bsp->mbs_copy_offset >= 0) &&
13056                             (bsp->mbs_copy_offset < un->un_tgt_blocksize));
13057                 }
13058 
13059                 shadow_bsp->mbs_copy_offset = bsp->mbs_copy_offset;
13060 
13061                 shadow_bsp->mbs_layer_index = bsp->mbs_layer_index = index;
13062 
13063                 /* Transfer the wmap (if any) to the shadow buf */
13064                 shadow_bsp->mbs_wmp = bsp->mbs_wmp;
13065                 bsp->mbs_wmp = NULL;
13066 
13067                 /*
13068                  * The shadow buf goes on from here in place of the
13069                  * original buf.
13070                  */
13071                 shadow_bsp->mbs_orig_bp = bp;
13072                 bp = shadow_bp;
13073         }
13074 
13075         SD_INFO(SD_LOG_IO_RMMEDIA, un,
13076             "sd_mapblocksize_iostart: tgt start block:0x%x\n", xp->xb_blkno);
13077         SD_INFO(SD_LOG_IO_RMMEDIA, un,
13078             "sd_mapblocksize_iostart: tgt request len:0x%x\n",
13079             request_bytes);
13080         SD_INFO(SD_LOG_IO_RMMEDIA, un,
13081             "sd_mapblocksize_iostart: shadow buf:0x%x\n", bp);
13082 
13083 done:
13084         SD_NEXT_IOSTART(index, un, bp);
13085 
13086         SD_TRACE(SD_LOG_IO_RMMEDIA, un,
13087             "sd_mapblocksize_iostart: exit: buf:0x%p\n", bp);
13088 }
13089 
13090 
13091 /*
13092  *    Function: sd_mapblocksize_iodone
13093  *
13094  * Description: Completion side processing for block-size mapping.
13095  *
13096  *     Context: May be called under interrupt context
13097  */
13098 
13099 static void
13100 sd_mapblocksize_iodone(int index, struct sd_lun *un, struct buf *bp)
13101 {
13102         struct sd_mapblocksize_info     *bsp;
13103         struct sd_xbuf  *xp;
13104         struct sd_xbuf  *orig_xp;       /* sd_xbuf for the original buf */
13105         struct buf      *orig_bp;       /* ptr to the original buf */
13106         offset_t        shadow_end;
13107         offset_t        request_end;
13108         offset_t        shadow_start;
13109         ssize_t         copy_offset;
13110         size_t          copy_length;
13111         size_t          shortfall;
13112         uint_t          is_write;       /* TRUE if this bp is a WRITE */
13113         uint_t          has_wmap;       /* TRUE is this bp has a wmap */
13114 
13115         ASSERT(un != NULL);
13116         ASSERT(bp != NULL);
13117 
13118         SD_TRACE(SD_LOG_IO_RMMEDIA, un,
13119             "sd_mapblocksize_iodone: entry: buf:0x%p\n", bp);
13120 
13121         /*
13122          * There is no shadow buf or layer-private data if the target is
13123          * using un->un_sys_blocksize as its block size or if bcount == 0.
13124          */
13125         if ((un->un_tgt_blocksize == DEV_BSIZE && !un->un_f_enable_rmw) ||
13126             (bp->b_bcount == 0)) {
13127                 goto exit;
13128         }
13129 
13130         xp = SD_GET_XBUF(bp);
13131         ASSERT(xp != NULL);
13132 
13133         /* Retrieve the pointer to the layer-private data area from the xbuf. */
13134         bsp = xp->xb_private;
13135 
13136         is_write = ((bp->b_flags & B_READ) == 0) ? TRUE : FALSE;
13137         has_wmap = (bsp->mbs_wmp != NULL) ? TRUE : FALSE;
13138 
13139         if (is_write) {
13140                 /*
13141                  * For a WRITE request we must free up the block range that
13142                  * we have locked up.  This holds regardless of whether this is
13143                  * an aligned write request or a read-modify-write request.
13144                  */
13145                 sd_range_unlock(un, bsp->mbs_wmp);
13146                 bsp->mbs_wmp = NULL;
13147         }
13148 
13149         if ((bp->b_iodone != (int(*)(struct buf *))sd_mapblocksize_iodone)) {
13150                 /*
13151                  * An aligned read or write command will have no shadow buf;
13152                  * there is not much else to do with it.
13153                  */
13154                 goto done;
13155         }
13156 
13157         orig_bp = bsp->mbs_orig_bp;
13158         ASSERT(orig_bp != NULL);
13159         orig_xp = SD_GET_XBUF(orig_bp);
13160         ASSERT(orig_xp != NULL);
13161         ASSERT(!mutex_owned(SD_MUTEX(un)));
13162 
13163         if (!is_write && has_wmap) {
13164                 /*
13165                  * A READ with a wmap means this is the READ phase of a
13166                  * read-modify-write. If an error occurred on the READ then
13167                  * we do not proceed with the WRITE phase or copy any data.
13168                  * Just release the write maps and return with an error.
13169                  */
13170                 if ((bp->b_resid != 0) || (bp->b_error != 0)) {
13171                         orig_bp->b_resid = orig_bp->b_bcount;
13172                         bioerror(orig_bp, bp->b_error);
13173                         sd_range_unlock(un, bsp->mbs_wmp);
13174                         goto freebuf_done;
13175                 }
13176         }
13177 
13178         /*
13179          * Here is where we set up to copy the data from the shadow buf
13180          * into the space associated with the original buf.
13181          *
13182          * To deal with the conversion between block sizes, these
13183          * computations treat the data as an array of bytes, with the
13184          * first byte (byte 0) corresponding to the first byte in the
13185          * first block on the disk.
13186          */
13187 
13188         /*
13189          * shadow_start and shadow_len indicate the location and size of
13190          * the data returned with the shadow IO request.
13191          */
13192         if (un->un_f_enable_rmw) {
13193                 shadow_start  = SD_SYSBLOCKS2BYTES((offset_t)xp->xb_blkno);
13194         } else {
13195                 shadow_start  = SD_TGTBLOCKS2BYTES(un, (offset_t)xp->xb_blkno);
13196         }
13197         shadow_end    = shadow_start + bp->b_bcount - bp->b_resid;
13198 
13199         /*
13200          * copy_offset gives the offset (in bytes) from the start of the first
13201          * block of the READ request to the beginning of the data.  We retrieve
13202          * this value from xb_pktp in the ORIGINAL xbuf, as it has been saved
13203          * there by sd_mapblockize_iostart(). copy_length gives the amount of
13204          * data to be copied (in bytes).
13205          */
13206         copy_offset  = bsp->mbs_copy_offset;
13207         if (un->un_f_enable_rmw) {
13208                 ASSERT((copy_offset >= 0) &&
13209                     (copy_offset < un->un_phy_blocksize));
13210         } else {
13211                 ASSERT((copy_offset >= 0) &&
13212                     (copy_offset < un->un_tgt_blocksize));
13213         }
13214 
13215         copy_length  = orig_bp->b_bcount;
13216         request_end  = shadow_start + copy_offset + orig_bp->b_bcount;
13217 
13218         /*
13219          * Set up the resid and error fields of orig_bp as appropriate.
13220          */
13221         if (shadow_end >= request_end) {
13222                 /* We got all the requested data; set resid to zero */
13223                 orig_bp->b_resid = 0;
13224         } else {
13225                 /*
13226                  * We failed to get enough data to fully satisfy the original
13227                  * request. Just copy back whatever data we got and set
13228                  * up the residual and error code as required.
13229                  *
13230                  * 'shortfall' is the amount by which the data received with the
13231                  * shadow buf has "fallen short" of the requested amount.
13232                  */
13233                 shortfall = (size_t)(request_end - shadow_end);
13234 
13235                 if (shortfall > orig_bp->b_bcount) {
13236                         /*
13237                          * We did not get enough data to even partially
13238                          * fulfill the original request.  The residual is
13239                          * equal to the amount requested.
13240                          */
13241                         orig_bp->b_resid = orig_bp->b_bcount;
13242                 } else {
13243                         /*
13244                          * We did not get all the data that we requested
13245                          * from the device, but we will try to return what
13246                          * portion we did get.
13247                          */
13248                         orig_bp->b_resid = shortfall;
13249                 }
13250                 ASSERT(copy_length >= orig_bp->b_resid);
13251                 copy_length  -= orig_bp->b_resid;
13252         }
13253 
13254         /* Propagate the error code from the shadow buf to the original buf */
13255         bioerror(orig_bp, bp->b_error);
13256 
13257         if (is_write) {
13258                 goto freebuf_done;      /* No data copying for a WRITE */
13259         }
13260 
13261         if (has_wmap) {
13262                 /*
13263                  * This is a READ command from the READ phase of a
13264                  * read-modify-write request. We have to copy the data given
13265                  * by the user OVER the data returned by the READ command,
13266                  * then convert the command from a READ to a WRITE and send
13267                  * it back to the target.
13268                  */
13269                 bcopy(orig_bp->b_un.b_addr, bp->b_un.b_addr + copy_offset,
13270                     copy_length);
13271 
13272                 bp->b_flags &= ~((int)B_READ);   /* Convert to a WRITE */
13273 
13274                 /*
13275                  * Dispatch the WRITE command to the taskq thread, which
13276                  * will in turn send the command to the target. When the
13277                  * WRITE command completes, we (sd_mapblocksize_iodone())
13278                  * will get called again as part of the iodone chain
13279                  * processing for it. Note that we will still be dealing
13280                  * with the shadow buf at that point.
13281                  */
13282                 if (taskq_dispatch(sd_wmr_tq, sd_read_modify_write_task, bp,
13283                     KM_NOSLEEP) != 0) {
13284                         /*
13285                          * Dispatch was successful so we are done. Return
13286                          * without going any higher up the iodone chain. Do
13287                          * not free up any layer-private data until after the
13288                          * WRITE completes.
13289                          */
13290                         return;
13291                 }
13292 
13293                 /*
13294                  * Dispatch of the WRITE command failed; set up the error
13295                  * condition and send this IO back up the iodone chain.
13296                  */
13297                 bioerror(orig_bp, EIO);
13298                 orig_bp->b_resid = orig_bp->b_bcount;
13299 
13300         } else {
13301                 /*
13302                  * This is a regular READ request (ie, not a RMW). Copy the
13303                  * data from the shadow buf into the original buf. The
13304                  * copy_offset compensates for any "misalignment" between the
13305                  * shadow buf (with its un->un_tgt_blocksize blocks) and the
13306                  * original buf (with its un->un_sys_blocksize blocks).
13307                  */
13308                 bcopy(bp->b_un.b_addr + copy_offset, orig_bp->b_un.b_addr,
13309                     copy_length);
13310         }
13311 
13312 freebuf_done:
13313 
13314         /*
13315          * At this point we still have both the shadow buf AND the original
13316          * buf to deal with, as well as the layer-private data area in each.
13317          * Local variables are as follows:
13318          *
13319          * bp -- points to shadow buf
13320          * xp -- points to xbuf of shadow buf
13321          * bsp -- points to layer-private data area of shadow buf
13322          * orig_bp -- points to original buf
13323          *
13324          * First free the shadow buf and its associated xbuf, then free the
13325          * layer-private data area from the shadow buf. There is no need to
13326          * restore xb_private in the shadow xbuf.
13327          */
13328         sd_shadow_buf_free(bp);
13329         kmem_free(bsp, sizeof (struct sd_mapblocksize_info));
13330 
13331         /*
13332          * Now update the local variables to point to the original buf, xbuf,
13333          * and layer-private area.
13334          */
13335         bp = orig_bp;
13336         xp = SD_GET_XBUF(bp);
13337         ASSERT(xp != NULL);
13338         ASSERT(xp == orig_xp);
13339         bsp = xp->xb_private;
13340         ASSERT(bsp != NULL);
13341 
13342 done:
13343         /*
13344          * Restore xb_private to whatever it was set to by the next higher
13345          * layer in the chain, then free the layer-private data area.
13346          */
13347         xp->xb_private = bsp->mbs_oprivate;
13348         kmem_free(bsp, sizeof (struct sd_mapblocksize_info));
13349 
13350 exit:
13351         SD_TRACE(SD_LOG_IO_RMMEDIA, SD_GET_UN(bp),
13352             "sd_mapblocksize_iodone: calling SD_NEXT_IODONE: buf:0x%p\n", bp);
13353 
13354         SD_NEXT_IODONE(index, un, bp);
13355 }
13356 
13357 
13358 /*
13359  *    Function: sd_checksum_iostart
13360  *
13361  * Description: A stub function for a layer that's currently not used.
13362  *              For now just a placeholder.
13363  *
13364  *     Context: Kernel thread context
13365  */
13366 
13367 static void
13368 sd_checksum_iostart(int index, struct sd_lun *un, struct buf *bp)
13369 {
13370         ASSERT(un != NULL);
13371         ASSERT(bp != NULL);
13372         ASSERT(!mutex_owned(SD_MUTEX(un)));
13373         SD_NEXT_IOSTART(index, un, bp);
13374 }
13375 
13376 
13377 /*
13378  *    Function: sd_checksum_iodone
13379  *
13380  * Description: A stub function for a layer that's currently not used.
13381  *              For now just a placeholder.
13382  *
13383  *     Context: May be called under interrupt context
13384  */
13385 
13386 static void
13387 sd_checksum_iodone(int index, struct sd_lun *un, struct buf *bp)
13388 {
13389         ASSERT(un != NULL);
13390         ASSERT(bp != NULL);
13391         ASSERT(!mutex_owned(SD_MUTEX(un)));
13392         SD_NEXT_IODONE(index, un, bp);
13393 }
13394 
13395 
13396 /*
13397  *    Function: sd_checksum_uscsi_iostart
13398  *
13399  * Description: A stub function for a layer that's currently not used.
13400  *              For now just a placeholder.
13401  *
13402  *     Context: Kernel thread context
13403  */
13404 
13405 static void
13406 sd_checksum_uscsi_iostart(int index, struct sd_lun *un, struct buf *bp)
13407 {
13408         ASSERT(un != NULL);
13409         ASSERT(bp != NULL);
13410         ASSERT(!mutex_owned(SD_MUTEX(un)));
13411         SD_NEXT_IOSTART(index, un, bp);
13412 }
13413 
13414 
13415 /*
13416  *    Function: sd_checksum_uscsi_iodone
13417  *
13418  * Description: A stub function for a layer that's currently not used.
13419  *              For now just a placeholder.
13420  *
13421  *     Context: May be called under interrupt context
13422  */
13423 
13424 static void
13425 sd_checksum_uscsi_iodone(int index, struct sd_lun *un, struct buf *bp)
13426 {
13427         ASSERT(un != NULL);
13428         ASSERT(bp != NULL);
13429         ASSERT(!mutex_owned(SD_MUTEX(un)));
13430         SD_NEXT_IODONE(index, un, bp);
13431 }
13432 
13433 
13434 /*
13435  *    Function: sd_pm_iostart
13436  *
13437  * Description: iostart-side routine for Power mangement.
13438  *
13439  *     Context: Kernel thread context
13440  */
13441 
13442 static void
13443 sd_pm_iostart(int index, struct sd_lun *un, struct buf *bp)
13444 {
13445         ASSERT(un != NULL);
13446         ASSERT(bp != NULL);
13447         ASSERT(!mutex_owned(SD_MUTEX(un)));
13448         ASSERT(!mutex_owned(&un->un_pm_mutex));
13449 
13450         SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_iostart: entry\n");
13451 
13452         if (sd_pm_entry(un) != DDI_SUCCESS) {
13453                 /*
13454                  * Set up to return the failed buf back up the 'iodone'
13455                  * side of the calling chain.
13456                  */
13457                 bioerror(bp, EIO);
13458                 bp->b_resid = bp->b_bcount;
13459 
13460                 SD_BEGIN_IODONE(index, un, bp);
13461 
13462                 SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_iostart: exit\n");
13463                 return;
13464         }
13465 
13466         SD_NEXT_IOSTART(index, un, bp);
13467 
13468         SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_iostart: exit\n");
13469 }
13470 
13471 
13472 /*
13473  *    Function: sd_pm_iodone
13474  *
13475  * Description: iodone-side routine for power mangement.
13476  *
13477  *     Context: may be called from interrupt context
13478  */
13479 
13480 static void
13481 sd_pm_iodone(int index, struct sd_lun *un, struct buf *bp)
13482 {
13483         ASSERT(un != NULL);
13484         ASSERT(bp != NULL);
13485         ASSERT(!mutex_owned(&un->un_pm_mutex));
13486 
13487         SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_iodone: entry\n");
13488 
13489         /*
13490          * After attach the following flag is only read, so don't
13491          * take the penalty of acquiring a mutex for it.
13492          */
13493         if (un->un_f_pm_is_enabled == TRUE) {
13494                 sd_pm_exit(un);
13495         }
13496 
13497         SD_NEXT_IODONE(index, un, bp);
13498 
13499         SD_TRACE(SD_LOG_IO_PM, un, "sd_pm_iodone: exit\n");
13500 }
13501 
13502 
13503 /*
13504  *    Function: sd_core_iostart
13505  *
13506  * Description: Primary driver function for enqueuing buf(9S) structs from
13507  *              the system and initiating IO to the target device
13508  *
13509  *     Context: Kernel thread context. Can sleep.
13510  *
13511  * Assumptions:  - The given xp->xb_blkno is absolute
13512  *                 (ie, relative to the start of the device).
13513  *               - The IO is to be done using the native blocksize of
13514  *                 the device, as specified in un->un_tgt_blocksize.
13515  */
13516 /* ARGSUSED */
13517 static void
13518 sd_core_iostart(int index, struct sd_lun *un, struct buf *bp)
13519 {
13520         struct sd_xbuf *xp;
13521 
13522         ASSERT(un != NULL);
13523         ASSERT(bp != NULL);
13524         ASSERT(!mutex_owned(SD_MUTEX(un)));
13525         ASSERT(bp->b_resid == 0);
13526 
13527         SD_TRACE(SD_LOG_IO_CORE, un, "sd_core_iostart: entry: bp:0x%p\n", bp);
13528 
13529         xp = SD_GET_XBUF(bp);
13530         ASSERT(xp != NULL);
13531 
13532         mutex_enter(SD_MUTEX(un));
13533 
13534         /*
13535          * If we are currently in the failfast state, fail any new IO
13536          * that has B_FAILFAST set, then return.
13537          */
13538         if ((bp->b_flags & B_FAILFAST) &&
13539             (un->un_failfast_state == SD_FAILFAST_ACTIVE)) {
13540                 mutex_exit(SD_MUTEX(un));
13541                 bioerror(bp, EIO);
13542                 bp->b_resid = bp->b_bcount;
13543                 SD_BEGIN_IODONE(index, un, bp);
13544                 return;
13545         }
13546 
13547         if (SD_IS_DIRECT_PRIORITY(xp)) {
13548                 /*
13549                  * Priority command -- transport it immediately.
13550                  *
13551                  * Note: We may want to assert that USCSI_DIAGNOSE is set,
13552                  * because all direct priority commands should be associated
13553                  * with error recovery actions which we don't want to retry.
13554                  */
13555                 sd_start_cmds(un, bp);
13556         } else {
13557                 /*
13558                  * Normal command -- add it to the wait queue, then start
13559                  * transporting commands from the wait queue.
13560                  */
13561                 sd_add_buf_to_waitq(un, bp);
13562                 SD_UPDATE_KSTATS(un, kstat_waitq_enter, bp);
13563                 sd_start_cmds(un, NULL);
13564         }
13565 
13566         mutex_exit(SD_MUTEX(un));
13567 
13568         SD_TRACE(SD_LOG_IO_CORE, un, "sd_core_iostart: exit: bp:0x%p\n", bp);
13569 }
13570 
13571 
13572 /*
13573  *    Function: sd_init_cdb_limits
13574  *
13575  * Description: This is to handle scsi_pkt initialization differences
13576  *              between the driver platforms.
13577  *
13578  *              Legacy behaviors:
13579  *
13580  *              If the block number or the sector count exceeds the
13581  *              capabilities of a Group 0 command, shift over to a
13582  *              Group 1 command. We don't blindly use Group 1
13583  *              commands because a) some drives (CDC Wren IVs) get a
13584  *              bit confused, and b) there is probably a fair amount
13585  *              of speed difference for a target to receive and decode
13586  *              a 10 byte command instead of a 6 byte command.
13587  *
13588  *              The xfer time difference of 6 vs 10 byte CDBs is
13589  *              still significant so this code is still worthwhile.
13590  *              10 byte CDBs are very inefficient with the fas HBA driver
13591  *              and older disks. Each CDB byte took 1 usec with some
13592  *              popular disks.
13593  *
13594  *     Context: Must be called at attach time
13595  */
13596 
13597 static void
13598 sd_init_cdb_limits(struct sd_lun *un)
13599 {
13600         int hba_cdb_limit;
13601 
13602         /*
13603          * Use CDB_GROUP1 commands for most devices except for
13604          * parallel SCSI fixed drives in which case we get better
13605          * performance using CDB_GROUP0 commands (where applicable).
13606          */
13607         un->un_mincdb = SD_CDB_GROUP1;
13608 #if !defined(__fibre)
13609         if (!un->un_f_is_fibre && !un->un_f_cfg_is_atapi && !ISROD(un) &&
13610             !un->un_f_has_removable_media) {
13611                 un->un_mincdb = SD_CDB_GROUP0;
13612         }
13613 #endif
13614 
13615         /*
13616          * Try to read the max-cdb-length supported by HBA.
13617          */
13618         un->un_max_hba_cdb = scsi_ifgetcap(SD_ADDRESS(un), "max-cdb-length", 1);
13619         if (0 >= un->un_max_hba_cdb) {
13620                 un->un_max_hba_cdb = CDB_GROUP4;
13621                 hba_cdb_limit = SD_CDB_GROUP4;
13622         } else if (0 < un->un_max_hba_cdb &&
13623             un->un_max_hba_cdb < CDB_GROUP1) {
13624                 hba_cdb_limit = SD_CDB_GROUP0;
13625         } else if (CDB_GROUP1 <= un->un_max_hba_cdb &&
13626             un->un_max_hba_cdb < CDB_GROUP5) {
13627                 hba_cdb_limit = SD_CDB_GROUP1;
13628         } else if (CDB_GROUP5 <= un->un_max_hba_cdb &&
13629             un->un_max_hba_cdb < CDB_GROUP4) {
13630                 hba_cdb_limit = SD_CDB_GROUP5;
13631         } else {
13632                 hba_cdb_limit = SD_CDB_GROUP4;
13633         }
13634 
13635         /*
13636          * Use CDB_GROUP5 commands for removable devices.  Use CDB_GROUP4
13637          * commands for fixed disks unless we are building for a 32 bit
13638          * kernel.
13639          */
13640 #ifdef _LP64
13641         un->un_maxcdb = (un->un_f_has_removable_media) ? SD_CDB_GROUP5 :
13642             min(hba_cdb_limit, SD_CDB_GROUP4);
13643 #else
13644         un->un_maxcdb = (un->un_f_has_removable_media) ? SD_CDB_GROUP5 :
13645             min(hba_cdb_limit, SD_CDB_GROUP1);
13646 #endif
13647 
13648         un->un_status_len = (int)((un->un_f_arq_enabled == TRUE)
13649             ? sizeof (struct scsi_arq_status) : 1);
13650         un->un_cmd_timeout = (ushort_t)sd_io_time;
13651         un->un_uscsi_timeout = ((ISCD(un)) ? 2 : 1) * un->un_cmd_timeout;
13652 }
13653 
13654 
13655 /*
13656  *    Function: sd_initpkt_for_buf
13657  *
13658  * Description: Allocate and initialize for transport a scsi_pkt struct,
13659  *              based upon the info specified in the given buf struct.
13660  *
13661  *              Assumes the xb_blkno in the request is absolute (ie,
13662  *              relative to the start of the device (NOT partition!).
13663  *              Also assumes that the request is using the native block
13664  *              size of the device (as returned by the READ CAPACITY
13665  *              command).
13666  *
13667  * Return Code: SD_PKT_ALLOC_SUCCESS
13668  *              SD_PKT_ALLOC_FAILURE
13669  *              SD_PKT_ALLOC_FAILURE_NO_DMA
13670  *              SD_PKT_ALLOC_FAILURE_CDB_TOO_SMALL
13671  *
13672  *     Context: Kernel thread and may be called from software interrupt context
13673  *              as part of a sdrunout callback. This function may not block or
13674  *              call routines that block
13675  */
13676 
13677 static int
13678 sd_initpkt_for_buf(struct buf *bp, struct scsi_pkt **pktpp)
13679 {
13680         struct sd_xbuf  *xp;
13681         struct scsi_pkt *pktp = NULL;
13682         struct sd_lun   *un;
13683         size_t          blockcount;
13684         daddr_t         startblock;
13685         int             rval;
13686         int             cmd_flags;
13687 
13688         ASSERT(bp != NULL);
13689         ASSERT(pktpp != NULL);
13690         xp = SD_GET_XBUF(bp);
13691         ASSERT(xp != NULL);
13692         un = SD_GET_UN(bp);
13693         ASSERT(un != NULL);
13694         ASSERT(mutex_owned(SD_MUTEX(un)));
13695         ASSERT(bp->b_resid == 0);
13696 
13697         SD_TRACE(SD_LOG_IO_CORE, un,
13698             "sd_initpkt_for_buf: entry: buf:0x%p\n", bp);
13699 
13700         mutex_exit(SD_MUTEX(un));
13701 
13702 #if defined(__i386) || defined(__amd64) /* DMAFREE for x86 only */
13703         if (xp->xb_pkt_flags & SD_XB_DMA_FREED) {
13704                 /*
13705                  * Already have a scsi_pkt -- just need DMA resources.
13706                  * We must recompute the CDB in case the mapping returns
13707                  * a nonzero pkt_resid.
13708                  * Note: if this is a portion of a PKT_DMA_PARTIAL transfer
13709                  * that is being retried, the unmap/remap of the DMA resouces
13710                  * will result in the entire transfer starting over again
13711                  * from the very first block.
13712                  */
13713                 ASSERT(xp->xb_pktp != NULL);
13714                 pktp = xp->xb_pktp;
13715         } else {
13716                 pktp = NULL;
13717         }
13718 #endif /* __i386 || __amd64 */
13719 
13720         startblock = xp->xb_blkno;   /* Absolute block num. */
13721         blockcount = SD_BYTES2TGTBLOCKS(un, bp->b_bcount);
13722 
13723         cmd_flags = un->un_pkt_flags | (xp->xb_pkt_flags & SD_XB_INITPKT_MASK);
13724 
13725         /*
13726          * sd_setup_rw_pkt will determine the appropriate CDB group to use,
13727          * call scsi_init_pkt, and build the CDB.
13728          */
13729         rval = sd_setup_rw_pkt(un, &pktp, bp,
13730             cmd_flags, sdrunout, (caddr_t)un,
13731             startblock, blockcount);
13732 
13733         if (rval == 0) {
13734                 /*
13735                  * Success.
13736                  *
13737                  * If partial DMA is being used and required for this transfer.
13738                  * set it up here.
13739                  */
13740                 if ((un->un_pkt_flags & PKT_DMA_PARTIAL) != 0 &&
13741                     (pktp->pkt_resid != 0)) {
13742 
13743                         /*
13744                          * Save the CDB length and pkt_resid for the
13745                          * next xfer
13746                          */
13747                         xp->xb_dma_resid = pktp->pkt_resid;
13748 
13749                         /* rezero resid */
13750                         pktp->pkt_resid = 0;
13751 
13752                 } else {
13753                         xp->xb_dma_resid = 0;
13754                 }
13755 
13756                 pktp->pkt_flags = un->un_tagflags;
13757                 pktp->pkt_time  = un->un_cmd_timeout;
13758                 pktp->pkt_comp  = sdintr;
13759 
13760                 pktp->pkt_private = bp;
13761                 *pktpp = pktp;
13762 
13763                 SD_TRACE(SD_LOG_IO_CORE, un,
13764                     "sd_initpkt_for_buf: exit: buf:0x%p\n", bp);
13765 
13766 #if defined(__i386) || defined(__amd64) /* DMAFREE for x86 only */
13767                 xp->xb_pkt_flags &= ~SD_XB_DMA_FREED;
13768 #endif
13769 
13770                 mutex_enter(SD_MUTEX(un));
13771                 return (SD_PKT_ALLOC_SUCCESS);
13772 
13773         }
13774 
13775         /*
13776          * SD_PKT_ALLOC_FAILURE is the only expected failure code
13777          * from sd_setup_rw_pkt.
13778          */
13779         ASSERT(rval == SD_PKT_ALLOC_FAILURE);
13780 
13781         if (rval == SD_PKT_ALLOC_FAILURE) {
13782                 *pktpp = NULL;
13783                 /*
13784                  * Set the driver state to RWAIT to indicate the driver
13785                  * is waiting on resource allocations. The driver will not
13786                  * suspend, pm_suspend, or detatch while the state is RWAIT.
13787                  */
13788                 mutex_enter(SD_MUTEX(un));
13789                 New_state(un, SD_STATE_RWAIT);
13790 
13791                 SD_ERROR(SD_LOG_IO_CORE, un,
13792                     "sd_initpkt_for_buf: No pktp. exit bp:0x%p\n", bp);
13793 
13794                 if ((bp->b_flags & B_ERROR) != 0) {
13795                         return (SD_PKT_ALLOC_FAILURE_NO_DMA);
13796                 }
13797                 return (SD_PKT_ALLOC_FAILURE);
13798         } else {
13799                 /*
13800                  * PKT_ALLOC_FAILURE_CDB_TOO_SMALL
13801                  *
13802                  * This should never happen.  Maybe someone messed with the
13803                  * kernel's minphys?
13804                  */
13805                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
13806                     "Request rejected: too large for CDB: "
13807                     "lba:0x%08lx  len:0x%08lx\n", startblock, blockcount);
13808                 SD_ERROR(SD_LOG_IO_CORE, un,
13809                     "sd_initpkt_for_buf: No cp. exit bp:0x%p\n", bp);
13810                 mutex_enter(SD_MUTEX(un));
13811                 return (SD_PKT_ALLOC_FAILURE_CDB_TOO_SMALL);
13812 
13813         }
13814 }
13815 
13816 
13817 /*
13818  *    Function: sd_destroypkt_for_buf
13819  *
13820  * Description: Free the scsi_pkt(9S) for the given bp (buf IO processing).
13821  *
13822  *     Context: Kernel thread or interrupt context
13823  */
13824 
13825 static void
13826 sd_destroypkt_for_buf(struct buf *bp)
13827 {
13828         ASSERT(bp != NULL);
13829         ASSERT(SD_GET_UN(bp) != NULL);
13830 
13831         SD_TRACE(SD_LOG_IO_CORE, SD_GET_UN(bp),
13832             "sd_destroypkt_for_buf: entry: buf:0x%p\n", bp);
13833 
13834         ASSERT(SD_GET_PKTP(bp) != NULL);
13835         scsi_destroy_pkt(SD_GET_PKTP(bp));
13836 
13837         SD_TRACE(SD_LOG_IO_CORE, SD_GET_UN(bp),
13838             "sd_destroypkt_for_buf: exit: buf:0x%p\n", bp);
13839 }
13840 
13841 /*
13842  *    Function: sd_setup_rw_pkt
13843  *
13844  * Description: Determines appropriate CDB group for the requested LBA
13845  *              and transfer length, calls scsi_init_pkt, and builds
13846  *              the CDB.  Do not use for partial DMA transfers except
13847  *              for the initial transfer since the CDB size must
13848  *              remain constant.
13849  *
13850  *     Context: Kernel thread and may be called from software interrupt
13851  *              context as part of a sdrunout callback. This function may not
13852  *              block or call routines that block
13853  */
13854 
13855 
13856 int
13857 sd_setup_rw_pkt(struct sd_lun *un,
13858     struct scsi_pkt **pktpp, struct buf *bp, int flags,
13859     int (*callback)(caddr_t), caddr_t callback_arg,
13860     diskaddr_t lba, uint32_t blockcount)
13861 {
13862         struct scsi_pkt *return_pktp;
13863         union scsi_cdb *cdbp;
13864         struct sd_cdbinfo *cp = NULL;
13865         int i;
13866 
13867         /*
13868          * See which size CDB to use, based upon the request.
13869          */
13870         for (i = un->un_mincdb; i <= un->un_maxcdb; i++) {
13871 
13872                 /*
13873                  * Check lba and block count against sd_cdbtab limits.
13874                  * In the partial DMA case, we have to use the same size
13875                  * CDB for all the transfers.  Check lba + blockcount
13876                  * against the max LBA so we know that segment of the
13877                  * transfer can use the CDB we select.
13878                  */
13879                 if ((lba + blockcount - 1 <= sd_cdbtab[i].sc_maxlba) &&
13880                     (blockcount <= sd_cdbtab[i].sc_maxlen)) {
13881 
13882                         /*
13883                          * The command will fit into the CDB type
13884                          * specified by sd_cdbtab[i].
13885                          */
13886                         cp = sd_cdbtab + i;
13887 
13888                         /*
13889                          * Call scsi_init_pkt so we can fill in the
13890                          * CDB.
13891                          */
13892                         return_pktp = scsi_init_pkt(SD_ADDRESS(un), *pktpp,
13893                             bp, cp->sc_grpcode, un->un_status_len, 0,
13894                             flags, callback, callback_arg);
13895 
13896                         if (return_pktp != NULL) {
13897 
13898                                 /*
13899                                  * Return new value of pkt
13900                                  */
13901                                 *pktpp = return_pktp;
13902 
13903                                 /*
13904                                  * To be safe, zero the CDB insuring there is
13905                                  * no leftover data from a previous command.
13906                                  */
13907                                 bzero(return_pktp->pkt_cdbp, cp->sc_grpcode);
13908 
13909                                 /*
13910                                  * Handle partial DMA mapping
13911                                  */
13912                                 if (return_pktp->pkt_resid != 0) {
13913 
13914                                         /*
13915                                          * Not going to xfer as many blocks as
13916                                          * originally expected
13917                                          */
13918                                         blockcount -=
13919                                             SD_BYTES2TGTBLOCKS(un,
13920                                             return_pktp->pkt_resid);
13921                                 }
13922 
13923                                 cdbp = (union scsi_cdb *)return_pktp->pkt_cdbp;
13924 
13925                                 /*
13926                                  * Set command byte based on the CDB
13927                                  * type we matched.
13928                                  */
13929                                 cdbp->scc_cmd = cp->sc_grpmask |
13930                                     ((bp->b_flags & B_READ) ?
13931                                     SCMD_READ : SCMD_WRITE);
13932 
13933                                 SD_FILL_SCSI1_LUN(un, return_pktp);
13934 
13935                                 /*
13936                                  * Fill in LBA and length
13937                                  */
13938                                 ASSERT((cp->sc_grpcode == CDB_GROUP1) ||
13939                                     (cp->sc_grpcode == CDB_GROUP4) ||
13940                                     (cp->sc_grpcode == CDB_GROUP0) ||
13941                                     (cp->sc_grpcode == CDB_GROUP5));
13942 
13943                                 if (cp->sc_grpcode == CDB_GROUP1) {
13944                                         FORMG1ADDR(cdbp, lba);
13945                                         FORMG1COUNT(cdbp, blockcount);
13946                                         return (0);
13947                                 } else if (cp->sc_grpcode == CDB_GROUP4) {
13948                                         FORMG4LONGADDR(cdbp, lba);
13949                                         FORMG4COUNT(cdbp, blockcount);
13950                                         return (0);
13951                                 } else if (cp->sc_grpcode == CDB_GROUP0) {
13952                                         FORMG0ADDR(cdbp, lba);
13953                                         FORMG0COUNT(cdbp, blockcount);
13954                                         return (0);
13955                                 } else if (cp->sc_grpcode == CDB_GROUP5) {
13956                                         FORMG5ADDR(cdbp, lba);
13957                                         FORMG5COUNT(cdbp, blockcount);
13958                                         return (0);
13959                                 }
13960 
13961                                 /*
13962                                  * It should be impossible to not match one
13963                                  * of the CDB types above, so we should never
13964                                  * reach this point.  Set the CDB command byte
13965                                  * to test-unit-ready to avoid writing
13966                                  * to somewhere we don't intend.
13967                                  */
13968                                 cdbp->scc_cmd = SCMD_TEST_UNIT_READY;
13969                                 return (SD_PKT_ALLOC_FAILURE_CDB_TOO_SMALL);
13970                         } else {
13971                                 /*
13972                                  * Couldn't get scsi_pkt
13973                                  */
13974                                 return (SD_PKT_ALLOC_FAILURE);
13975                         }
13976                 }
13977         }
13978 
13979         /*
13980          * None of the available CDB types were suitable.  This really
13981          * should never happen:  on a 64 bit system we support
13982          * READ16/WRITE16 which will hold an entire 64 bit disk address
13983          * and on a 32 bit system we will refuse to bind to a device
13984          * larger than 2TB so addresses will never be larger than 32 bits.
13985          */
13986         return (SD_PKT_ALLOC_FAILURE_CDB_TOO_SMALL);
13987 }
13988 
13989 /*
13990  *    Function: sd_setup_next_rw_pkt
13991  *
13992  * Description: Setup packet for partial DMA transfers, except for the
13993  *              initial transfer.  sd_setup_rw_pkt should be used for
13994  *              the initial transfer.
13995  *
13996  *     Context: Kernel thread and may be called from interrupt context.
13997  */
13998 
13999 int
14000 sd_setup_next_rw_pkt(struct sd_lun *un,
14001     struct scsi_pkt *pktp, struct buf *bp,
14002     diskaddr_t lba, uint32_t blockcount)
14003 {
14004         uchar_t com;
14005         union scsi_cdb *cdbp;
14006         uchar_t cdb_group_id;
14007 
14008         ASSERT(pktp != NULL);
14009         ASSERT(pktp->pkt_cdbp != NULL);
14010 
14011         cdbp = (union scsi_cdb *)pktp->pkt_cdbp;
14012         com = cdbp->scc_cmd;
14013         cdb_group_id = CDB_GROUPID(com);
14014 
14015         ASSERT((cdb_group_id == CDB_GROUPID_0) ||
14016             (cdb_group_id == CDB_GROUPID_1) ||
14017             (cdb_group_id == CDB_GROUPID_4) ||
14018             (cdb_group_id == CDB_GROUPID_5));
14019 
14020         /*
14021          * Move pkt to the next portion of the xfer.
14022          * func is NULL_FUNC so we do not have to release
14023          * the disk mutex here.
14024          */
14025         if (scsi_init_pkt(SD_ADDRESS(un), pktp, bp, 0, 0, 0, 0,
14026             NULL_FUNC, NULL) == pktp) {
14027                 /* Success.  Handle partial DMA */
14028                 if (pktp->pkt_resid != 0) {
14029                         blockcount -=
14030                             SD_BYTES2TGTBLOCKS(un, pktp->pkt_resid);
14031                 }
14032 
14033                 cdbp->scc_cmd = com;
14034                 SD_FILL_SCSI1_LUN(un, pktp);
14035                 if (cdb_group_id == CDB_GROUPID_1) {
14036                         FORMG1ADDR(cdbp, lba);
14037                         FORMG1COUNT(cdbp, blockcount);
14038                         return (0);
14039                 } else if (cdb_group_id == CDB_GROUPID_4) {
14040                         FORMG4LONGADDR(cdbp, lba);
14041                         FORMG4COUNT(cdbp, blockcount);
14042                         return (0);
14043                 } else if (cdb_group_id == CDB_GROUPID_0) {
14044                         FORMG0ADDR(cdbp, lba);
14045                         FORMG0COUNT(cdbp, blockcount);
14046                         return (0);
14047                 } else if (cdb_group_id == CDB_GROUPID_5) {
14048                         FORMG5ADDR(cdbp, lba);
14049                         FORMG5COUNT(cdbp, blockcount);
14050                         return (0);
14051                 }
14052 
14053                 /* Unreachable */
14054                 return (SD_PKT_ALLOC_FAILURE_CDB_TOO_SMALL);
14055         }
14056 
14057         /*
14058          * Error setting up next portion of cmd transfer.
14059          * Something is definitely very wrong and this
14060          * should not happen.
14061          */
14062         return (SD_PKT_ALLOC_FAILURE);
14063 }
14064 
14065 /*
14066  *    Function: sd_initpkt_for_uscsi
14067  *
14068  * Description: Allocate and initialize for transport a scsi_pkt struct,
14069  *              based upon the info specified in the given uscsi_cmd struct.
14070  *
14071  * Return Code: SD_PKT_ALLOC_SUCCESS
14072  *              SD_PKT_ALLOC_FAILURE
14073  *              SD_PKT_ALLOC_FAILURE_NO_DMA
14074  *              SD_PKT_ALLOC_FAILURE_CDB_TOO_SMALL
14075  *
14076  *     Context: Kernel thread and may be called from software interrupt context
14077  *              as part of a sdrunout callback. This function may not block or
14078  *              call routines that block
14079  */
14080 
14081 static int
14082 sd_initpkt_for_uscsi(struct buf *bp, struct scsi_pkt **pktpp)
14083 {
14084         struct uscsi_cmd *uscmd;
14085         struct sd_xbuf  *xp;
14086         struct scsi_pkt *pktp;
14087         struct sd_lun   *un;
14088         uint32_t        flags = 0;
14089 
14090         ASSERT(bp != NULL);
14091         ASSERT(pktpp != NULL);
14092         xp = SD_GET_XBUF(bp);
14093         ASSERT(xp != NULL);
14094         un = SD_GET_UN(bp);
14095         ASSERT(un != NULL);
14096         ASSERT(mutex_owned(SD_MUTEX(un)));
14097 
14098         /* The pointer to the uscsi_cmd struct is expected in xb_pktinfo */
14099         uscmd = (struct uscsi_cmd *)xp->xb_pktinfo;
14100         ASSERT(uscmd != NULL);
14101 
14102         SD_TRACE(SD_LOG_IO_CORE, un,
14103             "sd_initpkt_for_uscsi: entry: buf:0x%p\n", bp);
14104 
14105         /*
14106          * Allocate the scsi_pkt for the command.
14107          * Note: If PKT_DMA_PARTIAL flag is set, scsi_vhci binds a path
14108          *       during scsi_init_pkt time and will continue to use the
14109          *       same path as long as the same scsi_pkt is used without
14110          *       intervening scsi_dma_free(). Since uscsi command does
14111          *       not call scsi_dmafree() before retry failed command, it
14112          *       is necessary to make sure PKT_DMA_PARTIAL flag is NOT
14113          *       set such that scsi_vhci can use other available path for
14114          *       retry. Besides, ucsci command does not allow DMA breakup,
14115          *       so there is no need to set PKT_DMA_PARTIAL flag.
14116          */
14117         if (uscmd->uscsi_rqlen > SENSE_LENGTH) {
14118                 pktp = scsi_init_pkt(SD_ADDRESS(un), NULL,
14119                     ((bp->b_bcount != 0) ? bp : NULL), uscmd->uscsi_cdblen,
14120                     ((int)(uscmd->uscsi_rqlen) + sizeof (struct scsi_arq_status)
14121                     - sizeof (struct scsi_extended_sense)), 0,
14122                     (un->un_pkt_flags & ~PKT_DMA_PARTIAL) | PKT_XARQ,
14123                     sdrunout, (caddr_t)un);
14124         } else {
14125                 pktp = scsi_init_pkt(SD_ADDRESS(un), NULL,
14126                     ((bp->b_bcount != 0) ? bp : NULL), uscmd->uscsi_cdblen,
14127                     sizeof (struct scsi_arq_status), 0,
14128                     (un->un_pkt_flags & ~PKT_DMA_PARTIAL),
14129                     sdrunout, (caddr_t)un);
14130         }
14131 
14132         if (pktp == NULL) {
14133                 *pktpp = NULL;
14134                 /*
14135                  * Set the driver state to RWAIT to indicate the driver
14136                  * is waiting on resource allocations. The driver will not
14137                  * suspend, pm_suspend, or detatch while the state is RWAIT.
14138                  */
14139                 New_state(un, SD_STATE_RWAIT);
14140 
14141                 SD_ERROR(SD_LOG_IO_CORE, un,
14142                     "sd_initpkt_for_uscsi: No pktp. exit bp:0x%p\n", bp);
14143 
14144                 if ((bp->b_flags & B_ERROR) != 0) {
14145                         return (SD_PKT_ALLOC_FAILURE_NO_DMA);
14146                 }
14147                 return (SD_PKT_ALLOC_FAILURE);
14148         }
14149 
14150         /*
14151          * We do not do DMA breakup for USCSI commands, so return failure
14152          * here if all the needed DMA resources were not allocated.
14153          */
14154         if ((un->un_pkt_flags & PKT_DMA_PARTIAL) &&
14155             (bp->b_bcount != 0) && (pktp->pkt_resid != 0)) {
14156                 scsi_destroy_pkt(pktp);
14157                 SD_ERROR(SD_LOG_IO_CORE, un, "sd_initpkt_for_uscsi: "
14158                     "No partial DMA for USCSI. exit: buf:0x%p\n", bp);
14159                 return (SD_PKT_ALLOC_FAILURE_PKT_TOO_SMALL);
14160         }
14161 
14162         /* Init the cdb from the given uscsi struct */
14163         (void) scsi_setup_cdb((union scsi_cdb *)pktp->pkt_cdbp,
14164             uscmd->uscsi_cdb[0], 0, 0, 0);
14165 
14166         SD_FILL_SCSI1_LUN(un, pktp);
14167 
14168         /*
14169          * Set up the optional USCSI flags. See the uscsi (7I) man page
14170          * for listing of the supported flags.
14171          */
14172 
14173         if (uscmd->uscsi_flags & USCSI_SILENT) {
14174                 flags |= FLAG_SILENT;
14175         }
14176 
14177         if (uscmd->uscsi_flags & USCSI_DIAGNOSE) {
14178                 flags |= FLAG_DIAGNOSE;
14179         }
14180 
14181         if (uscmd->uscsi_flags & USCSI_ISOLATE) {
14182                 flags |= FLAG_ISOLATE;
14183         }
14184 
14185         if (un->un_f_is_fibre == FALSE) {
14186                 if (uscmd->uscsi_flags & USCSI_RENEGOT) {
14187                         flags |= FLAG_RENEGOTIATE_WIDE_SYNC;
14188                 }
14189         }
14190 
14191         /*
14192          * Set the pkt flags here so we save time later.
14193          * Note: These flags are NOT in the uscsi man page!!!
14194          */
14195         if (uscmd->uscsi_flags & USCSI_HEAD) {
14196                 flags |= FLAG_HEAD;
14197         }
14198 
14199         if (uscmd->uscsi_flags & USCSI_NOINTR) {
14200                 flags |= FLAG_NOINTR;
14201         }
14202 
14203         /*
14204          * For tagged queueing, things get a bit complicated.
14205          * Check first for head of queue and last for ordered queue.
14206          * If neither head nor order, use the default driver tag flags.
14207          */
14208         if ((uscmd->uscsi_flags & USCSI_NOTAG) == 0) {
14209                 if (uscmd->uscsi_flags & USCSI_HTAG) {
14210                         flags |= FLAG_HTAG;
14211                 } else if (uscmd->uscsi_flags & USCSI_OTAG) {
14212                         flags |= FLAG_OTAG;
14213                 } else {
14214                         flags |= un->un_tagflags & FLAG_TAGMASK;
14215                 }
14216         }
14217 
14218         if (uscmd->uscsi_flags & USCSI_NODISCON) {
14219                 flags = (flags & ~FLAG_TAGMASK) | FLAG_NODISCON;
14220         }
14221 
14222         pktp->pkt_flags = flags;
14223 
14224         /* Transfer uscsi information to scsi_pkt */
14225         (void) scsi_uscsi_pktinit(uscmd, pktp);
14226 
14227         /* Copy the caller's CDB into the pkt... */
14228         bcopy(uscmd->uscsi_cdb, pktp->pkt_cdbp, uscmd->uscsi_cdblen);
14229 
14230         if (uscmd->uscsi_timeout == 0) {
14231                 pktp->pkt_time = un->un_uscsi_timeout;
14232         } else {
14233                 pktp->pkt_time = uscmd->uscsi_timeout;
14234         }
14235 
14236         /* need it later to identify USCSI request in sdintr */
14237         xp->xb_pkt_flags |= SD_XB_USCSICMD;
14238 
14239         xp->xb_sense_resid = uscmd->uscsi_rqresid;
14240 
14241         pktp->pkt_private = bp;
14242         pktp->pkt_comp = sdintr;
14243         *pktpp = pktp;
14244 
14245         SD_TRACE(SD_LOG_IO_CORE, un,
14246             "sd_initpkt_for_uscsi: exit: buf:0x%p\n", bp);
14247 
14248         return (SD_PKT_ALLOC_SUCCESS);
14249 }
14250 
14251 
14252 /*
14253  *    Function: sd_destroypkt_for_uscsi
14254  *
14255  * Description: Free the scsi_pkt(9S) struct for the given bp, for uscsi
14256  *              IOs.. Also saves relevant info into the associated uscsi_cmd
14257  *              struct.
14258  *
14259  *     Context: May be called under interrupt context
14260  */
14261 
14262 static void
14263 sd_destroypkt_for_uscsi(struct buf *bp)
14264 {
14265         struct uscsi_cmd *uscmd;
14266         struct sd_xbuf  *xp;
14267         struct scsi_pkt *pktp;
14268         struct sd_lun   *un;
14269         struct sd_uscsi_info *suip;
14270 
14271         ASSERT(bp != NULL);
14272         xp = SD_GET_XBUF(bp);
14273         ASSERT(xp != NULL);
14274         un = SD_GET_UN(bp);
14275         ASSERT(un != NULL);
14276         ASSERT(!mutex_owned(SD_MUTEX(un)));
14277         pktp = SD_GET_PKTP(bp);
14278         ASSERT(pktp != NULL);
14279 
14280         SD_TRACE(SD_LOG_IO_CORE, un,
14281             "sd_destroypkt_for_uscsi: entry: buf:0x%p\n", bp);
14282 
14283         /* The pointer to the uscsi_cmd struct is expected in xb_pktinfo */
14284         uscmd = (struct uscsi_cmd *)xp->xb_pktinfo;
14285         ASSERT(uscmd != NULL);
14286 
14287         /* Save the status and the residual into the uscsi_cmd struct */
14288         uscmd->uscsi_status = ((*(pktp)->pkt_scbp) & STATUS_MASK);
14289         uscmd->uscsi_resid  = bp->b_resid;
14290 
14291         /* Transfer scsi_pkt information to uscsi */
14292         (void) scsi_uscsi_pktfini(pktp, uscmd);
14293 
14294         /*
14295          * If enabled, copy any saved sense data into the area specified
14296          * by the uscsi command.
14297          */
14298         if (((uscmd->uscsi_flags & USCSI_RQENABLE) != 0) &&
14299             (uscmd->uscsi_rqlen != 0) && (uscmd->uscsi_rqbuf != NULL)) {
14300                 /*
14301                  * Note: uscmd->uscsi_rqbuf should always point to a buffer
14302                  * at least SENSE_LENGTH bytes in size (see sd_send_scsi_cmd())
14303                  */
14304                 uscmd->uscsi_rqstatus = xp->xb_sense_status;
14305                 uscmd->uscsi_rqresid  = xp->xb_sense_resid;
14306                 if (uscmd->uscsi_rqlen > SENSE_LENGTH) {
14307                         bcopy(xp->xb_sense_data, uscmd->uscsi_rqbuf,
14308                             MAX_SENSE_LENGTH);
14309                 } else {
14310                         bcopy(xp->xb_sense_data, uscmd->uscsi_rqbuf,
14311                             SENSE_LENGTH);
14312                 }
14313         }
14314         /*
14315          * The following assignments are for SCSI FMA.
14316          */
14317         ASSERT(xp->xb_private != NULL);
14318         suip = (struct sd_uscsi_info *)xp->xb_private;
14319         suip->ui_pkt_reason = pktp->pkt_reason;
14320         suip->ui_pkt_state = pktp->pkt_state;
14321         suip->ui_pkt_statistics = pktp->pkt_statistics;
14322         suip->ui_lba = (uint64_t)SD_GET_BLKNO(bp);
14323 
14324         /* We are done with the scsi_pkt; free it now */
14325         ASSERT(SD_GET_PKTP(bp) != NULL);
14326         scsi_destroy_pkt(SD_GET_PKTP(bp));
14327 
14328         SD_TRACE(SD_LOG_IO_CORE, un,
14329             "sd_destroypkt_for_uscsi: exit: buf:0x%p\n", bp);
14330 }
14331 
14332 
14333 /*
14334  *    Function: sd_bioclone_alloc
14335  *
14336  * Description: Allocate a buf(9S) and init it as per the given buf
14337  *              and the various arguments.  The associated sd_xbuf
14338  *              struct is (nearly) duplicated.  The struct buf *bp
14339  *              argument is saved in new_xp->xb_private.
14340  *
14341  *   Arguments: bp - ptr the the buf(9S) to be "shadowed"
14342  *              datalen - size of data area for the shadow bp
14343  *              blkno - starting LBA
14344  *              func - function pointer for b_iodone in the shadow buf. (May
14345  *                      be NULL if none.)
14346  *
14347  * Return Code: Pointer to allocates buf(9S) struct
14348  *
14349  *     Context: Can sleep.
14350  */
14351 
14352 static struct buf *
14353 sd_bioclone_alloc(struct buf *bp, size_t datalen,
14354         daddr_t blkno, int (*func)(struct buf *))
14355 {
14356         struct  sd_lun  *un;
14357         struct  sd_xbuf *xp;
14358         struct  sd_xbuf *new_xp;
14359         struct  buf     *new_bp;
14360 
14361         ASSERT(bp != NULL);
14362         xp = SD_GET_XBUF(bp);
14363         ASSERT(xp != NULL);
14364         un = SD_GET_UN(bp);
14365         ASSERT(un != NULL);
14366         ASSERT(!mutex_owned(SD_MUTEX(un)));
14367 
14368         new_bp = bioclone(bp, 0, datalen, SD_GET_DEV(un), blkno, func,
14369             NULL, KM_SLEEP);
14370 
14371         new_bp->b_lblkno     = blkno;
14372 
14373         /*
14374          * Allocate an xbuf for the shadow bp and copy the contents of the
14375          * original xbuf into it.
14376          */
14377         new_xp = kmem_alloc(sizeof (struct sd_xbuf), KM_SLEEP);
14378         bcopy(xp, new_xp, sizeof (struct sd_xbuf));
14379 
14380         /*
14381          * The given bp is automatically saved in the xb_private member
14382          * of the new xbuf.  Callers are allowed to depend on this.
14383          */
14384         new_xp->xb_private = bp;
14385 
14386         new_bp->b_private  = new_xp;
14387 
14388         return (new_bp);
14389 }
14390 
14391 /*
14392  *    Function: sd_shadow_buf_alloc
14393  *
14394  * Description: Allocate a buf(9S) and init it as per the given buf
14395  *              and the various arguments.  The associated sd_xbuf
14396  *              struct is (nearly) duplicated.  The struct buf *bp
14397  *              argument is saved in new_xp->xb_private.
14398  *
14399  *   Arguments: bp - ptr the the buf(9S) to be "shadowed"
14400  *              datalen - size of data area for the shadow bp
14401  *              bflags - B_READ or B_WRITE (pseudo flag)
14402  *              blkno - starting LBA
14403  *              func - function pointer for b_iodone in the shadow buf. (May
14404  *                      be NULL if none.)
14405  *
14406  * Return Code: Pointer to allocates buf(9S) struct
14407  *
14408  *     Context: Can sleep.
14409  */
14410 
14411 static struct buf *
14412 sd_shadow_buf_alloc(struct buf *bp, size_t datalen, uint_t bflags,
14413         daddr_t blkno, int (*func)(struct buf *))
14414 {
14415         struct  sd_lun  *un;
14416         struct  sd_xbuf *xp;
14417         struct  sd_xbuf *new_xp;
14418         struct  buf     *new_bp;
14419 
14420         ASSERT(bp != NULL);
14421         xp = SD_GET_XBUF(bp);
14422         ASSERT(xp != NULL);
14423         un = SD_GET_UN(bp);
14424         ASSERT(un != NULL);
14425         ASSERT(!mutex_owned(SD_MUTEX(un)));
14426 
14427         if (bp->b_flags & (B_PAGEIO | B_PHYS)) {
14428                 bp_mapin(bp);
14429         }
14430 
14431         bflags &= (B_READ | B_WRITE);
14432 #if defined(__i386) || defined(__amd64)
14433         new_bp = getrbuf(KM_SLEEP);
14434         new_bp->b_un.b_addr = kmem_zalloc(datalen, KM_SLEEP);
14435         new_bp->b_bcount = datalen;
14436         new_bp->b_flags = bflags |
14437             (bp->b_flags & ~(B_PAGEIO | B_PHYS | B_REMAPPED | B_SHADOW));
14438 #else
14439         new_bp = scsi_alloc_consistent_buf(SD_ADDRESS(un), NULL,
14440             datalen, bflags, SLEEP_FUNC, NULL);
14441 #endif
14442         new_bp->av_forw      = NULL;
14443         new_bp->av_back      = NULL;
14444         new_bp->b_dev        = bp->b_dev;
14445         new_bp->b_blkno      = blkno;
14446         new_bp->b_iodone = func;
14447         new_bp->b_edev       = bp->b_edev;
14448         new_bp->b_resid      = 0;
14449 
14450         /* We need to preserve the B_FAILFAST flag */
14451         if (bp->b_flags & B_FAILFAST) {
14452                 new_bp->b_flags |= B_FAILFAST;
14453         }
14454 
14455         /*
14456          * Allocate an xbuf for the shadow bp and copy the contents of the
14457          * original xbuf into it.
14458          */
14459         new_xp = kmem_alloc(sizeof (struct sd_xbuf), KM_SLEEP);
14460         bcopy(xp, new_xp, sizeof (struct sd_xbuf));
14461 
14462         /* Need later to copy data between the shadow buf & original buf! */
14463         new_xp->xb_pkt_flags |= PKT_CONSISTENT;
14464 
14465         /*
14466          * The given bp is automatically saved in the xb_private member
14467          * of the new xbuf.  Callers are allowed to depend on this.
14468          */
14469         new_xp->xb_private = bp;
14470 
14471         new_bp->b_private  = new_xp;
14472 
14473         return (new_bp);
14474 }
14475 
14476 /*
14477  *    Function: sd_bioclone_free
14478  *
14479  * Description: Deallocate a buf(9S) that was used for 'shadow' IO operations
14480  *              in the larger than partition operation.
14481  *
14482  *     Context: May be called under interrupt context
14483  */
14484 
14485 static void
14486 sd_bioclone_free(struct buf *bp)
14487 {
14488         struct sd_xbuf  *xp;
14489 
14490         ASSERT(bp != NULL);
14491         xp = SD_GET_XBUF(bp);
14492         ASSERT(xp != NULL);
14493 
14494         /*
14495          * Call bp_mapout() before freeing the buf,  in case a lower
14496          * layer or HBA  had done a bp_mapin().  we must do this here
14497          * as we are the "originator" of the shadow buf.
14498          */
14499         bp_mapout(bp);
14500 
14501         /*
14502          * Null out b_iodone before freeing the bp, to ensure that the driver
14503          * never gets confused by a stale value in this field. (Just a little
14504          * extra defensiveness here.)
14505          */
14506         bp->b_iodone = NULL;
14507 
14508         freerbuf(bp);
14509 
14510         kmem_free(xp, sizeof (struct sd_xbuf));
14511 }
14512 
14513 /*
14514  *    Function: sd_shadow_buf_free
14515  *
14516  * Description: Deallocate a buf(9S) that was used for 'shadow' IO operations.
14517  *
14518  *     Context: May be called under interrupt context
14519  */
14520 
14521 static void
14522 sd_shadow_buf_free(struct buf *bp)
14523 {
14524         struct sd_xbuf  *xp;
14525 
14526         ASSERT(bp != NULL);
14527         xp = SD_GET_XBUF(bp);
14528         ASSERT(xp != NULL);
14529 
14530 #if defined(__sparc)
14531         /*
14532          * Call bp_mapout() before freeing the buf,  in case a lower
14533          * layer or HBA  had done a bp_mapin().  we must do this here
14534          * as we are the "originator" of the shadow buf.
14535          */
14536         bp_mapout(bp);
14537 #endif
14538 
14539         /*
14540          * Null out b_iodone before freeing the bp, to ensure that the driver
14541          * never gets confused by a stale value in this field. (Just a little
14542          * extra defensiveness here.)
14543          */
14544         bp->b_iodone = NULL;
14545 
14546 #if defined(__i386) || defined(__amd64)
14547         kmem_free(bp->b_un.b_addr, bp->b_bcount);
14548         freerbuf(bp);
14549 #else
14550         scsi_free_consistent_buf(bp);
14551 #endif
14552 
14553         kmem_free(xp, sizeof (struct sd_xbuf));
14554 }
14555 
14556 
14557 /*
14558  *    Function: sd_print_transport_rejected_message
14559  *
14560  * Description: This implements the ludicrously complex rules for printing
14561  *              a "transport rejected" message.  This is to address the
14562  *              specific problem of having a flood of this error message
14563  *              produced when a failover occurs.
14564  *
14565  *     Context: Any.
14566  */
14567 
14568 static void
14569 sd_print_transport_rejected_message(struct sd_lun *un, struct sd_xbuf *xp,
14570         int code)
14571 {
14572         ASSERT(un != NULL);
14573         ASSERT(mutex_owned(SD_MUTEX(un)));
14574         ASSERT(xp != NULL);
14575 
14576         /*
14577          * Print the "transport rejected" message under the following
14578          * conditions:
14579          *
14580          * - Whenever the SD_LOGMASK_DIAG bit of sd_level_mask is set
14581          * - The error code from scsi_transport() is NOT a TRAN_FATAL_ERROR.
14582          * - If the error code IS a TRAN_FATAL_ERROR, then the message is
14583          *   printed the FIRST time a TRAN_FATAL_ERROR is returned from
14584          *   scsi_transport(9F) (which indicates that the target might have
14585          *   gone off-line).  This uses the un->un_tran_fatal_count
14586          *   count, which is incremented whenever a TRAN_FATAL_ERROR is
14587          *   received, and reset to zero whenver a TRAN_ACCEPT is returned
14588          *   from scsi_transport().
14589          *
14590          * The FLAG_SILENT in the scsi_pkt must be CLEARED in ALL of
14591          * the preceeding cases in order for the message to be printed.
14592          */
14593         if (((xp->xb_pktp->pkt_flags & FLAG_SILENT) == 0) &&
14594             (SD_FM_LOG(un) == SD_FM_LOG_NSUP)) {
14595                 if ((sd_level_mask & SD_LOGMASK_DIAG) ||
14596                     (code != TRAN_FATAL_ERROR) ||
14597                     (un->un_tran_fatal_count == 1)) {
14598                         switch (code) {
14599                         case TRAN_BADPKT:
14600                                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
14601                                     "transport rejected bad packet\n");
14602                                 break;
14603                         case TRAN_FATAL_ERROR:
14604                                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
14605                                     "transport rejected fatal error\n");
14606                                 break;
14607                         default:
14608                                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
14609                                     "transport rejected (%d)\n", code);
14610                                 break;
14611                         }
14612                 }
14613         }
14614 }
14615 
14616 
14617 /*
14618  *    Function: sd_add_buf_to_waitq
14619  *
14620  * Description: Add the given buf(9S) struct to the wait queue for the
14621  *              instance.  If sorting is enabled, then the buf is added
14622  *              to the queue via an elevator sort algorithm (a la
14623  *              disksort(9F)).  The SD_GET_BLKNO(bp) is used as the sort key.
14624  *              If sorting is not enabled, then the buf is just added
14625  *              to the end of the wait queue.
14626  *
14627  * Return Code: void
14628  *
14629  *     Context: Does not sleep/block, therefore technically can be called
14630  *              from any context.  However if sorting is enabled then the
14631  *              execution time is indeterminate, and may take long if
14632  *              the wait queue grows large.
14633  */
14634 
14635 static void
14636 sd_add_buf_to_waitq(struct sd_lun *un, struct buf *bp)
14637 {
14638         struct buf *ap;
14639 
14640         ASSERT(bp != NULL);
14641         ASSERT(un != NULL);
14642         ASSERT(mutex_owned(SD_MUTEX(un)));
14643 
14644         /* If the queue is empty, add the buf as the only entry & return. */
14645         if (un->un_waitq_headp == NULL) {
14646                 ASSERT(un->un_waitq_tailp == NULL);
14647                 un->un_waitq_headp = un->un_waitq_tailp = bp;
14648                 bp->av_forw = NULL;
14649                 return;
14650         }
14651 
14652         ASSERT(un->un_waitq_tailp != NULL);
14653 
14654         /*
14655          * If sorting is disabled, just add the buf to the tail end of
14656          * the wait queue and return.
14657          */
14658         if (un->un_f_disksort_disabled || un->un_f_enable_rmw) {
14659                 un->un_waitq_tailp->av_forw = bp;
14660                 un->un_waitq_tailp = bp;
14661                 bp->av_forw = NULL;
14662                 return;
14663         }
14664 
14665         /*
14666          * Sort thru the list of requests currently on the wait queue
14667          * and add the new buf request at the appropriate position.
14668          *
14669          * The un->un_waitq_headp is an activity chain pointer on which
14670          * we keep two queues, sorted in ascending SD_GET_BLKNO() order. The
14671          * first queue holds those requests which are positioned after
14672          * the current SD_GET_BLKNO() (in the first request); the second holds
14673          * requests which came in after their SD_GET_BLKNO() number was passed.
14674          * Thus we implement a one way scan, retracting after reaching
14675          * the end of the drive to the first request on the second
14676          * queue, at which time it becomes the first queue.
14677          * A one-way scan is natural because of the way UNIX read-ahead
14678          * blocks are allocated.
14679          *
14680          * If we lie after the first request, then we must locate the
14681          * second request list and add ourselves to it.
14682          */
14683         ap = un->un_waitq_headp;
14684         if (SD_GET_BLKNO(bp) < SD_GET_BLKNO(ap)) {
14685                 while (ap->av_forw != NULL) {
14686                         /*
14687                          * Look for an "inversion" in the (normally
14688                          * ascending) block numbers. This indicates
14689                          * the start of the second request list.
14690                          */
14691                         if (SD_GET_BLKNO(ap->av_forw) < SD_GET_BLKNO(ap)) {
14692                                 /*
14693                                  * Search the second request list for the
14694                                  * first request at a larger block number.
14695                                  * We go before that; however if there is
14696                                  * no such request, we go at the end.
14697                                  */
14698                                 do {
14699                                         if (SD_GET_BLKNO(bp) <
14700                                             SD_GET_BLKNO(ap->av_forw)) {
14701                                                 goto insert;
14702                                         }
14703                                         ap = ap->av_forw;
14704                                 } while (ap->av_forw != NULL);
14705                                 goto insert;            /* after last */
14706                         }
14707                         ap = ap->av_forw;
14708                 }
14709 
14710                 /*
14711                  * No inversions... we will go after the last, and
14712                  * be the first request in the second request list.
14713                  */
14714                 goto insert;
14715         }
14716 
14717         /*
14718          * Request is at/after the current request...
14719          * sort in the first request list.
14720          */
14721         while (ap->av_forw != NULL) {
14722                 /*
14723                  * We want to go after the current request (1) if
14724                  * there is an inversion after it (i.e. it is the end
14725                  * of the first request list), or (2) if the next
14726                  * request is a larger block no. than our request.
14727                  */
14728                 if ((SD_GET_BLKNO(ap->av_forw) < SD_GET_BLKNO(ap)) ||
14729                     (SD_GET_BLKNO(bp) < SD_GET_BLKNO(ap->av_forw))) {
14730                         goto insert;
14731                 }
14732                 ap = ap->av_forw;
14733         }
14734 
14735         /*
14736          * Neither a second list nor a larger request, therefore
14737          * we go at the end of the first list (which is the same
14738          * as the end of the whole schebang).
14739          */
14740 insert:
14741         bp->av_forw = ap->av_forw;
14742         ap->av_forw = bp;
14743 
14744         /*
14745          * If we inserted onto the tail end of the waitq, make sure the
14746          * tail pointer is updated.
14747          */
14748         if (ap == un->un_waitq_tailp) {
14749                 un->un_waitq_tailp = bp;
14750         }
14751 }
14752 
14753 
14754 /*
14755  *    Function: sd_start_cmds
14756  *
14757  * Description: Remove and transport cmds from the driver queues.
14758  *
14759  *   Arguments: un - pointer to the unit (soft state) struct for the target.
14760  *
14761  *              immed_bp - ptr to a buf to be transported immediately. Only
14762  *              the immed_bp is transported; bufs on the waitq are not
14763  *              processed and the un_retry_bp is not checked.  If immed_bp is
14764  *              NULL, then normal queue processing is performed.
14765  *
14766  *     Context: May be called from kernel thread context, interrupt context,
14767  *              or runout callback context. This function may not block or
14768  *              call routines that block.
14769  */
14770 
14771 static void
14772 sd_start_cmds(struct sd_lun *un, struct buf *immed_bp)
14773 {
14774         struct  sd_xbuf *xp;
14775         struct  buf     *bp;
14776         void    (*statp)(kstat_io_t *);
14777 #if defined(__i386) || defined(__amd64) /* DMAFREE for x86 only */
14778         void    (*saved_statp)(kstat_io_t *);
14779 #endif
14780         int     rval;
14781         struct sd_fm_internal *sfip = NULL;
14782 
14783         ASSERT(un != NULL);
14784         ASSERT(mutex_owned(SD_MUTEX(un)));
14785         ASSERT(un->un_ncmds_in_transport >= 0);
14786         ASSERT(un->un_throttle >= 0);
14787 
14788         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_start_cmds: entry\n");
14789 
14790         do {
14791 #if defined(__i386) || defined(__amd64) /* DMAFREE for x86 only */
14792                 saved_statp = NULL;
14793 #endif
14794 
14795                 /*
14796                  * If we are syncing or dumping, fail the command to
14797                  * avoid recursively calling back into scsi_transport().
14798                  * The dump I/O itself uses a separate code path so this
14799                  * only prevents non-dump I/O from being sent while dumping.
14800                  * File system sync takes place before dumping begins.
14801                  * During panic, filesystem I/O is allowed provided
14802                  * un_in_callback is <= 1.  This is to prevent recursion
14803                  * such as sd_start_cmds -> scsi_transport -> sdintr ->
14804                  * sd_start_cmds and so on.  See panic.c for more information
14805                  * about the states the system can be in during panic.
14806                  */
14807                 if ((un->un_state == SD_STATE_DUMPING) ||
14808                     (ddi_in_panic() && (un->un_in_callback > 1))) {
14809                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
14810                             "sd_start_cmds: panicking\n");
14811                         goto exit;
14812                 }
14813 
14814                 if ((bp = immed_bp) != NULL) {
14815                         /*
14816                          * We have a bp that must be transported immediately.
14817                          * It's OK to transport the immed_bp here without doing
14818                          * the throttle limit check because the immed_bp is
14819                          * always used in a retry/recovery case. This means
14820                          * that we know we are not at the throttle limit by
14821                          * virtue of the fact that to get here we must have
14822                          * already gotten a command back via sdintr(). This also
14823                          * relies on (1) the command on un_retry_bp preventing
14824                          * further commands from the waitq from being issued;
14825                          * and (2) the code in sd_retry_command checking the
14826                          * throttle limit before issuing a delayed or immediate
14827                          * retry. This holds even if the throttle limit is
14828                          * currently ratcheted down from its maximum value.
14829                          */
14830                         statp = kstat_runq_enter;
14831                         if (bp == un->un_retry_bp) {
14832                                 ASSERT((un->un_retry_statp == NULL) ||
14833                                     (un->un_retry_statp == kstat_waitq_enter) ||
14834                                     (un->un_retry_statp ==
14835                                     kstat_runq_back_to_waitq));
14836                                 /*
14837                                  * If the waitq kstat was incremented when
14838                                  * sd_set_retry_bp() queued this bp for a retry,
14839                                  * then we must set up statp so that the waitq
14840                                  * count will get decremented correctly below.
14841                                  * Also we must clear un->un_retry_statp to
14842                                  * ensure that we do not act on a stale value
14843                                  * in this field.
14844                                  */
14845                                 if ((un->un_retry_statp == kstat_waitq_enter) ||
14846                                     (un->un_retry_statp ==
14847                                     kstat_runq_back_to_waitq)) {
14848                                         statp = kstat_waitq_to_runq;
14849                                 }
14850 #if defined(__i386) || defined(__amd64) /* DMAFREE for x86 only */
14851                                 saved_statp = un->un_retry_statp;
14852 #endif
14853                                 un->un_retry_statp = NULL;
14854 
14855                                 SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un,
14856                                     "sd_start_cmds: un:0x%p: GOT retry_bp:0x%p "
14857                                     "un_throttle:%d un_ncmds_in_transport:%d\n",
14858                                     un, un->un_retry_bp, un->un_throttle,
14859                                     un->un_ncmds_in_transport);
14860                         } else {
14861                                 SD_TRACE(SD_LOG_IO_CORE, un, "sd_start_cmds: "
14862                                     "processing priority bp:0x%p\n", bp);
14863                         }
14864 
14865                 } else if ((bp = un->un_waitq_headp) != NULL) {
14866                         /*
14867                          * A command on the waitq is ready to go, but do not
14868                          * send it if:
14869                          *
14870                          * (1) the throttle limit has been reached, or
14871                          * (2) a retry is pending, or
14872                          * (3) a START_STOP_UNIT callback pending, or
14873                          * (4) a callback for a SD_PATH_DIRECT_PRIORITY
14874                          *      command is pending.
14875                          *
14876                          * For all of these conditions, IO processing will
14877                          * restart after the condition is cleared.
14878                          */
14879                         if (un->un_ncmds_in_transport >= un->un_throttle) {
14880                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
14881                                     "sd_start_cmds: exiting, "
14882                                     "throttle limit reached!\n");
14883                                 goto exit;
14884                         }
14885                         if (un->un_retry_bp != NULL) {
14886                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
14887                                     "sd_start_cmds: exiting, retry pending!\n");
14888                                 goto exit;
14889                         }
14890                         if (un->un_startstop_timeid != NULL) {
14891                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
14892                                     "sd_start_cmds: exiting, "
14893                                     "START_STOP pending!\n");
14894                                 goto exit;
14895                         }
14896                         if (un->un_direct_priority_timeid != NULL) {
14897                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
14898                                     "sd_start_cmds: exiting, "
14899                                     "SD_PATH_DIRECT_PRIORITY cmd. pending!\n");
14900                                 goto exit;
14901                         }
14902 
14903                         /* Dequeue the command */
14904                         un->un_waitq_headp = bp->av_forw;
14905                         if (un->un_waitq_headp == NULL) {
14906                                 un->un_waitq_tailp = NULL;
14907                         }
14908                         bp->av_forw = NULL;
14909                         statp = kstat_waitq_to_runq;
14910                         SD_TRACE(SD_LOG_IO_CORE, un,
14911                             "sd_start_cmds: processing waitq bp:0x%p\n", bp);
14912 
14913                 } else {
14914                         /* No work to do so bail out now */
14915                         SD_TRACE(SD_LOG_IO_CORE, un,
14916                             "sd_start_cmds: no more work, exiting!\n");
14917                         goto exit;
14918                 }
14919 
14920                 /*
14921                  * Reset the state to normal. This is the mechanism by which
14922                  * the state transitions from either SD_STATE_RWAIT or
14923                  * SD_STATE_OFFLINE to SD_STATE_NORMAL.
14924                  * If state is SD_STATE_PM_CHANGING then this command is
14925                  * part of the device power control and the state must
14926                  * not be put back to normal. Doing so would would
14927                  * allow new commands to proceed when they shouldn't,
14928                  * the device may be going off.
14929                  */
14930                 if ((un->un_state != SD_STATE_SUSPENDED) &&
14931                     (un->un_state != SD_STATE_PM_CHANGING)) {
14932                         New_state(un, SD_STATE_NORMAL);
14933                 }
14934 
14935                 xp = SD_GET_XBUF(bp);
14936                 ASSERT(xp != NULL);
14937 
14938 #if defined(__i386) || defined(__amd64) /* DMAFREE for x86 only */
14939                 /*
14940                  * Allocate the scsi_pkt if we need one, or attach DMA
14941                  * resources if we have a scsi_pkt that needs them. The
14942                  * latter should only occur for commands that are being
14943                  * retried.
14944                  */
14945                 if ((xp->xb_pktp == NULL) ||
14946                     ((xp->xb_pkt_flags & SD_XB_DMA_FREED) != 0)) {
14947 #else
14948                 if (xp->xb_pktp == NULL) {
14949 #endif
14950                         /*
14951                          * There is no scsi_pkt allocated for this buf. Call
14952                          * the initpkt function to allocate & init one.
14953                          *
14954                          * The scsi_init_pkt runout callback functionality is
14955                          * implemented as follows:
14956                          *
14957                          * 1) The initpkt function always calls
14958                          *    scsi_init_pkt(9F) with sdrunout specified as the
14959                          *    callback routine.
14960                          * 2) A successful packet allocation is initialized and
14961                          *    the I/O is transported.
14962                          * 3) The I/O associated with an allocation resource
14963                          *    failure is left on its queue to be retried via
14964                          *    runout or the next I/O.
14965                          * 4) The I/O associated with a DMA error is removed
14966                          *    from the queue and failed with EIO. Processing of
14967                          *    the transport queues is also halted to be
14968                          *    restarted via runout or the next I/O.
14969                          * 5) The I/O associated with a CDB size or packet
14970                          *    size error is removed from the queue and failed
14971                          *    with EIO. Processing of the transport queues is
14972                          *    continued.
14973                          *
14974                          * Note: there is no interface for canceling a runout
14975                          * callback. To prevent the driver from detaching or
14976                          * suspending while a runout is pending the driver
14977                          * state is set to SD_STATE_RWAIT
14978                          *
14979                          * Note: using the scsi_init_pkt callback facility can
14980                          * result in an I/O request persisting at the head of
14981                          * the list which cannot be satisfied even after
14982                          * multiple retries. In the future the driver may
14983                          * implement some kind of maximum runout count before
14984                          * failing an I/O.
14985                          *
14986                          * Note: the use of funcp below may seem superfluous,
14987                          * but it helps warlock figure out the correct
14988                          * initpkt function calls (see [s]sd.wlcmd).
14989                          */
14990                         struct scsi_pkt *pktp;
14991                         int (*funcp)(struct buf *bp, struct scsi_pkt **pktp);
14992 
14993                         ASSERT(bp != un->un_rqs_bp);
14994 
14995                         funcp = sd_initpkt_map[xp->xb_chain_iostart];
14996                         switch ((*funcp)(bp, &pktp)) {
14997                         case  SD_PKT_ALLOC_SUCCESS:
14998                                 xp->xb_pktp = pktp;
14999                                 SD_TRACE(SD_LOG_IO_CORE, un,
15000                                     "sd_start_cmd: SD_PKT_ALLOC_SUCCESS 0x%p\n",
15001                                     pktp);
15002                                 goto got_pkt;
15003 
15004                         case SD_PKT_ALLOC_FAILURE:
15005                                 /*
15006                                  * Temporary (hopefully) resource depletion.
15007                                  * Since retries and RQS commands always have a
15008                                  * scsi_pkt allocated, these cases should never
15009                                  * get here. So the only cases this needs to
15010                                  * handle is a bp from the waitq (which we put
15011                                  * back onto the waitq for sdrunout), or a bp
15012                                  * sent as an immed_bp (which we just fail).
15013                                  */
15014                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15015                                     "sd_start_cmds: SD_PKT_ALLOC_FAILURE\n");
15016 
15017 #if defined(__i386) || defined(__amd64) /* DMAFREE for x86 only */
15018 
15019                                 if (bp == immed_bp) {
15020                                         /*
15021                                          * If SD_XB_DMA_FREED is clear, then
15022                                          * this is a failure to allocate a
15023                                          * scsi_pkt, and we must fail the
15024                                          * command.
15025                                          */
15026                                         if ((xp->xb_pkt_flags &
15027                                             SD_XB_DMA_FREED) == 0) {
15028                                                 break;
15029                                         }
15030 
15031                                         /*
15032                                          * If this immediate command is NOT our
15033                                          * un_retry_bp, then we must fail it.
15034                                          */
15035                                         if (bp != un->un_retry_bp) {
15036                                                 break;
15037                                         }
15038 
15039                                         /*
15040                                          * We get here if this cmd is our
15041                                          * un_retry_bp that was DMAFREED, but
15042                                          * scsi_init_pkt() failed to reallocate
15043                                          * DMA resources when we attempted to
15044                                          * retry it. This can happen when an
15045                                          * mpxio failover is in progress, but
15046                                          * we don't want to just fail the
15047                                          * command in this case.
15048                                          *
15049                                          * Use timeout(9F) to restart it after
15050                                          * a 100ms delay.  We don't want to
15051                                          * let sdrunout() restart it, because
15052                                          * sdrunout() is just supposed to start
15053                                          * commands that are sitting on the
15054                                          * wait queue.  The un_retry_bp stays
15055                                          * set until the command completes, but
15056                                          * sdrunout can be called many times
15057                                          * before that happens.  Since sdrunout
15058                                          * cannot tell if the un_retry_bp is
15059                                          * already in the transport, it could
15060                                          * end up calling scsi_transport() for
15061                                          * the un_retry_bp multiple times.
15062                                          *
15063                                          * Also: don't schedule the callback
15064                                          * if some other callback is already
15065                                          * pending.
15066                                          */
15067                                         if (un->un_retry_statp == NULL) {
15068                                                 /*
15069                                                  * restore the kstat pointer to
15070                                                  * keep kstat counts coherent
15071                                                  * when we do retry the command.
15072                                                  */
15073                                                 un->un_retry_statp =
15074                                                     saved_statp;
15075                                         }
15076 
15077                                         if ((un->un_startstop_timeid == NULL) &&
15078                                             (un->un_retry_timeid == NULL) &&
15079                                             (un->un_direct_priority_timeid ==
15080                                             NULL)) {
15081 
15082                                                 un->un_retry_timeid =
15083                                                     timeout(
15084                                                     sd_start_retry_command,
15085                                                     un, SD_RESTART_TIMEOUT);
15086                                         }
15087                                         goto exit;
15088                                 }
15089 
15090 #else
15091                                 if (bp == immed_bp) {
15092                                         break;  /* Just fail the command */
15093                                 }
15094 #endif
15095 
15096                                 /* Add the buf back to the head of the waitq */
15097                                 bp->av_forw = un->un_waitq_headp;
15098                                 un->un_waitq_headp = bp;
15099                                 if (un->un_waitq_tailp == NULL) {
15100                                         un->un_waitq_tailp = bp;
15101                                 }
15102                                 goto exit;
15103 
15104                         case SD_PKT_ALLOC_FAILURE_NO_DMA:
15105                                 /*
15106                                  * HBA DMA resource failure. Fail the command
15107                                  * and continue processing of the queues.
15108                                  */
15109                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15110                                     "sd_start_cmds: "
15111                                     "SD_PKT_ALLOC_FAILURE_NO_DMA\n");
15112                                 break;
15113 
15114                         case SD_PKT_ALLOC_FAILURE_PKT_TOO_SMALL:
15115                                 /*
15116                                  * Note:x86: Partial DMA mapping not supported
15117                                  * for USCSI commands, and all the needed DMA
15118                                  * resources were not allocated.
15119                                  */
15120                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15121                                     "sd_start_cmds: "
15122                                     "SD_PKT_ALLOC_FAILURE_PKT_TOO_SMALL\n");
15123                                 break;
15124 
15125                         case SD_PKT_ALLOC_FAILURE_CDB_TOO_SMALL:
15126                                 /*
15127                                  * Note:x86: Request cannot fit into CDB based
15128                                  * on lba and len.
15129                                  */
15130                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15131                                     "sd_start_cmds: "
15132                                     "SD_PKT_ALLOC_FAILURE_CDB_TOO_SMALL\n");
15133                                 break;
15134 
15135                         default:
15136                                 /* Should NEVER get here! */
15137                                 panic("scsi_initpkt error");
15138                                 /*NOTREACHED*/
15139                         }
15140 
15141                         /*
15142                          * Fatal error in allocating a scsi_pkt for this buf.
15143                          * Update kstats & return the buf with an error code.
15144                          * We must use sd_return_failed_command_no_restart() to
15145                          * avoid a recursive call back into sd_start_cmds().
15146                          * However this also means that we must keep processing
15147                          * the waitq here in order to avoid stalling.
15148                          */
15149                         if (statp == kstat_waitq_to_runq) {
15150                                 SD_UPDATE_KSTATS(un, kstat_waitq_exit, bp);
15151                         }
15152                         sd_return_failed_command_no_restart(un, bp, EIO);
15153                         if (bp == immed_bp) {
15154                                 /* immed_bp is gone by now, so clear this */
15155                                 immed_bp = NULL;
15156                         }
15157                         continue;
15158                 }
15159 got_pkt:
15160                 if (bp == immed_bp) {
15161                         /* goto the head of the class.... */
15162                         xp->xb_pktp->pkt_flags |= FLAG_HEAD;
15163                 }
15164 
15165                 un->un_ncmds_in_transport++;
15166                 SD_UPDATE_KSTATS(un, statp, bp);
15167 
15168                 /*
15169                  * Call scsi_transport() to send the command to the target.
15170                  * According to SCSA architecture, we must drop the mutex here
15171                  * before calling scsi_transport() in order to avoid deadlock.
15172                  * Note that the scsi_pkt's completion routine can be executed
15173                  * (from interrupt context) even before the call to
15174                  * scsi_transport() returns.
15175                  */
15176                 SD_TRACE(SD_LOG_IO_CORE, un,
15177                     "sd_start_cmds: calling scsi_transport()\n");
15178                 DTRACE_PROBE1(scsi__transport__dispatch, struct buf *, bp);
15179 
15180                 mutex_exit(SD_MUTEX(un));
15181                 rval = scsi_transport(xp->xb_pktp);
15182                 mutex_enter(SD_MUTEX(un));
15183 
15184                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15185                     "sd_start_cmds: scsi_transport() returned %d\n", rval);
15186 
15187                 switch (rval) {
15188                 case TRAN_ACCEPT:
15189                         /* Clear this with every pkt accepted by the HBA */
15190                         un->un_tran_fatal_count = 0;
15191                         break;  /* Success; try the next cmd (if any) */
15192 
15193                 case TRAN_BUSY:
15194                         un->un_ncmds_in_transport--;
15195                         ASSERT(un->un_ncmds_in_transport >= 0);
15196 
15197                         /*
15198                          * Don't retry request sense, the sense data
15199                          * is lost when another request is sent.
15200                          * Free up the rqs buf and retry
15201                          * the original failed cmd.  Update kstat.
15202                          */
15203                         if (bp == un->un_rqs_bp) {
15204                                 SD_UPDATE_KSTATS(un, kstat_runq_exit, bp);
15205                                 bp = sd_mark_rqs_idle(un, xp);
15206                                 sd_retry_command(un, bp, SD_RETRIES_STANDARD,
15207                                     NULL, NULL, EIO, un->un_busy_timeout / 500,
15208                                     kstat_waitq_enter);
15209                                 goto exit;
15210                         }
15211 
15212 #if defined(__i386) || defined(__amd64) /* DMAFREE for x86 only */
15213                         /*
15214                          * Free the DMA resources for the  scsi_pkt. This will
15215                          * allow mpxio to select another path the next time
15216                          * we call scsi_transport() with this scsi_pkt.
15217                          * See sdintr() for the rationalization behind this.
15218                          */
15219                         if ((un->un_f_is_fibre == TRUE) &&
15220                             ((xp->xb_pkt_flags & SD_XB_USCSICMD) == 0) &&
15221                             ((xp->xb_pktp->pkt_flags & FLAG_SENSING) == 0)) {
15222                                 scsi_dmafree(xp->xb_pktp);
15223                                 xp->xb_pkt_flags |= SD_XB_DMA_FREED;
15224                         }
15225 #endif
15226 
15227                         if (SD_IS_DIRECT_PRIORITY(SD_GET_XBUF(bp))) {
15228                                 /*
15229                                  * Commands that are SD_PATH_DIRECT_PRIORITY
15230                                  * are for error recovery situations. These do
15231                                  * not use the normal command waitq, so if they
15232                                  * get a TRAN_BUSY we cannot put them back onto
15233                                  * the waitq for later retry. One possible
15234                                  * problem is that there could already be some
15235                                  * other command on un_retry_bp that is waiting
15236                                  * for this one to complete, so we would be
15237                                  * deadlocked if we put this command back onto
15238                                  * the waitq for later retry (since un_retry_bp
15239                                  * must complete before the driver gets back to
15240                                  * commands on the waitq).
15241                                  *
15242                                  * To avoid deadlock we must schedule a callback
15243                                  * that will restart this command after a set
15244                                  * interval.  This should keep retrying for as
15245                                  * long as the underlying transport keeps
15246                                  * returning TRAN_BUSY (just like for other
15247                                  * commands).  Use the same timeout interval as
15248                                  * for the ordinary TRAN_BUSY retry.
15249                                  */
15250                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15251                                     "sd_start_cmds: scsi_transport() returned "
15252                                     "TRAN_BUSY for DIRECT_PRIORITY cmd!\n");
15253 
15254                                 SD_UPDATE_KSTATS(un, kstat_runq_exit, bp);
15255                                 un->un_direct_priority_timeid =
15256                                     timeout(sd_start_direct_priority_command,
15257                                     bp, un->un_busy_timeout / 500);
15258 
15259                                 goto exit;
15260                         }
15261 
15262                         /*
15263                          * For TRAN_BUSY, we want to reduce the throttle value,
15264                          * unless we are retrying a command.
15265                          */
15266                         if (bp != un->un_retry_bp) {
15267                                 sd_reduce_throttle(un, SD_THROTTLE_TRAN_BUSY);
15268                         }
15269 
15270                         /*
15271                          * Set up the bp to be tried again 10 ms later.
15272                          * Note:x86: Is there a timeout value in the sd_lun
15273                          * for this condition?
15274                          */
15275                         sd_set_retry_bp(un, bp, un->un_busy_timeout / 500,
15276                             kstat_runq_back_to_waitq);
15277                         goto exit;
15278 
15279                 case TRAN_FATAL_ERROR:
15280                         un->un_tran_fatal_count++;
15281                         /* FALLTHRU */
15282 
15283                 case TRAN_BADPKT:
15284                 default:
15285                         un->un_ncmds_in_transport--;
15286                         ASSERT(un->un_ncmds_in_transport >= 0);
15287 
15288                         /*
15289                          * If this is our REQUEST SENSE command with a
15290                          * transport error, we must get back the pointers
15291                          * to the original buf, and mark the REQUEST
15292                          * SENSE command as "available".
15293                          */
15294                         if (bp == un->un_rqs_bp) {
15295                                 bp = sd_mark_rqs_idle(un, xp);
15296                                 xp = SD_GET_XBUF(bp);
15297                         } else {
15298                                 /*
15299                                  * Legacy behavior: do not update transport
15300                                  * error count for request sense commands.
15301                                  */
15302                                 SD_UPDATE_ERRSTATS(un, sd_transerrs);
15303                         }
15304 
15305                         SD_UPDATE_KSTATS(un, kstat_runq_exit, bp);
15306                         sd_print_transport_rejected_message(un, xp, rval);
15307 
15308                         /*
15309                          * This command will be terminated by SD driver due
15310                          * to a fatal transport error. We should post
15311                          * ereport.io.scsi.cmd.disk.tran with driver-assessment
15312                          * of "fail" for any command to indicate this
15313                          * situation.
15314                          */
15315                         if (xp->xb_ena > 0) {
15316                                 ASSERT(un->un_fm_private != NULL);
15317                                 sfip = un->un_fm_private;
15318                                 sfip->fm_ssc.ssc_flags |= SSC_FLAGS_TRAN_ABORT;
15319                                 sd_ssc_extract_info(&sfip->fm_ssc, un,
15320                                     xp->xb_pktp, bp, xp);
15321                                 sd_ssc_post(&sfip->fm_ssc, SD_FM_DRV_FATAL);
15322                         }
15323 
15324                         /*
15325                          * We must use sd_return_failed_command_no_restart() to
15326                          * avoid a recursive call back into sd_start_cmds().
15327                          * However this also means that we must keep processing
15328                          * the waitq here in order to avoid stalling.
15329                          */
15330                         sd_return_failed_command_no_restart(un, bp, EIO);
15331 
15332                         /*
15333                          * Notify any threads waiting in sd_ddi_suspend() that
15334                          * a command completion has occurred.
15335                          */
15336                         if (un->un_state == SD_STATE_SUSPENDED) {
15337                                 cv_broadcast(&un->un_disk_busy_cv);
15338                         }
15339 
15340                         if (bp == immed_bp) {
15341                                 /* immed_bp is gone by now, so clear this */
15342                                 immed_bp = NULL;
15343                         }
15344                         break;
15345                 }
15346 
15347         } while (immed_bp == NULL);
15348 
15349 exit:
15350         ASSERT(mutex_owned(SD_MUTEX(un)));
15351         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_start_cmds: exit\n");
15352 }
15353 
15354 
15355 /*
15356  *    Function: sd_return_command
15357  *
15358  * Description: Returns a command to its originator (with or without an
15359  *              error).  Also starts commands waiting to be transported
15360  *              to the target.
15361  *
15362  *     Context: May be called from interrupt, kernel, or timeout context
15363  */
15364 
15365 static void
15366 sd_return_command(struct sd_lun *un, struct buf *bp)
15367 {
15368         struct sd_xbuf *xp;
15369         struct scsi_pkt *pktp;
15370         struct sd_fm_internal *sfip;
15371 
15372         ASSERT(bp != NULL);
15373         ASSERT(un != NULL);
15374         ASSERT(mutex_owned(SD_MUTEX(un)));
15375         ASSERT(bp != un->un_rqs_bp);
15376         xp = SD_GET_XBUF(bp);
15377         ASSERT(xp != NULL);
15378 
15379         pktp = SD_GET_PKTP(bp);
15380         sfip = (struct sd_fm_internal *)un->un_fm_private;
15381         ASSERT(sfip != NULL);
15382 
15383         SD_TRACE(SD_LOG_IO_CORE, un, "sd_return_command: entry\n");
15384 
15385         /*
15386          * Note: check for the "sdrestart failed" case.
15387          */
15388         if ((un->un_partial_dma_supported == 1) &&
15389             ((xp->xb_pkt_flags & SD_XB_USCSICMD) != SD_XB_USCSICMD) &&
15390             (geterror(bp) == 0) && (xp->xb_dma_resid != 0) &&
15391             (xp->xb_pktp->pkt_resid == 0)) {
15392 
15393                 if (sd_setup_next_xfer(un, bp, pktp, xp) != 0) {
15394                         /*
15395                          * Successfully set up next portion of cmd
15396                          * transfer, try sending it
15397                          */
15398                         sd_retry_command(un, bp, SD_RETRIES_NOCHECK,
15399                             NULL, NULL, 0, (clock_t)0, NULL);
15400                         sd_start_cmds(un, NULL);
15401                         return; /* Note:x86: need a return here? */
15402                 }
15403         }
15404 
15405         /*
15406          * If this is the failfast bp, clear it from un_failfast_bp. This
15407          * can happen if upon being re-tried the failfast bp either
15408          * succeeded or encountered another error (possibly even a different
15409          * error than the one that precipitated the failfast state, but in
15410          * that case it would have had to exhaust retries as well). Regardless,
15411          * this should not occur whenever the instance is in the active
15412          * failfast state.
15413          */
15414         if (bp == un->un_failfast_bp) {
15415                 ASSERT(un->un_failfast_state == SD_FAILFAST_INACTIVE);
15416                 un->un_failfast_bp = NULL;
15417         }
15418 
15419         /*
15420          * Clear the failfast state upon successful completion of ANY cmd.
15421          */
15422         if (bp->b_error == 0) {
15423                 un->un_failfast_state = SD_FAILFAST_INACTIVE;
15424                 /*
15425                  * If this is a successful command, but used to be retried,
15426                  * we will take it as a recovered command and post an
15427                  * ereport with driver-assessment of "recovered".
15428                  */
15429                 if (xp->xb_ena > 0) {
15430                         sd_ssc_extract_info(&sfip->fm_ssc, un, pktp, bp, xp);
15431                         sd_ssc_post(&sfip->fm_ssc, SD_FM_DRV_RECOVERY);
15432                 }
15433         } else {
15434                 /*
15435                  * If this is a failed non-USCSI command we will post an
15436                  * ereport with driver-assessment set accordingly("fail" or
15437                  * "fatal").
15438                  */
15439                 if (!(xp->xb_pkt_flags & SD_XB_USCSICMD)) {
15440                         sd_ssc_extract_info(&sfip->fm_ssc, un, pktp, bp, xp);
15441                         sd_ssc_post(&sfip->fm_ssc, SD_FM_DRV_FATAL);
15442                 }
15443         }
15444 
15445         /*
15446          * This is used if the command was retried one or more times. Show that
15447          * we are done with it, and allow processing of the waitq to resume.
15448          */
15449         if (bp == un->un_retry_bp) {
15450                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15451                     "sd_return_command: un:0x%p: "
15452                     "RETURNING retry_bp:0x%p\n", un, un->un_retry_bp);
15453                 un->un_retry_bp = NULL;
15454                 un->un_retry_statp = NULL;
15455         }
15456 
15457         SD_UPDATE_RDWR_STATS(un, bp);
15458         SD_UPDATE_PARTITION_STATS(un, bp);
15459 
15460         switch (un->un_state) {
15461         case SD_STATE_SUSPENDED:
15462                 /*
15463                  * Notify any threads waiting in sd_ddi_suspend() that
15464                  * a command completion has occurred.
15465                  */
15466                 cv_broadcast(&un->un_disk_busy_cv);
15467                 break;
15468         default:
15469                 sd_start_cmds(un, NULL);
15470                 break;
15471         }
15472 
15473         /* Return this command up the iodone chain to its originator. */
15474         mutex_exit(SD_MUTEX(un));
15475 
15476         (*(sd_destroypkt_map[xp->xb_chain_iodone]))(bp);
15477         xp->xb_pktp = NULL;
15478 
15479         SD_BEGIN_IODONE(xp->xb_chain_iodone, un, bp);
15480 
15481         ASSERT(!mutex_owned(SD_MUTEX(un)));
15482         mutex_enter(SD_MUTEX(un));
15483 
15484         SD_TRACE(SD_LOG_IO_CORE, un, "sd_return_command: exit\n");
15485 }
15486 
15487 
15488 /*
15489  *    Function: sd_return_failed_command
15490  *
15491  * Description: Command completion when an error occurred.
15492  *
15493  *     Context: May be called from interrupt context
15494  */
15495 
15496 static void
15497 sd_return_failed_command(struct sd_lun *un, struct buf *bp, int errcode)
15498 {
15499         ASSERT(bp != NULL);
15500         ASSERT(un != NULL);
15501         ASSERT(mutex_owned(SD_MUTEX(un)));
15502 
15503         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15504             "sd_return_failed_command: entry\n");
15505 
15506         /*
15507          * b_resid could already be nonzero due to a partial data
15508          * transfer, so do not change it here.
15509          */
15510         SD_BIOERROR(bp, errcode);
15511 
15512         sd_return_command(un, bp);
15513         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15514             "sd_return_failed_command: exit\n");
15515 }
15516 
15517 
15518 /*
15519  *    Function: sd_return_failed_command_no_restart
15520  *
15521  * Description: Same as sd_return_failed_command, but ensures that no
15522  *              call back into sd_start_cmds will be issued.
15523  *
15524  *     Context: May be called from interrupt context
15525  */
15526 
15527 static void
15528 sd_return_failed_command_no_restart(struct sd_lun *un, struct buf *bp,
15529         int errcode)
15530 {
15531         struct sd_xbuf *xp;
15532 
15533         ASSERT(bp != NULL);
15534         ASSERT(un != NULL);
15535         ASSERT(mutex_owned(SD_MUTEX(un)));
15536         xp = SD_GET_XBUF(bp);
15537         ASSERT(xp != NULL);
15538         ASSERT(errcode != 0);
15539 
15540         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15541             "sd_return_failed_command_no_restart: entry\n");
15542 
15543         /*
15544          * b_resid could already be nonzero due to a partial data
15545          * transfer, so do not change it here.
15546          */
15547         SD_BIOERROR(bp, errcode);
15548 
15549         /*
15550          * If this is the failfast bp, clear it. This can happen if the
15551          * failfast bp encounterd a fatal error when we attempted to
15552          * re-try it (such as a scsi_transport(9F) failure).  However
15553          * we should NOT be in an active failfast state if the failfast
15554          * bp is not NULL.
15555          */
15556         if (bp == un->un_failfast_bp) {
15557                 ASSERT(un->un_failfast_state == SD_FAILFAST_INACTIVE);
15558                 un->un_failfast_bp = NULL;
15559         }
15560 
15561         if (bp == un->un_retry_bp) {
15562                 /*
15563                  * This command was retried one or more times. Show that we are
15564                  * done with it, and allow processing of the waitq to resume.
15565                  */
15566                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15567                     "sd_return_failed_command_no_restart: "
15568                     " un:0x%p: RETURNING retry_bp:0x%p\n", un, un->un_retry_bp);
15569                 un->un_retry_bp = NULL;
15570                 un->un_retry_statp = NULL;
15571         }
15572 
15573         SD_UPDATE_RDWR_STATS(un, bp);
15574         SD_UPDATE_PARTITION_STATS(un, bp);
15575 
15576         mutex_exit(SD_MUTEX(un));
15577 
15578         if (xp->xb_pktp != NULL) {
15579                 (*(sd_destroypkt_map[xp->xb_chain_iodone]))(bp);
15580                 xp->xb_pktp = NULL;
15581         }
15582 
15583         SD_BEGIN_IODONE(xp->xb_chain_iodone, un, bp);
15584 
15585         mutex_enter(SD_MUTEX(un));
15586 
15587         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15588             "sd_return_failed_command_no_restart: exit\n");
15589 }
15590 
15591 
15592 /*
15593  *    Function: sd_retry_command
15594  *
15595  * Description: queue up a command for retry, or (optionally) fail it
15596  *              if retry counts are exhausted.
15597  *
15598  *   Arguments: un - Pointer to the sd_lun struct for the target.
15599  *
15600  *              bp - Pointer to the buf for the command to be retried.
15601  *
15602  *              retry_check_flag - Flag to see which (if any) of the retry
15603  *                 counts should be decremented/checked. If the indicated
15604  *                 retry count is exhausted, then the command will not be
15605  *                 retried; it will be failed instead. This should use a
15606  *                 value equal to one of the following:
15607  *
15608  *                      SD_RETRIES_NOCHECK
15609  *                      SD_RESD_RETRIES_STANDARD
15610  *                      SD_RETRIES_VICTIM
15611  *
15612  *                 Optionally may be bitwise-OR'ed with SD_RETRIES_ISOLATE
15613  *                 if the check should be made to see of FLAG_ISOLATE is set
15614  *                 in the pkt. If FLAG_ISOLATE is set, then the command is
15615  *                 not retried, it is simply failed.
15616  *
15617  *              user_funcp - Ptr to function to call before dispatching the
15618  *                 command. May be NULL if no action needs to be performed.
15619  *                 (Primarily intended for printing messages.)
15620  *
15621  *              user_arg - Optional argument to be passed along to
15622  *                 the user_funcp call.
15623  *
15624  *              failure_code - errno return code to set in the bp if the
15625  *                 command is going to be failed.
15626  *
15627  *              retry_delay - Retry delay interval in (clock_t) units. May
15628  *                 be zero which indicates that the retry should be retried
15629  *                 immediately (ie, without an intervening delay).
15630  *
15631  *              statp - Ptr to kstat function to be updated if the command
15632  *                 is queued for a delayed retry. May be NULL if no kstat
15633  *                 update is desired.
15634  *
15635  *     Context: May be called from interrupt context.
15636  */
15637 
15638 static void
15639 sd_retry_command(struct sd_lun *un, struct buf *bp, int retry_check_flag,
15640         void (*user_funcp)(struct sd_lun *un, struct buf *bp, void *argp, int
15641         code), void *user_arg, int failure_code,  clock_t retry_delay,
15642         void (*statp)(kstat_io_t *))
15643 {
15644         struct sd_xbuf  *xp;
15645         struct scsi_pkt *pktp;
15646         struct sd_fm_internal *sfip;
15647 
15648         ASSERT(un != NULL);
15649         ASSERT(mutex_owned(SD_MUTEX(un)));
15650         ASSERT(bp != NULL);
15651         xp = SD_GET_XBUF(bp);
15652         ASSERT(xp != NULL);
15653         pktp = SD_GET_PKTP(bp);
15654         ASSERT(pktp != NULL);
15655 
15656         sfip = (struct sd_fm_internal *)un->un_fm_private;
15657         ASSERT(sfip != NULL);
15658 
15659         SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un,
15660             "sd_retry_command: entry: bp:0x%p xp:0x%p\n", bp, xp);
15661 
15662         /*
15663          * If we are syncing or dumping, fail the command to avoid
15664          * recursively calling back into scsi_transport().
15665          */
15666         if (ddi_in_panic()) {
15667                 goto fail_command_no_log;
15668         }
15669 
15670         /*
15671          * We should never be be retrying a command with FLAG_DIAGNOSE set, so
15672          * log an error and fail the command.
15673          */
15674         if ((pktp->pkt_flags & FLAG_DIAGNOSE) != 0) {
15675                 scsi_log(SD_DEVINFO(un), sd_label, CE_NOTE,
15676                     "ERROR, retrying FLAG_DIAGNOSE command.\n");
15677                 sd_dump_memory(un, SD_LOG_IO, "CDB",
15678                     (uchar_t *)pktp->pkt_cdbp, CDB_SIZE, SD_LOG_HEX);
15679                 sd_dump_memory(un, SD_LOG_IO, "Sense Data",
15680                     (uchar_t *)xp->xb_sense_data, SENSE_LENGTH, SD_LOG_HEX);
15681                 goto fail_command;
15682         }
15683 
15684         /*
15685          * If we are suspended, then put the command onto head of the
15686          * wait queue since we don't want to start more commands, and
15687          * clear the un_retry_bp. Next time when we are resumed, will
15688          * handle the command in the wait queue.
15689          */
15690         switch (un->un_state) {
15691         case SD_STATE_SUSPENDED:
15692         case SD_STATE_DUMPING:
15693                 bp->av_forw = un->un_waitq_headp;
15694                 un->un_waitq_headp = bp;
15695                 if (un->un_waitq_tailp == NULL) {
15696                         un->un_waitq_tailp = bp;
15697                 }
15698                 if (bp == un->un_retry_bp) {
15699                         un->un_retry_bp = NULL;
15700                         un->un_retry_statp = NULL;
15701                 }
15702                 SD_UPDATE_KSTATS(un, kstat_waitq_enter, bp);
15703                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_retry_command: "
15704                     "exiting; cmd bp:0x%p requeued for SUSPEND/DUMP\n", bp);
15705                 return;
15706         default:
15707                 break;
15708         }
15709 
15710         /*
15711          * If the caller wants us to check FLAG_ISOLATE, then see if that
15712          * is set; if it is then we do not want to retry the command.
15713          * Normally, FLAG_ISOLATE is only used with USCSI cmds.
15714          */
15715         if ((retry_check_flag & SD_RETRIES_ISOLATE) != 0) {
15716                 if ((pktp->pkt_flags & FLAG_ISOLATE) != 0) {
15717                         goto fail_command;
15718                 }
15719         }
15720 
15721 
15722         /*
15723          * If SD_RETRIES_FAILFAST is set, it indicates that either a
15724          * command timeout or a selection timeout has occurred. This means
15725          * that we were unable to establish an kind of communication with
15726          * the target, and subsequent retries and/or commands are likely
15727          * to encounter similar results and take a long time to complete.
15728          *
15729          * If this is a failfast error condition, we need to update the
15730          * failfast state, even if this bp does not have B_FAILFAST set.
15731          */
15732         if (retry_check_flag & SD_RETRIES_FAILFAST) {
15733                 if (un->un_failfast_state == SD_FAILFAST_ACTIVE) {
15734                         ASSERT(un->un_failfast_bp == NULL);
15735                         /*
15736                          * If we are already in the active failfast state, and
15737                          * another failfast error condition has been detected,
15738                          * then fail this command if it has B_FAILFAST set.
15739                          * If B_FAILFAST is clear, then maintain the legacy
15740                          * behavior of retrying heroically, even tho this will
15741                          * take a lot more time to fail the command.
15742                          */
15743                         if (bp->b_flags & B_FAILFAST) {
15744                                 goto fail_command;
15745                         }
15746                 } else {
15747                         /*
15748                          * We're not in the active failfast state, but we
15749                          * have a failfast error condition, so we must begin
15750                          * transition to the next state. We do this regardless
15751                          * of whether or not this bp has B_FAILFAST set.
15752                          */
15753                         if (un->un_failfast_bp == NULL) {
15754                                 /*
15755                                  * This is the first bp to meet a failfast
15756                                  * condition so save it on un_failfast_bp &
15757                                  * do normal retry processing. Do not enter
15758                                  * active failfast state yet. This marks
15759                                  * entry into the "failfast pending" state.
15760                                  */
15761                                 un->un_failfast_bp = bp;
15762 
15763                         } else if (un->un_failfast_bp == bp) {
15764                                 /*
15765                                  * This is the second time *this* bp has
15766                                  * encountered a failfast error condition,
15767                                  * so enter active failfast state & flush
15768                                  * queues as appropriate.
15769                                  */
15770                                 un->un_failfast_state = SD_FAILFAST_ACTIVE;
15771                                 un->un_failfast_bp = NULL;
15772                                 sd_failfast_flushq(un);
15773 
15774                                 /*
15775                                  * Fail this bp now if B_FAILFAST set;
15776                                  * otherwise continue with retries. (It would
15777                                  * be pretty ironic if this bp succeeded on a
15778                                  * subsequent retry after we just flushed all
15779                                  * the queues).
15780                                  */
15781                                 if (bp->b_flags & B_FAILFAST) {
15782                                         goto fail_command;
15783                                 }
15784 
15785 #if !defined(lint) && !defined(__lint)
15786                         } else {
15787                                 /*
15788                                  * If neither of the preceeding conditionals
15789                                  * was true, it means that there is some
15790                                  * *other* bp that has met an inital failfast
15791                                  * condition and is currently either being
15792                                  * retried or is waiting to be retried. In
15793                                  * that case we should perform normal retry
15794                                  * processing on *this* bp, since there is a
15795                                  * chance that the current failfast condition
15796                                  * is transient and recoverable. If that does
15797                                  * not turn out to be the case, then retries
15798                                  * will be cleared when the wait queue is
15799                                  * flushed anyway.
15800                                  */
15801 #endif
15802                         }
15803                 }
15804         } else {
15805                 /*
15806                  * SD_RETRIES_FAILFAST is clear, which indicates that we
15807                  * likely were able to at least establish some level of
15808                  * communication with the target and subsequent commands
15809                  * and/or retries are likely to get through to the target,
15810                  * In this case we want to be aggressive about clearing
15811                  * the failfast state. Note that this does not affect
15812                  * the "failfast pending" condition.
15813                  */
15814                 un->un_failfast_state = SD_FAILFAST_INACTIVE;
15815         }
15816 
15817 
15818         /*
15819          * Check the specified retry count to see if we can still do
15820          * any retries with this pkt before we should fail it.
15821          */
15822         switch (retry_check_flag & SD_RETRIES_MASK) {
15823         case SD_RETRIES_VICTIM:
15824                 /*
15825                  * Check the victim retry count. If exhausted, then fall
15826                  * thru & check against the standard retry count.
15827                  */
15828                 if (xp->xb_victim_retry_count < un->un_victim_retry_count) {
15829                         /* Increment count & proceed with the retry */
15830                         xp->xb_victim_retry_count++;
15831                         break;
15832                 }
15833                 /* Victim retries exhausted, fall back to std. retries... */
15834                 /* FALLTHRU */
15835 
15836         case SD_RETRIES_STANDARD:
15837                 if (xp->xb_retry_count >= un->un_retry_count) {
15838                         /* Retries exhausted, fail the command */
15839                         SD_TRACE(SD_LOG_IO_CORE, un,
15840                             "sd_retry_command: retries exhausted!\n");
15841                         /*
15842                          * update b_resid for failed SCMD_READ & SCMD_WRITE
15843                          * commands with nonzero pkt_resid.
15844                          */
15845                         if ((pktp->pkt_reason == CMD_CMPLT) &&
15846                             (SD_GET_PKT_STATUS(pktp) == STATUS_GOOD) &&
15847                             (pktp->pkt_resid != 0)) {
15848                                 uchar_t op = SD_GET_PKT_OPCODE(pktp) & 0x1F;
15849                                 if ((op == SCMD_READ) || (op == SCMD_WRITE)) {
15850                                         SD_UPDATE_B_RESID(bp, pktp);
15851                                 }
15852                         }
15853                         goto fail_command;
15854                 }
15855                 xp->xb_retry_count++;
15856                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15857                     "sd_retry_command: retry count:%d\n", xp->xb_retry_count);
15858                 break;
15859 
15860         case SD_RETRIES_UA:
15861                 if (xp->xb_ua_retry_count >= sd_ua_retry_count) {
15862                         /* Retries exhausted, fail the command */
15863                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
15864                             "Unit Attention retries exhausted. "
15865                             "Check the target.\n");
15866                         goto fail_command;
15867                 }
15868                 xp->xb_ua_retry_count++;
15869                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15870                     "sd_retry_command: retry count:%d\n",
15871                     xp->xb_ua_retry_count);
15872                 break;
15873 
15874         case SD_RETRIES_BUSY:
15875                 if (xp->xb_retry_count >= un->un_busy_retry_count) {
15876                         /* Retries exhausted, fail the command */
15877                         SD_TRACE(SD_LOG_IO_CORE, un,
15878                             "sd_retry_command: retries exhausted!\n");
15879                         goto fail_command;
15880                 }
15881                 xp->xb_retry_count++;
15882                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15883                     "sd_retry_command: retry count:%d\n", xp->xb_retry_count);
15884                 break;
15885 
15886         case SD_RETRIES_NOCHECK:
15887         default:
15888                 /* No retry count to check. Just proceed with the retry */
15889                 break;
15890         }
15891 
15892         xp->xb_pktp->pkt_flags |= FLAG_HEAD;
15893 
15894         /*
15895          * If this is a non-USCSI command being retried
15896          * during execution last time, we should post an ereport with
15897          * driver-assessment of the value "retry".
15898          * For partial DMA, request sense and STATUS_QFULL, there are no
15899          * hardware errors, we bypass ereport posting.
15900          */
15901         if (failure_code != 0) {
15902                 if (!(xp->xb_pkt_flags & SD_XB_USCSICMD)) {
15903                         sd_ssc_extract_info(&sfip->fm_ssc, un, pktp, bp, xp);
15904                         sd_ssc_post(&sfip->fm_ssc, SD_FM_DRV_RETRY);
15905                 }
15906         }
15907 
15908         /*
15909          * If we were given a zero timeout, we must attempt to retry the
15910          * command immediately (ie, without a delay).
15911          */
15912         if (retry_delay == 0) {
15913                 /*
15914                  * Check some limiting conditions to see if we can actually
15915                  * do the immediate retry.  If we cannot, then we must
15916                  * fall back to queueing up a delayed retry.
15917                  */
15918                 if (un->un_ncmds_in_transport >= un->un_throttle) {
15919                         /*
15920                          * We are at the throttle limit for the target,
15921                          * fall back to delayed retry.
15922                          */
15923                         retry_delay = un->un_busy_timeout;
15924                         statp = kstat_waitq_enter;
15925                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15926                             "sd_retry_command: immed. retry hit "
15927                             "throttle!\n");
15928                 } else {
15929                         /*
15930                          * We're clear to proceed with the immediate retry.
15931                          * First call the user-provided function (if any)
15932                          */
15933                         if (user_funcp != NULL) {
15934                                 (*user_funcp)(un, bp, user_arg,
15935                                     SD_IMMEDIATE_RETRY_ISSUED);
15936 #ifdef __lock_lint
15937                                 sd_print_incomplete_msg(un, bp, user_arg,
15938                                     SD_IMMEDIATE_RETRY_ISSUED);
15939                                 sd_print_cmd_incomplete_msg(un, bp, user_arg,
15940                                     SD_IMMEDIATE_RETRY_ISSUED);
15941                                 sd_print_sense_failed_msg(un, bp, user_arg,
15942                                     SD_IMMEDIATE_RETRY_ISSUED);
15943 #endif
15944                         }
15945 
15946                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15947                             "sd_retry_command: issuing immediate retry\n");
15948 
15949                         /*
15950                          * Call sd_start_cmds() to transport the command to
15951                          * the target.
15952                          */
15953                         sd_start_cmds(un, bp);
15954 
15955                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15956                             "sd_retry_command exit\n");
15957                         return;
15958                 }
15959         }
15960 
15961         /*
15962          * Set up to retry the command after a delay.
15963          * First call the user-provided function (if any)
15964          */
15965         if (user_funcp != NULL) {
15966                 (*user_funcp)(un, bp, user_arg, SD_DELAYED_RETRY_ISSUED);
15967         }
15968 
15969         sd_set_retry_bp(un, bp, retry_delay, statp);
15970 
15971         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_retry_command: exit\n");
15972         return;
15973 
15974 fail_command:
15975 
15976         if (user_funcp != NULL) {
15977                 (*user_funcp)(un, bp, user_arg, SD_NO_RETRY_ISSUED);
15978         }
15979 
15980 fail_command_no_log:
15981 
15982         SD_INFO(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
15983             "sd_retry_command: returning failed command\n");
15984 
15985         sd_return_failed_command(un, bp, failure_code);
15986 
15987         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_retry_command: exit\n");
15988 }
15989 
15990 
15991 /*
15992  *    Function: sd_set_retry_bp
15993  *
15994  * Description: Set up the given bp for retry.
15995  *
15996  *   Arguments: un - ptr to associated softstate
15997  *              bp - ptr to buf(9S) for the command
15998  *              retry_delay - time interval before issuing retry (may be 0)
15999  *              statp - optional pointer to kstat function
16000  *
16001  *     Context: May be called under interrupt context
16002  */
16003 
16004 static void
16005 sd_set_retry_bp(struct sd_lun *un, struct buf *bp, clock_t retry_delay,
16006         void (*statp)(kstat_io_t *))
16007 {
16008         ASSERT(un != NULL);
16009         ASSERT(mutex_owned(SD_MUTEX(un)));
16010         ASSERT(bp != NULL);
16011 
16012         SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un,
16013             "sd_set_retry_bp: entry: un:0x%p bp:0x%p\n", un, bp);
16014 
16015         /*
16016          * Indicate that the command is being retried. This will not allow any
16017          * other commands on the wait queue to be transported to the target
16018          * until this command has been completed (success or failure). The
16019          * "retry command" is not transported to the target until the given
16020          * time delay expires, unless the user specified a 0 retry_delay.
16021          *
16022          * Note: the timeout(9F) callback routine is what actually calls
16023          * sd_start_cmds() to transport the command, with the exception of a
16024          * zero retry_delay. The only current implementor of a zero retry delay
16025          * is the case where a START_STOP_UNIT is sent to spin-up a device.
16026          */
16027         if (un->un_retry_bp == NULL) {
16028                 ASSERT(un->un_retry_statp == NULL);
16029                 un->un_retry_bp = bp;
16030 
16031                 /*
16032                  * If the user has not specified a delay the command should
16033                  * be queued and no timeout should be scheduled.
16034                  */
16035                 if (retry_delay == 0) {
16036                         /*
16037                          * Save the kstat pointer that will be used in the
16038                          * call to SD_UPDATE_KSTATS() below, so that
16039                          * sd_start_cmds() can correctly decrement the waitq
16040                          * count when it is time to transport this command.
16041                          */
16042                         un->un_retry_statp = statp;
16043                         goto done;
16044                 }
16045         }
16046 
16047         if (un->un_retry_bp == bp) {
16048                 /*
16049                  * Save the kstat pointer that will be used in the call to
16050                  * SD_UPDATE_KSTATS() below, so that sd_start_cmds() can
16051                  * correctly decrement the waitq count when it is time to
16052                  * transport this command.
16053                  */
16054                 un->un_retry_statp = statp;
16055 
16056                 /*
16057                  * Schedule a timeout if:
16058                  *   1) The user has specified a delay.
16059                  *   2) There is not a START_STOP_UNIT callback pending.
16060                  *
16061                  * If no delay has been specified, then it is up to the caller
16062                  * to ensure that IO processing continues without stalling.
16063                  * Effectively, this means that the caller will issue the
16064                  * required call to sd_start_cmds(). The START_STOP_UNIT
16065                  * callback does this after the START STOP UNIT command has
16066                  * completed. In either of these cases we should not schedule
16067                  * a timeout callback here.  Also don't schedule the timeout if
16068                  * an SD_PATH_DIRECT_PRIORITY command is waiting to restart.
16069                  */
16070                 if ((retry_delay != 0) && (un->un_startstop_timeid == NULL) &&
16071                     (un->un_direct_priority_timeid == NULL)) {
16072                         un->un_retry_timeid =
16073                             timeout(sd_start_retry_command, un, retry_delay);
16074                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16075                             "sd_set_retry_bp: setting timeout: un: 0x%p"
16076                             " bp:0x%p un_retry_timeid:0x%p\n",
16077                             un, bp, un->un_retry_timeid);
16078                 }
16079         } else {
16080                 /*
16081                  * We only get in here if there is already another command
16082                  * waiting to be retried.  In this case, we just put the
16083                  * given command onto the wait queue, so it can be transported
16084                  * after the current retry command has completed.
16085                  *
16086                  * Also we have to make sure that if the command at the head
16087                  * of the wait queue is the un_failfast_bp, that we do not
16088                  * put ahead of it any other commands that are to be retried.
16089                  */
16090                 if ((un->un_failfast_bp != NULL) &&
16091                     (un->un_failfast_bp == un->un_waitq_headp)) {
16092                         /*
16093                          * Enqueue this command AFTER the first command on
16094                          * the wait queue (which is also un_failfast_bp).
16095                          */
16096                         bp->av_forw = un->un_waitq_headp->av_forw;
16097                         un->un_waitq_headp->av_forw = bp;
16098                         if (un->un_waitq_headp == un->un_waitq_tailp) {
16099                                 un->un_waitq_tailp = bp;
16100                         }
16101                 } else {
16102                         /* Enqueue this command at the head of the waitq. */
16103                         bp->av_forw = un->un_waitq_headp;
16104                         un->un_waitq_headp = bp;
16105                         if (un->un_waitq_tailp == NULL) {
16106                                 un->un_waitq_tailp = bp;
16107                         }
16108                 }
16109 
16110                 if (statp == NULL) {
16111                         statp = kstat_waitq_enter;
16112                 }
16113                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16114                     "sd_set_retry_bp: un:0x%p already delayed retry\n", un);
16115         }
16116 
16117 done:
16118         if (statp != NULL) {
16119                 SD_UPDATE_KSTATS(un, statp, bp);
16120         }
16121 
16122         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16123             "sd_set_retry_bp: exit un:0x%p\n", un);
16124 }
16125 
16126 
16127 /*
16128  *    Function: sd_start_retry_command
16129  *
16130  * Description: Start the command that has been waiting on the target's
16131  *              retry queue.  Called from timeout(9F) context after the
16132  *              retry delay interval has expired.
16133  *
16134  *   Arguments: arg - pointer to associated softstate for the device.
16135  *
16136  *     Context: timeout(9F) thread context.  May not sleep.
16137  */
16138 
16139 static void
16140 sd_start_retry_command(void *arg)
16141 {
16142         struct sd_lun *un = arg;
16143 
16144         ASSERT(un != NULL);
16145         ASSERT(!mutex_owned(SD_MUTEX(un)));
16146 
16147         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16148             "sd_start_retry_command: entry\n");
16149 
16150         mutex_enter(SD_MUTEX(un));
16151 
16152         un->un_retry_timeid = NULL;
16153 
16154         if (un->un_retry_bp != NULL) {
16155                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16156                     "sd_start_retry_command: un:0x%p STARTING bp:0x%p\n",
16157                     un, un->un_retry_bp);
16158                 sd_start_cmds(un, un->un_retry_bp);
16159         }
16160 
16161         mutex_exit(SD_MUTEX(un));
16162 
16163         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16164             "sd_start_retry_command: exit\n");
16165 }
16166 
16167 /*
16168  *    Function: sd_rmw_msg_print_handler
16169  *
16170  * Description: If RMW mode is enabled and warning message is triggered
16171  *              print I/O count during a fixed interval.
16172  *
16173  *   Arguments: arg - pointer to associated softstate for the device.
16174  *
16175  *     Context: timeout(9F) thread context. May not sleep.
16176  */
16177 static void
16178 sd_rmw_msg_print_handler(void *arg)
16179 {
16180         struct sd_lun *un = arg;
16181 
16182         ASSERT(un != NULL);
16183         ASSERT(!mutex_owned(SD_MUTEX(un)));
16184 
16185         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16186             "sd_rmw_msg_print_handler: entry\n");
16187 
16188         mutex_enter(SD_MUTEX(un));
16189 
16190         if (un->un_rmw_incre_count > 0) {
16191                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
16192                     "%"PRIu64" I/O requests are not aligned with %d disk "
16193                     "sector size in %ld seconds. They are handled through "
16194                     "Read Modify Write but the performance is very low!\n",
16195                     un->un_rmw_incre_count, un->un_tgt_blocksize,
16196                     drv_hztousec(SD_RMW_MSG_PRINT_TIMEOUT) / 1000000);
16197                 un->un_rmw_incre_count = 0;
16198                 un->un_rmw_msg_timeid = timeout(sd_rmw_msg_print_handler,
16199                     un, SD_RMW_MSG_PRINT_TIMEOUT);
16200         } else {
16201                 un->un_rmw_msg_timeid = NULL;
16202         }
16203 
16204         mutex_exit(SD_MUTEX(un));
16205 
16206         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16207             "sd_rmw_msg_print_handler: exit\n");
16208 }
16209 
16210 /*
16211  *    Function: sd_start_direct_priority_command
16212  *
16213  * Description: Used to re-start an SD_PATH_DIRECT_PRIORITY command that had
16214  *              received TRAN_BUSY when we called scsi_transport() to send it
16215  *              to the underlying HBA. This function is called from timeout(9F)
16216  *              context after the delay interval has expired.
16217  *
16218  *   Arguments: arg - pointer to associated buf(9S) to be restarted.
16219  *
16220  *     Context: timeout(9F) thread context.  May not sleep.
16221  */
16222 
16223 static void
16224 sd_start_direct_priority_command(void *arg)
16225 {
16226         struct buf      *priority_bp = arg;
16227         struct sd_lun   *un;
16228 
16229         ASSERT(priority_bp != NULL);
16230         un = SD_GET_UN(priority_bp);
16231         ASSERT(un != NULL);
16232         ASSERT(!mutex_owned(SD_MUTEX(un)));
16233 
16234         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16235             "sd_start_direct_priority_command: entry\n");
16236 
16237         mutex_enter(SD_MUTEX(un));
16238         un->un_direct_priority_timeid = NULL;
16239         sd_start_cmds(un, priority_bp);
16240         mutex_exit(SD_MUTEX(un));
16241 
16242         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16243             "sd_start_direct_priority_command: exit\n");
16244 }
16245 
16246 
16247 /*
16248  *    Function: sd_send_request_sense_command
16249  *
16250  * Description: Sends a REQUEST SENSE command to the target
16251  *
16252  *     Context: May be called from interrupt context.
16253  */
16254 
16255 static void
16256 sd_send_request_sense_command(struct sd_lun *un, struct buf *bp,
16257         struct scsi_pkt *pktp)
16258 {
16259         ASSERT(bp != NULL);
16260         ASSERT(un != NULL);
16261         ASSERT(mutex_owned(SD_MUTEX(un)));
16262 
16263         SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un, "sd_send_request_sense_command: "
16264             "entry: buf:0x%p\n", bp);
16265 
16266         /*
16267          * If we are syncing or dumping, then fail the command to avoid a
16268          * recursive callback into scsi_transport(). Also fail the command
16269          * if we are suspended (legacy behavior).
16270          */
16271         if (ddi_in_panic() || (un->un_state == SD_STATE_SUSPENDED) ||
16272             (un->un_state == SD_STATE_DUMPING)) {
16273                 sd_return_failed_command(un, bp, EIO);
16274                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16275                     "sd_send_request_sense_command: syncing/dumping, exit\n");
16276                 return;
16277         }
16278 
16279         /*
16280          * Retry the failed command and don't issue the request sense if:
16281          *    1) the sense buf is busy
16282          *    2) we have 1 or more outstanding commands on the target
16283          *    (the sense data will be cleared or invalidated any way)
16284          *
16285          * Note: There could be an issue with not checking a retry limit here,
16286          * the problem is determining which retry limit to check.
16287          */
16288         if ((un->un_sense_isbusy != 0) || (un->un_ncmds_in_transport > 0)) {
16289                 /* Don't retry if the command is flagged as non-retryable */
16290                 if ((pktp->pkt_flags & FLAG_DIAGNOSE) == 0) {
16291                         sd_retry_command(un, bp, SD_RETRIES_NOCHECK,
16292                             NULL, NULL, 0, un->un_busy_timeout,
16293                             kstat_waitq_enter);
16294                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16295                             "sd_send_request_sense_command: "
16296                             "at full throttle, retrying exit\n");
16297                 } else {
16298                         sd_return_failed_command(un, bp, EIO);
16299                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16300                             "sd_send_request_sense_command: "
16301                             "at full throttle, non-retryable exit\n");
16302                 }
16303                 return;
16304         }
16305 
16306         sd_mark_rqs_busy(un, bp);
16307         sd_start_cmds(un, un->un_rqs_bp);
16308 
16309         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16310             "sd_send_request_sense_command: exit\n");
16311 }
16312 
16313 
16314 /*
16315  *    Function: sd_mark_rqs_busy
16316  *
16317  * Description: Indicate that the request sense bp for this instance is
16318  *              in use.
16319  *
16320  *     Context: May be called under interrupt context
16321  */
16322 
16323 static void
16324 sd_mark_rqs_busy(struct sd_lun *un, struct buf *bp)
16325 {
16326         struct sd_xbuf  *sense_xp;
16327 
16328         ASSERT(un != NULL);
16329         ASSERT(bp != NULL);
16330         ASSERT(mutex_owned(SD_MUTEX(un)));
16331         ASSERT(un->un_sense_isbusy == 0);
16332 
16333         SD_TRACE(SD_LOG_IO_CORE, un, "sd_mark_rqs_busy: entry: "
16334             "buf:0x%p xp:0x%p un:0x%p\n", bp, SD_GET_XBUF(bp), un);
16335 
16336         sense_xp = SD_GET_XBUF(un->un_rqs_bp);
16337         ASSERT(sense_xp != NULL);
16338 
16339         SD_INFO(SD_LOG_IO, un,
16340             "sd_mark_rqs_busy: entry: sense_xp:0x%p\n", sense_xp);
16341 
16342         ASSERT(sense_xp->xb_pktp != NULL);
16343         ASSERT((sense_xp->xb_pktp->pkt_flags & (FLAG_SENSING | FLAG_HEAD))
16344             == (FLAG_SENSING | FLAG_HEAD));
16345 
16346         un->un_sense_isbusy = 1;
16347         un->un_rqs_bp->b_resid = 0;
16348         sense_xp->xb_pktp->pkt_resid  = 0;
16349         sense_xp->xb_pktp->pkt_reason = 0;
16350 
16351         /* So we can get back the bp at interrupt time! */
16352         sense_xp->xb_sense_bp = bp;
16353 
16354         bzero(un->un_rqs_bp->b_un.b_addr, SENSE_LENGTH);
16355 
16356         /*
16357          * Mark this buf as awaiting sense data. (This is already set in
16358          * the pkt_flags for the RQS packet.)
16359          */
16360         ((SD_GET_XBUF(bp))->xb_pktp)->pkt_flags |= FLAG_SENSING;
16361 
16362         /* Request sense down same path */
16363         if (scsi_pkt_allocated_correctly((SD_GET_XBUF(bp))->xb_pktp) &&
16364             ((SD_GET_XBUF(bp))->xb_pktp)->pkt_path_instance)
16365                 sense_xp->xb_pktp->pkt_path_instance =
16366                     ((SD_GET_XBUF(bp))->xb_pktp)->pkt_path_instance;
16367 
16368         sense_xp->xb_retry_count     = 0;
16369         sense_xp->xb_victim_retry_count = 0;
16370         sense_xp->xb_ua_retry_count  = 0;
16371         sense_xp->xb_nr_retry_count  = 0;
16372         sense_xp->xb_dma_resid  = 0;
16373 
16374         /* Clean up the fields for auto-request sense */
16375         sense_xp->xb_sense_status = 0;
16376         sense_xp->xb_sense_state  = 0;
16377         sense_xp->xb_sense_resid  = 0;
16378         bzero(sense_xp->xb_sense_data, sizeof (sense_xp->xb_sense_data));
16379 
16380         SD_TRACE(SD_LOG_IO_CORE, un, "sd_mark_rqs_busy: exit\n");
16381 }
16382 
16383 
16384 /*
16385  *    Function: sd_mark_rqs_idle
16386  *
16387  * Description: SD_MUTEX must be held continuously through this routine
16388  *              to prevent reuse of the rqs struct before the caller can
16389  *              complete it's processing.
16390  *
16391  * Return Code: Pointer to the RQS buf
16392  *
16393  *     Context: May be called under interrupt context
16394  */
16395 
16396 static struct buf *
16397 sd_mark_rqs_idle(struct sd_lun *un, struct sd_xbuf *sense_xp)
16398 {
16399         struct buf *bp;
16400         ASSERT(un != NULL);
16401         ASSERT(sense_xp != NULL);
16402         ASSERT(mutex_owned(SD_MUTEX(un)));
16403         ASSERT(un->un_sense_isbusy != 0);
16404 
16405         un->un_sense_isbusy = 0;
16406         bp = sense_xp->xb_sense_bp;
16407         sense_xp->xb_sense_bp = NULL;
16408 
16409         /* This pkt is no longer interested in getting sense data */
16410         ((SD_GET_XBUF(bp))->xb_pktp)->pkt_flags &= ~FLAG_SENSING;
16411 
16412         return (bp);
16413 }
16414 
16415 
16416 
16417 /*
16418  *    Function: sd_alloc_rqs
16419  *
16420  * Description: Set up the unit to receive auto request sense data
16421  *
16422  * Return Code: DDI_SUCCESS or DDI_FAILURE
16423  *
16424  *     Context: Called under attach(9E) context
16425  */
16426 
16427 static int
16428 sd_alloc_rqs(struct scsi_device *devp, struct sd_lun *un)
16429 {
16430         struct sd_xbuf *xp;
16431 
16432         ASSERT(un != NULL);
16433         ASSERT(!mutex_owned(SD_MUTEX(un)));
16434         ASSERT(un->un_rqs_bp == NULL);
16435         ASSERT(un->un_rqs_pktp == NULL);
16436 
16437         /*
16438          * First allocate the required buf and scsi_pkt structs, then set up
16439          * the CDB in the scsi_pkt for a REQUEST SENSE command.
16440          */
16441         un->un_rqs_bp = scsi_alloc_consistent_buf(&devp->sd_address, NULL,
16442             MAX_SENSE_LENGTH, B_READ, SLEEP_FUNC, NULL);
16443         if (un->un_rqs_bp == NULL) {
16444                 return (DDI_FAILURE);
16445         }
16446 
16447         un->un_rqs_pktp = scsi_init_pkt(&devp->sd_address, NULL, un->un_rqs_bp,
16448             CDB_GROUP0, 1, 0, PKT_CONSISTENT, SLEEP_FUNC, NULL);
16449 
16450         if (un->un_rqs_pktp == NULL) {
16451                 sd_free_rqs(un);
16452                 return (DDI_FAILURE);
16453         }
16454 
16455         /* Set up the CDB in the scsi_pkt for a REQUEST SENSE command. */
16456         (void) scsi_setup_cdb((union scsi_cdb *)un->un_rqs_pktp->pkt_cdbp,
16457             SCMD_REQUEST_SENSE, 0, MAX_SENSE_LENGTH, 0);
16458 
16459         SD_FILL_SCSI1_LUN(un, un->un_rqs_pktp);
16460 
16461         /* Set up the other needed members in the ARQ scsi_pkt. */
16462         un->un_rqs_pktp->pkt_comp   = sdintr;
16463         un->un_rqs_pktp->pkt_time   = sd_io_time;
16464         un->un_rqs_pktp->pkt_flags |=
16465             (FLAG_SENSING | FLAG_HEAD); /* (1222170) */
16466 
16467         /*
16468          * Allocate  & init the sd_xbuf struct for the RQS command. Do not
16469          * provide any intpkt, destroypkt routines as we take care of
16470          * scsi_pkt allocation/freeing here and in sd_free_rqs().
16471          */
16472         xp = kmem_alloc(sizeof (struct sd_xbuf), KM_SLEEP);
16473         sd_xbuf_init(un, un->un_rqs_bp, xp, SD_CHAIN_NULL, NULL);
16474         xp->xb_pktp = un->un_rqs_pktp;
16475         SD_INFO(SD_LOG_ATTACH_DETACH, un,
16476             "sd_alloc_rqs: un 0x%p, rqs  xp 0x%p,  pkt 0x%p,  buf 0x%p\n",
16477             un, xp, un->un_rqs_pktp, un->un_rqs_bp);
16478 
16479         /*
16480          * Save the pointer to the request sense private bp so it can
16481          * be retrieved in sdintr.
16482          */
16483         un->un_rqs_pktp->pkt_private = un->un_rqs_bp;
16484         ASSERT(un->un_rqs_bp->b_private == xp);
16485 
16486         /*
16487          * See if the HBA supports auto-request sense for the specified
16488          * target/lun. If it does, then try to enable it (if not already
16489          * enabled).
16490          *
16491          * Note: For some HBAs (ifp & sf), scsi_ifsetcap will always return
16492          * failure, while for other HBAs (pln) scsi_ifsetcap will always
16493          * return success.  However, in both of these cases ARQ is always
16494          * enabled and scsi_ifgetcap will always return true. The best approach
16495          * is to issue the scsi_ifgetcap() first, then try the scsi_ifsetcap().
16496          *
16497          * The 3rd case is the HBA (adp) always return enabled on
16498          * scsi_ifgetgetcap even when it's not enable, the best approach
16499          * is issue a scsi_ifsetcap then a scsi_ifgetcap
16500          * Note: this case is to circumvent the Adaptec bug. (x86 only)
16501          */
16502 
16503         if (un->un_f_is_fibre == TRUE) {
16504                 un->un_f_arq_enabled = TRUE;
16505         } else {
16506 #if defined(__i386) || defined(__amd64)
16507                 /*
16508                  * Circumvent the Adaptec bug, remove this code when
16509                  * the bug is fixed
16510                  */
16511                 (void) scsi_ifsetcap(SD_ADDRESS(un), "auto-rqsense", 1, 1);
16512 #endif
16513                 switch (scsi_ifgetcap(SD_ADDRESS(un), "auto-rqsense", 1)) {
16514                 case 0:
16515                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
16516                             "sd_alloc_rqs: HBA supports ARQ\n");
16517                         /*
16518                          * ARQ is supported by this HBA but currently is not
16519                          * enabled. Attempt to enable it and if successful then
16520                          * mark this instance as ARQ enabled.
16521                          */
16522                         if (scsi_ifsetcap(SD_ADDRESS(un), "auto-rqsense", 1, 1)
16523                             == 1) {
16524                                 /* Successfully enabled ARQ in the HBA */
16525                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
16526                                     "sd_alloc_rqs: ARQ enabled\n");
16527                                 un->un_f_arq_enabled = TRUE;
16528                         } else {
16529                                 /* Could not enable ARQ in the HBA */
16530                                 SD_INFO(SD_LOG_ATTACH_DETACH, un,
16531                                     "sd_alloc_rqs: failed ARQ enable\n");
16532                                 un->un_f_arq_enabled = FALSE;
16533                         }
16534                         break;
16535                 case 1:
16536                         /*
16537                          * ARQ is supported by this HBA and is already enabled.
16538                          * Just mark ARQ as enabled for this instance.
16539                          */
16540                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
16541                             "sd_alloc_rqs: ARQ already enabled\n");
16542                         un->un_f_arq_enabled = TRUE;
16543                         break;
16544                 default:
16545                         /*
16546                          * ARQ is not supported by this HBA; disable it for this
16547                          * instance.
16548                          */
16549                         SD_INFO(SD_LOG_ATTACH_DETACH, un,
16550                             "sd_alloc_rqs: HBA does not support ARQ\n");
16551                         un->un_f_arq_enabled = FALSE;
16552                         break;
16553                 }
16554         }
16555 
16556         return (DDI_SUCCESS);
16557 }
16558 
16559 
16560 /*
16561  *    Function: sd_free_rqs
16562  *
16563  * Description: Cleanup for the pre-instance RQS command.
16564  *
16565  *     Context: Kernel thread context
16566  */
16567 
16568 static void
16569 sd_free_rqs(struct sd_lun *un)
16570 {
16571         ASSERT(un != NULL);
16572 
16573         SD_TRACE(SD_LOG_IO_CORE, un, "sd_free_rqs: entry\n");
16574 
16575         /*
16576          * If consistent memory is bound to a scsi_pkt, the pkt
16577          * has to be destroyed *before* freeing the consistent memory.
16578          * Don't change the sequence of this operations.
16579          * scsi_destroy_pkt() might access memory, which isn't allowed,
16580          * after it was freed in scsi_free_consistent_buf().
16581          */
16582         if (un->un_rqs_pktp != NULL) {
16583                 scsi_destroy_pkt(un->un_rqs_pktp);
16584                 un->un_rqs_pktp = NULL;
16585         }
16586 
16587         if (un->un_rqs_bp != NULL) {
16588                 struct sd_xbuf *xp = SD_GET_XBUF(un->un_rqs_bp);
16589                 if (xp != NULL) {
16590                         kmem_free(xp, sizeof (struct sd_xbuf));
16591                 }
16592                 scsi_free_consistent_buf(un->un_rqs_bp);
16593                 un->un_rqs_bp = NULL;
16594         }
16595         SD_TRACE(SD_LOG_IO_CORE, un, "sd_free_rqs: exit\n");
16596 }
16597 
16598 
16599 
16600 /*
16601  *    Function: sd_reduce_throttle
16602  *
16603  * Description: Reduces the maximum # of outstanding commands on a
16604  *              target to the current number of outstanding commands.
16605  *              Queues a tiemout(9F) callback to restore the limit
16606  *              after a specified interval has elapsed.
16607  *              Typically used when we get a TRAN_BUSY return code
16608  *              back from scsi_transport().
16609  *
16610  *   Arguments: un - ptr to the sd_lun softstate struct
16611  *              throttle_type: SD_THROTTLE_TRAN_BUSY or SD_THROTTLE_QFULL
16612  *
16613  *     Context: May be called from interrupt context
16614  */
16615 
16616 static void
16617 sd_reduce_throttle(struct sd_lun *un, int throttle_type)
16618 {
16619         ASSERT(un != NULL);
16620         ASSERT(mutex_owned(SD_MUTEX(un)));
16621         ASSERT(un->un_ncmds_in_transport >= 0);
16622 
16623         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_reduce_throttle: "
16624             "entry: un:0x%p un_throttle:%d un_ncmds_in_transport:%d\n",
16625             un, un->un_throttle, un->un_ncmds_in_transport);
16626 
16627         if (un->un_throttle > 1) {
16628                 if (un->un_f_use_adaptive_throttle == TRUE) {
16629                         switch (throttle_type) {
16630                         case SD_THROTTLE_TRAN_BUSY:
16631                                 if (un->un_busy_throttle == 0) {
16632                                         un->un_busy_throttle = un->un_throttle;
16633                                 }
16634                                 break;
16635                         case SD_THROTTLE_QFULL:
16636                                 un->un_busy_throttle = 0;
16637                                 break;
16638                         default:
16639                                 ASSERT(FALSE);
16640                         }
16641 
16642                         if (un->un_ncmds_in_transport > 0) {
16643                                 un->un_throttle = un->un_ncmds_in_transport;
16644                         }
16645 
16646                 } else {
16647                         if (un->un_ncmds_in_transport == 0) {
16648                                 un->un_throttle = 1;
16649                         } else {
16650                                 un->un_throttle = un->un_ncmds_in_transport;
16651                         }
16652                 }
16653         }
16654 
16655         /* Reschedule the timeout if none is currently active */
16656         if (un->un_reset_throttle_timeid == NULL) {
16657                 un->un_reset_throttle_timeid = timeout(sd_restore_throttle,
16658                     un, SD_THROTTLE_RESET_INTERVAL);
16659                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16660                     "sd_reduce_throttle: timeout scheduled!\n");
16661         }
16662 
16663         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_reduce_throttle: "
16664             "exit: un:0x%p un_throttle:%d\n", un, un->un_throttle);
16665 }
16666 
16667 
16668 
16669 /*
16670  *    Function: sd_restore_throttle
16671  *
16672  * Description: Callback function for timeout(9F).  Resets the current
16673  *              value of un->un_throttle to its default.
16674  *
16675  *   Arguments: arg - pointer to associated softstate for the device.
16676  *
16677  *     Context: May be called from interrupt context
16678  */
16679 
16680 static void
16681 sd_restore_throttle(void *arg)
16682 {
16683         struct sd_lun   *un = arg;
16684 
16685         ASSERT(un != NULL);
16686         ASSERT(!mutex_owned(SD_MUTEX(un)));
16687 
16688         mutex_enter(SD_MUTEX(un));
16689 
16690         SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un, "sd_restore_throttle: "
16691             "entry: un:0x%p un_throttle:%d\n", un, un->un_throttle);
16692 
16693         un->un_reset_throttle_timeid = NULL;
16694 
16695         if (un->un_f_use_adaptive_throttle == TRUE) {
16696                 /*
16697                  * If un_busy_throttle is nonzero, then it contains the
16698                  * value that un_throttle was when we got a TRAN_BUSY back
16699                  * from scsi_transport(). We want to revert back to this
16700                  * value.
16701                  *
16702                  * In the QFULL case, the throttle limit will incrementally
16703                  * increase until it reaches max throttle.
16704                  */
16705                 if (un->un_busy_throttle > 0) {
16706                         un->un_throttle = un->un_busy_throttle;
16707                         un->un_busy_throttle = 0;
16708                 } else {
16709                         /*
16710                          * increase throttle by 10% open gate slowly, schedule
16711                          * another restore if saved throttle has not been
16712                          * reached
16713                          */
16714                         short throttle;
16715                         if (sd_qfull_throttle_enable) {
16716                                 throttle = un->un_throttle +
16717                                     max((un->un_throttle / 10), 1);
16718                                 un->un_throttle =
16719                                     (throttle < un->un_saved_throttle) ?
16720                                     throttle : un->un_saved_throttle;
16721                                 if (un->un_throttle < un->un_saved_throttle) {
16722                                         un->un_reset_throttle_timeid =
16723                                             timeout(sd_restore_throttle,
16724                                             un,
16725                                             SD_QFULL_THROTTLE_RESET_INTERVAL);
16726                                 }
16727                         }
16728                 }
16729 
16730                 /*
16731                  * If un_throttle has fallen below the low-water mark, we
16732                  * restore the maximum value here (and allow it to ratchet
16733                  * down again if necessary).
16734                  */
16735                 if (un->un_throttle < un->un_min_throttle) {
16736                         un->un_throttle = un->un_saved_throttle;
16737                 }
16738         } else {
16739                 SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un, "sd_restore_throttle: "
16740                     "restoring limit from 0x%x to 0x%x\n",
16741                     un->un_throttle, un->un_saved_throttle);
16742                 un->un_throttle = un->un_saved_throttle;
16743         }
16744 
16745         SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un,
16746             "sd_restore_throttle: calling sd_start_cmds!\n");
16747 
16748         sd_start_cmds(un, NULL);
16749 
16750         SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un,
16751             "sd_restore_throttle: exit: un:0x%p un_throttle:%d\n",
16752             un, un->un_throttle);
16753 
16754         mutex_exit(SD_MUTEX(un));
16755 
16756         SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un, "sd_restore_throttle: exit\n");
16757 }
16758 
16759 /*
16760  *    Function: sdrunout
16761  *
16762  * Description: Callback routine for scsi_init_pkt when a resource allocation
16763  *              fails.
16764  *
16765  *   Arguments: arg - a pointer to the sd_lun unit struct for the particular
16766  *              soft state instance.
16767  *
16768  * Return Code: The scsi_init_pkt routine allows for the callback function to
16769  *              return a 0 indicating the callback should be rescheduled or a 1
16770  *              indicating not to reschedule. This routine always returns 1
16771  *              because the driver always provides a callback function to
16772  *              scsi_init_pkt. This results in a callback always being scheduled
16773  *              (via the scsi_init_pkt callback implementation) if a resource
16774  *              failure occurs.
16775  *
16776  *     Context: This callback function may not block or call routines that block
16777  *
16778  *        Note: Using the scsi_init_pkt callback facility can result in an I/O
16779  *              request persisting at the head of the list which cannot be
16780  *              satisfied even after multiple retries. In the future the driver
16781  *              may implement some time of maximum runout count before failing
16782  *              an I/O.
16783  */
16784 
16785 static int
16786 sdrunout(caddr_t arg)
16787 {
16788         struct sd_lun   *un = (struct sd_lun *)arg;
16789 
16790         ASSERT(un != NULL);
16791         ASSERT(!mutex_owned(SD_MUTEX(un)));
16792 
16793         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sdrunout: entry\n");
16794 
16795         mutex_enter(SD_MUTEX(un));
16796         sd_start_cmds(un, NULL);
16797         mutex_exit(SD_MUTEX(un));
16798         /*
16799          * This callback routine always returns 1 (i.e. do not reschedule)
16800          * because we always specify sdrunout as the callback handler for
16801          * scsi_init_pkt inside the call to sd_start_cmds.
16802          */
16803         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sdrunout: exit\n");
16804         return (1);
16805 }
16806 
16807 
16808 /*
16809  *    Function: sdintr
16810  *
16811  * Description: Completion callback routine for scsi_pkt(9S) structs
16812  *              sent to the HBA driver via scsi_transport(9F).
16813  *
16814  *     Context: Interrupt context
16815  */
16816 
16817 static void
16818 sdintr(struct scsi_pkt *pktp)
16819 {
16820         struct buf      *bp;
16821         struct sd_xbuf  *xp;
16822         struct sd_lun   *un;
16823         size_t          actual_len;
16824         sd_ssc_t        *sscp;
16825 
16826         ASSERT(pktp != NULL);
16827         bp = (struct buf *)pktp->pkt_private;
16828         ASSERT(bp != NULL);
16829         xp = SD_GET_XBUF(bp);
16830         ASSERT(xp != NULL);
16831         ASSERT(xp->xb_pktp != NULL);
16832         un = SD_GET_UN(bp);
16833         ASSERT(un != NULL);
16834         ASSERT(!mutex_owned(SD_MUTEX(un)));
16835 
16836 #ifdef SD_FAULT_INJECTION
16837 
16838         SD_INFO(SD_LOG_IOERR, un, "sdintr: sdintr calling Fault injection\n");
16839         /* SD FaultInjection */
16840         sd_faultinjection(pktp);
16841 
16842 #endif /* SD_FAULT_INJECTION */
16843 
16844         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sdintr: entry: buf:0x%p,"
16845             " xp:0x%p, un:0x%p\n", bp, xp, un);
16846 
16847         mutex_enter(SD_MUTEX(un));
16848 
16849         ASSERT(un->un_fm_private != NULL);
16850         sscp = &((struct sd_fm_internal *)(un->un_fm_private))->fm_ssc;
16851         ASSERT(sscp != NULL);
16852 
16853         /* Reduce the count of the #commands currently in transport */
16854         un->un_ncmds_in_transport--;
16855         ASSERT(un->un_ncmds_in_transport >= 0);
16856 
16857         /* Increment counter to indicate that the callback routine is active */
16858         un->un_in_callback++;
16859 
16860         SD_UPDATE_KSTATS(un, kstat_runq_exit, bp);
16861 
16862 #ifdef  SDDEBUG
16863         if (bp == un->un_retry_bp) {
16864                 SD_TRACE(SD_LOG_IO | SD_LOG_ERROR, un, "sdintr: "
16865                     "un:0x%p: GOT retry_bp:0x%p un_ncmds_in_transport:%d\n",
16866                     un, un->un_retry_bp, un->un_ncmds_in_transport);
16867         }
16868 #endif
16869 
16870         /*
16871          * If pkt_reason is CMD_DEV_GONE, fail the command, and update the media
16872          * state if needed.
16873          */
16874         if (pktp->pkt_reason == CMD_DEV_GONE) {
16875                 /* Prevent multiple console messages for the same failure. */
16876                 if (un->un_last_pkt_reason != CMD_DEV_GONE) {
16877                         un->un_last_pkt_reason = CMD_DEV_GONE;
16878                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
16879                             "Command failed to complete...Device is gone\n");
16880                 }
16881                 if (un->un_mediastate != DKIO_DEV_GONE) {
16882                         un->un_mediastate = DKIO_DEV_GONE;
16883                         cv_broadcast(&un->un_state_cv);
16884                 }
16885                 /*
16886                  * If the command happens to be the REQUEST SENSE command,
16887                  * free up the rqs buf and fail the original command.
16888                  */
16889                 if (bp == un->un_rqs_bp) {
16890                         bp = sd_mark_rqs_idle(un, xp);
16891                 }
16892                 sd_return_failed_command(un, bp, EIO);
16893                 goto exit;
16894         }
16895 
16896         if (pktp->pkt_state & STATE_XARQ_DONE) {
16897                 SD_TRACE(SD_LOG_COMMON, un,
16898                     "sdintr: extra sense data received. pkt=%p\n", pktp);
16899         }
16900 
16901         /*
16902          * First see if the pkt has auto-request sense data with it....
16903          * Look at the packet state first so we don't take a performance
16904          * hit looking at the arq enabled flag unless absolutely necessary.
16905          */
16906         if ((pktp->pkt_state & STATE_ARQ_DONE) &&
16907             (un->un_f_arq_enabled == TRUE)) {
16908                 /*
16909                  * The HBA did an auto request sense for this command so check
16910                  * for FLAG_DIAGNOSE. If set this indicates a uscsi or internal
16911                  * driver command that should not be retried.
16912                  */
16913                 if ((pktp->pkt_flags & FLAG_DIAGNOSE) != 0) {
16914                         /*
16915                          * Save the relevant sense info into the xp for the
16916                          * original cmd.
16917                          */
16918                         struct scsi_arq_status *asp;
16919                         asp = (struct scsi_arq_status *)(pktp->pkt_scbp);
16920                         xp->xb_sense_status =
16921                             *((uchar_t *)(&(asp->sts_rqpkt_status)));
16922                         xp->xb_sense_state  = asp->sts_rqpkt_state;
16923                         xp->xb_sense_resid  = asp->sts_rqpkt_resid;
16924                         if (pktp->pkt_state & STATE_XARQ_DONE) {
16925                                 actual_len = MAX_SENSE_LENGTH -
16926                                     xp->xb_sense_resid;
16927                                 bcopy(&asp->sts_sensedata, xp->xb_sense_data,
16928                                     MAX_SENSE_LENGTH);
16929                         } else {
16930                                 if (xp->xb_sense_resid > SENSE_LENGTH) {
16931                                         actual_len = MAX_SENSE_LENGTH -
16932                                             xp->xb_sense_resid;
16933                                 } else {
16934                                         actual_len = SENSE_LENGTH -
16935                                             xp->xb_sense_resid;
16936                                 }
16937                                 if (xp->xb_pkt_flags & SD_XB_USCSICMD) {
16938                                         if ((((struct uscsi_cmd *)
16939                                             (xp->xb_pktinfo))->uscsi_rqlen) >
16940                                             actual_len) {
16941                                                 xp->xb_sense_resid =
16942                                                     (((struct uscsi_cmd *)
16943                                                     (xp->xb_pktinfo))->
16944                                                     uscsi_rqlen) - actual_len;
16945                                         } else {
16946                                                 xp->xb_sense_resid = 0;
16947                                         }
16948                                 }
16949                                 bcopy(&asp->sts_sensedata, xp->xb_sense_data,
16950                                     SENSE_LENGTH);
16951                         }
16952 
16953                         /* fail the command */
16954                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16955                             "sdintr: arq done and FLAG_DIAGNOSE set\n");
16956                         sd_return_failed_command(un, bp, EIO);
16957                         goto exit;
16958                 }
16959 
16960 #if (defined(__i386) || defined(__amd64))       /* DMAFREE for x86 only */
16961                 /*
16962                  * We want to either retry or fail this command, so free
16963                  * the DMA resources here.  If we retry the command then
16964                  * the DMA resources will be reallocated in sd_start_cmds().
16965                  * Note that when PKT_DMA_PARTIAL is used, this reallocation
16966                  * causes the *entire* transfer to start over again from the
16967                  * beginning of the request, even for PARTIAL chunks that
16968                  * have already transferred successfully.
16969                  */
16970                 if ((un->un_f_is_fibre == TRUE) &&
16971                     ((xp->xb_pkt_flags & SD_XB_USCSICMD) == 0) &&
16972                     ((pktp->pkt_flags & FLAG_SENSING) == 0))  {
16973                         scsi_dmafree(pktp);
16974                         xp->xb_pkt_flags |= SD_XB_DMA_FREED;
16975                 }
16976 #endif
16977 
16978                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16979                     "sdintr: arq done, sd_handle_auto_request_sense\n");
16980 
16981                 sd_handle_auto_request_sense(un, bp, xp, pktp);
16982                 goto exit;
16983         }
16984 
16985         /* Next see if this is the REQUEST SENSE pkt for the instance */
16986         if (pktp->pkt_flags & FLAG_SENSING)  {
16987                 /* This pktp is from the unit's REQUEST_SENSE command */
16988                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
16989                     "sdintr: sd_handle_request_sense\n");
16990                 sd_handle_request_sense(un, bp, xp, pktp);
16991                 goto exit;
16992         }
16993 
16994         /*
16995          * Check to see if the command successfully completed as requested;
16996          * this is the most common case (and also the hot performance path).
16997          *
16998          * Requirements for successful completion are:
16999          * pkt_reason is CMD_CMPLT and packet status is status good.
17000          * In addition:
17001          * - A residual of zero indicates successful completion no matter what
17002          *   the command is.
17003          * - If the residual is not zero and the command is not a read or
17004          *   write, then it's still defined as successful completion. In other
17005          *   words, if the command is a read or write the residual must be
17006          *   zero for successful completion.
17007          * - If the residual is not zero and the command is a read or
17008          *   write, and it's a USCSICMD, then it's still defined as
17009          *   successful completion.
17010          */
17011         if ((pktp->pkt_reason == CMD_CMPLT) &&
17012             (SD_GET_PKT_STATUS(pktp) == STATUS_GOOD)) {
17013 
17014                 /*
17015                  * Since this command is returned with a good status, we
17016                  * can reset the count for Sonoma failover.
17017                  */
17018                 un->un_sonoma_failure_count = 0;
17019 
17020                 /*
17021                  * Return all USCSI commands on good status
17022                  */
17023                 if (pktp->pkt_resid == 0) {
17024                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17025                             "sdintr: returning command for resid == 0\n");
17026                 } else if (((SD_GET_PKT_OPCODE(pktp) & 0x1F) != SCMD_READ) &&
17027                     ((SD_GET_PKT_OPCODE(pktp) & 0x1F) != SCMD_WRITE)) {
17028                         SD_UPDATE_B_RESID(bp, pktp);
17029                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17030                             "sdintr: returning command for resid != 0\n");
17031                 } else if (xp->xb_pkt_flags & SD_XB_USCSICMD) {
17032                         SD_UPDATE_B_RESID(bp, pktp);
17033                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17034                             "sdintr: returning uscsi command\n");
17035                 } else {
17036                         goto not_successful;
17037                 }
17038                 sd_return_command(un, bp);
17039 
17040                 /*
17041                  * Decrement counter to indicate that the callback routine
17042                  * is done.
17043                  */
17044                 un->un_in_callback--;
17045                 ASSERT(un->un_in_callback >= 0);
17046                 mutex_exit(SD_MUTEX(un));
17047 
17048                 return;
17049         }
17050 
17051 not_successful:
17052 
17053 #if (defined(__i386) || defined(__amd64))       /* DMAFREE for x86 only */
17054         /*
17055          * The following is based upon knowledge of the underlying transport
17056          * and its use of DMA resources.  This code should be removed when
17057          * PKT_DMA_PARTIAL support is taken out of the disk driver in favor
17058          * of the new PKT_CMD_BREAKUP protocol. See also sd_initpkt_for_buf()
17059          * and sd_start_cmds().
17060          *
17061          * Free any DMA resources associated with this command if there
17062          * is a chance it could be retried or enqueued for later retry.
17063          * If we keep the DMA binding then mpxio cannot reissue the
17064          * command on another path whenever a path failure occurs.
17065          *
17066          * Note that when PKT_DMA_PARTIAL is used, free/reallocation
17067          * causes the *entire* transfer to start over again from the
17068          * beginning of the request, even for PARTIAL chunks that
17069          * have already transferred successfully.
17070          *
17071          * This is only done for non-uscsi commands (and also skipped for the
17072          * driver's internal RQS command). Also just do this for Fibre Channel
17073          * devices as these are the only ones that support mpxio.
17074          */
17075         if ((un->un_f_is_fibre == TRUE) &&
17076             ((xp->xb_pkt_flags & SD_XB_USCSICMD) == 0) &&
17077             ((pktp->pkt_flags & FLAG_SENSING) == 0))  {
17078                 scsi_dmafree(pktp);
17079                 xp->xb_pkt_flags |= SD_XB_DMA_FREED;
17080         }
17081 #endif
17082 
17083         /*
17084          * The command did not successfully complete as requested so check
17085          * for FLAG_DIAGNOSE. If set this indicates a uscsi or internal
17086          * driver command that should not be retried so just return. If
17087          * FLAG_DIAGNOSE is not set the error will be processed below.
17088          */
17089         if ((pktp->pkt_flags & FLAG_DIAGNOSE) != 0) {
17090                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17091                     "sdintr: FLAG_DIAGNOSE: sd_return_failed_command\n");
17092                 /*
17093                  * Issue a request sense if a check condition caused the error
17094                  * (we handle the auto request sense case above), otherwise
17095                  * just fail the command.
17096                  */
17097                 if ((pktp->pkt_reason == CMD_CMPLT) &&
17098                     (SD_GET_PKT_STATUS(pktp) == STATUS_CHECK)) {
17099                         sd_send_request_sense_command(un, bp, pktp);
17100                 } else {
17101                         sd_return_failed_command(un, bp, EIO);
17102                 }
17103                 goto exit;
17104         }
17105 
17106         /*
17107          * The command did not successfully complete as requested so process
17108          * the error, retry, and/or attempt recovery.
17109          */
17110         switch (pktp->pkt_reason) {
17111         case CMD_CMPLT:
17112                 switch (SD_GET_PKT_STATUS(pktp)) {
17113                 case STATUS_GOOD:
17114                         /*
17115                          * The command completed successfully with a non-zero
17116                          * residual
17117                          */
17118                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17119                             "sdintr: STATUS_GOOD \n");
17120                         sd_pkt_status_good(un, bp, xp, pktp);
17121                         break;
17122 
17123                 case STATUS_CHECK:
17124                 case STATUS_TERMINATED:
17125                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17126                             "sdintr: STATUS_TERMINATED | STATUS_CHECK\n");
17127                         sd_pkt_status_check_condition(un, bp, xp, pktp);
17128                         break;
17129 
17130                 case STATUS_BUSY:
17131                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17132                             "sdintr: STATUS_BUSY\n");
17133                         sd_pkt_status_busy(un, bp, xp, pktp);
17134                         break;
17135 
17136                 case STATUS_RESERVATION_CONFLICT:
17137                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17138                             "sdintr: STATUS_RESERVATION_CONFLICT\n");
17139                         sd_pkt_status_reservation_conflict(un, bp, xp, pktp);
17140                         break;
17141 
17142                 case STATUS_QFULL:
17143                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17144                             "sdintr: STATUS_QFULL\n");
17145                         sd_pkt_status_qfull(un, bp, xp, pktp);
17146                         break;
17147 
17148                 case STATUS_MET:
17149                 case STATUS_INTERMEDIATE:
17150                 case STATUS_SCSI2:
17151                 case STATUS_INTERMEDIATE_MET:
17152                 case STATUS_ACA_ACTIVE:
17153                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
17154                             "Unexpected SCSI status received: 0x%x\n",
17155                             SD_GET_PKT_STATUS(pktp));
17156                         /*
17157                          * Mark the ssc_flags when detected invalid status
17158                          * code for non-USCSI command.
17159                          */
17160                         if (!(xp->xb_pkt_flags & SD_XB_USCSICMD)) {
17161                                 sd_ssc_set_info(sscp, SSC_FLAGS_INVALID_STATUS,
17162                                     0, "stat-code");
17163                         }
17164                         sd_return_failed_command(un, bp, EIO);
17165                         break;
17166 
17167                 default:
17168                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
17169                             "Invalid SCSI status received: 0x%x\n",
17170                             SD_GET_PKT_STATUS(pktp));
17171                         if (!(xp->xb_pkt_flags & SD_XB_USCSICMD)) {
17172                                 sd_ssc_set_info(sscp, SSC_FLAGS_INVALID_STATUS,
17173                                     0, "stat-code");
17174                         }
17175                         sd_return_failed_command(un, bp, EIO);
17176                         break;
17177 
17178                 }
17179                 break;
17180 
17181         case CMD_INCOMPLETE:
17182                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17183                     "sdintr:  CMD_INCOMPLETE\n");
17184                 sd_pkt_reason_cmd_incomplete(un, bp, xp, pktp);
17185                 break;
17186         case CMD_TRAN_ERR:
17187                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17188                     "sdintr: CMD_TRAN_ERR\n");
17189                 sd_pkt_reason_cmd_tran_err(un, bp, xp, pktp);
17190                 break;
17191         case CMD_RESET:
17192                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17193                     "sdintr: CMD_RESET \n");
17194                 sd_pkt_reason_cmd_reset(un, bp, xp, pktp);
17195                 break;
17196         case CMD_ABORTED:
17197                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17198                     "sdintr: CMD_ABORTED \n");
17199                 sd_pkt_reason_cmd_aborted(un, bp, xp, pktp);
17200                 break;
17201         case CMD_TIMEOUT:
17202                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17203                     "sdintr: CMD_TIMEOUT\n");
17204                 sd_pkt_reason_cmd_timeout(un, bp, xp, pktp);
17205                 break;
17206         case CMD_UNX_BUS_FREE:
17207                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17208                     "sdintr: CMD_UNX_BUS_FREE \n");
17209                 sd_pkt_reason_cmd_unx_bus_free(un, bp, xp, pktp);
17210                 break;
17211         case CMD_TAG_REJECT:
17212                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17213                     "sdintr: CMD_TAG_REJECT\n");
17214                 sd_pkt_reason_cmd_tag_reject(un, bp, xp, pktp);
17215                 break;
17216         default:
17217                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
17218                     "sdintr: default\n");
17219                 /*
17220                  * Mark the ssc_flags for detecting invliad pkt_reason.
17221                  */
17222                 if (!(xp->xb_pkt_flags & SD_XB_USCSICMD)) {
17223                         sd_ssc_set_info(sscp, SSC_FLAGS_INVALID_PKT_REASON,
17224                             0, "pkt-reason");
17225                 }
17226                 sd_pkt_reason_default(un, bp, xp, pktp);
17227                 break;
17228         }
17229 
17230 exit:
17231         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sdintr: exit\n");
17232 
17233         /* Decrement counter to indicate that the callback routine is done. */
17234         un->un_in_callback--;
17235         ASSERT(un->un_in_callback >= 0);
17236 
17237         /*
17238          * At this point, the pkt has been dispatched, ie, it is either
17239          * being re-tried or has been returned to its caller and should
17240          * not be referenced.
17241          */
17242 
17243         mutex_exit(SD_MUTEX(un));
17244 }
17245 
17246 
17247 /*
17248  *    Function: sd_print_incomplete_msg
17249  *
17250  * Description: Prints the error message for a CMD_INCOMPLETE error.
17251  *
17252  *   Arguments: un - ptr to associated softstate for the device.
17253  *              bp - ptr to the buf(9S) for the command.
17254  *              arg - message string ptr
17255  *              code - SD_DELAYED_RETRY_ISSUED, SD_IMMEDIATE_RETRY_ISSUED,
17256  *                      or SD_NO_RETRY_ISSUED.
17257  *
17258  *     Context: May be called under interrupt context
17259  */
17260 
17261 static void
17262 sd_print_incomplete_msg(struct sd_lun *un, struct buf *bp, void *arg, int code)
17263 {
17264         struct scsi_pkt *pktp;
17265         char    *msgp;
17266         char    *cmdp = arg;
17267 
17268         ASSERT(un != NULL);
17269         ASSERT(mutex_owned(SD_MUTEX(un)));
17270         ASSERT(bp != NULL);
17271         ASSERT(arg != NULL);
17272         pktp = SD_GET_PKTP(bp);
17273         ASSERT(pktp != NULL);
17274 
17275         switch (code) {
17276         case SD_DELAYED_RETRY_ISSUED:
17277         case SD_IMMEDIATE_RETRY_ISSUED:
17278                 msgp = "retrying";
17279                 break;
17280         case SD_NO_RETRY_ISSUED:
17281         default:
17282                 msgp = "giving up";
17283                 break;
17284         }
17285 
17286         if ((pktp->pkt_flags & FLAG_SILENT) == 0) {
17287                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
17288                     "incomplete %s- %s\n", cmdp, msgp);
17289         }
17290 }
17291 
17292 
17293 
17294 /*
17295  *    Function: sd_pkt_status_good
17296  *
17297  * Description: Processing for a STATUS_GOOD code in pkt_status.
17298  *
17299  *     Context: May be called under interrupt context
17300  */
17301 
17302 static void
17303 sd_pkt_status_good(struct sd_lun *un, struct buf *bp,
17304         struct sd_xbuf *xp, struct scsi_pkt *pktp)
17305 {
17306         char    *cmdp;
17307 
17308         ASSERT(un != NULL);
17309         ASSERT(mutex_owned(SD_MUTEX(un)));
17310         ASSERT(bp != NULL);
17311         ASSERT(xp != NULL);
17312         ASSERT(pktp != NULL);
17313         ASSERT(pktp->pkt_reason == CMD_CMPLT);
17314         ASSERT(SD_GET_PKT_STATUS(pktp) == STATUS_GOOD);
17315         ASSERT(pktp->pkt_resid != 0);
17316 
17317         SD_TRACE(SD_LOG_IO_CORE, un, "sd_pkt_status_good: entry\n");
17318 
17319         SD_UPDATE_ERRSTATS(un, sd_harderrs);
17320         switch (SD_GET_PKT_OPCODE(pktp) & 0x1F) {
17321         case SCMD_READ:
17322                 cmdp = "read";
17323                 break;
17324         case SCMD_WRITE:
17325                 cmdp = "write";
17326                 break;
17327         default:
17328                 SD_UPDATE_B_RESID(bp, pktp);
17329                 sd_return_command(un, bp);
17330                 SD_TRACE(SD_LOG_IO_CORE, un, "sd_pkt_status_good: exit\n");
17331                 return;
17332         }
17333 
17334         /*
17335          * See if we can retry the read/write, preferrably immediately.
17336          * If retries are exhaused, then sd_retry_command() will update
17337          * the b_resid count.
17338          */
17339         sd_retry_command(un, bp, SD_RETRIES_STANDARD, sd_print_incomplete_msg,
17340             cmdp, EIO, (clock_t)0, NULL);
17341 
17342         SD_TRACE(SD_LOG_IO_CORE, un, "sd_pkt_status_good: exit\n");
17343 }
17344 
17345 
17346 
17347 
17348 
17349 /*
17350  *    Function: sd_handle_request_sense
17351  *
17352  * Description: Processing for non-auto Request Sense command.
17353  *
17354  *   Arguments: un - ptr to associated softstate
17355  *              sense_bp - ptr to buf(9S) for the RQS command
17356  *              sense_xp - ptr to the sd_xbuf for the RQS command
17357  *              sense_pktp - ptr to the scsi_pkt(9S) for the RQS command
17358  *
17359  *     Context: May be called under interrupt context
17360  */
17361 
17362 static void
17363 sd_handle_request_sense(struct sd_lun *un, struct buf *sense_bp,
17364         struct sd_xbuf *sense_xp, struct scsi_pkt *sense_pktp)
17365 {
17366         struct buf      *cmd_bp;        /* buf for the original command */
17367         struct sd_xbuf  *cmd_xp;        /* sd_xbuf for the original command */
17368         struct scsi_pkt *cmd_pktp;      /* pkt for the original command */
17369         size_t          actual_len;     /* actual sense data length */
17370 
17371         ASSERT(un != NULL);
17372         ASSERT(mutex_owned(SD_MUTEX(un)));
17373         ASSERT(sense_bp != NULL);
17374         ASSERT(sense_xp != NULL);
17375         ASSERT(sense_pktp != NULL);
17376 
17377         /*
17378          * Note the sense_bp, sense_xp, and sense_pktp here are for the
17379          * RQS command and not the original command.
17380          */
17381         ASSERT(sense_pktp == un->un_rqs_pktp);
17382         ASSERT(sense_bp   == un->un_rqs_bp);
17383         ASSERT((sense_pktp->pkt_flags & (FLAG_SENSING | FLAG_HEAD)) ==
17384             (FLAG_SENSING | FLAG_HEAD));
17385         ASSERT((((SD_GET_XBUF(sense_xp->xb_sense_bp))->xb_pktp->pkt_flags) &
17386             FLAG_SENSING) == FLAG_SENSING);
17387 
17388         /* These are the bp, xp, and pktp for the original command */
17389         cmd_bp = sense_xp->xb_sense_bp;
17390         cmd_xp = SD_GET_XBUF(cmd_bp);
17391         cmd_pktp = SD_GET_PKTP(cmd_bp);
17392 
17393         if (sense_pktp->pkt_reason != CMD_CMPLT) {
17394                 /*
17395                  * The REQUEST SENSE command failed.  Release the REQUEST
17396                  * SENSE command for re-use, get back the bp for the original
17397                  * command, and attempt to re-try the original command if
17398                  * FLAG_DIAGNOSE is not set in the original packet.
17399                  */
17400                 SD_UPDATE_ERRSTATS(un, sd_harderrs);
17401                 if ((cmd_pktp->pkt_flags & FLAG_DIAGNOSE) == 0) {
17402                         cmd_bp = sd_mark_rqs_idle(un, sense_xp);
17403                         sd_retry_command(un, cmd_bp, SD_RETRIES_STANDARD,
17404                             NULL, NULL, EIO, (clock_t)0, NULL);
17405                         return;
17406                 }
17407         }
17408 
17409         /*
17410          * Save the relevant sense info into the xp for the original cmd.
17411          *
17412          * Note: if the request sense failed the state info will be zero
17413          * as set in sd_mark_rqs_busy()
17414          */
17415         cmd_xp->xb_sense_status = *(sense_pktp->pkt_scbp);
17416         cmd_xp->xb_sense_state  = sense_pktp->pkt_state;
17417         actual_len = MAX_SENSE_LENGTH - sense_pktp->pkt_resid;
17418         if ((cmd_xp->xb_pkt_flags & SD_XB_USCSICMD) &&
17419             (((struct uscsi_cmd *)cmd_xp->xb_pktinfo)->uscsi_rqlen >
17420             SENSE_LENGTH)) {
17421                 bcopy(sense_bp->b_un.b_addr, cmd_xp->xb_sense_data,
17422                     MAX_SENSE_LENGTH);
17423                 cmd_xp->xb_sense_resid = sense_pktp->pkt_resid;
17424         } else {
17425                 bcopy(sense_bp->b_un.b_addr, cmd_xp->xb_sense_data,
17426                     SENSE_LENGTH);
17427                 if (actual_len < SENSE_LENGTH) {
17428                         cmd_xp->xb_sense_resid = SENSE_LENGTH - actual_len;
17429                 } else {
17430                         cmd_xp->xb_sense_resid = 0;
17431                 }
17432         }
17433 
17434         /*
17435          *  Free up the RQS command....
17436          *  NOTE:
17437          *      Must do this BEFORE calling sd_validate_sense_data!
17438          *      sd_validate_sense_data may return the original command in
17439          *      which case the pkt will be freed and the flags can no
17440          *      longer be touched.
17441          *      SD_MUTEX is held through this process until the command
17442          *      is dispatched based upon the sense data, so there are
17443          *      no race conditions.
17444          */
17445         (void) sd_mark_rqs_idle(un, sense_xp);
17446 
17447         /*
17448          * For a retryable command see if we have valid sense data, if so then
17449          * turn it over to sd_decode_sense() to figure out the right course of
17450          * action. Just fail a non-retryable command.
17451          */
17452         if ((cmd_pktp->pkt_flags & FLAG_DIAGNOSE) == 0) {
17453                 if (sd_validate_sense_data(un, cmd_bp, cmd_xp, actual_len) ==
17454                     SD_SENSE_DATA_IS_VALID) {
17455                         sd_decode_sense(un, cmd_bp, cmd_xp, cmd_pktp);
17456                 }
17457         } else {
17458                 SD_DUMP_MEMORY(un, SD_LOG_IO_CORE, "Failed CDB",
17459                     (uchar_t *)cmd_pktp->pkt_cdbp, CDB_SIZE, SD_LOG_HEX);
17460                 SD_DUMP_MEMORY(un, SD_LOG_IO_CORE, "Sense Data",
17461                     (uchar_t *)cmd_xp->xb_sense_data, SENSE_LENGTH, SD_LOG_HEX);
17462                 sd_return_failed_command(un, cmd_bp, EIO);
17463         }
17464 }
17465 
17466 
17467 
17468 
17469 /*
17470  *    Function: sd_handle_auto_request_sense
17471  *
17472  * Description: Processing for auto-request sense information.
17473  *
17474  *   Arguments: un - ptr to associated softstate
17475  *              bp - ptr to buf(9S) for the command
17476  *              xp - ptr to the sd_xbuf for the command
17477  *              pktp - ptr to the scsi_pkt(9S) for the command
17478  *
17479  *     Context: May be called under interrupt context
17480  */
17481 
17482 static void
17483 sd_handle_auto_request_sense(struct sd_lun *un, struct buf *bp,
17484         struct sd_xbuf *xp, struct scsi_pkt *pktp)
17485 {
17486         struct scsi_arq_status *asp;
17487         size_t actual_len;
17488 
17489         ASSERT(un != NULL);
17490         ASSERT(mutex_owned(SD_MUTEX(un)));
17491         ASSERT(bp != NULL);
17492         ASSERT(xp != NULL);
17493         ASSERT(pktp != NULL);
17494         ASSERT(pktp != un->un_rqs_pktp);
17495         ASSERT(bp   != un->un_rqs_bp);
17496 
17497         /*
17498          * For auto-request sense, we get a scsi_arq_status back from
17499          * the HBA, with the sense data in the sts_sensedata member.
17500          * The pkt_scbp of the packet points to this scsi_arq_status.
17501          */
17502         asp = (struct scsi_arq_status *)(pktp->pkt_scbp);
17503 
17504         if (asp->sts_rqpkt_reason != CMD_CMPLT) {
17505                 /*
17506                  * The auto REQUEST SENSE failed; see if we can re-try
17507                  * the original command.
17508                  */
17509                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
17510                     "auto request sense failed (reason=%s)\n",
17511                     scsi_rname(asp->sts_rqpkt_reason));
17512 
17513                 sd_reset_target(un, pktp);
17514 
17515                 sd_retry_command(un, bp, SD_RETRIES_STANDARD,
17516                     NULL, NULL, EIO, (clock_t)0, NULL);
17517                 return;
17518         }
17519 
17520         /* Save the relevant sense info into the xp for the original cmd. */
17521         xp->xb_sense_status = *((uchar_t *)(&(asp->sts_rqpkt_status)));
17522         xp->xb_sense_state  = asp->sts_rqpkt_state;
17523         xp->xb_sense_resid  = asp->sts_rqpkt_resid;
17524         if (xp->xb_sense_state & STATE_XARQ_DONE) {
17525                 actual_len = MAX_SENSE_LENGTH - xp->xb_sense_resid;
17526                 bcopy(&asp->sts_sensedata, xp->xb_sense_data,
17527                     MAX_SENSE_LENGTH);
17528         } else {
17529                 if (xp->xb_sense_resid > SENSE_LENGTH) {
17530                         actual_len = MAX_SENSE_LENGTH - xp->xb_sense_resid;
17531                 } else {
17532                         actual_len = SENSE_LENGTH - xp->xb_sense_resid;
17533                 }
17534                 if (xp->xb_pkt_flags & SD_XB_USCSICMD) {
17535                         if ((((struct uscsi_cmd *)
17536                             (xp->xb_pktinfo))->uscsi_rqlen) > actual_len) {
17537                                 xp->xb_sense_resid = (((struct uscsi_cmd *)
17538                                     (xp->xb_pktinfo))->uscsi_rqlen) -
17539                                     actual_len;
17540                         } else {
17541                                 xp->xb_sense_resid = 0;
17542                         }
17543                 }
17544                 bcopy(&asp->sts_sensedata, xp->xb_sense_data, SENSE_LENGTH);
17545         }
17546 
17547         /*
17548          * See if we have valid sense data, if so then turn it over to
17549          * sd_decode_sense() to figure out the right course of action.
17550          */
17551         if (sd_validate_sense_data(un, bp, xp, actual_len) ==
17552             SD_SENSE_DATA_IS_VALID) {
17553                 sd_decode_sense(un, bp, xp, pktp);
17554         }
17555 }
17556 
17557 
17558 /*
17559  *    Function: sd_print_sense_failed_msg
17560  *
17561  * Description: Print log message when RQS has failed.
17562  *
17563  *   Arguments: un - ptr to associated softstate
17564  *              bp - ptr to buf(9S) for the command
17565  *              arg - generic message string ptr
17566  *              code - SD_IMMEDIATE_RETRY_ISSUED, SD_DELAYED_RETRY_ISSUED,
17567  *                      or SD_NO_RETRY_ISSUED
17568  *
17569  *     Context: May be called from interrupt context
17570  */
17571 
17572 static void
17573 sd_print_sense_failed_msg(struct sd_lun *un, struct buf *bp, void *arg,
17574         int code)
17575 {
17576         char    *msgp = arg;
17577 
17578         ASSERT(un != NULL);
17579         ASSERT(mutex_owned(SD_MUTEX(un)));
17580         ASSERT(bp != NULL);
17581 
17582         if ((code == SD_NO_RETRY_ISSUED) && (msgp != NULL)) {
17583                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN, msgp);
17584         }
17585 }
17586 
17587 
17588 /*
17589  *    Function: sd_validate_sense_data
17590  *
17591  * Description: Check the given sense data for validity.
17592  *              If the sense data is not valid, the command will
17593  *              be either failed or retried!
17594  *
17595  * Return Code: SD_SENSE_DATA_IS_INVALID
17596  *              SD_SENSE_DATA_IS_VALID
17597  *
17598  *     Context: May be called from interrupt context
17599  */
17600 
17601 static int
17602 sd_validate_sense_data(struct sd_lun *un, struct buf *bp, struct sd_xbuf *xp,
17603         size_t actual_len)
17604 {
17605         struct scsi_extended_sense *esp;
17606         struct  scsi_pkt *pktp;
17607         char    *msgp = NULL;
17608         sd_ssc_t *sscp;
17609 
17610         ASSERT(un != NULL);
17611         ASSERT(mutex_owned(SD_MUTEX(un)));
17612         ASSERT(bp != NULL);
17613         ASSERT(bp != un->un_rqs_bp);
17614         ASSERT(xp != NULL);
17615         ASSERT(un->un_fm_private != NULL);
17616 
17617         pktp = SD_GET_PKTP(bp);
17618         ASSERT(pktp != NULL);
17619 
17620         sscp = &((struct sd_fm_internal *)(un->un_fm_private))->fm_ssc;
17621         ASSERT(sscp != NULL);
17622 
17623         /*
17624          * Check the status of the RQS command (auto or manual).
17625          */
17626         switch (xp->xb_sense_status & STATUS_MASK) {
17627         case STATUS_GOOD:
17628                 break;
17629 
17630         case STATUS_RESERVATION_CONFLICT:
17631                 sd_pkt_status_reservation_conflict(un, bp, xp, pktp);
17632                 return (SD_SENSE_DATA_IS_INVALID);
17633 
17634         case STATUS_BUSY:
17635                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
17636                     "Busy Status on REQUEST SENSE\n");
17637                 sd_retry_command(un, bp, SD_RETRIES_BUSY, NULL,
17638                     NULL, EIO, un->un_busy_timeout / 500, kstat_waitq_enter);
17639                 return (SD_SENSE_DATA_IS_INVALID);
17640 
17641         case STATUS_QFULL:
17642                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
17643                     "QFULL Status on REQUEST SENSE\n");
17644                 sd_retry_command(un, bp, SD_RETRIES_STANDARD, NULL,
17645                     NULL, EIO, un->un_busy_timeout / 500, kstat_waitq_enter);
17646                 return (SD_SENSE_DATA_IS_INVALID);
17647 
17648         case STATUS_CHECK:
17649         case STATUS_TERMINATED:
17650                 msgp = "Check Condition on REQUEST SENSE\n";
17651                 goto sense_failed;
17652 
17653         default:
17654                 msgp = "Not STATUS_GOOD on REQUEST_SENSE\n";
17655                 goto sense_failed;
17656         }
17657 
17658         /*
17659          * See if we got the minimum required amount of sense data.
17660          * Note: We are assuming the returned sense data is SENSE_LENGTH bytes
17661          * or less.
17662          */
17663         if (((xp->xb_sense_state & STATE_XFERRED_DATA) == 0) ||
17664             (actual_len == 0)) {
17665                 msgp = "Request Sense couldn't get sense data\n";
17666                 goto sense_failed;
17667         }
17668 
17669         if (actual_len < SUN_MIN_SENSE_LENGTH) {
17670                 msgp = "Not enough sense information\n";
17671                 /* Mark the ssc_flags for detecting invalid sense data */
17672                 if (!(xp->xb_pkt_flags & SD_XB_USCSICMD)) {
17673                         sd_ssc_set_info(sscp, SSC_FLAGS_INVALID_SENSE, 0,
17674                             "sense-data");
17675                 }
17676                 goto sense_failed;
17677         }
17678 
17679         /*
17680          * We require the extended sense data
17681          */
17682         esp = (struct scsi_extended_sense *)xp->xb_sense_data;
17683         if (esp->es_class != CLASS_EXTENDED_SENSE) {
17684                 if ((pktp->pkt_flags & FLAG_SILENT) == 0) {
17685                         static char tmp[8];
17686                         static char buf[148];
17687                         char *p = (char *)(xp->xb_sense_data);
17688                         int i;
17689 
17690                         mutex_enter(&sd_sense_mutex);
17691                         (void) strcpy(buf, "undecodable sense information:");
17692                         for (i = 0; i < actual_len; i++) {
17693                                 (void) sprintf(tmp, " 0x%x", *(p++)&0xff);
17694                                 (void) strcpy(&buf[strlen(buf)], tmp);
17695                         }
17696                         i = strlen(buf);
17697                         (void) strcpy(&buf[i], "-(assumed fatal)\n");
17698 
17699                         if (SD_FM_LOG(un) == SD_FM_LOG_NSUP) {
17700                                 scsi_log(SD_DEVINFO(un), sd_label,
17701                                     CE_WARN, buf);
17702                         }
17703                         mutex_exit(&sd_sense_mutex);
17704                 }
17705 
17706                 /* Mark the ssc_flags for detecting invalid sense data */
17707                 if (!(xp->xb_pkt_flags & SD_XB_USCSICMD)) {
17708                         sd_ssc_set_info(sscp, SSC_FLAGS_INVALID_SENSE, 0,
17709                             "sense-data");
17710                 }
17711 
17712                 /* Note: Legacy behavior, fail the command with no retry */
17713                 sd_return_failed_command(un, bp, EIO);
17714                 return (SD_SENSE_DATA_IS_INVALID);
17715         }
17716 
17717         /*
17718          * Check that es_code is valid (es_class concatenated with es_code
17719          * make up the "response code" field.  es_class will always be 7, so
17720          * make sure es_code is 0, 1, 2, 3 or 0xf.  es_code will indicate the
17721          * format.
17722          */
17723         if ((esp->es_code != CODE_FMT_FIXED_CURRENT) &&
17724             (esp->es_code != CODE_FMT_FIXED_DEFERRED) &&
17725             (esp->es_code != CODE_FMT_DESCR_CURRENT) &&
17726             (esp->es_code != CODE_FMT_DESCR_DEFERRED) &&
17727             (esp->es_code != CODE_FMT_VENDOR_SPECIFIC)) {
17728                 /* Mark the ssc_flags for detecting invalid sense data */
17729                 if (!(xp->xb_pkt_flags & SD_XB_USCSICMD)) {
17730                         sd_ssc_set_info(sscp, SSC_FLAGS_INVALID_SENSE, 0,
17731                             "sense-data");
17732                 }
17733                 goto sense_failed;
17734         }
17735 
17736         return (SD_SENSE_DATA_IS_VALID);
17737 
17738 sense_failed:
17739         /*
17740          * If the request sense failed (for whatever reason), attempt
17741          * to retry the original command.
17742          */
17743 #if defined(__i386) || defined(__amd64)
17744         /*
17745          * SD_RETRY_DELAY is conditionally compile (#if fibre) in
17746          * sddef.h for Sparc platform, and x86 uses 1 binary
17747          * for both SCSI/FC.
17748          * The SD_RETRY_DELAY value need to be adjusted here
17749          * when SD_RETRY_DELAY change in sddef.h
17750          */
17751         sd_retry_command(un, bp, SD_RETRIES_STANDARD,
17752             sd_print_sense_failed_msg, msgp, EIO,
17753             un->un_f_is_fibre?drv_usectohz(100000):(clock_t)0, NULL);
17754 #else
17755         sd_retry_command(un, bp, SD_RETRIES_STANDARD,
17756             sd_print_sense_failed_msg, msgp, EIO, SD_RETRY_DELAY, NULL);
17757 #endif
17758 
17759         return (SD_SENSE_DATA_IS_INVALID);
17760 }
17761 
17762 /*
17763  *    Function: sd_decode_sense
17764  *
17765  * Description: Take recovery action(s) when SCSI Sense Data is received.
17766  *
17767  *     Context: Interrupt context.
17768  */
17769 
17770 static void
17771 sd_decode_sense(struct sd_lun *un, struct buf *bp, struct sd_xbuf *xp,
17772         struct scsi_pkt *pktp)
17773 {
17774         uint8_t sense_key;
17775 
17776         ASSERT(un != NULL);
17777         ASSERT(mutex_owned(SD_MUTEX(un)));
17778         ASSERT(bp != NULL);
17779         ASSERT(bp != un->un_rqs_bp);
17780         ASSERT(xp != NULL);
17781         ASSERT(pktp != NULL);
17782 
17783         sense_key = scsi_sense_key(xp->xb_sense_data);
17784 
17785         switch (sense_key) {
17786         case KEY_NO_SENSE:
17787                 sd_sense_key_no_sense(un, bp, xp, pktp);
17788                 break;
17789         case KEY_RECOVERABLE_ERROR:
17790                 sd_sense_key_recoverable_error(un, xp->xb_sense_data,
17791                     bp, xp, pktp);
17792                 break;
17793         case KEY_NOT_READY:
17794                 sd_sense_key_not_ready(un, xp->xb_sense_data,
17795                     bp, xp, pktp);
17796                 break;
17797         case KEY_MEDIUM_ERROR:
17798         case KEY_HARDWARE_ERROR:
17799                 sd_sense_key_medium_or_hardware_error(un,
17800                     xp->xb_sense_data, bp, xp, pktp);
17801                 break;
17802         case KEY_ILLEGAL_REQUEST:
17803                 sd_sense_key_illegal_request(un, bp, xp, pktp);
17804                 break;
17805         case KEY_UNIT_ATTENTION:
17806                 sd_sense_key_unit_attention(un, xp->xb_sense_data,
17807                     bp, xp, pktp);
17808                 break;
17809         case KEY_WRITE_PROTECT:
17810         case KEY_VOLUME_OVERFLOW:
17811         case KEY_MISCOMPARE:
17812                 sd_sense_key_fail_command(un, bp, xp, pktp);
17813                 break;
17814         case KEY_BLANK_CHECK:
17815                 sd_sense_key_blank_check(un, bp, xp, pktp);
17816                 break;
17817         case KEY_ABORTED_COMMAND:
17818                 sd_sense_key_aborted_command(un, bp, xp, pktp);
17819                 break;
17820         case KEY_VENDOR_UNIQUE:
17821         case KEY_COPY_ABORTED:
17822         case KEY_EQUAL:
17823         case KEY_RESERVED:
17824         default:
17825                 sd_sense_key_default(un, xp->xb_sense_data,
17826                     bp, xp, pktp);
17827                 break;
17828         }
17829 }
17830 
17831 
17832 /*
17833  *    Function: sd_dump_memory
17834  *
17835  * Description: Debug logging routine to print the contents of a user provided
17836  *              buffer. The output of the buffer is broken up into 256 byte
17837  *              segments due to a size constraint of the scsi_log.
17838  *              implementation.
17839  *
17840  *   Arguments: un - ptr to softstate
17841  *              comp - component mask
17842  *              title - "title" string to preceed data when printed
17843  *              data - ptr to data block to be printed
17844  *              len - size of data block to be printed
17845  *              fmt - SD_LOG_HEX (use 0x%02x format) or SD_LOG_CHAR (use %c)
17846  *
17847  *     Context: May be called from interrupt context
17848  */
17849 
17850 #define SD_DUMP_MEMORY_BUF_SIZE 256
17851 
17852 static char *sd_dump_format_string[] = {
17853                 " 0x%02x",
17854                 " %c"
17855 };
17856 
17857 static void
17858 sd_dump_memory(struct sd_lun *un, uint_t comp, char *title, uchar_t *data,
17859     int len, int fmt)
17860 {
17861         int     i, j;
17862         int     avail_count;
17863         int     start_offset;
17864         int     end_offset;
17865         size_t  entry_len;
17866         char    *bufp;
17867         char    *local_buf;
17868         char    *format_string;
17869 
17870         ASSERT((fmt == SD_LOG_HEX) || (fmt == SD_LOG_CHAR));
17871 
17872         /*
17873          * In the debug version of the driver, this function is called from a
17874          * number of places which are NOPs in the release driver.
17875          * The debug driver therefore has additional methods of filtering
17876          * debug output.
17877          */
17878 #ifdef SDDEBUG
17879         /*
17880          * In the debug version of the driver we can reduce the amount of debug
17881          * messages by setting sd_error_level to something other than
17882          * SCSI_ERR_ALL and clearing bits in sd_level_mask and
17883          * sd_component_mask.
17884          */
17885         if (((sd_level_mask & (SD_LOGMASK_DUMP_MEM | SD_LOGMASK_DIAG)) == 0) ||
17886             (sd_error_level != SCSI_ERR_ALL)) {
17887                 return;
17888         }
17889         if (((sd_component_mask & comp) == 0) ||
17890             (sd_error_level != SCSI_ERR_ALL)) {
17891                 return;
17892         }
17893 #else
17894         if (sd_error_level != SCSI_ERR_ALL) {
17895                 return;
17896         }
17897 #endif
17898 
17899         local_buf = kmem_zalloc(SD_DUMP_MEMORY_BUF_SIZE, KM_SLEEP);
17900         bufp = local_buf;
17901         /*
17902          * Available length is the length of local_buf[], minus the
17903          * length of the title string, minus one for the ":", minus
17904          * one for the newline, minus one for the NULL terminator.
17905          * This gives the #bytes available for holding the printed
17906          * values from the given data buffer.
17907          */
17908         if (fmt == SD_LOG_HEX) {
17909                 format_string = sd_dump_format_string[0];
17910         } else /* SD_LOG_CHAR */ {
17911                 format_string = sd_dump_format_string[1];
17912         }
17913         /*
17914          * Available count is the number of elements from the given
17915          * data buffer that we can fit into the available length.
17916          * This is based upon the size of the format string used.
17917          * Make one entry and find it's size.
17918          */
17919         (void) sprintf(bufp, format_string, data[0]);
17920         entry_len = strlen(bufp);
17921         avail_count = (SD_DUMP_MEMORY_BUF_SIZE - strlen(title) - 3) / entry_len;
17922 
17923         j = 0;
17924         while (j < len) {
17925                 bufp = local_buf;
17926                 bzero(bufp, SD_DUMP_MEMORY_BUF_SIZE);
17927                 start_offset = j;
17928 
17929                 end_offset = start_offset + avail_count;
17930 
17931                 (void) sprintf(bufp, "%s:", title);
17932                 bufp += strlen(bufp);
17933                 for (i = start_offset; ((i < end_offset) && (j < len));
17934                     i++, j++) {
17935                         (void) sprintf(bufp, format_string, data[i]);
17936                         bufp += entry_len;
17937                 }
17938                 (void) sprintf(bufp, "\n");
17939 
17940                 scsi_log(SD_DEVINFO(un), sd_label, CE_NOTE, "%s", local_buf);
17941         }
17942         kmem_free(local_buf, SD_DUMP_MEMORY_BUF_SIZE);
17943 }
17944 
17945 /*
17946  *    Function: sd_print_sense_msg
17947  *
17948  * Description: Log a message based upon the given sense data.
17949  *
17950  *   Arguments: un - ptr to associated softstate
17951  *              bp - ptr to buf(9S) for the command
17952  *              arg - ptr to associate sd_sense_info struct
17953  *              code - SD_IMMEDIATE_RETRY_ISSUED, SD_DELAYED_RETRY_ISSUED,
17954  *                      or SD_NO_RETRY_ISSUED
17955  *
17956  *     Context: May be called from interrupt context
17957  */
17958 
17959 static void
17960 sd_print_sense_msg(struct sd_lun *un, struct buf *bp, void *arg, int code)
17961 {
17962         struct sd_xbuf  *xp;
17963         struct scsi_pkt *pktp;
17964         uint8_t *sensep;
17965         daddr_t request_blkno;
17966         diskaddr_t err_blkno;
17967         int severity;
17968         int pfa_flag;
17969         extern struct scsi_key_strings scsi_cmds[];
17970 
17971         ASSERT(un != NULL);
17972         ASSERT(mutex_owned(SD_MUTEX(un)));
17973         ASSERT(bp != NULL);
17974         xp = SD_GET_XBUF(bp);
17975         ASSERT(xp != NULL);
17976         pktp = SD_GET_PKTP(bp);
17977         ASSERT(pktp != NULL);
17978         ASSERT(arg != NULL);
17979 
17980         severity = ((struct sd_sense_info *)(arg))->ssi_severity;
17981         pfa_flag = ((struct sd_sense_info *)(arg))->ssi_pfa_flag;
17982 
17983         if ((code == SD_DELAYED_RETRY_ISSUED) ||
17984             (code == SD_IMMEDIATE_RETRY_ISSUED)) {
17985                 severity = SCSI_ERR_RETRYABLE;
17986         }
17987 
17988         /* Use absolute block number for the request block number */
17989         request_blkno = xp->xb_blkno;
17990 
17991         /*
17992          * Now try to get the error block number from the sense data
17993          */
17994         sensep = xp->xb_sense_data;
17995 
17996         if (scsi_sense_info_uint64(sensep, SENSE_LENGTH,
17997             (uint64_t *)&err_blkno)) {
17998                 /*
17999                  * We retrieved the error block number from the information
18000                  * portion of the sense data.
18001                  *
18002                  * For USCSI commands we are better off using the error
18003                  * block no. as the requested block no. (This is the best
18004                  * we can estimate.)
18005                  */
18006                 if ((SD_IS_BUFIO(xp) == FALSE) &&
18007                     ((pktp->pkt_flags & FLAG_SILENT) == 0)) {
18008                         request_blkno = err_blkno;
18009                 }
18010         } else {
18011                 /*
18012                  * Without the es_valid bit set (for fixed format) or an
18013                  * information descriptor (for descriptor format) we cannot
18014                  * be certain of the error blkno, so just use the
18015                  * request_blkno.
18016                  */
18017                 err_blkno = (diskaddr_t)request_blkno;
18018         }
18019 
18020         /*
18021          * The following will log the buffer contents for the release driver
18022          * if the SD_LOGMASK_DIAG bit of sd_level_mask is set, or the error
18023          * level is set to verbose.
18024          */
18025         sd_dump_memory(un, SD_LOG_IO, "Failed CDB",
18026             (uchar_t *)pktp->pkt_cdbp, CDB_SIZE, SD_LOG_HEX);
18027         sd_dump_memory(un, SD_LOG_IO, "Sense Data",
18028             (uchar_t *)sensep, SENSE_LENGTH, SD_LOG_HEX);
18029 
18030         if (pfa_flag == FALSE) {
18031                 /* This is normally only set for USCSI */
18032                 if ((pktp->pkt_flags & FLAG_SILENT) != 0) {
18033                         return;
18034                 }
18035 
18036                 if ((SD_IS_BUFIO(xp) == TRUE) &&
18037                     (((sd_level_mask & SD_LOGMASK_DIAG) == 0) &&
18038                     (severity < sd_error_level))) {
18039                         return;
18040                 }
18041         }
18042         /*
18043          * Check for Sonoma Failover and keep a count of how many failed I/O's
18044          */
18045         if ((SD_IS_LSI(un)) &&
18046             (scsi_sense_key(sensep) == KEY_ILLEGAL_REQUEST) &&
18047             (scsi_sense_asc(sensep) == 0x94) &&
18048             (scsi_sense_ascq(sensep) == 0x01)) {
18049                 un->un_sonoma_failure_count++;
18050                 if (un->un_sonoma_failure_count > 1) {
18051                         return;
18052                 }
18053         }
18054 
18055         if (SD_FM_LOG(un) == SD_FM_LOG_NSUP ||
18056             ((scsi_sense_key(sensep) == KEY_RECOVERABLE_ERROR) &&
18057             (pktp->pkt_resid == 0))) {
18058                 scsi_vu_errmsg(SD_SCSI_DEVP(un), pktp, sd_label, severity,
18059                     request_blkno, err_blkno, scsi_cmds,
18060                     (struct scsi_extended_sense *)sensep,
18061                     un->un_additional_codes, NULL);
18062         }
18063 }
18064 
18065 /*
18066  *    Function: sd_sense_key_no_sense
18067  *
18068  * Description: Recovery action when sense data was not received.
18069  *
18070  *     Context: May be called from interrupt context
18071  */
18072 
18073 static void
18074 sd_sense_key_no_sense(struct sd_lun *un, struct buf *bp,
18075         struct sd_xbuf *xp, struct scsi_pkt *pktp)
18076 {
18077         struct sd_sense_info    si;
18078 
18079         ASSERT(un != NULL);
18080         ASSERT(mutex_owned(SD_MUTEX(un)));
18081         ASSERT(bp != NULL);
18082         ASSERT(xp != NULL);
18083         ASSERT(pktp != NULL);
18084 
18085         si.ssi_severity = SCSI_ERR_FATAL;
18086         si.ssi_pfa_flag = FALSE;
18087 
18088         SD_UPDATE_ERRSTATS(un, sd_softerrs);
18089 
18090         sd_retry_command(un, bp, SD_RETRIES_STANDARD, sd_print_sense_msg,
18091             &si, EIO, (clock_t)0, NULL);
18092 }
18093 
18094 
18095 /*
18096  *    Function: sd_sense_key_recoverable_error
18097  *
18098  * Description: Recovery actions for a SCSI "Recovered Error" sense key.
18099  *
18100  *     Context: May be called from interrupt context
18101  */
18102 
18103 static void
18104 sd_sense_key_recoverable_error(struct sd_lun *un,
18105         uint8_t *sense_datap,
18106         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp)
18107 {
18108         struct sd_sense_info    si;
18109         uint8_t asc = scsi_sense_asc(sense_datap);
18110 
18111         ASSERT(un != NULL);
18112         ASSERT(mutex_owned(SD_MUTEX(un)));
18113         ASSERT(bp != NULL);
18114         ASSERT(xp != NULL);
18115         ASSERT(pktp != NULL);
18116 
18117         /*
18118          * 0x5D: FAILURE PREDICTION THRESHOLD EXCEEDED
18119          */
18120         if ((asc == 0x5D) && (sd_report_pfa != 0)) {
18121                 SD_UPDATE_ERRSTATS(un, sd_rq_pfa_err);
18122                 si.ssi_severity = SCSI_ERR_INFO;
18123                 si.ssi_pfa_flag = TRUE;
18124         } else {
18125                 SD_UPDATE_ERRSTATS(un, sd_softerrs);
18126                 SD_UPDATE_ERRSTATS(un, sd_rq_recov_err);
18127                 si.ssi_severity = SCSI_ERR_RECOVERED;
18128                 si.ssi_pfa_flag = FALSE;
18129         }
18130 
18131         if (pktp->pkt_resid == 0) {
18132                 sd_print_sense_msg(un, bp, &si, SD_NO_RETRY_ISSUED);
18133                 sd_return_command(un, bp);
18134                 return;
18135         }
18136 
18137         sd_retry_command(un, bp, SD_RETRIES_STANDARD, sd_print_sense_msg,
18138             &si, EIO, (clock_t)0, NULL);
18139 }
18140 
18141 
18142 
18143 
18144 /*
18145  *    Function: sd_sense_key_not_ready
18146  *
18147  * Description: Recovery actions for a SCSI "Not Ready" sense key.
18148  *
18149  *     Context: May be called from interrupt context
18150  */
18151 
18152 static void
18153 sd_sense_key_not_ready(struct sd_lun *un,
18154         uint8_t *sense_datap,
18155         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp)
18156 {
18157         struct sd_sense_info    si;
18158         uint8_t asc = scsi_sense_asc(sense_datap);
18159         uint8_t ascq = scsi_sense_ascq(sense_datap);
18160 
18161         ASSERT(un != NULL);
18162         ASSERT(mutex_owned(SD_MUTEX(un)));
18163         ASSERT(bp != NULL);
18164         ASSERT(xp != NULL);
18165         ASSERT(pktp != NULL);
18166 
18167         si.ssi_severity = SCSI_ERR_FATAL;
18168         si.ssi_pfa_flag = FALSE;
18169 
18170         /*
18171          * Update error stats after first NOT READY error. Disks may have
18172          * been powered down and may need to be restarted.  For CDROMs,
18173          * report NOT READY errors only if media is present.
18174          */
18175         if ((ISCD(un) && (asc == 0x3A)) ||
18176             (xp->xb_nr_retry_count > 0)) {
18177                 SD_UPDATE_ERRSTATS(un, sd_harderrs);
18178                 SD_UPDATE_ERRSTATS(un, sd_rq_ntrdy_err);
18179         }
18180 
18181         /*
18182          * Just fail if the "not ready" retry limit has been reached.
18183          */
18184         if (xp->xb_nr_retry_count >= un->un_notready_retry_count) {
18185                 /* Special check for error message printing for removables. */
18186                 if (un->un_f_has_removable_media && (asc == 0x04) &&
18187                     (ascq >= 0x04)) {
18188                         si.ssi_severity = SCSI_ERR_ALL;
18189                 }
18190                 goto fail_command;
18191         }
18192 
18193         /*
18194          * Check the ASC and ASCQ in the sense data as needed, to determine
18195          * what to do.
18196          */
18197         switch (asc) {
18198         case 0x04:      /* LOGICAL UNIT NOT READY */
18199                 /*
18200                  * disk drives that don't spin up result in a very long delay
18201                  * in format without warning messages. We will log a message
18202                  * if the error level is set to verbose.
18203                  */
18204                 if (sd_error_level < SCSI_ERR_RETRYABLE) {
18205                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
18206                             "logical unit not ready, resetting disk\n");
18207                 }
18208 
18209                 /*
18210                  * There are different requirements for CDROMs and disks for
18211                  * the number of retries.  If a CD-ROM is giving this, it is
18212                  * probably reading TOC and is in the process of getting
18213                  * ready, so we should keep on trying for a long time to make
18214                  * sure that all types of media are taken in account (for
18215                  * some media the drive takes a long time to read TOC).  For
18216                  * disks we do not want to retry this too many times as this
18217                  * can cause a long hang in format when the drive refuses to
18218                  * spin up (a very common failure).
18219                  */
18220                 switch (ascq) {
18221                 case 0x00:  /* LUN NOT READY, CAUSE NOT REPORTABLE */
18222                         /*
18223                          * Disk drives frequently refuse to spin up which
18224                          * results in a very long hang in format without
18225                          * warning messages.
18226                          *
18227                          * Note: This code preserves the legacy behavior of
18228                          * comparing xb_nr_retry_count against zero for fibre
18229                          * channel targets instead of comparing against the
18230                          * un_reset_retry_count value.  The reason for this
18231                          * discrepancy has been so utterly lost beneath the
18232                          * Sands of Time that even Indiana Jones could not
18233                          * find it.
18234                          */
18235                         if (un->un_f_is_fibre == TRUE) {
18236                                 if (((sd_level_mask & SD_LOGMASK_DIAG) ||
18237                                     (xp->xb_nr_retry_count > 0)) &&
18238                                     (un->un_startstop_timeid == NULL)) {
18239                                         scsi_log(SD_DEVINFO(un), sd_label,
18240                                             CE_WARN, "logical unit not ready, "
18241                                             "resetting disk\n");
18242                                         sd_reset_target(un, pktp);
18243                                 }
18244                         } else {
18245                                 if (((sd_level_mask & SD_LOGMASK_DIAG) ||
18246                                     (xp->xb_nr_retry_count >
18247                                     un->un_reset_retry_count)) &&
18248                                     (un->un_startstop_timeid == NULL)) {
18249                                         scsi_log(SD_DEVINFO(un), sd_label,
18250                                             CE_WARN, "logical unit not ready, "
18251                                             "resetting disk\n");
18252                                         sd_reset_target(un, pktp);
18253                                 }
18254                         }
18255                         break;
18256 
18257                 case 0x01:  /* LUN IS IN PROCESS OF BECOMING READY */
18258                         /*
18259                          * If the target is in the process of becoming
18260                          * ready, just proceed with the retry. This can
18261                          * happen with CD-ROMs that take a long time to
18262                          * read TOC after a power cycle or reset.
18263                          */
18264                         goto do_retry;
18265 
18266                 case 0x02:  /* LUN NOT READY, INITITIALIZING CMD REQUIRED */
18267                         break;
18268 
18269                 case 0x03:  /* LUN NOT READY, MANUAL INTERVENTION REQUIRED */
18270                         /*
18271                          * Retries cannot help here so just fail right away.
18272                          */
18273                         goto fail_command;
18274 
18275                 case 0x88:
18276                         /*
18277                          * Vendor-unique code for T3/T4: it indicates a
18278                          * path problem in a mutipathed config, but as far as
18279                          * the target driver is concerned it equates to a fatal
18280                          * error, so we should just fail the command right away
18281                          * (without printing anything to the console). If this
18282                          * is not a T3/T4, fall thru to the default recovery
18283                          * action.
18284                          * T3/T4 is FC only, don't need to check is_fibre
18285                          */
18286                         if (SD_IS_T3(un) || SD_IS_T4(un)) {
18287                                 sd_return_failed_command(un, bp, EIO);
18288                                 return;
18289                         }
18290                         /* FALLTHRU */
18291 
18292                 case 0x04:  /* LUN NOT READY, FORMAT IN PROGRESS */
18293                 case 0x05:  /* LUN NOT READY, REBUILD IN PROGRESS */
18294                 case 0x06:  /* LUN NOT READY, RECALCULATION IN PROGRESS */
18295                 case 0x07:  /* LUN NOT READY, OPERATION IN PROGRESS */
18296                 case 0x08:  /* LUN NOT READY, LONG WRITE IN PROGRESS */
18297                 default:    /* Possible future codes in SCSI spec? */
18298                         /*
18299                          * For removable-media devices, do not retry if
18300                          * ASCQ > 2 as these result mostly from USCSI commands
18301                          * on MMC devices issued to check status of an
18302                          * operation initiated in immediate mode.  Also for
18303                          * ASCQ >= 4 do not print console messages as these
18304                          * mainly represent a user-initiated operation
18305                          * instead of a system failure.
18306                          */
18307                         if (un->un_f_has_removable_media) {
18308                                 si.ssi_severity = SCSI_ERR_ALL;
18309                                 goto fail_command;
18310                         }
18311                         break;
18312                 }
18313 
18314                 /*
18315                  * As part of our recovery attempt for the NOT READY
18316                  * condition, we issue a START STOP UNIT command. However
18317                  * we want to wait for a short delay before attempting this
18318                  * as there may still be more commands coming back from the
18319                  * target with the check condition. To do this we use
18320                  * timeout(9F) to call sd_start_stop_unit_callback() after
18321                  * the delay interval expires. (sd_start_stop_unit_callback()
18322                  * dispatches sd_start_stop_unit_task(), which will issue
18323                  * the actual START STOP UNIT command. The delay interval
18324                  * is one-half of the delay that we will use to retry the
18325                  * command that generated the NOT READY condition.
18326                  *
18327                  * Note that we could just dispatch sd_start_stop_unit_task()
18328                  * from here and allow it to sleep for the delay interval,
18329                  * but then we would be tying up the taskq thread
18330                  * uncesessarily for the duration of the delay.
18331                  *
18332                  * Do not issue the START STOP UNIT if the current command
18333                  * is already a START STOP UNIT.
18334                  */
18335                 if (pktp->pkt_cdbp[0] == SCMD_START_STOP) {
18336                         break;
18337                 }
18338 
18339                 /*
18340                  * Do not schedule the timeout if one is already pending.
18341                  */
18342                 if (un->un_startstop_timeid != NULL) {
18343                         SD_INFO(SD_LOG_ERROR, un,
18344                             "sd_sense_key_not_ready: restart already issued to"
18345                             " %s%d\n", ddi_driver_name(SD_DEVINFO(un)),
18346                             ddi_get_instance(SD_DEVINFO(un)));
18347                         break;
18348                 }
18349 
18350                 /*
18351                  * Schedule the START STOP UNIT command, then queue the command
18352                  * for a retry.
18353                  *
18354                  * Note: A timeout is not scheduled for this retry because we
18355                  * want the retry to be serial with the START_STOP_UNIT. The
18356                  * retry will be started when the START_STOP_UNIT is completed
18357                  * in sd_start_stop_unit_task.
18358                  */
18359                 un->un_startstop_timeid = timeout(sd_start_stop_unit_callback,
18360                     un, un->un_busy_timeout / 2);
18361                 xp->xb_nr_retry_count++;
18362                 sd_set_retry_bp(un, bp, 0, kstat_waitq_enter);
18363                 return;
18364 
18365         case 0x05:      /* LOGICAL UNIT DOES NOT RESPOND TO SELECTION */
18366                 if (sd_error_level < SCSI_ERR_RETRYABLE) {
18367                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
18368                             "unit does not respond to selection\n");
18369                 }
18370                 break;
18371 
18372         case 0x3A:      /* MEDIUM NOT PRESENT */
18373                 if (sd_error_level >= SCSI_ERR_FATAL) {
18374                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
18375                             "Caddy not inserted in drive\n");
18376                 }
18377 
18378                 sr_ejected(un);
18379                 un->un_mediastate = DKIO_EJECTED;
18380                 /* The state has changed, inform the media watch routines */
18381                 cv_broadcast(&un->un_state_cv);
18382                 /* Just fail if no media is present in the drive. */
18383                 goto fail_command;
18384 
18385         default:
18386                 if (sd_error_level < SCSI_ERR_RETRYABLE) {
18387                         scsi_log(SD_DEVINFO(un), sd_label, CE_NOTE,
18388                             "Unit not Ready. Additional sense code 0x%x\n",
18389                             asc);
18390                 }
18391                 break;
18392         }
18393 
18394 do_retry:
18395 
18396         /*
18397          * Retry the command, as some targets may report NOT READY for
18398          * several seconds after being reset.
18399          */
18400         xp->xb_nr_retry_count++;
18401         si.ssi_severity = SCSI_ERR_RETRYABLE;
18402         sd_retry_command(un, bp, SD_RETRIES_NOCHECK, sd_print_sense_msg,
18403             &si, EIO, un->un_busy_timeout, NULL);
18404 
18405         return;
18406 
18407 fail_command:
18408         sd_print_sense_msg(un, bp, &si, SD_NO_RETRY_ISSUED);
18409         sd_return_failed_command(un, bp, EIO);
18410 }
18411 
18412 
18413 
18414 /*
18415  *    Function: sd_sense_key_medium_or_hardware_error
18416  *
18417  * Description: Recovery actions for a SCSI "Medium Error" or "Hardware Error"
18418  *              sense key.
18419  *
18420  *     Context: May be called from interrupt context
18421  */
18422 
18423 static void
18424 sd_sense_key_medium_or_hardware_error(struct sd_lun *un,
18425         uint8_t *sense_datap,
18426         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp)
18427 {
18428         struct sd_sense_info    si;
18429         uint8_t sense_key = scsi_sense_key(sense_datap);
18430         uint8_t asc = scsi_sense_asc(sense_datap);
18431 
18432         ASSERT(un != NULL);
18433         ASSERT(mutex_owned(SD_MUTEX(un)));
18434         ASSERT(bp != NULL);
18435         ASSERT(xp != NULL);
18436         ASSERT(pktp != NULL);
18437 
18438         si.ssi_severity = SCSI_ERR_FATAL;
18439         si.ssi_pfa_flag = FALSE;
18440 
18441         if (sense_key == KEY_MEDIUM_ERROR) {
18442                 SD_UPDATE_ERRSTATS(un, sd_rq_media_err);
18443         }
18444 
18445         SD_UPDATE_ERRSTATS(un, sd_harderrs);
18446 
18447         if ((un->un_reset_retry_count != 0) &&
18448             (xp->xb_retry_count == un->un_reset_retry_count)) {
18449                 mutex_exit(SD_MUTEX(un));
18450                 /* Do NOT do a RESET_ALL here: too intrusive. (4112858) */
18451                 if (un->un_f_allow_bus_device_reset == TRUE) {
18452 
18453                         boolean_t try_resetting_target = B_TRUE;
18454 
18455                         /*
18456                          * We need to be able to handle specific ASC when we are
18457                          * handling a KEY_HARDWARE_ERROR. In particular
18458                          * taking the default action of resetting the target may
18459                          * not be the appropriate way to attempt recovery.
18460                          * Resetting a target because of a single LUN failure
18461                          * victimizes all LUNs on that target.
18462                          *
18463                          * This is true for the LSI arrays, if an LSI
18464                          * array controller returns an ASC of 0x84 (LUN Dead) we
18465                          * should trust it.
18466                          */
18467 
18468                         if (sense_key == KEY_HARDWARE_ERROR) {
18469                                 switch (asc) {
18470                                 case 0x84:
18471                                         if (SD_IS_LSI(un)) {
18472                                                 try_resetting_target = B_FALSE;
18473                                         }
18474                                         break;
18475                                 default:
18476                                         break;
18477                                 }
18478                         }
18479 
18480                         if (try_resetting_target == B_TRUE) {
18481                                 int reset_retval = 0;
18482                                 if (un->un_f_lun_reset_enabled == TRUE) {
18483                                         SD_TRACE(SD_LOG_IO_CORE, un,
18484                                             "sd_sense_key_medium_or_hardware_"
18485                                             "error: issuing RESET_LUN\n");
18486                                         reset_retval =
18487                                             scsi_reset(SD_ADDRESS(un),
18488                                             RESET_LUN);
18489                                 }
18490                                 if (reset_retval == 0) {
18491                                         SD_TRACE(SD_LOG_IO_CORE, un,
18492                                             "sd_sense_key_medium_or_hardware_"
18493                                             "error: issuing RESET_TARGET\n");
18494                                         (void) scsi_reset(SD_ADDRESS(un),
18495                                             RESET_TARGET);
18496                                 }
18497                         }
18498                 }
18499                 mutex_enter(SD_MUTEX(un));
18500         }
18501 
18502         /*
18503          * This really ought to be a fatal error, but we will retry anyway
18504          * as some drives report this as a spurious error.
18505          */
18506         sd_retry_command(un, bp, SD_RETRIES_STANDARD, sd_print_sense_msg,
18507             &si, EIO, (clock_t)0, NULL);
18508 }
18509 
18510 
18511 
18512 /*
18513  *    Function: sd_sense_key_illegal_request
18514  *
18515  * Description: Recovery actions for a SCSI "Illegal Request" sense key.
18516  *
18517  *     Context: May be called from interrupt context
18518  */
18519 
18520 static void
18521 sd_sense_key_illegal_request(struct sd_lun *un, struct buf *bp,
18522         struct sd_xbuf *xp, struct scsi_pkt *pktp)
18523 {
18524         struct sd_sense_info    si;
18525 
18526         ASSERT(un != NULL);
18527         ASSERT(mutex_owned(SD_MUTEX(un)));
18528         ASSERT(bp != NULL);
18529         ASSERT(xp != NULL);
18530         ASSERT(pktp != NULL);
18531 
18532         SD_UPDATE_ERRSTATS(un, sd_rq_illrq_err);
18533 
18534         si.ssi_severity = SCSI_ERR_INFO;
18535         si.ssi_pfa_flag = FALSE;
18536 
18537         /* Pointless to retry if the target thinks it's an illegal request */
18538         sd_print_sense_msg(un, bp, &si, SD_NO_RETRY_ISSUED);
18539         sd_return_failed_command(un, bp, EIO);
18540 }
18541 
18542 
18543 
18544 
18545 /*
18546  *    Function: sd_sense_key_unit_attention
18547  *
18548  * Description: Recovery actions for a SCSI "Unit Attention" sense key.
18549  *
18550  *     Context: May be called from interrupt context
18551  */
18552 
18553 static void
18554 sd_sense_key_unit_attention(struct sd_lun *un,
18555         uint8_t *sense_datap,
18556         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp)
18557 {
18558         /*
18559          * For UNIT ATTENTION we allow retries for one minute. Devices
18560          * like Sonoma can return UNIT ATTENTION close to a minute
18561          * under certain conditions.
18562          */
18563         int     retry_check_flag = SD_RETRIES_UA;
18564         boolean_t       kstat_updated = B_FALSE;
18565         struct  sd_sense_info           si;
18566         uint8_t asc = scsi_sense_asc(sense_datap);
18567         uint8_t ascq = scsi_sense_ascq(sense_datap);
18568 
18569         ASSERT(un != NULL);
18570         ASSERT(mutex_owned(SD_MUTEX(un)));
18571         ASSERT(bp != NULL);
18572         ASSERT(xp != NULL);
18573         ASSERT(pktp != NULL);
18574 
18575         si.ssi_severity = SCSI_ERR_INFO;
18576         si.ssi_pfa_flag = FALSE;
18577 
18578 
18579         switch (asc) {
18580         case 0x5D:  /* FAILURE PREDICTION THRESHOLD EXCEEDED */
18581                 if (sd_report_pfa != 0) {
18582                         SD_UPDATE_ERRSTATS(un, sd_rq_pfa_err);
18583                         si.ssi_pfa_flag = TRUE;
18584                         retry_check_flag = SD_RETRIES_STANDARD;
18585                         goto do_retry;
18586                 }
18587 
18588                 break;
18589 
18590         case 0x29:  /* POWER ON, RESET, OR BUS DEVICE RESET OCCURRED */
18591                 if ((un->un_resvd_status & SD_RESERVE) == SD_RESERVE) {
18592                         un->un_resvd_status |=
18593                             (SD_LOST_RESERVE | SD_WANT_RESERVE);
18594                 }
18595 #ifdef _LP64
18596                 if (un->un_blockcount + 1 > SD_GROUP1_MAX_ADDRESS) {
18597                         if (taskq_dispatch(sd_tq, sd_reenable_dsense_task,
18598                             un, KM_NOSLEEP) == 0) {
18599                                 /*
18600                                  * If we can't dispatch the task we'll just
18601                                  * live without descriptor sense.  We can
18602                                  * try again on the next "unit attention"
18603                                  */
18604                                 SD_ERROR(SD_LOG_ERROR, un,
18605                                     "sd_sense_key_unit_attention: "
18606                                     "Could not dispatch "
18607                                     "sd_reenable_dsense_task\n");
18608                         }
18609                 }
18610 #endif /* _LP64 */
18611                 /* FALLTHRU */
18612 
18613         case 0x28: /* NOT READY TO READY CHANGE, MEDIUM MAY HAVE CHANGED */
18614                 if (!un->un_f_has_removable_media) {
18615                         break;
18616                 }
18617 
18618                 /*
18619                  * When we get a unit attention from a removable-media device,
18620                  * it may be in a state that will take a long time to recover
18621                  * (e.g., from a reset).  Since we are executing in interrupt
18622                  * context here, we cannot wait around for the device to come
18623                  * back. So hand this command off to sd_media_change_task()
18624                  * for deferred processing under taskq thread context. (Note
18625                  * that the command still may be failed if a problem is
18626                  * encountered at a later time.)
18627                  */
18628                 if (taskq_dispatch(sd_tq, sd_media_change_task, pktp,
18629                     KM_NOSLEEP) == 0) {
18630                         /*
18631                          * Cannot dispatch the request so fail the command.
18632                          */
18633                         SD_UPDATE_ERRSTATS(un, sd_harderrs);
18634                         SD_UPDATE_ERRSTATS(un, sd_rq_nodev_err);
18635                         si.ssi_severity = SCSI_ERR_FATAL;
18636                         sd_print_sense_msg(un, bp, &si, SD_NO_RETRY_ISSUED);
18637                         sd_return_failed_command(un, bp, EIO);
18638                 }
18639 
18640                 /*
18641                  * If failed to dispatch sd_media_change_task(), we already
18642                  * updated kstat. If succeed to dispatch sd_media_change_task(),
18643                  * we should update kstat later if it encounters an error. So,
18644                  * we update kstat_updated flag here.
18645                  */
18646                 kstat_updated = B_TRUE;
18647 
18648                 /*
18649                  * Either the command has been successfully dispatched to a
18650                  * task Q for retrying, or the dispatch failed. In either case
18651                  * do NOT retry again by calling sd_retry_command. This sets up
18652                  * two retries of the same command and when one completes and
18653                  * frees the resources the other will access freed memory,
18654                  * a bad thing.
18655                  */
18656                 return;
18657 
18658         default:
18659                 break;
18660         }
18661 
18662         /*
18663          * ASC  ASCQ
18664          *  2A   09     Capacity data has changed
18665          *  2A   01     Mode parameters changed
18666          *  3F   0E     Reported luns data has changed
18667          * Arrays that support logical unit expansion should report
18668          * capacity changes(2Ah/09). Mode parameters changed and
18669          * reported luns data has changed are the approximation.
18670          */
18671         if (((asc == 0x2a) && (ascq == 0x09)) ||
18672             ((asc == 0x2a) && (ascq == 0x01)) ||
18673             ((asc == 0x3f) && (ascq == 0x0e))) {
18674                 if (taskq_dispatch(sd_tq, sd_target_change_task, un,
18675                     KM_NOSLEEP) == 0) {
18676                         SD_ERROR(SD_LOG_ERROR, un,
18677                             "sd_sense_key_unit_attention: "
18678                             "Could not dispatch sd_target_change_task\n");
18679                 }
18680         }
18681 
18682         /*
18683          * Update kstat if we haven't done that.
18684          */
18685         if (!kstat_updated) {
18686                 SD_UPDATE_ERRSTATS(un, sd_harderrs);
18687                 SD_UPDATE_ERRSTATS(un, sd_rq_nodev_err);
18688         }
18689 
18690 do_retry:
18691         sd_retry_command(un, bp, retry_check_flag, sd_print_sense_msg, &si,
18692             EIO, SD_UA_RETRY_DELAY, NULL);
18693 }
18694 
18695 
18696 
18697 /*
18698  *    Function: sd_sense_key_fail_command
18699  *
18700  * Description: Use to fail a command when we don't like the sense key that
18701  *              was returned.
18702  *
18703  *     Context: May be called from interrupt context
18704  */
18705 
18706 static void
18707 sd_sense_key_fail_command(struct sd_lun *un, struct buf *bp,
18708         struct sd_xbuf *xp, struct scsi_pkt *pktp)
18709 {
18710         struct sd_sense_info    si;
18711 
18712         ASSERT(un != NULL);
18713         ASSERT(mutex_owned(SD_MUTEX(un)));
18714         ASSERT(bp != NULL);
18715         ASSERT(xp != NULL);
18716         ASSERT(pktp != NULL);
18717 
18718         si.ssi_severity = SCSI_ERR_FATAL;
18719         si.ssi_pfa_flag = FALSE;
18720 
18721         sd_print_sense_msg(un, bp, &si, SD_NO_RETRY_ISSUED);
18722         sd_return_failed_command(un, bp, EIO);
18723 }
18724 
18725 
18726 
18727 /*
18728  *    Function: sd_sense_key_blank_check
18729  *
18730  * Description: Recovery actions for a SCSI "Blank Check" sense key.
18731  *              Has no monetary connotation.
18732  *
18733  *     Context: May be called from interrupt context
18734  */
18735 
18736 static void
18737 sd_sense_key_blank_check(struct sd_lun *un, struct buf *bp,
18738         struct sd_xbuf *xp, struct scsi_pkt *pktp)
18739 {
18740         struct sd_sense_info    si;
18741 
18742         ASSERT(un != NULL);
18743         ASSERT(mutex_owned(SD_MUTEX(un)));
18744         ASSERT(bp != NULL);
18745         ASSERT(xp != NULL);
18746         ASSERT(pktp != NULL);
18747 
18748         /*
18749          * Blank check is not fatal for removable devices, therefore
18750          * it does not require a console message.
18751          */
18752         si.ssi_severity = (un->un_f_has_removable_media) ? SCSI_ERR_ALL :
18753             SCSI_ERR_FATAL;
18754         si.ssi_pfa_flag = FALSE;
18755 
18756         sd_print_sense_msg(un, bp, &si, SD_NO_RETRY_ISSUED);
18757         sd_return_failed_command(un, bp, EIO);
18758 }
18759 
18760 
18761 
18762 
18763 /*
18764  *    Function: sd_sense_key_aborted_command
18765  *
18766  * Description: Recovery actions for a SCSI "Aborted Command" sense key.
18767  *
18768  *     Context: May be called from interrupt context
18769  */
18770 
18771 static void
18772 sd_sense_key_aborted_command(struct sd_lun *un, struct buf *bp,
18773         struct sd_xbuf *xp, struct scsi_pkt *pktp)
18774 {
18775         struct sd_sense_info    si;
18776 
18777         ASSERT(un != NULL);
18778         ASSERT(mutex_owned(SD_MUTEX(un)));
18779         ASSERT(bp != NULL);
18780         ASSERT(xp != NULL);
18781         ASSERT(pktp != NULL);
18782 
18783         si.ssi_severity = SCSI_ERR_FATAL;
18784         si.ssi_pfa_flag = FALSE;
18785 
18786         SD_UPDATE_ERRSTATS(un, sd_harderrs);
18787 
18788         /*
18789          * This really ought to be a fatal error, but we will retry anyway
18790          * as some drives report this as a spurious error.
18791          */
18792         sd_retry_command(un, bp, SD_RETRIES_STANDARD, sd_print_sense_msg,
18793             &si, EIO, drv_usectohz(100000), NULL);
18794 }
18795 
18796 
18797 
18798 /*
18799  *    Function: sd_sense_key_default
18800  *
18801  * Description: Default recovery action for several SCSI sense keys (basically
18802  *              attempts a retry).
18803  *
18804  *     Context: May be called from interrupt context
18805  */
18806 
18807 static void
18808 sd_sense_key_default(struct sd_lun *un,
18809         uint8_t *sense_datap,
18810         struct buf *bp, struct sd_xbuf *xp, struct scsi_pkt *pktp)
18811 {
18812         struct sd_sense_info    si;
18813         uint8_t sense_key = scsi_sense_key(sense_datap);
18814 
18815         ASSERT(un != NULL);
18816         ASSERT(mutex_owned(SD_MUTEX(un)));
18817         ASSERT(bp != NULL);
18818         ASSERT(xp != NULL);
18819         ASSERT(pktp != NULL);
18820 
18821         SD_UPDATE_ERRSTATS(un, sd_harderrs);
18822 
18823         /*
18824          * Undecoded sense key. Attempt retries and hope that will fix
18825          * the problem.  Otherwise, we're dead.
18826          */
18827         if ((pktp->pkt_flags & FLAG_SILENT) == 0) {
18828                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
18829                     "Unhandled Sense Key '%s'\n", sense_keys[sense_key]);
18830         }
18831 
18832         si.ssi_severity = SCSI_ERR_FATAL;
18833         si.ssi_pfa_flag = FALSE;
18834 
18835         sd_retry_command(un, bp, SD_RETRIES_STANDARD, sd_print_sense_msg,
18836             &si, EIO, (clock_t)0, NULL);
18837 }
18838 
18839 
18840 
18841 /*
18842  *    Function: sd_print_retry_msg
18843  *
18844  * Description: Print a message indicating the retry action being taken.
18845  *
18846  *   Arguments: un - ptr to associated softstate
18847  *              bp - ptr to buf(9S) for the command
18848  *              arg - not used.
18849  *              flag - SD_IMMEDIATE_RETRY_ISSUED, SD_DELAYED_RETRY_ISSUED,
18850  *                      or SD_NO_RETRY_ISSUED
18851  *
18852  *     Context: May be called from interrupt context
18853  */
18854 /* ARGSUSED */
18855 static void
18856 sd_print_retry_msg(struct sd_lun *un, struct buf *bp, void *arg, int flag)
18857 {
18858         struct sd_xbuf  *xp;
18859         struct scsi_pkt *pktp;
18860         char *reasonp;
18861         char *msgp;
18862 
18863         ASSERT(un != NULL);
18864         ASSERT(mutex_owned(SD_MUTEX(un)));
18865         ASSERT(bp != NULL);
18866         pktp = SD_GET_PKTP(bp);
18867         ASSERT(pktp != NULL);
18868         xp = SD_GET_XBUF(bp);
18869         ASSERT(xp != NULL);
18870 
18871         ASSERT(!mutex_owned(&un->un_pm_mutex));
18872         mutex_enter(&un->un_pm_mutex);
18873         if ((un->un_state == SD_STATE_SUSPENDED) ||
18874             (SD_DEVICE_IS_IN_LOW_POWER(un)) ||
18875             (pktp->pkt_flags & FLAG_SILENT)) {
18876                 mutex_exit(&un->un_pm_mutex);
18877                 goto update_pkt_reason;
18878         }
18879         mutex_exit(&un->un_pm_mutex);
18880 
18881         /*
18882          * Suppress messages if they are all the same pkt_reason; with
18883          * TQ, many (up to 256) are returned with the same pkt_reason.
18884          * If we are in panic, then suppress the retry messages.
18885          */
18886         switch (flag) {
18887         case SD_NO_RETRY_ISSUED:
18888                 msgp = "giving up";
18889                 break;
18890         case SD_IMMEDIATE_RETRY_ISSUED:
18891         case SD_DELAYED_RETRY_ISSUED:
18892                 if (ddi_in_panic() || (un->un_state == SD_STATE_OFFLINE) ||
18893                     ((pktp->pkt_reason == un->un_last_pkt_reason) &&
18894                     (sd_error_level != SCSI_ERR_ALL))) {
18895                         return;
18896                 }
18897                 msgp = "retrying command";
18898                 break;
18899         default:
18900                 goto update_pkt_reason;
18901         }
18902 
18903         reasonp = (((pktp->pkt_statistics & STAT_PERR) != 0) ? "parity error" :
18904             scsi_rname(pktp->pkt_reason));
18905 
18906         if (SD_FM_LOG(un) == SD_FM_LOG_NSUP) {
18907                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
18908                     "SCSI transport failed: reason '%s': %s\n", reasonp, msgp);
18909         }
18910 
18911 update_pkt_reason:
18912         /*
18913          * Update un->un_last_pkt_reason with the value in pktp->pkt_reason.
18914          * This is to prevent multiple console messages for the same failure
18915          * condition.  Note that un->un_last_pkt_reason is NOT restored if &
18916          * when the command is retried successfully because there still may be
18917          * more commands coming back with the same value of pktp->pkt_reason.
18918          */
18919         if ((pktp->pkt_reason != CMD_CMPLT) || (xp->xb_retry_count == 0)) {
18920                 un->un_last_pkt_reason = pktp->pkt_reason;
18921         }
18922 }
18923 
18924 
18925 /*
18926  *    Function: sd_print_cmd_incomplete_msg
18927  *
18928  * Description: Message logging fn. for a SCSA "CMD_INCOMPLETE" pkt_reason.
18929  *
18930  *   Arguments: un - ptr to associated softstate
18931  *              bp - ptr to buf(9S) for the command
18932  *              arg - passed to sd_print_retry_msg()
18933  *              code - SD_IMMEDIATE_RETRY_ISSUED, SD_DELAYED_RETRY_ISSUED,
18934  *                      or SD_NO_RETRY_ISSUED
18935  *
18936  *     Context: May be called from interrupt context
18937  */
18938 
18939 static void
18940 sd_print_cmd_incomplete_msg(struct sd_lun *un, struct buf *bp, void *arg,
18941         int code)
18942 {
18943         dev_info_t      *dip;
18944 
18945         ASSERT(un != NULL);
18946         ASSERT(mutex_owned(SD_MUTEX(un)));
18947         ASSERT(bp != NULL);
18948 
18949         switch (code) {
18950         case SD_NO_RETRY_ISSUED:
18951                 /* Command was failed. Someone turned off this target? */
18952                 if (un->un_state != SD_STATE_OFFLINE) {
18953                         /*
18954                          * Suppress message if we are detaching and
18955                          * device has been disconnected
18956                          * Note that DEVI_IS_DEVICE_REMOVED is a consolidation
18957                          * private interface and not part of the DDI
18958                          */
18959                         dip = un->un_sd->sd_dev;
18960                         if (!(DEVI_IS_DETACHING(dip) &&
18961                             DEVI_IS_DEVICE_REMOVED(dip))) {
18962                                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
18963                                 "disk not responding to selection\n");
18964                         }
18965                         New_state(un, SD_STATE_OFFLINE);
18966                 }
18967                 break;
18968 
18969         case SD_DELAYED_RETRY_ISSUED:
18970         case SD_IMMEDIATE_RETRY_ISSUED:
18971         default:
18972                 /* Command was successfully queued for retry */
18973                 sd_print_retry_msg(un, bp, arg, code);
18974                 break;
18975         }
18976 }
18977 
18978 
18979 /*
18980  *    Function: sd_pkt_reason_cmd_incomplete
18981  *
18982  * Description: Recovery actions for a SCSA "CMD_INCOMPLETE" pkt_reason.
18983  *
18984  *     Context: May be called from interrupt context
18985  */
18986 
18987 static void
18988 sd_pkt_reason_cmd_incomplete(struct sd_lun *un, struct buf *bp,
18989         struct sd_xbuf *xp, struct scsi_pkt *pktp)
18990 {
18991         int flag = SD_RETRIES_STANDARD | SD_RETRIES_ISOLATE;
18992 
18993         ASSERT(un != NULL);
18994         ASSERT(mutex_owned(SD_MUTEX(un)));
18995         ASSERT(bp != NULL);
18996         ASSERT(xp != NULL);
18997         ASSERT(pktp != NULL);
18998 
18999         /* Do not do a reset if selection did not complete */
19000         /* Note: Should this not just check the bit? */
19001         if (pktp->pkt_state != STATE_GOT_BUS) {
19002                 SD_UPDATE_ERRSTATS(un, sd_transerrs);
19003                 sd_reset_target(un, pktp);
19004         }
19005 
19006         /*
19007          * If the target was not successfully selected, then set
19008          * SD_RETRIES_FAILFAST to indicate that we lost communication
19009          * with the target, and further retries and/or commands are
19010          * likely to take a long time.
19011          */
19012         if ((pktp->pkt_state & STATE_GOT_TARGET) == 0) {
19013                 flag |= SD_RETRIES_FAILFAST;
19014         }
19015 
19016         SD_UPDATE_RESERVATION_STATUS(un, pktp);
19017 
19018         sd_retry_command(un, bp, flag,
19019             sd_print_cmd_incomplete_msg, NULL, EIO, SD_RESTART_TIMEOUT, NULL);
19020 }
19021 
19022 
19023 
19024 /*
19025  *    Function: sd_pkt_reason_cmd_tran_err
19026  *
19027  * Description: Recovery actions for a SCSA "CMD_TRAN_ERR" pkt_reason.
19028  *
19029  *     Context: May be called from interrupt context
19030  */
19031 
19032 static void
19033 sd_pkt_reason_cmd_tran_err(struct sd_lun *un, struct buf *bp,
19034         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19035 {
19036         ASSERT(un != NULL);
19037         ASSERT(mutex_owned(SD_MUTEX(un)));
19038         ASSERT(bp != NULL);
19039         ASSERT(xp != NULL);
19040         ASSERT(pktp != NULL);
19041 
19042         /*
19043          * Do not reset if we got a parity error, or if
19044          * selection did not complete.
19045          */
19046         SD_UPDATE_ERRSTATS(un, sd_harderrs);
19047         /* Note: Should this not just check the bit for pkt_state? */
19048         if (((pktp->pkt_statistics & STAT_PERR) == 0) &&
19049             (pktp->pkt_state != STATE_GOT_BUS)) {
19050                 SD_UPDATE_ERRSTATS(un, sd_transerrs);
19051                 sd_reset_target(un, pktp);
19052         }
19053 
19054         SD_UPDATE_RESERVATION_STATUS(un, pktp);
19055 
19056         sd_retry_command(un, bp, (SD_RETRIES_STANDARD | SD_RETRIES_ISOLATE),
19057             sd_print_retry_msg, NULL, EIO, SD_RESTART_TIMEOUT, NULL);
19058 }
19059 
19060 
19061 
19062 /*
19063  *    Function: sd_pkt_reason_cmd_reset
19064  *
19065  * Description: Recovery actions for a SCSA "CMD_RESET" pkt_reason.
19066  *
19067  *     Context: May be called from interrupt context
19068  */
19069 
19070 static void
19071 sd_pkt_reason_cmd_reset(struct sd_lun *un, struct buf *bp,
19072         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19073 {
19074         ASSERT(un != NULL);
19075         ASSERT(mutex_owned(SD_MUTEX(un)));
19076         ASSERT(bp != NULL);
19077         ASSERT(xp != NULL);
19078         ASSERT(pktp != NULL);
19079 
19080         /* The target may still be running the command, so try to reset. */
19081         SD_UPDATE_ERRSTATS(un, sd_transerrs);
19082         sd_reset_target(un, pktp);
19083 
19084         SD_UPDATE_RESERVATION_STATUS(un, pktp);
19085 
19086         /*
19087          * If pkt_reason is CMD_RESET chances are that this pkt got
19088          * reset because another target on this bus caused it. The target
19089          * that caused it should get CMD_TIMEOUT with pkt_statistics
19090          * of STAT_TIMEOUT/STAT_DEV_RESET.
19091          */
19092 
19093         sd_retry_command(un, bp, (SD_RETRIES_VICTIM | SD_RETRIES_ISOLATE),
19094             sd_print_retry_msg, NULL, EIO, SD_RESTART_TIMEOUT, NULL);
19095 }
19096 
19097 
19098 
19099 
19100 /*
19101  *    Function: sd_pkt_reason_cmd_aborted
19102  *
19103  * Description: Recovery actions for a SCSA "CMD_ABORTED" pkt_reason.
19104  *
19105  *     Context: May be called from interrupt context
19106  */
19107 
19108 static void
19109 sd_pkt_reason_cmd_aborted(struct sd_lun *un, struct buf *bp,
19110         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19111 {
19112         ASSERT(un != NULL);
19113         ASSERT(mutex_owned(SD_MUTEX(un)));
19114         ASSERT(bp != NULL);
19115         ASSERT(xp != NULL);
19116         ASSERT(pktp != NULL);
19117 
19118         /* The target may still be running the command, so try to reset. */
19119         SD_UPDATE_ERRSTATS(un, sd_transerrs);
19120         sd_reset_target(un, pktp);
19121 
19122         SD_UPDATE_RESERVATION_STATUS(un, pktp);
19123 
19124         /*
19125          * If pkt_reason is CMD_ABORTED chances are that this pkt got
19126          * aborted because another target on this bus caused it. The target
19127          * that caused it should get CMD_TIMEOUT with pkt_statistics
19128          * of STAT_TIMEOUT/STAT_DEV_RESET.
19129          */
19130 
19131         sd_retry_command(un, bp, (SD_RETRIES_VICTIM | SD_RETRIES_ISOLATE),
19132             sd_print_retry_msg, NULL, EIO, SD_RESTART_TIMEOUT, NULL);
19133 }
19134 
19135 
19136 
19137 /*
19138  *    Function: sd_pkt_reason_cmd_timeout
19139  *
19140  * Description: Recovery actions for a SCSA "CMD_TIMEOUT" pkt_reason.
19141  *
19142  *     Context: May be called from interrupt context
19143  */
19144 
19145 static void
19146 sd_pkt_reason_cmd_timeout(struct sd_lun *un, struct buf *bp,
19147         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19148 {
19149         ASSERT(un != NULL);
19150         ASSERT(mutex_owned(SD_MUTEX(un)));
19151         ASSERT(bp != NULL);
19152         ASSERT(xp != NULL);
19153         ASSERT(pktp != NULL);
19154 
19155 
19156         SD_UPDATE_ERRSTATS(un, sd_transerrs);
19157         sd_reset_target(un, pktp);
19158 
19159         SD_UPDATE_RESERVATION_STATUS(un, pktp);
19160 
19161         /*
19162          * A command timeout indicates that we could not establish
19163          * communication with the target, so set SD_RETRIES_FAILFAST
19164          * as further retries/commands are likely to take a long time.
19165          */
19166         sd_retry_command(un, bp,
19167             (SD_RETRIES_STANDARD | SD_RETRIES_ISOLATE | SD_RETRIES_FAILFAST),
19168             sd_print_retry_msg, NULL, EIO, SD_RESTART_TIMEOUT, NULL);
19169 }
19170 
19171 
19172 
19173 /*
19174  *    Function: sd_pkt_reason_cmd_unx_bus_free
19175  *
19176  * Description: Recovery actions for a SCSA "CMD_UNX_BUS_FREE" pkt_reason.
19177  *
19178  *     Context: May be called from interrupt context
19179  */
19180 
19181 static void
19182 sd_pkt_reason_cmd_unx_bus_free(struct sd_lun *un, struct buf *bp,
19183         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19184 {
19185         void (*funcp)(struct sd_lun *un, struct buf *bp, void *arg, int code);
19186 
19187         ASSERT(un != NULL);
19188         ASSERT(mutex_owned(SD_MUTEX(un)));
19189         ASSERT(bp != NULL);
19190         ASSERT(xp != NULL);
19191         ASSERT(pktp != NULL);
19192 
19193         SD_UPDATE_ERRSTATS(un, sd_harderrs);
19194         SD_UPDATE_RESERVATION_STATUS(un, pktp);
19195 
19196         funcp = ((pktp->pkt_statistics & STAT_PERR) == 0) ?
19197             sd_print_retry_msg : NULL;
19198 
19199         sd_retry_command(un, bp, (SD_RETRIES_STANDARD | SD_RETRIES_ISOLATE),
19200             funcp, NULL, EIO, SD_RESTART_TIMEOUT, NULL);
19201 }
19202 
19203 
19204 /*
19205  *    Function: sd_pkt_reason_cmd_tag_reject
19206  *
19207  * Description: Recovery actions for a SCSA "CMD_TAG_REJECT" pkt_reason.
19208  *
19209  *     Context: May be called from interrupt context
19210  */
19211 
19212 static void
19213 sd_pkt_reason_cmd_tag_reject(struct sd_lun *un, struct buf *bp,
19214         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19215 {
19216         ASSERT(un != NULL);
19217         ASSERT(mutex_owned(SD_MUTEX(un)));
19218         ASSERT(bp != NULL);
19219         ASSERT(xp != NULL);
19220         ASSERT(pktp != NULL);
19221 
19222         SD_UPDATE_ERRSTATS(un, sd_harderrs);
19223         pktp->pkt_flags = 0;
19224         un->un_tagflags = 0;
19225         if (un->un_f_opt_queueing == TRUE) {
19226                 un->un_throttle = min(un->un_throttle, 3);
19227         } else {
19228                 un->un_throttle = 1;
19229         }
19230         mutex_exit(SD_MUTEX(un));
19231         (void) scsi_ifsetcap(SD_ADDRESS(un), "tagged-qing", 0, 1);
19232         mutex_enter(SD_MUTEX(un));
19233 
19234         SD_UPDATE_RESERVATION_STATUS(un, pktp);
19235 
19236         /* Legacy behavior not to check retry counts here. */
19237         sd_retry_command(un, bp, (SD_RETRIES_NOCHECK | SD_RETRIES_ISOLATE),
19238             sd_print_retry_msg, NULL, EIO, SD_RESTART_TIMEOUT, NULL);
19239 }
19240 
19241 
19242 /*
19243  *    Function: sd_pkt_reason_default
19244  *
19245  * Description: Default recovery actions for SCSA pkt_reason values that
19246  *              do not have more explicit recovery actions.
19247  *
19248  *     Context: May be called from interrupt context
19249  */
19250 
19251 static void
19252 sd_pkt_reason_default(struct sd_lun *un, struct buf *bp,
19253         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19254 {
19255         ASSERT(un != NULL);
19256         ASSERT(mutex_owned(SD_MUTEX(un)));
19257         ASSERT(bp != NULL);
19258         ASSERT(xp != NULL);
19259         ASSERT(pktp != NULL);
19260 
19261         SD_UPDATE_ERRSTATS(un, sd_transerrs);
19262         sd_reset_target(un, pktp);
19263 
19264         SD_UPDATE_RESERVATION_STATUS(un, pktp);
19265 
19266         sd_retry_command(un, bp, (SD_RETRIES_STANDARD | SD_RETRIES_ISOLATE),
19267             sd_print_retry_msg, NULL, EIO, SD_RESTART_TIMEOUT, NULL);
19268 }
19269 
19270 
19271 
19272 /*
19273  *    Function: sd_pkt_status_check_condition
19274  *
19275  * Description: Recovery actions for a "STATUS_CHECK" SCSI command status.
19276  *
19277  *     Context: May be called from interrupt context
19278  */
19279 
19280 static void
19281 sd_pkt_status_check_condition(struct sd_lun *un, struct buf *bp,
19282         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19283 {
19284         ASSERT(un != NULL);
19285         ASSERT(mutex_owned(SD_MUTEX(un)));
19286         ASSERT(bp != NULL);
19287         ASSERT(xp != NULL);
19288         ASSERT(pktp != NULL);
19289 
19290         SD_TRACE(SD_LOG_IO, un, "sd_pkt_status_check_condition: "
19291             "entry: buf:0x%p xp:0x%p\n", bp, xp);
19292 
19293         /*
19294          * If ARQ is NOT enabled, then issue a REQUEST SENSE command (the
19295          * command will be retried after the request sense). Otherwise, retry
19296          * the command. Note: we are issuing the request sense even though the
19297          * retry limit may have been reached for the failed command.
19298          */
19299         if (un->un_f_arq_enabled == FALSE) {
19300                 SD_INFO(SD_LOG_IO_CORE, un, "sd_pkt_status_check_condition: "
19301                     "no ARQ, sending request sense command\n");
19302                 sd_send_request_sense_command(un, bp, pktp);
19303         } else {
19304                 SD_INFO(SD_LOG_IO_CORE, un, "sd_pkt_status_check_condition: "
19305                     "ARQ,retrying request sense command\n");
19306 #if defined(__i386) || defined(__amd64)
19307                 /*
19308                  * The SD_RETRY_DELAY value need to be adjusted here
19309                  * when SD_RETRY_DELAY change in sddef.h
19310                  */
19311                 sd_retry_command(un, bp, SD_RETRIES_STANDARD, NULL, NULL, EIO,
19312                     un->un_f_is_fibre?drv_usectohz(100000):(clock_t)0,
19313                     NULL);
19314 #else
19315                 sd_retry_command(un, bp, SD_RETRIES_STANDARD, NULL, NULL,
19316                     EIO, SD_RETRY_DELAY, NULL);
19317 #endif
19318         }
19319 
19320         SD_TRACE(SD_LOG_IO_CORE, un, "sd_pkt_status_check_condition: exit\n");
19321 }
19322 
19323 
19324 /*
19325  *    Function: sd_pkt_status_busy
19326  *
19327  * Description: Recovery actions for a "STATUS_BUSY" SCSI command status.
19328  *
19329  *     Context: May be called from interrupt context
19330  */
19331 
19332 static void
19333 sd_pkt_status_busy(struct sd_lun *un, struct buf *bp, struct sd_xbuf *xp,
19334         struct scsi_pkt *pktp)
19335 {
19336         ASSERT(un != NULL);
19337         ASSERT(mutex_owned(SD_MUTEX(un)));
19338         ASSERT(bp != NULL);
19339         ASSERT(xp != NULL);
19340         ASSERT(pktp != NULL);
19341 
19342         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19343             "sd_pkt_status_busy: entry\n");
19344 
19345         /* If retries are exhausted, just fail the command. */
19346         if (xp->xb_retry_count >= un->un_busy_retry_count) {
19347                 scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
19348                     "device busy too long\n");
19349                 sd_return_failed_command(un, bp, EIO);
19350                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19351                     "sd_pkt_status_busy: exit\n");
19352                 return;
19353         }
19354         xp->xb_retry_count++;
19355 
19356         /*
19357          * Try to reset the target. However, we do not want to perform
19358          * more than one reset if the device continues to fail. The reset
19359          * will be performed when the retry count reaches the reset
19360          * threshold.  This threshold should be set such that at least
19361          * one retry is issued before the reset is performed.
19362          */
19363         if (xp->xb_retry_count ==
19364             ((un->un_reset_retry_count < 2) ? 2 : un->un_reset_retry_count)) {
19365                 int rval = 0;
19366                 mutex_exit(SD_MUTEX(un));
19367                 if (un->un_f_allow_bus_device_reset == TRUE) {
19368                         /*
19369                          * First try to reset the LUN; if we cannot then
19370                          * try to reset the target.
19371                          */
19372                         if (un->un_f_lun_reset_enabled == TRUE) {
19373                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19374                                     "sd_pkt_status_busy: RESET_LUN\n");
19375                                 rval = scsi_reset(SD_ADDRESS(un), RESET_LUN);
19376                         }
19377                         if (rval == 0) {
19378                                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19379                                     "sd_pkt_status_busy: RESET_TARGET\n");
19380                                 rval = scsi_reset(SD_ADDRESS(un), RESET_TARGET);
19381                         }
19382                 }
19383                 if (rval == 0) {
19384                         /*
19385                          * If the RESET_LUN and/or RESET_TARGET failed,
19386                          * try RESET_ALL
19387                          */
19388                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19389                             "sd_pkt_status_busy: RESET_ALL\n");
19390                         rval = scsi_reset(SD_ADDRESS(un), RESET_ALL);
19391                 }
19392                 mutex_enter(SD_MUTEX(un));
19393                 if (rval == 0) {
19394                         /*
19395                          * The RESET_LUN, RESET_TARGET, and/or RESET_ALL failed.
19396                          * At this point we give up & fail the command.
19397                          */
19398                         sd_return_failed_command(un, bp, EIO);
19399                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19400                             "sd_pkt_status_busy: exit (failed cmd)\n");
19401                         return;
19402                 }
19403         }
19404 
19405         /*
19406          * Retry the command. Be sure to specify SD_RETRIES_NOCHECK as
19407          * we have already checked the retry counts above.
19408          */
19409         sd_retry_command(un, bp, SD_RETRIES_NOCHECK, NULL, NULL,
19410             EIO, un->un_busy_timeout, NULL);
19411 
19412         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19413             "sd_pkt_status_busy: exit\n");
19414 }
19415 
19416 
19417 /*
19418  *    Function: sd_pkt_status_reservation_conflict
19419  *
19420  * Description: Recovery actions for a "STATUS_RESERVATION_CONFLICT" SCSI
19421  *              command status.
19422  *
19423  *     Context: May be called from interrupt context
19424  */
19425 
19426 static void
19427 sd_pkt_status_reservation_conflict(struct sd_lun *un, struct buf *bp,
19428         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19429 {
19430         ASSERT(un != NULL);
19431         ASSERT(mutex_owned(SD_MUTEX(un)));
19432         ASSERT(bp != NULL);
19433         ASSERT(xp != NULL);
19434         ASSERT(pktp != NULL);
19435 
19436         /*
19437          * If the command was PERSISTENT_RESERVATION_[IN|OUT] then reservation
19438          * conflict could be due to various reasons like incorrect keys, not
19439          * registered or not reserved etc. So, we return EACCES to the caller.
19440          */
19441         if (un->un_reservation_type == SD_SCSI3_RESERVATION) {
19442                 int cmd = SD_GET_PKT_OPCODE(pktp);
19443                 if ((cmd == SCMD_PERSISTENT_RESERVE_IN) ||
19444                     (cmd == SCMD_PERSISTENT_RESERVE_OUT)) {
19445                         sd_return_failed_command(un, bp, EACCES);
19446                         return;
19447                 }
19448         }
19449 
19450         un->un_resvd_status |= SD_RESERVATION_CONFLICT;
19451 
19452         if ((un->un_resvd_status & SD_FAILFAST) != 0) {
19453                 if (sd_failfast_enable != 0) {
19454                         /* By definition, we must panic here.... */
19455                         sd_panic_for_res_conflict(un);
19456                         /*NOTREACHED*/
19457                 }
19458                 SD_ERROR(SD_LOG_IO, un,
19459                     "sd_handle_resv_conflict: Disk Reserved\n");
19460                 sd_return_failed_command(un, bp, EACCES);
19461                 return;
19462         }
19463 
19464         /*
19465          * 1147670: retry only if sd_retry_on_reservation_conflict
19466          * property is set (default is 1). Retries will not succeed
19467          * on a disk reserved by another initiator. HA systems
19468          * may reset this via sd.conf to avoid these retries.
19469          *
19470          * Note: The legacy return code for this failure is EIO, however EACCES
19471          * seems more appropriate for a reservation conflict.
19472          */
19473         if (sd_retry_on_reservation_conflict == 0) {
19474                 SD_ERROR(SD_LOG_IO, un,
19475                     "sd_handle_resv_conflict: Device Reserved\n");
19476                 sd_return_failed_command(un, bp, EIO);
19477                 return;
19478         }
19479 
19480         /*
19481          * Retry the command if we can.
19482          *
19483          * Note: The legacy return code for this failure is EIO, however EACCES
19484          * seems more appropriate for a reservation conflict.
19485          */
19486         sd_retry_command(un, bp, SD_RETRIES_STANDARD, NULL, NULL, EIO,
19487             (clock_t)2, NULL);
19488 }
19489 
19490 
19491 
19492 /*
19493  *    Function: sd_pkt_status_qfull
19494  *
19495  * Description: Handle a QUEUE FULL condition from the target.  This can
19496  *              occur if the HBA does not handle the queue full condition.
19497  *              (Basically this means third-party HBAs as Sun HBAs will
19498  *              handle the queue full condition.)  Note that if there are
19499  *              some commands already in the transport, then the queue full
19500  *              has occurred because the queue for this nexus is actually
19501  *              full. If there are no commands in the transport, then the
19502  *              queue full is resulting from some other initiator or lun
19503  *              consuming all the resources at the target.
19504  *
19505  *     Context: May be called from interrupt context
19506  */
19507 
19508 static void
19509 sd_pkt_status_qfull(struct sd_lun *un, struct buf *bp,
19510         struct sd_xbuf *xp, struct scsi_pkt *pktp)
19511 {
19512         ASSERT(un != NULL);
19513         ASSERT(mutex_owned(SD_MUTEX(un)));
19514         ASSERT(bp != NULL);
19515         ASSERT(xp != NULL);
19516         ASSERT(pktp != NULL);
19517 
19518         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19519             "sd_pkt_status_qfull: entry\n");
19520 
19521         /*
19522          * Just lower the QFULL throttle and retry the command.  Note that
19523          * we do not limit the number of retries here.
19524          */
19525         sd_reduce_throttle(un, SD_THROTTLE_QFULL);
19526         sd_retry_command(un, bp, SD_RETRIES_NOCHECK, NULL, NULL, 0,
19527             SD_RESTART_TIMEOUT, NULL);
19528 
19529         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19530             "sd_pkt_status_qfull: exit\n");
19531 }
19532 
19533 
19534 /*
19535  *    Function: sd_reset_target
19536  *
19537  * Description: Issue a scsi_reset(9F), with either RESET_LUN,
19538  *              RESET_TARGET, or RESET_ALL.
19539  *
19540  *     Context: May be called under interrupt context.
19541  */
19542 
19543 static void
19544 sd_reset_target(struct sd_lun *un, struct scsi_pkt *pktp)
19545 {
19546         int rval = 0;
19547 
19548         ASSERT(un != NULL);
19549         ASSERT(mutex_owned(SD_MUTEX(un)));
19550         ASSERT(pktp != NULL);
19551 
19552         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_reset_target: entry\n");
19553 
19554         /*
19555          * No need to reset if the transport layer has already done so.
19556          */
19557         if ((pktp->pkt_statistics &
19558             (STAT_BUS_RESET | STAT_DEV_RESET | STAT_ABORTED)) != 0) {
19559                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19560                     "sd_reset_target: no reset\n");
19561                 return;
19562         }
19563 
19564         mutex_exit(SD_MUTEX(un));
19565 
19566         if (un->un_f_allow_bus_device_reset == TRUE) {
19567                 if (un->un_f_lun_reset_enabled == TRUE) {
19568                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19569                             "sd_reset_target: RESET_LUN\n");
19570                         rval = scsi_reset(SD_ADDRESS(un), RESET_LUN);
19571                 }
19572                 if (rval == 0) {
19573                         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19574                             "sd_reset_target: RESET_TARGET\n");
19575                         rval = scsi_reset(SD_ADDRESS(un), RESET_TARGET);
19576                 }
19577         }
19578 
19579         if (rval == 0) {
19580                 SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
19581                     "sd_reset_target: RESET_ALL\n");
19582                 (void) scsi_reset(SD_ADDRESS(un), RESET_ALL);
19583         }
19584 
19585         mutex_enter(SD_MUTEX(un));
19586 
19587         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un, "sd_reset_target: exit\n");
19588 }
19589 
19590 /*
19591  *    Function: sd_target_change_task
19592  *
19593  * Description: Handle dynamic target change
19594  *
19595  *     Context: Executes in a taskq() thread context
19596  */
19597 static void
19598 sd_target_change_task(void *arg)
19599 {
19600         struct sd_lun           *un = arg;
19601         uint64_t                capacity;
19602         diskaddr_t              label_cap;
19603         uint_t                  lbasize;
19604         sd_ssc_t                *ssc;
19605 
19606         ASSERT(un != NULL);
19607         ASSERT(!mutex_owned(SD_MUTEX(un)));
19608 
19609         if ((un->un_f_blockcount_is_valid == FALSE) ||
19610             (un->un_f_tgt_blocksize_is_valid == FALSE)) {
19611                 return;
19612         }
19613 
19614         ssc = sd_ssc_init(un);
19615 
19616         if (sd_send_scsi_READ_CAPACITY(ssc, &capacity,
19617             &lbasize, SD_PATH_DIRECT) != 0) {
19618                 SD_ERROR(SD_LOG_ERROR, un,
19619                     "sd_target_change_task: fail to read capacity\n");
19620                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
19621                 goto task_exit;
19622         }
19623 
19624         mutex_enter(SD_MUTEX(un));
19625         if (capacity <= un->un_blockcount) {
19626                 mutex_exit(SD_MUTEX(un));
19627                 goto task_exit;
19628         }
19629 
19630         sd_update_block_info(un, lbasize, capacity);
19631         mutex_exit(SD_MUTEX(un));
19632 
19633         /*
19634          * If lun is EFI labeled and lun capacity is greater than the
19635          * capacity contained in the label, log a sys event.
19636          */
19637         if (cmlb_efi_label_capacity(un->un_cmlbhandle, &label_cap,
19638             (void*)SD_PATH_DIRECT) == 0) {
19639                 mutex_enter(SD_MUTEX(un));
19640                 if (un->un_f_blockcount_is_valid &&
19641                     un->un_blockcount > label_cap) {
19642                         mutex_exit(SD_MUTEX(un));
19643                         sd_log_lun_expansion_event(un, KM_SLEEP);
19644                 } else {
19645                         mutex_exit(SD_MUTEX(un));
19646                 }
19647         }
19648 
19649 task_exit:
19650         sd_ssc_fini(ssc);
19651 }
19652 
19653 
19654 /*
19655  *    Function: sd_log_dev_status_event
19656  *
19657  * Description: Log EC_dev_status sysevent
19658  *
19659  *     Context: Never called from interrupt context
19660  */
19661 static void
19662 sd_log_dev_status_event(struct sd_lun *un, char *esc, int km_flag)
19663 {
19664         int err;
19665         char                    *path;
19666         nvlist_t                *attr_list;
19667 
19668         /* Allocate and build sysevent attribute list */
19669         err = nvlist_alloc(&attr_list, NV_UNIQUE_NAME_TYPE, km_flag);
19670         if (err != 0) {
19671                 SD_ERROR(SD_LOG_ERROR, un,
19672                     "sd_log_dev_status_event: fail to allocate space\n");
19673                 return;
19674         }
19675 
19676         path = kmem_alloc(MAXPATHLEN, km_flag);
19677         if (path == NULL) {
19678                 nvlist_free(attr_list);
19679                 SD_ERROR(SD_LOG_ERROR, un,
19680                     "sd_log_dev_status_event: fail to allocate space\n");
19681                 return;
19682         }
19683         /*
19684          * Add path attribute to identify the lun.
19685          * We are using minor node 'a' as the sysevent attribute.
19686          */
19687         (void) snprintf(path, MAXPATHLEN, "/devices");
19688         (void) ddi_pathname(SD_DEVINFO(un), path + strlen(path));
19689         (void) snprintf(path + strlen(path), MAXPATHLEN - strlen(path),
19690             ":a");
19691 
19692         err = nvlist_add_string(attr_list, DEV_PHYS_PATH, path);
19693         if (err != 0) {
19694                 nvlist_free(attr_list);
19695                 kmem_free(path, MAXPATHLEN);
19696                 SD_ERROR(SD_LOG_ERROR, un,
19697                     "sd_log_dev_status_event: fail to add attribute\n");
19698                 return;
19699         }
19700 
19701         /* Log dynamic lun expansion sysevent */
19702         err = ddi_log_sysevent(SD_DEVINFO(un), SUNW_VENDOR, EC_DEV_STATUS,
19703             esc, attr_list, NULL, km_flag);
19704         if (err != DDI_SUCCESS) {
19705                 SD_ERROR(SD_LOG_ERROR, un,
19706                     "sd_log_dev_status_event: fail to log sysevent\n");
19707         }
19708 
19709         nvlist_free(attr_list);
19710         kmem_free(path, MAXPATHLEN);
19711 }
19712 
19713 
19714 /*
19715  *    Function: sd_log_lun_expansion_event
19716  *
19717  * Description: Log lun expansion sys event
19718  *
19719  *     Context: Never called from interrupt context
19720  */
19721 static void
19722 sd_log_lun_expansion_event(struct sd_lun *un, int km_flag)
19723 {
19724         sd_log_dev_status_event(un, ESC_DEV_DLE, km_flag);
19725 }
19726 
19727 
19728 /*
19729  *    Function: sd_log_eject_request_event
19730  *
19731  * Description: Log eject request sysevent
19732  *
19733  *     Context: Never called from interrupt context
19734  */
19735 static void
19736 sd_log_eject_request_event(struct sd_lun *un, int km_flag)
19737 {
19738         sd_log_dev_status_event(un, ESC_DEV_EJECT_REQUEST, km_flag);
19739 }
19740 
19741 
19742 /*
19743  *    Function: sd_media_change_task
19744  *
19745  * Description: Recovery action for CDROM to become available.
19746  *
19747  *     Context: Executes in a taskq() thread context
19748  */
19749 
19750 static void
19751 sd_media_change_task(void *arg)
19752 {
19753         struct  scsi_pkt        *pktp = arg;
19754         struct  sd_lun          *un;
19755         struct  buf             *bp;
19756         struct  sd_xbuf         *xp;
19757         int     err             = 0;
19758         int     retry_count     = 0;
19759         int     retry_limit     = SD_UNIT_ATTENTION_RETRY/10;
19760         struct  sd_sense_info   si;
19761 
19762         ASSERT(pktp != NULL);
19763         bp = (struct buf *)pktp->pkt_private;
19764         ASSERT(bp != NULL);
19765         xp = SD_GET_XBUF(bp);
19766         ASSERT(xp != NULL);
19767         un = SD_GET_UN(bp);
19768         ASSERT(un != NULL);
19769         ASSERT(!mutex_owned(SD_MUTEX(un)));
19770         ASSERT(un->un_f_monitor_media_state);
19771 
19772         si.ssi_severity = SCSI_ERR_INFO;
19773         si.ssi_pfa_flag = FALSE;
19774 
19775         /*
19776          * When a reset is issued on a CDROM, it takes a long time to
19777          * recover. First few attempts to read capacity and other things
19778          * related to handling unit attention fail (with a ASC 0x4 and
19779          * ASCQ 0x1). In that case we want to do enough retries and we want
19780          * to limit the retries in other cases of genuine failures like
19781          * no media in drive.
19782          */
19783         while (retry_count++ < retry_limit) {
19784                 if ((err = sd_handle_mchange(un)) == 0) {
19785                         break;
19786                 }
19787                 if (err == EAGAIN) {
19788                         retry_limit = SD_UNIT_ATTENTION_RETRY;
19789                 }
19790                 /* Sleep for 0.5 sec. & try again */
19791                 delay(drv_usectohz(500000));
19792         }
19793 
19794         /*
19795          * Dispatch (retry or fail) the original command here,
19796          * along with appropriate console messages....
19797          *
19798          * Must grab the mutex before calling sd_retry_command,
19799          * sd_print_sense_msg and sd_return_failed_command.
19800          */
19801         mutex_enter(SD_MUTEX(un));
19802         if (err != SD_CMD_SUCCESS) {
19803                 SD_UPDATE_ERRSTATS(un, sd_harderrs);
19804                 SD_UPDATE_ERRSTATS(un, sd_rq_nodev_err);
19805                 si.ssi_severity = SCSI_ERR_FATAL;
19806                 sd_print_sense_msg(un, bp, &si, SD_NO_RETRY_ISSUED);
19807                 sd_return_failed_command(un, bp, EIO);
19808         } else {
19809                 sd_retry_command(un, bp, SD_RETRIES_UA, sd_print_sense_msg,
19810                     &si, EIO, (clock_t)0, NULL);
19811         }
19812         mutex_exit(SD_MUTEX(un));
19813 }
19814 
19815 
19816 
19817 /*
19818  *    Function: sd_handle_mchange
19819  *
19820  * Description: Perform geometry validation & other recovery when CDROM
19821  *              has been removed from drive.
19822  *
19823  * Return Code: 0 for success
19824  *              errno-type return code of either sd_send_scsi_DOORLOCK() or
19825  *              sd_send_scsi_READ_CAPACITY()
19826  *
19827  *     Context: Executes in a taskq() thread context
19828  */
19829 
19830 static int
19831 sd_handle_mchange(struct sd_lun *un)
19832 {
19833         uint64_t        capacity;
19834         uint32_t        lbasize;
19835         int             rval;
19836         sd_ssc_t        *ssc;
19837 
19838         ASSERT(!mutex_owned(SD_MUTEX(un)));
19839         ASSERT(un->un_f_monitor_media_state);
19840 
19841         ssc = sd_ssc_init(un);
19842         rval = sd_send_scsi_READ_CAPACITY(ssc, &capacity, &lbasize,
19843             SD_PATH_DIRECT_PRIORITY);
19844 
19845         if (rval != 0)
19846                 goto failed;
19847 
19848         mutex_enter(SD_MUTEX(un));
19849         sd_update_block_info(un, lbasize, capacity);
19850 
19851         if (un->un_errstats != NULL) {
19852                 struct  sd_errstats *stp =
19853                     (struct sd_errstats *)un->un_errstats->ks_data;
19854                 stp->sd_capacity.value.ui64 = (uint64_t)
19855                     ((uint64_t)un->un_blockcount *
19856                     (uint64_t)un->un_tgt_blocksize);
19857         }
19858 
19859         /*
19860          * Check if the media in the device is writable or not
19861          */
19862         if (ISCD(un)) {
19863                 sd_check_for_writable_cd(ssc, SD_PATH_DIRECT_PRIORITY);
19864         }
19865 
19866         /*
19867          * Note: Maybe let the strategy/partitioning chain worry about getting
19868          * valid geometry.
19869          */
19870         mutex_exit(SD_MUTEX(un));
19871         cmlb_invalidate(un->un_cmlbhandle, (void *)SD_PATH_DIRECT_PRIORITY);
19872 
19873 
19874         if (cmlb_validate(un->un_cmlbhandle, 0,
19875             (void *)SD_PATH_DIRECT_PRIORITY) != 0) {
19876                 sd_ssc_fini(ssc);
19877                 return (EIO);
19878         } else {
19879                 if (un->un_f_pkstats_enabled) {
19880                         sd_set_pstats(un);
19881                         SD_TRACE(SD_LOG_IO_PARTITION, un,
19882                             "sd_handle_mchange: un:0x%p pstats created and "
19883                             "set\n", un);
19884                 }
19885         }
19886 
19887         /*
19888          * Try to lock the door
19889          */
19890         rval = sd_send_scsi_DOORLOCK(ssc, SD_REMOVAL_PREVENT,
19891             SD_PATH_DIRECT_PRIORITY);
19892 failed:
19893         if (rval != 0)
19894                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
19895         sd_ssc_fini(ssc);
19896         return (rval);
19897 }
19898 
19899 
19900 /*
19901  *    Function: sd_send_scsi_DOORLOCK
19902  *
19903  * Description: Issue the scsi DOOR LOCK command
19904  *
19905  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
19906  *                      structure for this target.
19907  *              flag  - SD_REMOVAL_ALLOW
19908  *                      SD_REMOVAL_PREVENT
19909  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
19910  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
19911  *                      to use the USCSI "direct" chain and bypass the normal
19912  *                      command waitq. SD_PATH_DIRECT_PRIORITY is used when this
19913  *                      command is issued as part of an error recovery action.
19914  *
19915  * Return Code: 0   - Success
19916  *              errno return code from sd_ssc_send()
19917  *
19918  *     Context: Can sleep.
19919  */
19920 
19921 static int
19922 sd_send_scsi_DOORLOCK(sd_ssc_t *ssc, int flag, int path_flag)
19923 {
19924         struct scsi_extended_sense      sense_buf;
19925         union scsi_cdb          cdb;
19926         struct uscsi_cmd        ucmd_buf;
19927         int                     status;
19928         struct sd_lun           *un;
19929 
19930         ASSERT(ssc != NULL);
19931         un = ssc->ssc_un;
19932         ASSERT(un != NULL);
19933         ASSERT(!mutex_owned(SD_MUTEX(un)));
19934 
19935         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_DOORLOCK: entry: un:0x%p\n", un);
19936 
19937         /* already determined doorlock is not supported, fake success */
19938         if (un->un_f_doorlock_supported == FALSE) {
19939                 return (0);
19940         }
19941 
19942         /*
19943          * If we are ejecting and see an SD_REMOVAL_PREVENT
19944          * ignore the command so we can complete the eject
19945          * operation.
19946          */
19947         if (flag == SD_REMOVAL_PREVENT) {
19948                 mutex_enter(SD_MUTEX(un));
19949                 if (un->un_f_ejecting == TRUE) {
19950                         mutex_exit(SD_MUTEX(un));
19951                         return (EAGAIN);
19952                 }
19953                 mutex_exit(SD_MUTEX(un));
19954         }
19955 
19956         bzero(&cdb, sizeof (cdb));
19957         bzero(&ucmd_buf, sizeof (ucmd_buf));
19958 
19959         cdb.scc_cmd = SCMD_DOORLOCK;
19960         cdb.cdb_opaque[4] = (uchar_t)flag;
19961 
19962         ucmd_buf.uscsi_cdb      = (char *)&cdb;
19963         ucmd_buf.uscsi_cdblen   = CDB_GROUP0;
19964         ucmd_buf.uscsi_bufaddr  = NULL;
19965         ucmd_buf.uscsi_buflen   = 0;
19966         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
19967         ucmd_buf.uscsi_rqlen    = sizeof (sense_buf);
19968         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_SILENT;
19969         ucmd_buf.uscsi_timeout  = 15;
19970 
19971         SD_TRACE(SD_LOG_IO, un,
19972             "sd_send_scsi_DOORLOCK: returning sd_ssc_send\n");
19973 
19974         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
19975             UIO_SYSSPACE, path_flag);
19976 
19977         if (status == 0)
19978                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
19979 
19980         if ((status == EIO) && (ucmd_buf.uscsi_status == STATUS_CHECK) &&
19981             (ucmd_buf.uscsi_rqstatus == STATUS_GOOD) &&
19982             (scsi_sense_key((uint8_t *)&sense_buf) == KEY_ILLEGAL_REQUEST)) {
19983                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
19984 
19985                 /* fake success and skip subsequent doorlock commands */
19986                 un->un_f_doorlock_supported = FALSE;
19987                 return (0);
19988         }
19989 
19990         return (status);
19991 }
19992 
19993 /*
19994  *    Function: sd_send_scsi_READ_CAPACITY
19995  *
19996  * Description: This routine uses the scsi READ CAPACITY command to determine
19997  *              the device capacity in number of blocks and the device native
19998  *              block size. If this function returns a failure, then the
19999  *              values in *capp and *lbap are undefined.  If the capacity
20000  *              returned is 0xffffffff then the lun is too large for a
20001  *              normal READ CAPACITY command and the results of a
20002  *              READ CAPACITY 16 will be used instead.
20003  *
20004  *   Arguments: ssc   - ssc contains ptr to soft state struct for the target
20005  *              capp - ptr to unsigned 64-bit variable to receive the
20006  *                      capacity value from the command.
20007  *              lbap - ptr to unsigned 32-bit varaible to receive the
20008  *                      block size value from the command
20009  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
20010  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
20011  *                      to use the USCSI "direct" chain and bypass the normal
20012  *                      command waitq. SD_PATH_DIRECT_PRIORITY is used when this
20013  *                      command is issued as part of an error recovery action.
20014  *
20015  * Return Code: 0   - Success
20016  *              EIO - IO error
20017  *              EACCES - Reservation conflict detected
20018  *              EAGAIN - Device is becoming ready
20019  *              errno return code from sd_ssc_send()
20020  *
20021  *     Context: Can sleep.  Blocks until command completes.
20022  */
20023 
20024 #define SD_CAPACITY_SIZE        sizeof (struct scsi_capacity)
20025 
20026 static int
20027 sd_send_scsi_READ_CAPACITY(sd_ssc_t *ssc, uint64_t *capp, uint32_t *lbap,
20028         int path_flag)
20029 {
20030         struct  scsi_extended_sense     sense_buf;
20031         struct  uscsi_cmd       ucmd_buf;
20032         union   scsi_cdb        cdb;
20033         uint32_t                *capacity_buf;
20034         uint64_t                capacity;
20035         uint32_t                lbasize;
20036         uint32_t                pbsize;
20037         int                     status;
20038         struct sd_lun           *un;
20039 
20040         ASSERT(ssc != NULL);
20041 
20042         un = ssc->ssc_un;
20043         ASSERT(un != NULL);
20044         ASSERT(!mutex_owned(SD_MUTEX(un)));
20045         ASSERT(capp != NULL);
20046         ASSERT(lbap != NULL);
20047 
20048         SD_TRACE(SD_LOG_IO, un,
20049             "sd_send_scsi_READ_CAPACITY: entry: un:0x%p\n", un);
20050 
20051         /*
20052          * First send a READ_CAPACITY command to the target.
20053          * (This command is mandatory under SCSI-2.)
20054          *
20055          * Set up the CDB for the READ_CAPACITY command.  The Partial
20056          * Medium Indicator bit is cleared.  The address field must be
20057          * zero if the PMI bit is zero.
20058          */
20059         bzero(&cdb, sizeof (cdb));
20060         bzero(&ucmd_buf, sizeof (ucmd_buf));
20061 
20062         capacity_buf = kmem_zalloc(SD_CAPACITY_SIZE, KM_SLEEP);
20063 
20064         cdb.scc_cmd = SCMD_READ_CAPACITY;
20065 
20066         ucmd_buf.uscsi_cdb      = (char *)&cdb;
20067         ucmd_buf.uscsi_cdblen   = CDB_GROUP1;
20068         ucmd_buf.uscsi_bufaddr  = (caddr_t)capacity_buf;
20069         ucmd_buf.uscsi_buflen   = SD_CAPACITY_SIZE;
20070         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
20071         ucmd_buf.uscsi_rqlen    = sizeof (sense_buf);
20072         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_READ | USCSI_SILENT;
20073         ucmd_buf.uscsi_timeout  = 60;
20074 
20075         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
20076             UIO_SYSSPACE, path_flag);
20077 
20078         switch (status) {
20079         case 0:
20080                 /* Return failure if we did not get valid capacity data. */
20081                 if (ucmd_buf.uscsi_resid != 0) {
20082                         sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, -1,
20083                             "sd_send_scsi_READ_CAPACITY received invalid "
20084                             "capacity data");
20085                         kmem_free(capacity_buf, SD_CAPACITY_SIZE);
20086                         return (EIO);
20087                 }
20088                 /*
20089                  * Read capacity and block size from the READ CAPACITY 10 data.
20090                  * This data may be adjusted later due to device specific
20091                  * issues.
20092                  *
20093                  * According to the SCSI spec, the READ CAPACITY 10
20094                  * command returns the following:
20095                  *
20096                  *  bytes 0-3: Maximum logical block address available.
20097                  *              (MSB in byte:0 & LSB in byte:3)
20098                  *
20099                  *  bytes 4-7: Block length in bytes
20100                  *              (MSB in byte:4 & LSB in byte:7)
20101                  *
20102                  */
20103                 capacity = BE_32(capacity_buf[0]);
20104                 lbasize = BE_32(capacity_buf[1]);
20105 
20106                 /*
20107                  * Done with capacity_buf
20108                  */
20109                 kmem_free(capacity_buf, SD_CAPACITY_SIZE);
20110 
20111                 /*
20112                  * if the reported capacity is set to all 0xf's, then
20113                  * this disk is too large and requires SBC-2 commands.
20114                  * Reissue the request using READ CAPACITY 16.
20115                  */
20116                 if (capacity == 0xffffffff) {
20117                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
20118                         status = sd_send_scsi_READ_CAPACITY_16(ssc, &capacity,
20119                             &lbasize, &pbsize, path_flag);
20120                         if (status != 0) {
20121                                 return (status);
20122                         } else {
20123                                 goto rc16_done;
20124                         }
20125                 }
20126                 break;  /* Success! */
20127         case EIO:
20128                 switch (ucmd_buf.uscsi_status) {
20129                 case STATUS_RESERVATION_CONFLICT:
20130                         status = EACCES;
20131                         break;
20132                 case STATUS_CHECK:
20133                         /*
20134                          * Check condition; look for ASC/ASCQ of 0x04/0x01
20135                          * (LOGICAL UNIT IS IN PROCESS OF BECOMING READY)
20136                          */
20137                         if ((ucmd_buf.uscsi_rqstatus == STATUS_GOOD) &&
20138                             (scsi_sense_asc((uint8_t *)&sense_buf) == 0x04) &&
20139                             (scsi_sense_ascq((uint8_t *)&sense_buf) == 0x01)) {
20140                                 kmem_free(capacity_buf, SD_CAPACITY_SIZE);
20141                                 return (EAGAIN);
20142                         }
20143                         break;
20144                 default:
20145                         break;
20146                 }
20147                 /* FALLTHRU */
20148         default:
20149                 kmem_free(capacity_buf, SD_CAPACITY_SIZE);
20150                 return (status);
20151         }
20152 
20153         /*
20154          * Some ATAPI CD-ROM drives report inaccurate LBA size values
20155          * (2352 and 0 are common) so for these devices always force the value
20156          * to 2048 as required by the ATAPI specs.
20157          */
20158         if ((un->un_f_cfg_is_atapi == TRUE) && (ISCD(un))) {
20159                 lbasize = 2048;
20160         }
20161 
20162         /*
20163          * Get the maximum LBA value from the READ CAPACITY data.
20164          * Here we assume that the Partial Medium Indicator (PMI) bit
20165          * was cleared when issuing the command. This means that the LBA
20166          * returned from the device is the LBA of the last logical block
20167          * on the logical unit.  The actual logical block count will be
20168          * this value plus one.
20169          */
20170         capacity += 1;
20171 
20172         /*
20173          * Currently, for removable media, the capacity is saved in terms
20174          * of un->un_sys_blocksize, so scale the capacity value to reflect this.
20175          */
20176         if (un->un_f_has_removable_media)
20177                 capacity *= (lbasize / un->un_sys_blocksize);
20178 
20179 rc16_done:
20180 
20181         /*
20182          * Copy the values from the READ CAPACITY command into the space
20183          * provided by the caller.
20184          */
20185         *capp = capacity;
20186         *lbap = lbasize;
20187 
20188         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_READ_CAPACITY: "
20189             "capacity:0x%llx  lbasize:0x%x\n", capacity, lbasize);
20190 
20191         /*
20192          * Both the lbasize and capacity from the device must be nonzero,
20193          * otherwise we assume that the values are not valid and return
20194          * failure to the caller. (4203735)
20195          */
20196         if ((capacity == 0) || (lbasize == 0)) {
20197                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, -1,
20198                     "sd_send_scsi_READ_CAPACITY received invalid value "
20199                     "capacity %llu lbasize %d", capacity, lbasize);
20200                 return (EIO);
20201         }
20202         sd_ssc_assessment(ssc, SD_FMT_STANDARD);
20203         return (0);
20204 }
20205 
20206 /*
20207  *    Function: sd_send_scsi_READ_CAPACITY_16
20208  *
20209  * Description: This routine uses the scsi READ CAPACITY 16 command to
20210  *              determine the device capacity in number of blocks and the
20211  *              device native block size.  If this function returns a failure,
20212  *              then the values in *capp and *lbap are undefined.
20213  *              This routine should be called by sd_send_scsi_READ_CAPACITY
20214  *              which will apply any device specific adjustments to capacity
20215  *              and lbasize. One exception is it is also called by
20216  *              sd_get_media_info_ext. In that function, there is no need to
20217  *              adjust the capacity and lbasize.
20218  *
20219  *   Arguments: ssc   - ssc contains ptr to soft state struct for the target
20220  *              capp - ptr to unsigned 64-bit variable to receive the
20221  *                      capacity value from the command.
20222  *              lbap - ptr to unsigned 32-bit varaible to receive the
20223  *                      block size value from the command
20224  *              psp  - ptr to unsigned 32-bit variable to receive the
20225  *                      physical block size value from the command
20226  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
20227  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
20228  *                      to use the USCSI "direct" chain and bypass the normal
20229  *                      command waitq. SD_PATH_DIRECT_PRIORITY is used when
20230  *                      this command is issued as part of an error recovery
20231  *                      action.
20232  *
20233  * Return Code: 0   - Success
20234  *              EIO - IO error
20235  *              EACCES - Reservation conflict detected
20236  *              EAGAIN - Device is becoming ready
20237  *              errno return code from sd_ssc_send()
20238  *
20239  *     Context: Can sleep.  Blocks until command completes.
20240  */
20241 
20242 #define SD_CAPACITY_16_SIZE     sizeof (struct scsi_capacity_16)
20243 
20244 static int
20245 sd_send_scsi_READ_CAPACITY_16(sd_ssc_t *ssc, uint64_t *capp,
20246         uint32_t *lbap, uint32_t *psp, int path_flag)
20247 {
20248         struct  scsi_extended_sense     sense_buf;
20249         struct  uscsi_cmd       ucmd_buf;
20250         union   scsi_cdb        cdb;
20251         uint64_t                *capacity16_buf;
20252         uint64_t                capacity;
20253         uint32_t                lbasize;
20254         uint32_t                pbsize;
20255         uint32_t                lbpb_exp;
20256         int                     status;
20257         struct sd_lun           *un;
20258 
20259         ASSERT(ssc != NULL);
20260 
20261         un = ssc->ssc_un;
20262         ASSERT(un != NULL);
20263         ASSERT(!mutex_owned(SD_MUTEX(un)));
20264         ASSERT(capp != NULL);
20265         ASSERT(lbap != NULL);
20266 
20267         SD_TRACE(SD_LOG_IO, un,
20268             "sd_send_scsi_READ_CAPACITY: entry: un:0x%p\n", un);
20269 
20270         /*
20271          * First send a READ_CAPACITY_16 command to the target.
20272          *
20273          * Set up the CDB for the READ_CAPACITY_16 command.  The Partial
20274          * Medium Indicator bit is cleared.  The address field must be
20275          * zero if the PMI bit is zero.
20276          */
20277         bzero(&cdb, sizeof (cdb));
20278         bzero(&ucmd_buf, sizeof (ucmd_buf));
20279 
20280         capacity16_buf = kmem_zalloc(SD_CAPACITY_16_SIZE, KM_SLEEP);
20281 
20282         ucmd_buf.uscsi_cdb      = (char *)&cdb;
20283         ucmd_buf.uscsi_cdblen   = CDB_GROUP4;
20284         ucmd_buf.uscsi_bufaddr  = (caddr_t)capacity16_buf;
20285         ucmd_buf.uscsi_buflen   = SD_CAPACITY_16_SIZE;
20286         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
20287         ucmd_buf.uscsi_rqlen    = sizeof (sense_buf);
20288         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_READ | USCSI_SILENT;
20289         ucmd_buf.uscsi_timeout  = 60;
20290 
20291         /*
20292          * Read Capacity (16) is a Service Action In command.  One
20293          * command byte (0x9E) is overloaded for multiple operations,
20294          * with the second CDB byte specifying the desired operation
20295          */
20296         cdb.scc_cmd = SCMD_SVC_ACTION_IN_G4;
20297         cdb.cdb_opaque[1] = SSVC_ACTION_READ_CAPACITY_G4;
20298 
20299         /*
20300          * Fill in allocation length field
20301          */
20302         FORMG4COUNT(&cdb, ucmd_buf.uscsi_buflen);
20303 
20304         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
20305             UIO_SYSSPACE, path_flag);
20306 
20307         switch (status) {
20308         case 0:
20309                 /* Return failure if we did not get valid capacity data. */
20310                 if (ucmd_buf.uscsi_resid > 20) {
20311                         sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, -1,
20312                             "sd_send_scsi_READ_CAPACITY_16 received invalid "
20313                             "capacity data");
20314                         kmem_free(capacity16_buf, SD_CAPACITY_16_SIZE);
20315                         return (EIO);
20316                 }
20317 
20318                 /*
20319                  * Read capacity and block size from the READ CAPACITY 16 data.
20320                  * This data may be adjusted later due to device specific
20321                  * issues.
20322                  *
20323                  * According to the SCSI spec, the READ CAPACITY 16
20324                  * command returns the following:
20325                  *
20326                  *  bytes 0-7: Maximum logical block address available.
20327                  *              (MSB in byte:0 & LSB in byte:7)
20328                  *
20329                  *  bytes 8-11: Block length in bytes
20330                  *              (MSB in byte:8 & LSB in byte:11)
20331                  *
20332                  *  byte 13: LOGICAL BLOCKS PER PHYSICAL BLOCK EXPONENT
20333                  */
20334                 capacity = BE_64(capacity16_buf[0]);
20335                 lbasize = BE_32(*(uint32_t *)&capacity16_buf[1]);
20336                 lbpb_exp = (BE_64(capacity16_buf[1]) >> 16) & 0x0f;
20337 
20338                 pbsize = lbasize << lbpb_exp;
20339 
20340                 /*
20341                  * Done with capacity16_buf
20342                  */
20343                 kmem_free(capacity16_buf, SD_CAPACITY_16_SIZE);
20344 
20345                 /*
20346                  * if the reported capacity is set to all 0xf's, then
20347                  * this disk is too large.  This could only happen with
20348                  * a device that supports LBAs larger than 64 bits which
20349                  * are not defined by any current T10 standards.
20350                  */
20351                 if (capacity == 0xffffffffffffffff) {
20352                         sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, -1,
20353                             "disk is too large");
20354                         return (EIO);
20355                 }
20356                 break;  /* Success! */
20357         case EIO:
20358                 switch (ucmd_buf.uscsi_status) {
20359                 case STATUS_RESERVATION_CONFLICT:
20360                         status = EACCES;
20361                         break;
20362                 case STATUS_CHECK:
20363                         /*
20364                          * Check condition; look for ASC/ASCQ of 0x04/0x01
20365                          * (LOGICAL UNIT IS IN PROCESS OF BECOMING READY)
20366                          */
20367                         if ((ucmd_buf.uscsi_rqstatus == STATUS_GOOD) &&
20368                             (scsi_sense_asc((uint8_t *)&sense_buf) == 0x04) &&
20369                             (scsi_sense_ascq((uint8_t *)&sense_buf) == 0x01)) {
20370                                 kmem_free(capacity16_buf, SD_CAPACITY_16_SIZE);
20371                                 return (EAGAIN);
20372                         }
20373                         break;
20374                 default:
20375                         break;
20376                 }
20377                 /* FALLTHRU */
20378         default:
20379                 kmem_free(capacity16_buf, SD_CAPACITY_16_SIZE);
20380                 return (status);
20381         }
20382 
20383         /*
20384          * Some ATAPI CD-ROM drives report inaccurate LBA size values
20385          * (2352 and 0 are common) so for these devices always force the value
20386          * to 2048 as required by the ATAPI specs.
20387          */
20388         if ((un->un_f_cfg_is_atapi == TRUE) && (ISCD(un))) {
20389                 lbasize = 2048;
20390         }
20391 
20392         /*
20393          * Get the maximum LBA value from the READ CAPACITY 16 data.
20394          * Here we assume that the Partial Medium Indicator (PMI) bit
20395          * was cleared when issuing the command. This means that the LBA
20396          * returned from the device is the LBA of the last logical block
20397          * on the logical unit.  The actual logical block count will be
20398          * this value plus one.
20399          */
20400         capacity += 1;
20401 
20402         /*
20403          * Currently, for removable media, the capacity is saved in terms
20404          * of un->un_sys_blocksize, so scale the capacity value to reflect this.
20405          */
20406         if (un->un_f_has_removable_media)
20407                 capacity *= (lbasize / un->un_sys_blocksize);
20408 
20409         *capp = capacity;
20410         *lbap = lbasize;
20411         *psp = pbsize;
20412 
20413         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_READ_CAPACITY_16: "
20414             "capacity:0x%llx  lbasize:0x%x, pbsize: 0x%x\n",
20415             capacity, lbasize, pbsize);
20416 
20417         if ((capacity == 0) || (lbasize == 0) || (pbsize == 0)) {
20418                 sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, -1,
20419                     "sd_send_scsi_READ_CAPACITY_16 received invalid value "
20420                     "capacity %llu lbasize %d pbsize %d", capacity, lbasize);
20421                 return (EIO);
20422         }
20423 
20424         sd_ssc_assessment(ssc, SD_FMT_STANDARD);
20425         return (0);
20426 }
20427 
20428 
20429 /*
20430  *    Function: sd_send_scsi_START_STOP_UNIT
20431  *
20432  * Description: Issue a scsi START STOP UNIT command to the target.
20433  *
20434  *   Arguments: ssc    - ssc contatins pointer to driver soft state (unit)
20435  *                       structure for this target.
20436  *      pc_flag - SD_POWER_CONDITION
20437  *                SD_START_STOP
20438  *              flag  - SD_TARGET_START
20439  *                      SD_TARGET_STOP
20440  *                      SD_TARGET_EJECT
20441  *                      SD_TARGET_CLOSE
20442  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
20443  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
20444  *                      to use the USCSI "direct" chain and bypass the normal
20445  *                      command waitq. SD_PATH_DIRECT_PRIORITY is used when this
20446  *                      command is issued as part of an error recovery action.
20447  *
20448  * Return Code: 0   - Success
20449  *              EIO - IO error
20450  *              EACCES - Reservation conflict detected
20451  *              ENXIO  - Not Ready, medium not present
20452  *              errno return code from sd_ssc_send()
20453  *
20454  *     Context: Can sleep.
20455  */
20456 
20457 static int
20458 sd_send_scsi_START_STOP_UNIT(sd_ssc_t *ssc, int pc_flag, int flag,
20459     int path_flag)
20460 {
20461         struct  scsi_extended_sense     sense_buf;
20462         union scsi_cdb          cdb;
20463         struct uscsi_cmd        ucmd_buf;
20464         int                     status;
20465         struct sd_lun           *un;
20466 
20467         ASSERT(ssc != NULL);
20468         un = ssc->ssc_un;
20469         ASSERT(un != NULL);
20470         ASSERT(!mutex_owned(SD_MUTEX(un)));
20471 
20472         SD_TRACE(SD_LOG_IO, un,
20473             "sd_send_scsi_START_STOP_UNIT: entry: un:0x%p\n", un);
20474 
20475         if (un->un_f_check_start_stop &&
20476             (pc_flag == SD_START_STOP) &&
20477             ((flag == SD_TARGET_START) || (flag == SD_TARGET_STOP)) &&
20478             (un->un_f_start_stop_supported != TRUE)) {
20479                 return (0);
20480         }
20481 
20482         /*
20483          * If we are performing an eject operation and
20484          * we receive any command other than SD_TARGET_EJECT
20485          * we should immediately return.
20486          */
20487         if (flag != SD_TARGET_EJECT) {
20488                 mutex_enter(SD_MUTEX(un));
20489                 if (un->un_f_ejecting == TRUE) {
20490                         mutex_exit(SD_MUTEX(un));
20491                         return (EAGAIN);
20492                 }
20493                 mutex_exit(SD_MUTEX(un));
20494         }
20495 
20496         bzero(&cdb, sizeof (cdb));
20497         bzero(&ucmd_buf, sizeof (ucmd_buf));
20498         bzero(&sense_buf, sizeof (struct scsi_extended_sense));
20499 
20500         cdb.scc_cmd = SCMD_START_STOP;
20501         cdb.cdb_opaque[4] = (pc_flag == SD_POWER_CONDITION) ?
20502             (uchar_t)(flag << 4) : (uchar_t)flag;
20503 
20504         ucmd_buf.uscsi_cdb      = (char *)&cdb;
20505         ucmd_buf.uscsi_cdblen   = CDB_GROUP0;
20506         ucmd_buf.uscsi_bufaddr  = NULL;
20507         ucmd_buf.uscsi_buflen   = 0;
20508         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
20509         ucmd_buf.uscsi_rqlen    = sizeof (struct scsi_extended_sense);
20510         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_SILENT;
20511         ucmd_buf.uscsi_timeout  = 200;
20512 
20513         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
20514             UIO_SYSSPACE, path_flag);
20515 
20516         switch (status) {
20517         case 0:
20518                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
20519                 break;  /* Success! */
20520         case EIO:
20521                 switch (ucmd_buf.uscsi_status) {
20522                 case STATUS_RESERVATION_CONFLICT:
20523                         status = EACCES;
20524                         break;
20525                 case STATUS_CHECK:
20526                         if (ucmd_buf.uscsi_rqstatus == STATUS_GOOD) {
20527                                 switch (scsi_sense_key(
20528                                     (uint8_t *)&sense_buf)) {
20529                                 case KEY_ILLEGAL_REQUEST:
20530                                         status = ENOTSUP;
20531                                         break;
20532                                 case KEY_NOT_READY:
20533                                         if (scsi_sense_asc(
20534                                             (uint8_t *)&sense_buf)
20535                                             == 0x3A) {
20536                                                 status = ENXIO;
20537                                         }
20538                                         break;
20539                                 default:
20540                                         break;
20541                                 }
20542                         }
20543                         break;
20544                 default:
20545                         break;
20546                 }
20547                 break;
20548         default:
20549                 break;
20550         }
20551 
20552         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_START_STOP_UNIT: exit\n");
20553 
20554         return (status);
20555 }
20556 
20557 
20558 /*
20559  *    Function: sd_start_stop_unit_callback
20560  *
20561  * Description: timeout(9F) callback to begin recovery process for a
20562  *              device that has spun down.
20563  *
20564  *   Arguments: arg - pointer to associated softstate struct.
20565  *
20566  *     Context: Executes in a timeout(9F) thread context
20567  */
20568 
20569 static void
20570 sd_start_stop_unit_callback(void *arg)
20571 {
20572         struct sd_lun   *un = arg;
20573         ASSERT(un != NULL);
20574         ASSERT(!mutex_owned(SD_MUTEX(un)));
20575 
20576         SD_TRACE(SD_LOG_IO, un, "sd_start_stop_unit_callback: entry\n");
20577 
20578         (void) taskq_dispatch(sd_tq, sd_start_stop_unit_task, un, KM_NOSLEEP);
20579 }
20580 
20581 
20582 /*
20583  *    Function: sd_start_stop_unit_task
20584  *
20585  * Description: Recovery procedure when a drive is spun down.
20586  *
20587  *   Arguments: arg - pointer to associated softstate struct.
20588  *
20589  *     Context: Executes in a taskq() thread context
20590  */
20591 
20592 static void
20593 sd_start_stop_unit_task(void *arg)
20594 {
20595         struct sd_lun   *un = arg;
20596         sd_ssc_t        *ssc;
20597         int             power_level;
20598         int             rval;
20599 
20600         ASSERT(un != NULL);
20601         ASSERT(!mutex_owned(SD_MUTEX(un)));
20602 
20603         SD_TRACE(SD_LOG_IO, un, "sd_start_stop_unit_task: entry\n");
20604 
20605         /*
20606          * Some unformatted drives report not ready error, no need to
20607          * restart if format has been initiated.
20608          */
20609         mutex_enter(SD_MUTEX(un));
20610         if (un->un_f_format_in_progress == TRUE) {
20611                 mutex_exit(SD_MUTEX(un));
20612                 return;
20613         }
20614         mutex_exit(SD_MUTEX(un));
20615 
20616         ssc = sd_ssc_init(un);
20617         /*
20618          * When a START STOP command is issued from here, it is part of a
20619          * failure recovery operation and must be issued before any other
20620          * commands, including any pending retries. Thus it must be sent
20621          * using SD_PATH_DIRECT_PRIORITY. It doesn't matter if the spin up
20622          * succeeds or not, we will start I/O after the attempt.
20623          * If power condition is supported and the current power level
20624          * is capable of performing I/O, we should set the power condition
20625          * to that level. Otherwise, set the power condition to ACTIVE.
20626          */
20627         if (un->un_f_power_condition_supported) {
20628                 mutex_enter(SD_MUTEX(un));
20629                 ASSERT(SD_PM_IS_LEVEL_VALID(un, un->un_power_level));
20630                 power_level = sd_pwr_pc.ran_perf[un->un_power_level]
20631                     > 0 ? un->un_power_level : SD_SPINDLE_ACTIVE;
20632                 mutex_exit(SD_MUTEX(un));
20633                 rval = sd_send_scsi_START_STOP_UNIT(ssc, SD_POWER_CONDITION,
20634                     sd_pl2pc[power_level], SD_PATH_DIRECT_PRIORITY);
20635         } else {
20636                 rval = sd_send_scsi_START_STOP_UNIT(ssc, SD_START_STOP,
20637                     SD_TARGET_START, SD_PATH_DIRECT_PRIORITY);
20638         }
20639 
20640         if (rval != 0)
20641                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
20642         sd_ssc_fini(ssc);
20643         /*
20644          * The above call blocks until the START_STOP_UNIT command completes.
20645          * Now that it has completed, we must re-try the original IO that
20646          * received the NOT READY condition in the first place. There are
20647          * three possible conditions here:
20648          *
20649          *  (1) The original IO is on un_retry_bp.
20650          *  (2) The original IO is on the regular wait queue, and un_retry_bp
20651          *      is NULL.
20652          *  (3) The original IO is on the regular wait queue, and un_retry_bp
20653          *      points to some other, unrelated bp.
20654          *
20655          * For each case, we must call sd_start_cmds() with un_retry_bp
20656          * as the argument. If un_retry_bp is NULL, this will initiate
20657          * processing of the regular wait queue.  If un_retry_bp is not NULL,
20658          * then this will process the bp on un_retry_bp. That may or may not
20659          * be the original IO, but that does not matter: the important thing
20660          * is to keep the IO processing going at this point.
20661          *
20662          * Note: This is a very specific error recovery sequence associated
20663          * with a drive that is not spun up. We attempt a START_STOP_UNIT and
20664          * serialize the I/O with completion of the spin-up.
20665          */
20666         mutex_enter(SD_MUTEX(un));
20667         SD_TRACE(SD_LOG_IO_CORE | SD_LOG_ERROR, un,
20668             "sd_start_stop_unit_task: un:0x%p starting bp:0x%p\n",
20669             un, un->un_retry_bp);
20670         un->un_startstop_timeid = NULL;      /* Timeout is no longer pending */
20671         sd_start_cmds(un, un->un_retry_bp);
20672         mutex_exit(SD_MUTEX(un));
20673 
20674         SD_TRACE(SD_LOG_IO, un, "sd_start_stop_unit_task: exit\n");
20675 }
20676 
20677 
20678 /*
20679  *    Function: sd_send_scsi_INQUIRY
20680  *
20681  * Description: Issue the scsi INQUIRY command.
20682  *
20683  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
20684  *                      structure for this target.
20685  *              bufaddr
20686  *              buflen
20687  *              evpd
20688  *              page_code
20689  *              page_length
20690  *
20691  * Return Code: 0   - Success
20692  *              errno return code from sd_ssc_send()
20693  *
20694  *     Context: Can sleep. Does not return until command is completed.
20695  */
20696 
20697 static int
20698 sd_send_scsi_INQUIRY(sd_ssc_t *ssc, uchar_t *bufaddr, size_t buflen,
20699         uchar_t evpd, uchar_t page_code, size_t *residp)
20700 {
20701         union scsi_cdb          cdb;
20702         struct uscsi_cmd        ucmd_buf;
20703         int                     status;
20704         struct sd_lun           *un;
20705 
20706         ASSERT(ssc != NULL);
20707         un = ssc->ssc_un;
20708         ASSERT(un != NULL);
20709         ASSERT(!mutex_owned(SD_MUTEX(un)));
20710         ASSERT(bufaddr != NULL);
20711 
20712         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_INQUIRY: entry: un:0x%p\n", un);
20713 
20714         bzero(&cdb, sizeof (cdb));
20715         bzero(&ucmd_buf, sizeof (ucmd_buf));
20716         bzero(bufaddr, buflen);
20717 
20718         cdb.scc_cmd = SCMD_INQUIRY;
20719         cdb.cdb_opaque[1] = evpd;
20720         cdb.cdb_opaque[2] = page_code;
20721         FORMG0COUNT(&cdb, buflen);
20722 
20723         ucmd_buf.uscsi_cdb      = (char *)&cdb;
20724         ucmd_buf.uscsi_cdblen   = CDB_GROUP0;
20725         ucmd_buf.uscsi_bufaddr  = (caddr_t)bufaddr;
20726         ucmd_buf.uscsi_buflen   = buflen;
20727         ucmd_buf.uscsi_rqbuf    = NULL;
20728         ucmd_buf.uscsi_rqlen    = 0;
20729         ucmd_buf.uscsi_flags    = USCSI_READ | USCSI_SILENT;
20730         ucmd_buf.uscsi_timeout  = 200;  /* Excessive legacy value */
20731 
20732         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
20733             UIO_SYSSPACE, SD_PATH_DIRECT);
20734 
20735         /*
20736          * Only handle status == 0, the upper-level caller
20737          * will put different assessment based on the context.
20738          */
20739         if (status == 0)
20740                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
20741 
20742         if ((status == 0) && (residp != NULL)) {
20743                 *residp = ucmd_buf.uscsi_resid;
20744         }
20745 
20746         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_INQUIRY: exit\n");
20747 
20748         return (status);
20749 }
20750 
20751 
20752 /*
20753  *    Function: sd_send_scsi_TEST_UNIT_READY
20754  *
20755  * Description: Issue the scsi TEST UNIT READY command.
20756  *              This routine can be told to set the flag USCSI_DIAGNOSE to
20757  *              prevent retrying failed commands. Use this when the intent
20758  *              is either to check for device readiness, to clear a Unit
20759  *              Attention, or to clear any outstanding sense data.
20760  *              However under specific conditions the expected behavior
20761  *              is for retries to bring a device ready, so use the flag
20762  *              with caution.
20763  *
20764  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
20765  *                      structure for this target.
20766  *              flag:   SD_CHECK_FOR_MEDIA: return ENXIO if no media present
20767  *                      SD_DONT_RETRY_TUR: include uscsi flag USCSI_DIAGNOSE.
20768  *                      0: dont check for media present, do retries on cmd.
20769  *
20770  * Return Code: 0   - Success
20771  *              EIO - IO error
20772  *              EACCES - Reservation conflict detected
20773  *              ENXIO  - Not Ready, medium not present
20774  *              errno return code from sd_ssc_send()
20775  *
20776  *     Context: Can sleep. Does not return until command is completed.
20777  */
20778 
20779 static int
20780 sd_send_scsi_TEST_UNIT_READY(sd_ssc_t *ssc, int flag)
20781 {
20782         struct  scsi_extended_sense     sense_buf;
20783         union scsi_cdb          cdb;
20784         struct uscsi_cmd        ucmd_buf;
20785         int                     status;
20786         struct sd_lun           *un;
20787 
20788         ASSERT(ssc != NULL);
20789         un = ssc->ssc_un;
20790         ASSERT(un != NULL);
20791         ASSERT(!mutex_owned(SD_MUTEX(un)));
20792 
20793         SD_TRACE(SD_LOG_IO, un,
20794             "sd_send_scsi_TEST_UNIT_READY: entry: un:0x%p\n", un);
20795 
20796         /*
20797          * Some Seagate elite1 TQ devices get hung with disconnect/reconnect
20798          * timeouts when they receive a TUR and the queue is not empty. Check
20799          * the configuration flag set during attach (indicating the drive has
20800          * this firmware bug) and un_ncmds_in_transport before issuing the
20801          * TUR. If there are
20802          * pending commands return success, this is a bit arbitrary but is ok
20803          * for non-removables (i.e. the eliteI disks) and non-clustering
20804          * configurations.
20805          */
20806         if (un->un_f_cfg_tur_check == TRUE) {
20807                 mutex_enter(SD_MUTEX(un));
20808                 if (un->un_ncmds_in_transport != 0) {
20809                         mutex_exit(SD_MUTEX(un));
20810                         return (0);
20811                 }
20812                 mutex_exit(SD_MUTEX(un));
20813         }
20814 
20815         bzero(&cdb, sizeof (cdb));
20816         bzero(&ucmd_buf, sizeof (ucmd_buf));
20817         bzero(&sense_buf, sizeof (struct scsi_extended_sense));
20818 
20819         cdb.scc_cmd = SCMD_TEST_UNIT_READY;
20820 
20821         ucmd_buf.uscsi_cdb      = (char *)&cdb;
20822         ucmd_buf.uscsi_cdblen   = CDB_GROUP0;
20823         ucmd_buf.uscsi_bufaddr  = NULL;
20824         ucmd_buf.uscsi_buflen   = 0;
20825         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
20826         ucmd_buf.uscsi_rqlen    = sizeof (struct scsi_extended_sense);
20827         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_SILENT;
20828 
20829         /* Use flag USCSI_DIAGNOSE to prevent retries if it fails. */
20830         if ((flag & SD_DONT_RETRY_TUR) != 0) {
20831                 ucmd_buf.uscsi_flags |= USCSI_DIAGNOSE;
20832         }
20833         ucmd_buf.uscsi_timeout  = 60;
20834 
20835         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
20836             UIO_SYSSPACE, ((flag & SD_BYPASS_PM) ? SD_PATH_DIRECT :
20837             SD_PATH_STANDARD));
20838 
20839         switch (status) {
20840         case 0:
20841                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
20842                 break;  /* Success! */
20843         case EIO:
20844                 switch (ucmd_buf.uscsi_status) {
20845                 case STATUS_RESERVATION_CONFLICT:
20846                         status = EACCES;
20847                         break;
20848                 case STATUS_CHECK:
20849                         if ((flag & SD_CHECK_FOR_MEDIA) == 0) {
20850                                 break;
20851                         }
20852                         if ((ucmd_buf.uscsi_rqstatus == STATUS_GOOD) &&
20853                             (scsi_sense_key((uint8_t *)&sense_buf) ==
20854                             KEY_NOT_READY) &&
20855                             (scsi_sense_asc((uint8_t *)&sense_buf) == 0x3A)) {
20856                                 status = ENXIO;
20857                         }
20858                         break;
20859                 default:
20860                         break;
20861                 }
20862                 break;
20863         default:
20864                 break;
20865         }
20866 
20867         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_TEST_UNIT_READY: exit\n");
20868 
20869         return (status);
20870 }
20871 
20872 /*
20873  *    Function: sd_send_scsi_PERSISTENT_RESERVE_IN
20874  *
20875  * Description: Issue the scsi PERSISTENT RESERVE IN command.
20876  *
20877  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
20878  *                      structure for this target.
20879  *
20880  * Return Code: 0   - Success
20881  *              EACCES
20882  *              ENOTSUP
20883  *              errno return code from sd_ssc_send()
20884  *
20885  *     Context: Can sleep. Does not return until command is completed.
20886  */
20887 
20888 static int
20889 sd_send_scsi_PERSISTENT_RESERVE_IN(sd_ssc_t *ssc, uchar_t  usr_cmd,
20890         uint16_t data_len, uchar_t *data_bufp)
20891 {
20892         struct scsi_extended_sense      sense_buf;
20893         union scsi_cdb          cdb;
20894         struct uscsi_cmd        ucmd_buf;
20895         int                     status;
20896         int                     no_caller_buf = FALSE;
20897         struct sd_lun           *un;
20898 
20899         ASSERT(ssc != NULL);
20900         un = ssc->ssc_un;
20901         ASSERT(un != NULL);
20902         ASSERT(!mutex_owned(SD_MUTEX(un)));
20903         ASSERT((usr_cmd == SD_READ_KEYS) || (usr_cmd == SD_READ_RESV));
20904 
20905         SD_TRACE(SD_LOG_IO, un,
20906             "sd_send_scsi_PERSISTENT_RESERVE_IN: entry: un:0x%p\n", un);
20907 
20908         bzero(&cdb, sizeof (cdb));
20909         bzero(&ucmd_buf, sizeof (ucmd_buf));
20910         bzero(&sense_buf, sizeof (struct scsi_extended_sense));
20911         if (data_bufp == NULL) {
20912                 /* Allocate a default buf if the caller did not give one */
20913                 ASSERT(data_len == 0);
20914                 data_len  = MHIOC_RESV_KEY_SIZE;
20915                 data_bufp = kmem_zalloc(MHIOC_RESV_KEY_SIZE, KM_SLEEP);
20916                 no_caller_buf = TRUE;
20917         }
20918 
20919         cdb.scc_cmd = SCMD_PERSISTENT_RESERVE_IN;
20920         cdb.cdb_opaque[1] = usr_cmd;
20921         FORMG1COUNT(&cdb, data_len);
20922 
20923         ucmd_buf.uscsi_cdb      = (char *)&cdb;
20924         ucmd_buf.uscsi_cdblen   = CDB_GROUP1;
20925         ucmd_buf.uscsi_bufaddr  = (caddr_t)data_bufp;
20926         ucmd_buf.uscsi_buflen   = data_len;
20927         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
20928         ucmd_buf.uscsi_rqlen    = sizeof (struct scsi_extended_sense);
20929         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_READ | USCSI_SILENT;
20930         ucmd_buf.uscsi_timeout  = 60;
20931 
20932         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
20933             UIO_SYSSPACE, SD_PATH_STANDARD);
20934 
20935         switch (status) {
20936         case 0:
20937                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
20938 
20939                 break;  /* Success! */
20940         case EIO:
20941                 switch (ucmd_buf.uscsi_status) {
20942                 case STATUS_RESERVATION_CONFLICT:
20943                         status = EACCES;
20944                         break;
20945                 case STATUS_CHECK:
20946                         if ((ucmd_buf.uscsi_rqstatus == STATUS_GOOD) &&
20947                             (scsi_sense_key((uint8_t *)&sense_buf) ==
20948                             KEY_ILLEGAL_REQUEST)) {
20949                                 status = ENOTSUP;
20950                         }
20951                         break;
20952                 default:
20953                         break;
20954                 }
20955                 break;
20956         default:
20957                 break;
20958         }
20959 
20960         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_PERSISTENT_RESERVE_IN: exit\n");
20961 
20962         if (no_caller_buf == TRUE) {
20963                 kmem_free(data_bufp, data_len);
20964         }
20965 
20966         return (status);
20967 }
20968 
20969 
20970 /*
20971  *    Function: sd_send_scsi_PERSISTENT_RESERVE_OUT
20972  *
20973  * Description: This routine is the driver entry point for handling CD-ROM
20974  *              multi-host persistent reservation requests (MHIOCGRP_INKEYS,
20975  *              MHIOCGRP_INRESV) by sending the SCSI-3 PROUT commands to the
20976  *              device.
20977  *
20978  *   Arguments: ssc  -  ssc contains un - pointer to soft state struct
20979  *                      for the target.
20980  *              usr_cmd SCSI-3 reservation facility command (one of
20981  *                      SD_SCSI3_REGISTER, SD_SCSI3_RESERVE, SD_SCSI3_RELEASE,
20982  *                      SD_SCSI3_PREEMPTANDABORT, SD_SCSI3_CLEAR)
20983  *              usr_bufp - user provided pointer register, reserve descriptor or
20984  *                      preempt and abort structure (mhioc_register_t,
20985  *                      mhioc_resv_desc_t, mhioc_preemptandabort_t)
20986  *
20987  * Return Code: 0   - Success
20988  *              EACCES
20989  *              ENOTSUP
20990  *              errno return code from sd_ssc_send()
20991  *
20992  *     Context: Can sleep. Does not return until command is completed.
20993  */
20994 
20995 static int
20996 sd_send_scsi_PERSISTENT_RESERVE_OUT(sd_ssc_t *ssc, uchar_t usr_cmd,
20997         uchar_t *usr_bufp)
20998 {
20999         struct scsi_extended_sense      sense_buf;
21000         union scsi_cdb          cdb;
21001         struct uscsi_cmd        ucmd_buf;
21002         int                     status;
21003         uchar_t                 data_len = sizeof (sd_prout_t);
21004         sd_prout_t              *prp;
21005         struct sd_lun           *un;
21006 
21007         ASSERT(ssc != NULL);
21008         un = ssc->ssc_un;
21009         ASSERT(un != NULL);
21010         ASSERT(!mutex_owned(SD_MUTEX(un)));
21011         ASSERT(data_len == 24); /* required by scsi spec */
21012 
21013         SD_TRACE(SD_LOG_IO, un,
21014             "sd_send_scsi_PERSISTENT_RESERVE_OUT: entry: un:0x%p\n", un);
21015 
21016         if (usr_bufp == NULL) {
21017                 return (EINVAL);
21018         }
21019 
21020         bzero(&cdb, sizeof (cdb));
21021         bzero(&ucmd_buf, sizeof (ucmd_buf));
21022         bzero(&sense_buf, sizeof (struct scsi_extended_sense));
21023         prp = kmem_zalloc(data_len, KM_SLEEP);
21024 
21025         cdb.scc_cmd = SCMD_PERSISTENT_RESERVE_OUT;
21026         cdb.cdb_opaque[1] = usr_cmd;
21027         FORMG1COUNT(&cdb, data_len);
21028 
21029         ucmd_buf.uscsi_cdb      = (char *)&cdb;
21030         ucmd_buf.uscsi_cdblen   = CDB_GROUP1;
21031         ucmd_buf.uscsi_bufaddr  = (caddr_t)prp;
21032         ucmd_buf.uscsi_buflen   = data_len;
21033         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
21034         ucmd_buf.uscsi_rqlen    = sizeof (struct scsi_extended_sense);
21035         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_WRITE | USCSI_SILENT;
21036         ucmd_buf.uscsi_timeout  = 60;
21037 
21038         switch (usr_cmd) {
21039         case SD_SCSI3_REGISTER: {
21040                 mhioc_register_t *ptr = (mhioc_register_t *)usr_bufp;
21041 
21042                 bcopy(ptr->oldkey.key, prp->res_key, MHIOC_RESV_KEY_SIZE);
21043                 bcopy(ptr->newkey.key, prp->service_key,
21044                     MHIOC_RESV_KEY_SIZE);
21045                 prp->aptpl = ptr->aptpl;
21046                 break;
21047         }
21048         case SD_SCSI3_CLEAR: {
21049                 mhioc_resv_desc_t *ptr = (mhioc_resv_desc_t *)usr_bufp;
21050 
21051                 bcopy(ptr->key.key, prp->res_key, MHIOC_RESV_KEY_SIZE);
21052                 break;
21053         }
21054         case SD_SCSI3_RESERVE:
21055         case SD_SCSI3_RELEASE: {
21056                 mhioc_resv_desc_t *ptr = (mhioc_resv_desc_t *)usr_bufp;
21057 
21058                 bcopy(ptr->key.key, prp->res_key, MHIOC_RESV_KEY_SIZE);
21059                 prp->scope_address = BE_32(ptr->scope_specific_addr);
21060                 cdb.cdb_opaque[2] = ptr->type;
21061                 break;
21062         }
21063         case SD_SCSI3_PREEMPTANDABORT: {
21064                 mhioc_preemptandabort_t *ptr =
21065                     (mhioc_preemptandabort_t *)usr_bufp;
21066 
21067                 bcopy(ptr->resvdesc.key.key, prp->res_key, MHIOC_RESV_KEY_SIZE);
21068                 bcopy(ptr->victim_key.key, prp->service_key,
21069                     MHIOC_RESV_KEY_SIZE);
21070                 prp->scope_address = BE_32(ptr->resvdesc.scope_specific_addr);
21071                 cdb.cdb_opaque[2] = ptr->resvdesc.type;
21072                 ucmd_buf.uscsi_flags |= USCSI_HEAD;
21073                 break;
21074         }
21075         case SD_SCSI3_REGISTERANDIGNOREKEY:
21076         {
21077                 mhioc_registerandignorekey_t *ptr;
21078                 ptr = (mhioc_registerandignorekey_t *)usr_bufp;
21079                 bcopy(ptr->newkey.key,
21080                     prp->service_key, MHIOC_RESV_KEY_SIZE);
21081                 prp->aptpl = ptr->aptpl;
21082                 break;
21083         }
21084         default:
21085                 ASSERT(FALSE);
21086                 break;
21087         }
21088 
21089         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
21090             UIO_SYSSPACE, SD_PATH_STANDARD);
21091 
21092         switch (status) {
21093         case 0:
21094                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
21095                 break;  /* Success! */
21096         case EIO:
21097                 switch (ucmd_buf.uscsi_status) {
21098                 case STATUS_RESERVATION_CONFLICT:
21099                         status = EACCES;
21100                         break;
21101                 case STATUS_CHECK:
21102                         if ((ucmd_buf.uscsi_rqstatus == STATUS_GOOD) &&
21103                             (scsi_sense_key((uint8_t *)&sense_buf) ==
21104                             KEY_ILLEGAL_REQUEST)) {
21105                                 status = ENOTSUP;
21106                         }
21107                         break;
21108                 default:
21109                         break;
21110                 }
21111                 break;
21112         default:
21113                 break;
21114         }
21115 
21116         kmem_free(prp, data_len);
21117         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_PERSISTENT_RESERVE_OUT: exit\n");
21118         return (status);
21119 }
21120 
21121 
21122 /*
21123  *    Function: sd_send_scsi_SYNCHRONIZE_CACHE
21124  *
21125  * Description: Issues a scsi SYNCHRONIZE CACHE command to the target
21126  *
21127  *   Arguments: un - pointer to the target's soft state struct
21128  *              dkc - pointer to the callback structure
21129  *
21130  * Return Code: 0 - success
21131  *              errno-type error code
21132  *
21133  *     Context: kernel thread context only.
21134  *
21135  *  _______________________________________________________________
21136  * | dkc_flag &   | dkc_callback | DKIOCFLUSHWRITECACHE            |
21137  * |FLUSH_VOLATILE|              | operation                       |
21138  * |______________|______________|_________________________________|
21139  * | 0            | NULL         | Synchronous flush on both       |
21140  * |              |              | volatile and non-volatile cache |
21141  * |______________|______________|_________________________________|
21142  * | 1            | NULL         | Synchronous flush on volatile   |
21143  * |              |              | cache; disk drivers may suppress|
21144  * |              |              | flush if disk table indicates   |
21145  * |              |              | non-volatile cache              |
21146  * |______________|______________|_________________________________|
21147  * | 0            | !NULL        | Asynchronous flush on both      |
21148  * |              |              | volatile and non-volatile cache;|
21149  * |______________|______________|_________________________________|
21150  * | 1            | !NULL        | Asynchronous flush on volatile  |
21151  * |              |              | cache; disk drivers may suppress|
21152  * |              |              | flush if disk table indicates   |
21153  * |              |              | non-volatile cache              |
21154  * |______________|______________|_________________________________|
21155  *
21156  */
21157 
21158 static int
21159 sd_send_scsi_SYNCHRONIZE_CACHE(struct sd_lun *un, struct dk_callback *dkc)
21160 {
21161         struct sd_uscsi_info    *uip;
21162         struct uscsi_cmd        *uscmd;
21163         union scsi_cdb          *cdb;
21164         struct buf              *bp;
21165         int                     rval = 0;
21166         int                     is_async;
21167 
21168         SD_TRACE(SD_LOG_IO, un,
21169             "sd_send_scsi_SYNCHRONIZE_CACHE: entry: un:0x%p\n", un);
21170 
21171         ASSERT(un != NULL);
21172         ASSERT(!mutex_owned(SD_MUTEX(un)));
21173 
21174         if (dkc == NULL || dkc->dkc_callback == NULL) {
21175                 is_async = FALSE;
21176         } else {
21177                 is_async = TRUE;
21178         }
21179 
21180         mutex_enter(SD_MUTEX(un));
21181         /* check whether cache flush should be suppressed */
21182         if (un->un_f_suppress_cache_flush == TRUE) {
21183                 mutex_exit(SD_MUTEX(un));
21184                 /*
21185                  * suppress the cache flush if the device is told to do
21186                  * so by sd.conf or disk table
21187                  */
21188                 SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_SYNCHRONIZE_CACHE: \
21189                     skip the cache flush since suppress_cache_flush is %d!\n",
21190                     un->un_f_suppress_cache_flush);
21191 
21192                 if (is_async == TRUE) {
21193                         /* invoke callback for asynchronous flush */
21194                         (*dkc->dkc_callback)(dkc->dkc_cookie, 0);
21195                 }
21196                 return (rval);
21197         }
21198         mutex_exit(SD_MUTEX(un));
21199 
21200         /*
21201          * check dkc_flag & FLUSH_VOLATILE so SYNC_NV bit can be
21202          * set properly
21203          */
21204         cdb = kmem_zalloc(CDB_GROUP1, KM_SLEEP);
21205         cdb->scc_cmd = SCMD_SYNCHRONIZE_CACHE;
21206 
21207         mutex_enter(SD_MUTEX(un));
21208         if (dkc != NULL && un->un_f_sync_nv_supported &&
21209             (dkc->dkc_flag & FLUSH_VOLATILE)) {
21210                 /*
21211                  * if the device supports SYNC_NV bit, turn on
21212                  * the SYNC_NV bit to only flush volatile cache
21213                  */
21214                 cdb->cdb_un.tag |= SD_SYNC_NV_BIT;
21215         }
21216         mutex_exit(SD_MUTEX(un));
21217 
21218         /*
21219          * First get some memory for the uscsi_cmd struct and cdb
21220          * and initialize for SYNCHRONIZE_CACHE cmd.
21221          */
21222         uscmd = kmem_zalloc(sizeof (struct uscsi_cmd), KM_SLEEP);
21223         uscmd->uscsi_cdblen = CDB_GROUP1;
21224         uscmd->uscsi_cdb = (caddr_t)cdb;
21225         uscmd->uscsi_bufaddr = NULL;
21226         uscmd->uscsi_buflen = 0;
21227         uscmd->uscsi_rqbuf = kmem_zalloc(SENSE_LENGTH, KM_SLEEP);
21228         uscmd->uscsi_rqlen = SENSE_LENGTH;
21229         uscmd->uscsi_rqresid = SENSE_LENGTH;
21230         uscmd->uscsi_flags = USCSI_RQENABLE | USCSI_SILENT;
21231         uscmd->uscsi_timeout = sd_io_time;
21232 
21233         /*
21234          * Allocate an sd_uscsi_info struct and fill it with the info
21235          * needed by sd_initpkt_for_uscsi().  Then put the pointer into
21236          * b_private in the buf for sd_initpkt_for_uscsi().  Note that
21237          * since we allocate the buf here in this function, we do not
21238          * need to preserve the prior contents of b_private.
21239          * The sd_uscsi_info struct is also used by sd_uscsi_strategy()
21240          */
21241         uip = kmem_zalloc(sizeof (struct sd_uscsi_info), KM_SLEEP);
21242         uip->ui_flags = SD_PATH_DIRECT;
21243         uip->ui_cmdp  = uscmd;
21244 
21245         bp = getrbuf(KM_SLEEP);
21246         bp->b_private = uip;
21247 
21248         /*
21249          * Setup buffer to carry uscsi request.
21250          */
21251         bp->b_flags  = B_BUSY;
21252         bp->b_bcount = 0;
21253         bp->b_blkno  = 0;
21254 
21255         if (is_async == TRUE) {
21256                 bp->b_iodone = sd_send_scsi_SYNCHRONIZE_CACHE_biodone;
21257                 uip->ui_dkc = *dkc;
21258         }
21259 
21260         bp->b_edev = SD_GET_DEV(un);
21261         bp->b_dev = cmpdev(bp->b_edev);   /* maybe unnecessary? */
21262 
21263         /*
21264          * Unset un_f_sync_cache_required flag
21265          */
21266         mutex_enter(SD_MUTEX(un));
21267         un->un_f_sync_cache_required = FALSE;
21268         mutex_exit(SD_MUTEX(un));
21269 
21270         (void) sd_uscsi_strategy(bp);
21271 
21272         /*
21273          * If synchronous request, wait for completion
21274          * If async just return and let b_iodone callback
21275          * cleanup.
21276          * NOTE: On return, u_ncmds_in_driver will be decremented,
21277          * but it was also incremented in sd_uscsi_strategy(), so
21278          * we should be ok.
21279          */
21280         if (is_async == FALSE) {
21281                 (void) biowait(bp);
21282                 rval = sd_send_scsi_SYNCHRONIZE_CACHE_biodone(bp);
21283         }
21284 
21285         return (rval);
21286 }
21287 
21288 
21289 static int
21290 sd_send_scsi_SYNCHRONIZE_CACHE_biodone(struct buf *bp)
21291 {
21292         struct sd_uscsi_info *uip;
21293         struct uscsi_cmd *uscmd;
21294         uint8_t *sense_buf;
21295         struct sd_lun *un;
21296         int status;
21297         union scsi_cdb *cdb;
21298 
21299         uip = (struct sd_uscsi_info *)(bp->b_private);
21300         ASSERT(uip != NULL);
21301 
21302         uscmd = uip->ui_cmdp;
21303         ASSERT(uscmd != NULL);
21304 
21305         sense_buf = (uint8_t *)uscmd->uscsi_rqbuf;
21306         ASSERT(sense_buf != NULL);
21307 
21308         un = ddi_get_soft_state(sd_state, SD_GET_INSTANCE_FROM_BUF(bp));
21309         ASSERT(un != NULL);
21310 
21311         cdb = (union scsi_cdb *)uscmd->uscsi_cdb;
21312 
21313         status = geterror(bp);
21314         switch (status) {
21315         case 0:
21316                 break;  /* Success! */
21317         case EIO:
21318                 switch (uscmd->uscsi_status) {
21319                 case STATUS_RESERVATION_CONFLICT:
21320                         /* Ignore reservation conflict */
21321                         status = 0;
21322                         goto done;
21323 
21324                 case STATUS_CHECK:
21325                         if ((uscmd->uscsi_rqstatus == STATUS_GOOD) &&
21326                             (scsi_sense_key(sense_buf) ==
21327                             KEY_ILLEGAL_REQUEST)) {
21328                                 /* Ignore Illegal Request error */
21329                                 if (cdb->cdb_un.tag&SD_SYNC_NV_BIT) {
21330                                         mutex_enter(SD_MUTEX(un));
21331                                         un->un_f_sync_nv_supported = FALSE;
21332                                         mutex_exit(SD_MUTEX(un));
21333                                         status = 0;
21334                                         SD_TRACE(SD_LOG_IO, un,
21335                                             "un_f_sync_nv_supported \
21336                                             is set to false.\n");
21337                                         goto done;
21338                                 }
21339 
21340                                 mutex_enter(SD_MUTEX(un));
21341                                 un->un_f_sync_cache_supported = FALSE;
21342                                 mutex_exit(SD_MUTEX(un));
21343                                 SD_TRACE(SD_LOG_IO, un,
21344                                     "sd_send_scsi_SYNCHRONIZE_CACHE_biodone: \
21345                                     un_f_sync_cache_supported set to false \
21346                                     with asc = %x, ascq = %x\n",
21347                                     scsi_sense_asc(sense_buf),
21348                                     scsi_sense_ascq(sense_buf));
21349                                 status = ENOTSUP;
21350                                 goto done;
21351                         }
21352                         break;
21353                 default:
21354                         break;
21355                 }
21356                 /* FALLTHRU */
21357         default:
21358                 /*
21359                  * Turn on the un_f_sync_cache_required flag
21360                  * since the SYNC CACHE command failed
21361                  */
21362                 mutex_enter(SD_MUTEX(un));
21363                 un->un_f_sync_cache_required = TRUE;
21364                 mutex_exit(SD_MUTEX(un));
21365 
21366                 /*
21367                  * Don't log an error message if this device
21368                  * has removable media.
21369                  */
21370                 if (!un->un_f_has_removable_media) {
21371                         scsi_log(SD_DEVINFO(un), sd_label, CE_WARN,
21372                             "SYNCHRONIZE CACHE command failed (%d)\n", status);
21373                 }
21374                 break;
21375         }
21376 
21377 done:
21378         if (uip->ui_dkc.dkc_callback != NULL) {
21379                 (*uip->ui_dkc.dkc_callback)(uip->ui_dkc.dkc_cookie, status);
21380         }
21381 
21382         ASSERT((bp->b_flags & B_REMAPPED) == 0);
21383         freerbuf(bp);
21384         kmem_free(uip, sizeof (struct sd_uscsi_info));
21385         kmem_free(uscmd->uscsi_rqbuf, SENSE_LENGTH);
21386         kmem_free(uscmd->uscsi_cdb, (size_t)uscmd->uscsi_cdblen);
21387         kmem_free(uscmd, sizeof (struct uscsi_cmd));
21388 
21389         return (status);
21390 }
21391 
21392 
21393 /*
21394  *    Function: sd_send_scsi_GET_CONFIGURATION
21395  *
21396  * Description: Issues the get configuration command to the device.
21397  *              Called from sd_check_for_writable_cd & sd_get_media_info
21398  *              caller needs to ensure that buflen = SD_PROFILE_HEADER_LEN
21399  *   Arguments: ssc
21400  *              ucmdbuf
21401  *              rqbuf
21402  *              rqbuflen
21403  *              bufaddr
21404  *              buflen
21405  *              path_flag
21406  *
21407  * Return Code: 0   - Success
21408  *              errno return code from sd_ssc_send()
21409  *
21410  *     Context: Can sleep. Does not return until command is completed.
21411  *
21412  */
21413 
21414 static int
21415 sd_send_scsi_GET_CONFIGURATION(sd_ssc_t *ssc, struct uscsi_cmd *ucmdbuf,
21416         uchar_t *rqbuf, uint_t rqbuflen, uchar_t *bufaddr, uint_t buflen,
21417         int path_flag)
21418 {
21419         char    cdb[CDB_GROUP1];
21420         int     status;
21421         struct sd_lun   *un;
21422 
21423         ASSERT(ssc != NULL);
21424         un = ssc->ssc_un;
21425         ASSERT(un != NULL);
21426         ASSERT(!mutex_owned(SD_MUTEX(un)));
21427         ASSERT(bufaddr != NULL);
21428         ASSERT(ucmdbuf != NULL);
21429         ASSERT(rqbuf != NULL);
21430 
21431         SD_TRACE(SD_LOG_IO, un,
21432             "sd_send_scsi_GET_CONFIGURATION: entry: un:0x%p\n", un);
21433 
21434         bzero(cdb, sizeof (cdb));
21435         bzero(ucmdbuf, sizeof (struct uscsi_cmd));
21436         bzero(rqbuf, rqbuflen);
21437         bzero(bufaddr, buflen);
21438 
21439         /*
21440          * Set up cdb field for the get configuration command.
21441          */
21442         cdb[0] = SCMD_GET_CONFIGURATION;
21443         cdb[1] = 0x02;  /* Requested Type */
21444         cdb[8] = SD_PROFILE_HEADER_LEN;
21445         ucmdbuf->uscsi_cdb = cdb;
21446         ucmdbuf->uscsi_cdblen = CDB_GROUP1;
21447         ucmdbuf->uscsi_bufaddr = (caddr_t)bufaddr;
21448         ucmdbuf->uscsi_buflen = buflen;
21449         ucmdbuf->uscsi_timeout = sd_io_time;
21450         ucmdbuf->uscsi_rqbuf = (caddr_t)rqbuf;
21451         ucmdbuf->uscsi_rqlen = rqbuflen;
21452         ucmdbuf->uscsi_flags = USCSI_RQENABLE|USCSI_SILENT|USCSI_READ;
21453 
21454         status = sd_ssc_send(ssc, ucmdbuf, FKIOCTL,
21455             UIO_SYSSPACE, path_flag);
21456 
21457         switch (status) {
21458         case 0:
21459                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
21460                 break;  /* Success! */
21461         case EIO:
21462                 switch (ucmdbuf->uscsi_status) {
21463                 case STATUS_RESERVATION_CONFLICT:
21464                         status = EACCES;
21465                         break;
21466                 default:
21467                         break;
21468                 }
21469                 break;
21470         default:
21471                 break;
21472         }
21473 
21474         if (status == 0) {
21475                 SD_DUMP_MEMORY(un, SD_LOG_IO,
21476                     "sd_send_scsi_GET_CONFIGURATION: data",
21477                     (uchar_t *)bufaddr, SD_PROFILE_HEADER_LEN, SD_LOG_HEX);
21478         }
21479 
21480         SD_TRACE(SD_LOG_IO, un,
21481             "sd_send_scsi_GET_CONFIGURATION: exit\n");
21482 
21483         return (status);
21484 }
21485 
21486 /*
21487  *    Function: sd_send_scsi_feature_GET_CONFIGURATION
21488  *
21489  * Description: Issues the get configuration command to the device to
21490  *              retrieve a specific feature. Called from
21491  *              sd_check_for_writable_cd & sd_set_mmc_caps.
21492  *   Arguments: ssc
21493  *              ucmdbuf
21494  *              rqbuf
21495  *              rqbuflen
21496  *              bufaddr
21497  *              buflen
21498  *              feature
21499  *
21500  * Return Code: 0   - Success
21501  *              errno return code from sd_ssc_send()
21502  *
21503  *     Context: Can sleep. Does not return until command is completed.
21504  *
21505  */
21506 static int
21507 sd_send_scsi_feature_GET_CONFIGURATION(sd_ssc_t *ssc,
21508         struct uscsi_cmd *ucmdbuf, uchar_t *rqbuf, uint_t rqbuflen,
21509         uchar_t *bufaddr, uint_t buflen, char feature, int path_flag)
21510 {
21511         char    cdb[CDB_GROUP1];
21512         int     status;
21513         struct sd_lun   *un;
21514 
21515         ASSERT(ssc != NULL);
21516         un = ssc->ssc_un;
21517         ASSERT(un != NULL);
21518         ASSERT(!mutex_owned(SD_MUTEX(un)));
21519         ASSERT(bufaddr != NULL);
21520         ASSERT(ucmdbuf != NULL);
21521         ASSERT(rqbuf != NULL);
21522 
21523         SD_TRACE(SD_LOG_IO, un,
21524             "sd_send_scsi_feature_GET_CONFIGURATION: entry: un:0x%p\n", un);
21525 
21526         bzero(cdb, sizeof (cdb));
21527         bzero(ucmdbuf, sizeof (struct uscsi_cmd));
21528         bzero(rqbuf, rqbuflen);
21529         bzero(bufaddr, buflen);
21530 
21531         /*
21532          * Set up cdb field for the get configuration command.
21533          */
21534         cdb[0] = SCMD_GET_CONFIGURATION;
21535         cdb[1] = 0x02;  /* Requested Type */
21536         cdb[3] = feature;
21537         cdb[8] = buflen;
21538         ucmdbuf->uscsi_cdb = cdb;
21539         ucmdbuf->uscsi_cdblen = CDB_GROUP1;
21540         ucmdbuf->uscsi_bufaddr = (caddr_t)bufaddr;
21541         ucmdbuf->uscsi_buflen = buflen;
21542         ucmdbuf->uscsi_timeout = sd_io_time;
21543         ucmdbuf->uscsi_rqbuf = (caddr_t)rqbuf;
21544         ucmdbuf->uscsi_rqlen = rqbuflen;
21545         ucmdbuf->uscsi_flags = USCSI_RQENABLE|USCSI_SILENT|USCSI_READ;
21546 
21547         status = sd_ssc_send(ssc, ucmdbuf, FKIOCTL,
21548             UIO_SYSSPACE, path_flag);
21549 
21550         switch (status) {
21551         case 0:
21552 
21553                 break;  /* Success! */
21554         case EIO:
21555                 switch (ucmdbuf->uscsi_status) {
21556                 case STATUS_RESERVATION_CONFLICT:
21557                         status = EACCES;
21558                         break;
21559                 default:
21560                         break;
21561                 }
21562                 break;
21563         default:
21564                 break;
21565         }
21566 
21567         if (status == 0) {
21568                 SD_DUMP_MEMORY(un, SD_LOG_IO,
21569                     "sd_send_scsi_feature_GET_CONFIGURATION: data",
21570                     (uchar_t *)bufaddr, SD_PROFILE_HEADER_LEN, SD_LOG_HEX);
21571         }
21572 
21573         SD_TRACE(SD_LOG_IO, un,
21574             "sd_send_scsi_feature_GET_CONFIGURATION: exit\n");
21575 
21576         return (status);
21577 }
21578 
21579 
21580 /*
21581  *    Function: sd_send_scsi_MODE_SENSE
21582  *
21583  * Description: Utility function for issuing a scsi MODE SENSE command.
21584  *              Note: This routine uses a consistent implementation for Group0,
21585  *              Group1, and Group2 commands across all platforms. ATAPI devices
21586  *              use Group 1 Read/Write commands and Group 2 Mode Sense/Select
21587  *
21588  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
21589  *                      structure for this target.
21590  *              cdbsize - size CDB to be used (CDB_GROUP0 (6 byte), or
21591  *                        CDB_GROUP[1|2] (10 byte).
21592  *              bufaddr - buffer for page data retrieved from the target.
21593  *              buflen - size of page to be retrieved.
21594  *              page_code - page code of data to be retrieved from the target.
21595  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
21596  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
21597  *                      to use the USCSI "direct" chain and bypass the normal
21598  *                      command waitq.
21599  *
21600  * Return Code: 0   - Success
21601  *              errno return code from sd_ssc_send()
21602  *
21603  *     Context: Can sleep. Does not return until command is completed.
21604  */
21605 
21606 static int
21607 sd_send_scsi_MODE_SENSE(sd_ssc_t *ssc, int cdbsize, uchar_t *bufaddr,
21608         size_t buflen,  uchar_t page_code, int path_flag)
21609 {
21610         struct  scsi_extended_sense     sense_buf;
21611         union scsi_cdb          cdb;
21612         struct uscsi_cmd        ucmd_buf;
21613         int                     status;
21614         int                     headlen;
21615         struct sd_lun           *un;
21616 
21617         ASSERT(ssc != NULL);
21618         un = ssc->ssc_un;
21619         ASSERT(un != NULL);
21620         ASSERT(!mutex_owned(SD_MUTEX(un)));
21621         ASSERT(bufaddr != NULL);
21622         ASSERT((cdbsize == CDB_GROUP0) || (cdbsize == CDB_GROUP1) ||
21623             (cdbsize == CDB_GROUP2));
21624 
21625         SD_TRACE(SD_LOG_IO, un,
21626             "sd_send_scsi_MODE_SENSE: entry: un:0x%p\n", un);
21627 
21628         bzero(&cdb, sizeof (cdb));
21629         bzero(&ucmd_buf, sizeof (ucmd_buf));
21630         bzero(&sense_buf, sizeof (struct scsi_extended_sense));
21631         bzero(bufaddr, buflen);
21632 
21633         if (cdbsize == CDB_GROUP0) {
21634                 cdb.scc_cmd = SCMD_MODE_SENSE;
21635                 cdb.cdb_opaque[2] = page_code;
21636                 FORMG0COUNT(&cdb, buflen);
21637                 headlen = MODE_HEADER_LENGTH;
21638         } else {
21639                 cdb.scc_cmd = SCMD_MODE_SENSE_G1;
21640                 cdb.cdb_opaque[2] = page_code;
21641                 FORMG1COUNT(&cdb, buflen);
21642                 headlen = MODE_HEADER_LENGTH_GRP2;
21643         }
21644 
21645         ASSERT(headlen <= buflen);
21646         SD_FILL_SCSI1_LUN_CDB(un, &cdb);
21647 
21648         ucmd_buf.uscsi_cdb      = (char *)&cdb;
21649         ucmd_buf.uscsi_cdblen   = (uchar_t)cdbsize;
21650         ucmd_buf.uscsi_bufaddr  = (caddr_t)bufaddr;
21651         ucmd_buf.uscsi_buflen   = buflen;
21652         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
21653         ucmd_buf.uscsi_rqlen    = sizeof (struct scsi_extended_sense);
21654         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_READ | USCSI_SILENT;
21655         ucmd_buf.uscsi_timeout  = 60;
21656 
21657         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
21658             UIO_SYSSPACE, path_flag);
21659 
21660         switch (status) {
21661         case 0:
21662                 /*
21663                  * sr_check_wp() uses 0x3f page code and check the header of
21664                  * mode page to determine if target device is write-protected.
21665                  * But some USB devices return 0 bytes for 0x3f page code. For
21666                  * this case, make sure that mode page header is returned at
21667                  * least.
21668                  */
21669                 if (buflen - ucmd_buf.uscsi_resid <  headlen) {
21670                         status = EIO;
21671                         sd_ssc_set_info(ssc, SSC_FLAGS_INVALID_DATA, -1,
21672                             "mode page header is not returned");
21673                 }
21674                 break;  /* Success! */
21675         case EIO:
21676                 switch (ucmd_buf.uscsi_status) {
21677                 case STATUS_RESERVATION_CONFLICT:
21678                         status = EACCES;
21679                         break;
21680                 default:
21681                         break;
21682                 }
21683                 break;
21684         default:
21685                 break;
21686         }
21687 
21688         if (status == 0) {
21689                 SD_DUMP_MEMORY(un, SD_LOG_IO, "sd_send_scsi_MODE_SENSE: data",
21690                     (uchar_t *)bufaddr, buflen, SD_LOG_HEX);
21691         }
21692         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_MODE_SENSE: exit\n");
21693 
21694         return (status);
21695 }
21696 
21697 
21698 /*
21699  *    Function: sd_send_scsi_MODE_SELECT
21700  *
21701  * Description: Utility function for issuing a scsi MODE SELECT command.
21702  *              Note: This routine uses a consistent implementation for Group0,
21703  *              Group1, and Group2 commands across all platforms. ATAPI devices
21704  *              use Group 1 Read/Write commands and Group 2 Mode Sense/Select
21705  *
21706  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
21707  *                      structure for this target.
21708  *              cdbsize - size CDB to be used (CDB_GROUP0 (6 byte), or
21709  *                        CDB_GROUP[1|2] (10 byte).
21710  *              bufaddr - buffer for page data retrieved from the target.
21711  *              buflen - size of page to be retrieved.
21712  *              save_page - boolean to determin if SP bit should be set.
21713  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
21714  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
21715  *                      to use the USCSI "direct" chain and bypass the normal
21716  *                      command waitq.
21717  *
21718  * Return Code: 0   - Success
21719  *              errno return code from sd_ssc_send()
21720  *
21721  *     Context: Can sleep. Does not return until command is completed.
21722  */
21723 
21724 static int
21725 sd_send_scsi_MODE_SELECT(sd_ssc_t *ssc, int cdbsize, uchar_t *bufaddr,
21726         size_t buflen,  uchar_t save_page, int path_flag)
21727 {
21728         struct  scsi_extended_sense     sense_buf;
21729         union scsi_cdb          cdb;
21730         struct uscsi_cmd        ucmd_buf;
21731         int                     status;
21732         struct sd_lun           *un;
21733 
21734         ASSERT(ssc != NULL);
21735         un = ssc->ssc_un;
21736         ASSERT(un != NULL);
21737         ASSERT(!mutex_owned(SD_MUTEX(un)));
21738         ASSERT(bufaddr != NULL);
21739         ASSERT((cdbsize == CDB_GROUP0) || (cdbsize == CDB_GROUP1) ||
21740             (cdbsize == CDB_GROUP2));
21741 
21742         SD_TRACE(SD_LOG_IO, un,
21743             "sd_send_scsi_MODE_SELECT: entry: un:0x%p\n", un);
21744 
21745         bzero(&cdb, sizeof (cdb));
21746         bzero(&ucmd_buf, sizeof (ucmd_buf));
21747         bzero(&sense_buf, sizeof (struct scsi_extended_sense));
21748 
21749         /* Set the PF bit for many third party drives */
21750         cdb.cdb_opaque[1] = 0x10;
21751 
21752         /* Set the savepage(SP) bit if given */
21753         if (save_page == SD_SAVE_PAGE) {
21754                 cdb.cdb_opaque[1] |= 0x01;
21755         }
21756 
21757         if (cdbsize == CDB_GROUP0) {
21758                 cdb.scc_cmd = SCMD_MODE_SELECT;
21759                 FORMG0COUNT(&cdb, buflen);
21760         } else {
21761                 cdb.scc_cmd = SCMD_MODE_SELECT_G1;
21762                 FORMG1COUNT(&cdb, buflen);
21763         }
21764 
21765         SD_FILL_SCSI1_LUN_CDB(un, &cdb);
21766 
21767         ucmd_buf.uscsi_cdb      = (char *)&cdb;
21768         ucmd_buf.uscsi_cdblen   = (uchar_t)cdbsize;
21769         ucmd_buf.uscsi_bufaddr  = (caddr_t)bufaddr;
21770         ucmd_buf.uscsi_buflen   = buflen;
21771         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
21772         ucmd_buf.uscsi_rqlen    = sizeof (struct scsi_extended_sense);
21773         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_WRITE | USCSI_SILENT;
21774         ucmd_buf.uscsi_timeout  = 60;
21775 
21776         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
21777             UIO_SYSSPACE, path_flag);
21778 
21779         switch (status) {
21780         case 0:
21781                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
21782                 break;  /* Success! */
21783         case EIO:
21784                 switch (ucmd_buf.uscsi_status) {
21785                 case STATUS_RESERVATION_CONFLICT:
21786                         status = EACCES;
21787                         break;
21788                 default:
21789                         break;
21790                 }
21791                 break;
21792         default:
21793                 break;
21794         }
21795 
21796         if (status == 0) {
21797                 SD_DUMP_MEMORY(un, SD_LOG_IO, "sd_send_scsi_MODE_SELECT: data",
21798                     (uchar_t *)bufaddr, buflen, SD_LOG_HEX);
21799         }
21800         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_MODE_SELECT: exit\n");
21801 
21802         return (status);
21803 }
21804 
21805 
21806 /*
21807  *    Function: sd_send_scsi_RDWR
21808  *
21809  * Description: Issue a scsi READ or WRITE command with the given parameters.
21810  *
21811  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
21812  *                      structure for this target.
21813  *              cmd:     SCMD_READ or SCMD_WRITE
21814  *              bufaddr: Address of caller's buffer to receive the RDWR data
21815  *              buflen:  Length of caller's buffer receive the RDWR data.
21816  *              start_block: Block number for the start of the RDWR operation.
21817  *                       (Assumes target-native block size.)
21818  *              residp:  Pointer to variable to receive the redisual of the
21819  *                       RDWR operation (may be NULL of no residual requested).
21820  *              path_flag - SD_PATH_DIRECT to use the USCSI "direct" chain and
21821  *                      the normal command waitq, or SD_PATH_DIRECT_PRIORITY
21822  *                      to use the USCSI "direct" chain and bypass the normal
21823  *                      command waitq.
21824  *
21825  * Return Code: 0   - Success
21826  *              errno return code from sd_ssc_send()
21827  *
21828  *     Context: Can sleep. Does not return until command is completed.
21829  */
21830 
21831 static int
21832 sd_send_scsi_RDWR(sd_ssc_t *ssc, uchar_t cmd, void *bufaddr,
21833         size_t buflen, daddr_t start_block, int path_flag)
21834 {
21835         struct  scsi_extended_sense     sense_buf;
21836         union scsi_cdb          cdb;
21837         struct uscsi_cmd        ucmd_buf;
21838         uint32_t                block_count;
21839         int                     status;
21840         int                     cdbsize;
21841         uchar_t                 flag;
21842         struct sd_lun           *un;
21843 
21844         ASSERT(ssc != NULL);
21845         un = ssc->ssc_un;
21846         ASSERT(un != NULL);
21847         ASSERT(!mutex_owned(SD_MUTEX(un)));
21848         ASSERT(bufaddr != NULL);
21849         ASSERT((cmd == SCMD_READ) || (cmd == SCMD_WRITE));
21850 
21851         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_RDWR: entry: un:0x%p\n", un);
21852 
21853         if (un->un_f_tgt_blocksize_is_valid != TRUE) {
21854                 return (EINVAL);
21855         }
21856 
21857         mutex_enter(SD_MUTEX(un));
21858         block_count = SD_BYTES2TGTBLOCKS(un, buflen);
21859         mutex_exit(SD_MUTEX(un));
21860 
21861         flag = (cmd == SCMD_READ) ? USCSI_READ : USCSI_WRITE;
21862 
21863         SD_INFO(SD_LOG_IO, un, "sd_send_scsi_RDWR: "
21864             "bufaddr:0x%p buflen:0x%x start_block:0x%p block_count:0x%x\n",
21865             bufaddr, buflen, start_block, block_count);
21866 
21867         bzero(&cdb, sizeof (cdb));
21868         bzero(&ucmd_buf, sizeof (ucmd_buf));
21869         bzero(&sense_buf, sizeof (struct scsi_extended_sense));
21870 
21871         /* Compute CDB size to use */
21872         if (start_block > 0xffffffff)
21873                 cdbsize = CDB_GROUP4;
21874         else if ((start_block & 0xFFE00000) ||
21875             (un->un_f_cfg_is_atapi == TRUE))
21876                 cdbsize = CDB_GROUP1;
21877         else
21878                 cdbsize = CDB_GROUP0;
21879 
21880         switch (cdbsize) {
21881         case CDB_GROUP0:        /* 6-byte CDBs */
21882                 cdb.scc_cmd = cmd;
21883                 FORMG0ADDR(&cdb, start_block);
21884                 FORMG0COUNT(&cdb, block_count);
21885                 break;
21886         case CDB_GROUP1:        /* 10-byte CDBs */
21887                 cdb.scc_cmd = cmd | SCMD_GROUP1;
21888                 FORMG1ADDR(&cdb, start_block);
21889                 FORMG1COUNT(&cdb, block_count);
21890                 break;
21891         case CDB_GROUP4:        /* 16-byte CDBs */
21892                 cdb.scc_cmd = cmd | SCMD_GROUP4;
21893                 FORMG4LONGADDR(&cdb, (uint64_t)start_block);
21894                 FORMG4COUNT(&cdb, block_count);
21895                 break;
21896         case CDB_GROUP5:        /* 12-byte CDBs (currently unsupported) */
21897         default:
21898                 /* All others reserved */
21899                 return (EINVAL);
21900         }
21901 
21902         /* Set LUN bit(s) in CDB if this is a SCSI-1 device */
21903         SD_FILL_SCSI1_LUN_CDB(un, &cdb);
21904 
21905         ucmd_buf.uscsi_cdb      = (char *)&cdb;
21906         ucmd_buf.uscsi_cdblen   = (uchar_t)cdbsize;
21907         ucmd_buf.uscsi_bufaddr  = bufaddr;
21908         ucmd_buf.uscsi_buflen   = buflen;
21909         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
21910         ucmd_buf.uscsi_rqlen    = sizeof (struct scsi_extended_sense);
21911         ucmd_buf.uscsi_flags    = flag | USCSI_RQENABLE | USCSI_SILENT;
21912         ucmd_buf.uscsi_timeout  = 60;
21913         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
21914             UIO_SYSSPACE, path_flag);
21915 
21916         switch (status) {
21917         case 0:
21918                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
21919                 break;  /* Success! */
21920         case EIO:
21921                 switch (ucmd_buf.uscsi_status) {
21922                 case STATUS_RESERVATION_CONFLICT:
21923                         status = EACCES;
21924                         break;
21925                 default:
21926                         break;
21927                 }
21928                 break;
21929         default:
21930                 break;
21931         }
21932 
21933         if (status == 0) {
21934                 SD_DUMP_MEMORY(un, SD_LOG_IO, "sd_send_scsi_RDWR: data",
21935                     (uchar_t *)bufaddr, buflen, SD_LOG_HEX);
21936         }
21937 
21938         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_RDWR: exit\n");
21939 
21940         return (status);
21941 }
21942 
21943 
21944 /*
21945  *    Function: sd_send_scsi_LOG_SENSE
21946  *
21947  * Description: Issue a scsi LOG_SENSE command with the given parameters.
21948  *
21949  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
21950  *                      structure for this target.
21951  *
21952  * Return Code: 0   - Success
21953  *              errno return code from sd_ssc_send()
21954  *
21955  *     Context: Can sleep. Does not return until command is completed.
21956  */
21957 
21958 static int
21959 sd_send_scsi_LOG_SENSE(sd_ssc_t *ssc, uchar_t *bufaddr, uint16_t buflen,
21960         uchar_t page_code, uchar_t page_control, uint16_t param_ptr,
21961         int path_flag)
21962 
21963 {
21964         struct scsi_extended_sense      sense_buf;
21965         union scsi_cdb          cdb;
21966         struct uscsi_cmd        ucmd_buf;
21967         int                     status;
21968         struct sd_lun           *un;
21969 
21970         ASSERT(ssc != NULL);
21971         un = ssc->ssc_un;
21972         ASSERT(un != NULL);
21973         ASSERT(!mutex_owned(SD_MUTEX(un)));
21974 
21975         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_LOG_SENSE: entry: un:0x%p\n", un);
21976 
21977         bzero(&cdb, sizeof (cdb));
21978         bzero(&ucmd_buf, sizeof (ucmd_buf));
21979         bzero(&sense_buf, sizeof (struct scsi_extended_sense));
21980 
21981         cdb.scc_cmd = SCMD_LOG_SENSE_G1;
21982         cdb.cdb_opaque[2] = (page_control << 6) | page_code;
21983         cdb.cdb_opaque[5] = (uchar_t)((param_ptr & 0xFF00) >> 8);
21984         cdb.cdb_opaque[6] = (uchar_t)(param_ptr  & 0x00FF);
21985         FORMG1COUNT(&cdb, buflen);
21986 
21987         ucmd_buf.uscsi_cdb      = (char *)&cdb;
21988         ucmd_buf.uscsi_cdblen   = CDB_GROUP1;
21989         ucmd_buf.uscsi_bufaddr  = (caddr_t)bufaddr;
21990         ucmd_buf.uscsi_buflen   = buflen;
21991         ucmd_buf.uscsi_rqbuf    = (caddr_t)&sense_buf;
21992         ucmd_buf.uscsi_rqlen    = sizeof (struct scsi_extended_sense);
21993         ucmd_buf.uscsi_flags    = USCSI_RQENABLE | USCSI_READ | USCSI_SILENT;
21994         ucmd_buf.uscsi_timeout  = 60;
21995 
21996         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
21997             UIO_SYSSPACE, path_flag);
21998 
21999         switch (status) {
22000         case 0:
22001                 break;
22002         case EIO:
22003                 switch (ucmd_buf.uscsi_status) {
22004                 case STATUS_RESERVATION_CONFLICT:
22005                         status = EACCES;
22006                         break;
22007                 case STATUS_CHECK:
22008                         if ((ucmd_buf.uscsi_rqstatus == STATUS_GOOD) &&
22009                             (scsi_sense_key((uint8_t *)&sense_buf) ==
22010                                 KEY_ILLEGAL_REQUEST) &&
22011                             (scsi_sense_asc((uint8_t *)&sense_buf) == 0x24)) {
22012                                 /*
22013                                  * ASC 0x24: INVALID FIELD IN CDB
22014                                  */
22015                                 switch (page_code) {
22016                                 case START_STOP_CYCLE_PAGE:
22017                                         /*
22018                                          * The start stop cycle counter is
22019                                          * implemented as page 0x31 in earlier
22020                                          * generation disks. In new generation
22021                                          * disks the start stop cycle counter is
22022                                          * implemented as page 0xE. To properly
22023                                          * handle this case if an attempt for
22024                                          * log page 0xE is made and fails we
22025                                          * will try again using page 0x31.
22026                                          *
22027                                          * Network storage BU committed to
22028                                          * maintain the page 0x31 for this
22029                                          * purpose and will not have any other
22030                                          * page implemented with page code 0x31
22031                                          * until all disks transition to the
22032                                          * standard page.
22033                                          */
22034                                         mutex_enter(SD_MUTEX(un));
22035                                         un->un_start_stop_cycle_page =
22036                                             START_STOP_CYCLE_VU_PAGE;
22037                                         cdb.cdb_opaque[2] =
22038                                             (char)(page_control << 6) |
22039                                             un->un_start_stop_cycle_page;
22040                                         mutex_exit(SD_MUTEX(un));
22041                                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
22042                                         status = sd_ssc_send(
22043                                             ssc, &ucmd_buf, FKIOCTL,
22044                                             UIO_SYSSPACE, path_flag);
22045 
22046                                         break;
22047                                 case TEMPERATURE_PAGE:
22048                                         status = ENOTTY;
22049                                         break;
22050                                 default:
22051                                         break;
22052                                 }
22053                         }
22054                         break;
22055                 default:
22056                         break;
22057                 }
22058                 break;
22059         default:
22060                 break;
22061         }
22062 
22063         if (status == 0) {
22064                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
22065                 SD_DUMP_MEMORY(un, SD_LOG_IO, "sd_send_scsi_LOG_SENSE: data",
22066                     (uchar_t *)bufaddr, buflen, SD_LOG_HEX);
22067         }
22068 
22069         SD_TRACE(SD_LOG_IO, un, "sd_send_scsi_LOG_SENSE: exit\n");
22070 
22071         return (status);
22072 }
22073 
22074 
22075 /*
22076  *    Function: sd_send_scsi_GET_EVENT_STATUS_NOTIFICATION
22077  *
22078  * Description: Issue the scsi GET EVENT STATUS NOTIFICATION command.
22079  *
22080  *   Arguments: ssc   - ssc contains pointer to driver soft state (unit)
22081  *                      structure for this target.
22082  *              bufaddr
22083  *              buflen
22084  *              class_req
22085  *
22086  * Return Code: 0   - Success
22087  *              errno return code from sd_ssc_send()
22088  *
22089  *     Context: Can sleep. Does not return until command is completed.
22090  */
22091 
22092 static int
22093 sd_send_scsi_GET_EVENT_STATUS_NOTIFICATION(sd_ssc_t *ssc, uchar_t *bufaddr,
22094         size_t buflen, uchar_t class_req)
22095 {
22096         union scsi_cdb          cdb;
22097         struct uscsi_cmd        ucmd_buf;
22098         int                     status;
22099         struct sd_lun           *un;
22100 
22101         ASSERT(ssc != NULL);
22102         un = ssc->ssc_un;
22103         ASSERT(un != NULL);
22104         ASSERT(!mutex_owned(SD_MUTEX(un)));
22105         ASSERT(bufaddr != NULL);
22106 
22107         SD_TRACE(SD_LOG_IO, un,
22108             "sd_send_scsi_GET_EVENT_STATUS_NOTIFICATION: entry: un:0x%p\n", un);
22109 
22110         bzero(&cdb, sizeof (cdb));
22111         bzero(&ucmd_buf, sizeof (ucmd_buf));
22112         bzero(bufaddr, buflen);
22113 
22114         cdb.scc_cmd = SCMD_GET_EVENT_STATUS_NOTIFICATION;
22115         cdb.cdb_opaque[1] = 1; /* polled */
22116         cdb.cdb_opaque[4] = class_req;
22117         FORMG1COUNT(&cdb, buflen);
22118 
22119         ucmd_buf.uscsi_cdb      = (char *)&cdb;
22120         ucmd_buf.uscsi_cdblen   = CDB_GROUP1;
22121         ucmd_buf.uscsi_bufaddr  = (caddr_t)bufaddr;
22122         ucmd_buf.uscsi_buflen   = buflen;
22123         ucmd_buf.uscsi_rqbuf    = NULL;
22124         ucmd_buf.uscsi_rqlen    = 0;
22125         ucmd_buf.uscsi_flags    = USCSI_READ | USCSI_SILENT;
22126         ucmd_buf.uscsi_timeout  = 60;
22127 
22128         status = sd_ssc_send(ssc, &ucmd_buf, FKIOCTL,
22129             UIO_SYSSPACE, SD_PATH_DIRECT);
22130 
22131         /*
22132          * Only handle status == 0, the upper-level caller
22133          * will put different assessment based on the context.
22134          */
22135         if (status == 0) {
22136                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
22137 
22138                 if (ucmd_buf.uscsi_resid != 0) {
22139                         status = EIO;
22140                 }
22141         }
22142 
22143         SD_TRACE(SD_LOG_IO, un,
22144             "sd_send_scsi_GET_EVENT_STATUS_NOTIFICATION: exit\n");
22145 
22146         return (status);
22147 }
22148 
22149 
22150 static boolean_t
22151 sd_gesn_media_data_valid(uchar_t *data)
22152 {
22153         uint16_t                        len;
22154 
22155         len = (data[1] << 8) | data[0];
22156         return ((len >= 6) &&
22157             ((data[2] & SD_GESN_HEADER_NEA) == 0) &&
22158             ((data[2] & SD_GESN_HEADER_CLASS) == SD_GESN_MEDIA_CLASS) &&
22159             ((data[3] & (1 << SD_GESN_MEDIA_CLASS)) != 0));
22160 }
22161 
22162 
22163 /*
22164  *    Function: sdioctl
22165  *
22166  * Description: Driver's ioctl(9e) entry point function.
22167  *
22168  *   Arguments: dev     - device number
22169  *              cmd     - ioctl operation to be performed
22170  *              arg     - user argument, contains data to be set or reference
22171  *                        parameter for get
22172  *              flag    - bit flag, indicating open settings, 32/64 bit type
22173  *              cred_p  - user credential pointer
22174  *              rval_p  - calling process return value (OPT)
22175  *
22176  * Return Code: EINVAL
22177  *              ENOTTY
22178  *              ENXIO
22179  *              EIO
22180  *              EFAULT
22181  *              ENOTSUP
22182  *              EPERM
22183  *
22184  *     Context: Called from the device switch at normal priority.
22185  */
22186 
22187 static int
22188 sdioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cred_p, int *rval_p)
22189 {
22190         struct sd_lun   *un = NULL;
22191         int             err = 0;
22192         int             i = 0;
22193         cred_t          *cr;
22194         int             tmprval = EINVAL;
22195         boolean_t       is_valid;
22196         sd_ssc_t        *ssc;
22197 
22198         /*
22199          * All device accesses go thru sdstrategy where we check on suspend
22200          * status
22201          */
22202         if ((un = ddi_get_soft_state(sd_state, SDUNIT(dev))) == NULL) {
22203                 return (ENXIO);
22204         }
22205 
22206         ASSERT(!mutex_owned(SD_MUTEX(un)));
22207 
22208         /* Initialize sd_ssc_t for internal uscsi commands */
22209         ssc = sd_ssc_init(un);
22210 
22211         is_valid = SD_IS_VALID_LABEL(un);
22212 
22213         /*
22214          * Moved this wait from sd_uscsi_strategy to here for
22215          * reasons of deadlock prevention. Internal driver commands,
22216          * specifically those to change a devices power level, result
22217          * in a call to sd_uscsi_strategy.
22218          */
22219         mutex_enter(SD_MUTEX(un));
22220         while ((un->un_state == SD_STATE_SUSPENDED) ||
22221             (un->un_state == SD_STATE_PM_CHANGING)) {
22222                 cv_wait(&un->un_suspend_cv, SD_MUTEX(un));
22223         }
22224         /*
22225          * Twiddling the counter here protects commands from now
22226          * through to the top of sd_uscsi_strategy. Without the
22227          * counter inc. a power down, for example, could get in
22228          * after the above check for state is made and before
22229          * execution gets to the top of sd_uscsi_strategy.
22230          * That would cause problems.
22231          */
22232         un->un_ncmds_in_driver++;
22233 
22234         if (!is_valid &&
22235             (flag & (FNDELAY | FNONBLOCK))) {
22236                 switch (cmd) {
22237                 case DKIOCGGEOM:        /* SD_PATH_DIRECT */
22238                 case DKIOCGVTOC:
22239                 case DKIOCGEXTVTOC:
22240                 case DKIOCGAPART:
22241                 case DKIOCPARTINFO:
22242                 case DKIOCEXTPARTINFO:
22243                 case DKIOCSGEOM:
22244                 case DKIOCSAPART:
22245                 case DKIOCGETEFI:
22246                 case DKIOCPARTITION:
22247                 case DKIOCSVTOC:
22248                 case DKIOCSEXTVTOC:
22249                 case DKIOCSETEFI:
22250                 case DKIOCGMBOOT:
22251                 case DKIOCSMBOOT:
22252                 case DKIOCG_PHYGEOM:
22253                 case DKIOCG_VIRTGEOM:
22254 #if defined(__i386) || defined(__amd64)
22255                 case DKIOCSETEXTPART:
22256 #endif
22257                         /* let cmlb handle it */
22258                         goto skip_ready_valid;
22259 
22260                 case CDROMPAUSE:
22261                 case CDROMRESUME:
22262                 case CDROMPLAYMSF:
22263                 case CDROMPLAYTRKIND:
22264                 case CDROMREADTOCHDR:
22265                 case CDROMREADTOCENTRY:
22266                 case CDROMSTOP:
22267                 case CDROMSTART:
22268                 case CDROMVOLCTRL:
22269                 case CDROMSUBCHNL:
22270                 case CDROMREADMODE2:
22271                 case CDROMREADMODE1:
22272                 case CDROMREADOFFSET:
22273                 case CDROMSBLKMODE:
22274                 case CDROMGBLKMODE:
22275                 case CDROMGDRVSPEED:
22276                 case CDROMSDRVSPEED:
22277                 case CDROMCDDA:
22278                 case CDROMCDXA:
22279                 case CDROMSUBCODE:
22280                         if (!ISCD(un)) {
22281                                 un->un_ncmds_in_driver--;
22282                                 ASSERT(un->un_ncmds_in_driver >= 0);
22283                                 mutex_exit(SD_MUTEX(un));
22284                                 err = ENOTTY;
22285                                 goto done_without_assess;
22286                         }
22287                         break;
22288                 case FDEJECT:
22289                 case DKIOCEJECT:
22290                 case CDROMEJECT:
22291                         if (!un->un_f_eject_media_supported) {
22292                                 un->un_ncmds_in_driver--;
22293                                 ASSERT(un->un_ncmds_in_driver >= 0);
22294                                 mutex_exit(SD_MUTEX(un));
22295                                 err = ENOTTY;
22296                                 goto done_without_assess;
22297                         }
22298                         break;
22299                 case DKIOCFLUSHWRITECACHE:
22300                         mutex_exit(SD_MUTEX(un));
22301                         err = sd_send_scsi_TEST_UNIT_READY(ssc, 0);
22302                         if (err != 0) {
22303                                 mutex_enter(SD_MUTEX(un));
22304                                 un->un_ncmds_in_driver--;
22305                                 ASSERT(un->un_ncmds_in_driver >= 0);
22306                                 mutex_exit(SD_MUTEX(un));
22307                                 err = EIO;
22308                                 goto done_quick_assess;
22309                         }
22310                         mutex_enter(SD_MUTEX(un));
22311                         /* FALLTHROUGH */
22312                 case DKIOCREMOVABLE:
22313                 case DKIOCHOTPLUGGABLE:
22314                 case DKIOCINFO:
22315                 case DKIOCGMEDIAINFO:
22316                 case DKIOCGMEDIAINFOEXT:
22317                 case MHIOCENFAILFAST:
22318                 case MHIOCSTATUS:
22319                 case MHIOCTKOWN:
22320                 case MHIOCRELEASE:
22321                 case MHIOCGRP_INKEYS:
22322                 case MHIOCGRP_INRESV:
22323                 case MHIOCGRP_REGISTER:
22324                 case MHIOCGRP_CLEAR:
22325                 case MHIOCGRP_RESERVE:
22326                 case MHIOCGRP_PREEMPTANDABORT:
22327                 case MHIOCGRP_REGISTERANDIGNOREKEY:
22328                 case CDROMCLOSETRAY:
22329                 case USCSICMD:
22330                         goto skip_ready_valid;
22331                 default:
22332                         break;
22333                 }
22334 
22335                 mutex_exit(SD_MUTEX(un));
22336                 err = sd_ready_and_valid(ssc, SDPART(dev));
22337                 mutex_enter(SD_MUTEX(un));
22338 
22339                 if (err != SD_READY_VALID) {
22340                         switch (cmd) {
22341                         case DKIOCSTATE:
22342                         case CDROMGDRVSPEED:
22343                         case CDROMSDRVSPEED:
22344                         case FDEJECT:   /* for eject command */
22345                         case DKIOCEJECT:
22346                         case CDROMEJECT:
22347                         case DKIOCREMOVABLE:
22348                         case DKIOCHOTPLUGGABLE:
22349                                 break;
22350                         default:
22351                                 if (un->un_f_has_removable_media) {
22352                                         err = ENXIO;
22353                                 } else {
22354                                 /* Do not map SD_RESERVED_BY_OTHERS to EIO */
22355                                         if (err == SD_RESERVED_BY_OTHERS) {
22356                                                 err = EACCES;
22357                                         } else {
22358                                                 err = EIO;
22359                                         }
22360                                 }
22361                                 un->un_ncmds_in_driver--;
22362                                 ASSERT(un->un_ncmds_in_driver >= 0);
22363                                 mutex_exit(SD_MUTEX(un));
22364 
22365                                 goto done_without_assess;
22366                         }
22367                 }
22368         }
22369 
22370 skip_ready_valid:
22371         mutex_exit(SD_MUTEX(un));
22372 
22373         switch (cmd) {
22374         case DKIOCINFO:
22375                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCINFO\n");
22376                 err = sd_dkio_ctrl_info(dev, (caddr_t)arg, flag);
22377                 break;
22378 
22379         case DKIOCGMEDIAINFO:
22380                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCGMEDIAINFO\n");
22381                 err = sd_get_media_info(dev, (caddr_t)arg, flag);
22382                 break;
22383 
22384         case DKIOCGMEDIAINFOEXT:
22385                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCGMEDIAINFOEXT\n");
22386                 err = sd_get_media_info_ext(dev, (caddr_t)arg, flag);
22387                 break;
22388 
22389         case DKIOCGGEOM:
22390         case DKIOCGVTOC:
22391         case DKIOCGEXTVTOC:
22392         case DKIOCGAPART:
22393         case DKIOCPARTINFO:
22394         case DKIOCEXTPARTINFO:
22395         case DKIOCSGEOM:
22396         case DKIOCSAPART:
22397         case DKIOCGETEFI:
22398         case DKIOCPARTITION:
22399         case DKIOCSVTOC:
22400         case DKIOCSEXTVTOC:
22401         case DKIOCSETEFI:
22402         case DKIOCGMBOOT:
22403         case DKIOCSMBOOT:
22404         case DKIOCG_PHYGEOM:
22405         case DKIOCG_VIRTGEOM:
22406 #if defined(__i386) || defined(__amd64)
22407         case DKIOCSETEXTPART:
22408 #endif
22409                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOC %d\n", cmd);
22410 
22411                 /* TUR should spin up */
22412 
22413                 if (un->un_f_has_removable_media)
22414                         err = sd_send_scsi_TEST_UNIT_READY(ssc,
22415                             SD_CHECK_FOR_MEDIA);
22416 
22417                 else
22418                         err = sd_send_scsi_TEST_UNIT_READY(ssc, 0);
22419 
22420                 if (err != 0)
22421                         goto done_with_assess;
22422 
22423                 err = cmlb_ioctl(un->un_cmlbhandle, dev,
22424                     cmd, arg, flag, cred_p, rval_p, (void *)SD_PATH_DIRECT);
22425 
22426                 if ((err == 0) &&
22427                     ((cmd == DKIOCSETEFI) ||
22428                     (un->un_f_pkstats_enabled) &&
22429                     (cmd == DKIOCSAPART || cmd == DKIOCSVTOC ||
22430                     cmd == DKIOCSEXTVTOC))) {
22431 
22432                         tmprval = cmlb_validate(un->un_cmlbhandle, CMLB_SILENT,
22433                             (void *)SD_PATH_DIRECT);
22434                         if ((tmprval == 0) && un->un_f_pkstats_enabled) {
22435                                 sd_set_pstats(un);
22436                                 SD_TRACE(SD_LOG_IO_PARTITION, un,
22437                                     "sd_ioctl: un:0x%p pstats created and "
22438                                     "set\n", un);
22439                         }
22440                 }
22441 
22442                 if ((cmd == DKIOCSVTOC || cmd == DKIOCSEXTVTOC) ||
22443                     ((cmd == DKIOCSETEFI) && (tmprval == 0))) {
22444 
22445                         mutex_enter(SD_MUTEX(un));
22446                         if (un->un_f_devid_supported &&
22447                             (un->un_f_opt_fab_devid == TRUE)) {
22448                                 if (un->un_devid == NULL) {
22449                                         sd_register_devid(ssc, SD_DEVINFO(un),
22450                                             SD_TARGET_IS_UNRESERVED);
22451                                 } else {
22452                                         /*
22453                                          * The device id for this disk
22454                                          * has been fabricated. The
22455                                          * device id must be preserved
22456                                          * by writing it back out to
22457                                          * disk.
22458                                          */
22459                                         if (sd_write_deviceid(ssc) != 0) {
22460                                                 ddi_devid_free(un->un_devid);
22461                                                 un->un_devid = NULL;
22462                                         }
22463                                 }
22464                         }
22465                         mutex_exit(SD_MUTEX(un));
22466                 }
22467 
22468                 break;
22469 
22470         case DKIOCLOCK:
22471                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCLOCK\n");
22472                 err = sd_send_scsi_DOORLOCK(ssc, SD_REMOVAL_PREVENT,
22473                     SD_PATH_STANDARD);
22474                 goto done_with_assess;
22475 
22476         case DKIOCUNLOCK:
22477                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCUNLOCK\n");
22478                 err = sd_send_scsi_DOORLOCK(ssc, SD_REMOVAL_ALLOW,
22479                     SD_PATH_STANDARD);
22480                 goto done_with_assess;
22481 
22482         case DKIOCSTATE: {
22483                 enum dkio_state         state;
22484                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCSTATE\n");
22485 
22486                 if (ddi_copyin((void *)arg, &state, sizeof (int), flag) != 0) {
22487                         err = EFAULT;
22488                 } else {
22489                         err = sd_check_media(dev, state);
22490                         if (err == 0) {
22491                                 if (ddi_copyout(&un->un_mediastate, (void *)arg,
22492                                     sizeof (int), flag) != 0)
22493                                         err = EFAULT;
22494                         }
22495                 }
22496                 break;
22497         }
22498 
22499         case DKIOCREMOVABLE:
22500                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCREMOVABLE\n");
22501                 i = un->un_f_has_removable_media ? 1 : 0;
22502                 if (ddi_copyout(&i, (void *)arg, sizeof (int), flag) != 0) {
22503                         err = EFAULT;
22504                 } else {
22505                         err = 0;
22506                 }
22507                 break;
22508 
22509         case DKIOCHOTPLUGGABLE:
22510                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCHOTPLUGGABLE\n");
22511                 i = un->un_f_is_hotpluggable ? 1 : 0;
22512                 if (ddi_copyout(&i, (void *)arg, sizeof (int), flag) != 0) {
22513                         err = EFAULT;
22514                 } else {
22515                         err = 0;
22516                 }
22517                 break;
22518 
22519         case DKIOCREADONLY:
22520                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCREADONLY\n");
22521                 i = 0;
22522                 if ((ISCD(un) && !un->un_f_mmc_writable_media) ||
22523                     (sr_check_wp(dev) != 0)) {
22524                         i = 1;
22525                 }
22526                 if (ddi_copyout(&i, (void *)arg, sizeof (int), flag) != 0) {
22527                         err = EFAULT;
22528                 } else {
22529                         err = 0;
22530                 }
22531                 break;
22532 
22533         case DKIOCGTEMPERATURE:
22534                 SD_TRACE(SD_LOG_IOCTL, un, "DKIOCGTEMPERATURE\n");
22535                 err = sd_dkio_get_temp(dev, (caddr_t)arg, flag);
22536                 break;
22537 
22538         case MHIOCENFAILFAST:
22539                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCENFAILFAST\n");
22540                 if ((err = drv_priv(cred_p)) == 0) {
22541                         err = sd_mhdioc_failfast(dev, (caddr_t)arg, flag);
22542                 }
22543                 break;
22544 
22545         case MHIOCTKOWN:
22546                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCTKOWN\n");
22547                 if ((err = drv_priv(cred_p)) == 0) {
22548                         err = sd_mhdioc_takeown(dev, (caddr_t)arg, flag);
22549                 }
22550                 break;
22551 
22552         case MHIOCRELEASE:
22553                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCRELEASE\n");
22554                 if ((err = drv_priv(cred_p)) == 0) {
22555                         err = sd_mhdioc_release(dev);
22556                 }
22557                 break;
22558 
22559         case MHIOCSTATUS:
22560                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCSTATUS\n");
22561                 if ((err = drv_priv(cred_p)) == 0) {
22562                         switch (sd_send_scsi_TEST_UNIT_READY(ssc, 0)) {
22563                         case 0:
22564                                 err = 0;
22565                                 break;
22566                         case EACCES:
22567                                 *rval_p = 1;
22568                                 err = 0;
22569                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
22570                                 break;
22571                         default:
22572                                 err = EIO;
22573                                 goto done_with_assess;
22574                         }
22575                 }
22576                 break;
22577 
22578         case MHIOCQRESERVE:
22579                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCQRESERVE\n");
22580                 if ((err = drv_priv(cred_p)) == 0) {
22581                         err = sd_reserve_release(dev, SD_RESERVE);
22582                 }
22583                 break;
22584 
22585         case MHIOCREREGISTERDEVID:
22586                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCREREGISTERDEVID\n");
22587                 if (drv_priv(cred_p) == EPERM) {
22588                         err = EPERM;
22589                 } else if (!un->un_f_devid_supported) {
22590                         err = ENOTTY;
22591                 } else {
22592                         err = sd_mhdioc_register_devid(dev);
22593                 }
22594                 break;
22595 
22596         case MHIOCGRP_INKEYS:
22597                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCGRP_INKEYS\n");
22598                 if (((err = drv_priv(cred_p)) != EPERM) && arg != NULL) {
22599                         if (un->un_reservation_type == SD_SCSI2_RESERVATION) {
22600                                 err = ENOTSUP;
22601                         } else {
22602                                 err = sd_mhdioc_inkeys(dev, (caddr_t)arg,
22603                                     flag);
22604                         }
22605                 }
22606                 break;
22607 
22608         case MHIOCGRP_INRESV:
22609                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCGRP_INRESV\n");
22610                 if (((err = drv_priv(cred_p)) != EPERM) && arg != NULL) {
22611                         if (un->un_reservation_type == SD_SCSI2_RESERVATION) {
22612                                 err = ENOTSUP;
22613                         } else {
22614                                 err = sd_mhdioc_inresv(dev, (caddr_t)arg, flag);
22615                         }
22616                 }
22617                 break;
22618 
22619         case MHIOCGRP_REGISTER:
22620                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCGRP_REGISTER\n");
22621                 if ((err = drv_priv(cred_p)) != EPERM) {
22622                         if (un->un_reservation_type == SD_SCSI2_RESERVATION) {
22623                                 err = ENOTSUP;
22624                         } else if (arg != NULL) {
22625                                 mhioc_register_t reg;
22626                                 if (ddi_copyin((void *)arg, &reg,
22627                                     sizeof (mhioc_register_t), flag) != 0) {
22628                                         err = EFAULT;
22629                                 } else {
22630                                         err =
22631                                             sd_send_scsi_PERSISTENT_RESERVE_OUT(
22632                                             ssc, SD_SCSI3_REGISTER,
22633                                             (uchar_t *)&reg);
22634                                         if (err != 0)
22635                                                 goto done_with_assess;
22636                                 }
22637                         }
22638                 }
22639                 break;
22640 
22641         case MHIOCGRP_CLEAR:
22642                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCGRP_CLEAR\n");
22643                 if ((err = drv_priv(cred_p)) != EPERM) {
22644                         if (un->un_reservation_type == SD_SCSI2_RESERVATION) {
22645                                 err = ENOTSUP;
22646                         } else if (arg != NULL) {
22647                                 mhioc_register_t reg;
22648                                 if (ddi_copyin((void *)arg, &reg,
22649                                     sizeof (mhioc_register_t), flag) != 0) {
22650                                         err = EFAULT;
22651                                 } else {
22652                                         err =
22653                                             sd_send_scsi_PERSISTENT_RESERVE_OUT(
22654                                             ssc, SD_SCSI3_CLEAR,
22655                                             (uchar_t *)&reg);
22656                                         if (err != 0)
22657                                                 goto done_with_assess;
22658                                 }
22659                         }
22660                 }
22661                 break;
22662 
22663         case MHIOCGRP_RESERVE:
22664                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCGRP_RESERVE\n");
22665                 if ((err = drv_priv(cred_p)) != EPERM) {
22666                         if (un->un_reservation_type == SD_SCSI2_RESERVATION) {
22667                                 err = ENOTSUP;
22668                         } else if (arg != NULL) {
22669                                 mhioc_resv_desc_t resv_desc;
22670                                 if (ddi_copyin((void *)arg, &resv_desc,
22671                                     sizeof (mhioc_resv_desc_t), flag) != 0) {
22672                                         err = EFAULT;
22673                                 } else {
22674                                         err =
22675                                             sd_send_scsi_PERSISTENT_RESERVE_OUT(
22676                                             ssc, SD_SCSI3_RESERVE,
22677                                             (uchar_t *)&resv_desc);
22678                                         if (err != 0)
22679                                                 goto done_with_assess;
22680                                 }
22681                         }
22682                 }
22683                 break;
22684 
22685         case MHIOCGRP_PREEMPTANDABORT:
22686                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCGRP_PREEMPTANDABORT\n");
22687                 if ((err = drv_priv(cred_p)) != EPERM) {
22688                         if (un->un_reservation_type == SD_SCSI2_RESERVATION) {
22689                                 err = ENOTSUP;
22690                         } else if (arg != NULL) {
22691                                 mhioc_preemptandabort_t preempt_abort;
22692                                 if (ddi_copyin((void *)arg, &preempt_abort,
22693                                     sizeof (mhioc_preemptandabort_t),
22694                                     flag) != 0) {
22695                                         err = EFAULT;
22696                                 } else {
22697                                         err =
22698                                             sd_send_scsi_PERSISTENT_RESERVE_OUT(
22699                                             ssc, SD_SCSI3_PREEMPTANDABORT,
22700                                             (uchar_t *)&preempt_abort);
22701                                         if (err != 0)
22702                                                 goto done_with_assess;
22703                                 }
22704                         }
22705                 }
22706                 break;
22707 
22708         case MHIOCGRP_REGISTERANDIGNOREKEY:
22709                 SD_TRACE(SD_LOG_IOCTL, un, "MHIOCGRP_REGISTERANDIGNOREKEY\n");
22710                 if ((err = drv_priv(cred_p)) != EPERM) {
22711                         if (un->un_reservation_type == SD_SCSI2_RESERVATION) {
22712                                 err = ENOTSUP;
22713                         } else if (arg != NULL) {
22714                                 mhioc_registerandignorekey_t r_and_i;
22715                                 if (ddi_copyin((void *)arg, (void *)&r_and_i,
22716                                     sizeof (mhioc_registerandignorekey_t),
22717                                     flag) != 0) {
22718                                         err = EFAULT;
22719                                 } else {
22720                                         err =
22721                                             sd_send_scsi_PERSISTENT_RESERVE_OUT(
22722                                             ssc, SD_SCSI3_REGISTERANDIGNOREKEY,
22723                                             (uchar_t *)&r_and_i);
22724                                         if (err != 0)
22725                                                 goto done_with_assess;
22726                                 }
22727                         }
22728                 }
22729                 break;
22730 
22731         case USCSICMD:
22732                 SD_TRACE(SD_LOG_IOCTL, un, "USCSICMD\n");
22733                 cr = ddi_get_cred();
22734                 if ((drv_priv(cred_p) != 0) && (drv_priv(cr) != 0)) {
22735                         err = EPERM;
22736                 } else {
22737                         enum uio_seg    uioseg;
22738 
22739                         uioseg = (flag & FKIOCTL) ? UIO_SYSSPACE :
22740                             UIO_USERSPACE;
22741                         if (un->un_f_format_in_progress == TRUE) {
22742                                 err = EAGAIN;
22743                                 break;
22744                         }
22745 
22746                         err = sd_ssc_send(ssc,
22747                             (struct uscsi_cmd *)arg,
22748                             flag, uioseg, SD_PATH_STANDARD);
22749                         if (err != 0)
22750                                 goto done_with_assess;
22751                         else
22752                                 sd_ssc_assessment(ssc, SD_FMT_STANDARD);
22753                 }
22754                 break;
22755 
22756         case CDROMPAUSE:
22757         case CDROMRESUME:
22758                 SD_TRACE(SD_LOG_IOCTL, un, "PAUSE-RESUME\n");
22759                 if (!ISCD(un)) {
22760                         err = ENOTTY;
22761                 } else {
22762                         err = sr_pause_resume(dev, cmd);
22763                 }
22764                 break;
22765 
22766         case CDROMPLAYMSF:
22767                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMPLAYMSF\n");
22768                 if (!ISCD(un)) {
22769                         err = ENOTTY;
22770                 } else {
22771                         err = sr_play_msf(dev, (caddr_t)arg, flag);
22772                 }
22773                 break;
22774 
22775         case CDROMPLAYTRKIND:
22776                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMPLAYTRKIND\n");
22777 #if defined(__i386) || defined(__amd64)
22778                 /*
22779                  * not supported on ATAPI CD drives, use CDROMPLAYMSF instead
22780                  */
22781                 if (!ISCD(un) || (un->un_f_cfg_is_atapi == TRUE)) {
22782 #else
22783                 if (!ISCD(un)) {
22784 #endif
22785                         err = ENOTTY;
22786                 } else {
22787                         err = sr_play_trkind(dev, (caddr_t)arg, flag);
22788                 }
22789                 break;
22790 
22791         case CDROMREADTOCHDR:
22792                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMREADTOCHDR\n");
22793                 if (!ISCD(un)) {
22794                         err = ENOTTY;
22795                 } else {
22796                         err = sr_read_tochdr(dev, (caddr_t)arg, flag);
22797                 }
22798                 break;
22799 
22800         case CDROMREADTOCENTRY:
22801                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMREADTOCENTRY\n");
22802                 if (!ISCD(un)) {
22803                         err = ENOTTY;
22804                 } else {
22805                         err = sr_read_tocentry(dev, (caddr_t)arg, flag);
22806                 }
22807                 break;
22808 
22809         case CDROMSTOP:
22810                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMSTOP\n");
22811                 if (!ISCD(un)) {
22812                         err = ENOTTY;
22813                 } else {
22814                         err = sd_send_scsi_START_STOP_UNIT(ssc, SD_START_STOP,
22815                             SD_TARGET_STOP, SD_PATH_STANDARD);
22816                         goto done_with_assess;
22817                 }
22818                 break;
22819 
22820         case CDROMSTART:
22821                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMSTART\n");
22822                 if (!ISCD(un)) {
22823                         err = ENOTTY;
22824                 } else {
22825                         err = sd_send_scsi_START_STOP_UNIT(ssc, SD_START_STOP,
22826                             SD_TARGET_START, SD_PATH_STANDARD);
22827                         goto done_with_assess;
22828                 }
22829                 break;
22830 
22831         case CDROMCLOSETRAY:
22832                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMCLOSETRAY\n");
22833                 if (!ISCD(un)) {
22834                         err = ENOTTY;
22835                 } else {
22836                         err = sd_send_scsi_START_STOP_UNIT(ssc, SD_START_STOP,
22837                             SD_TARGET_CLOSE, SD_PATH_STANDARD);
22838                         goto done_with_assess;
22839                 }
22840                 break;
22841 
22842         case FDEJECT:   /* for eject command */
22843         case DKIOCEJECT:
22844         case CDROMEJECT:
22845                 SD_TRACE(SD_LOG_IOCTL, un, "EJECT\n");
22846                 if (!un->un_f_eject_media_supported) {
22847                         err = ENOTTY;
22848                 } else {
22849                         err = sr_eject(dev);
22850                 }
22851                 break;
22852 
22853         case CDROMVOLCTRL:
22854                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMVOLCTRL\n");
22855                 if (!ISCD(un)) {
22856                         err = ENOTTY;
22857                 } else {
22858                         err = sr_volume_ctrl(dev, (caddr_t)arg, flag);
22859                 }
22860                 break;
22861 
22862         case CDROMSUBCHNL:
22863                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMSUBCHNL\n");
22864                 if (!ISCD(un)) {
22865                         err = ENOTTY;
22866                 } else {
22867                         err = sr_read_subchannel(dev, (caddr_t)arg, flag);
22868                 }
22869                 break;
22870 
22871         case CDROMREADMODE2:
22872                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMREADMODE2\n");
22873                 if (!ISCD(un)) {
22874                         err = ENOTTY;
22875                 } else if (un->un_f_cfg_is_atapi == TRUE) {
22876                         /*
22877                          * If the drive supports READ CD, use that instead of
22878                          * switching the LBA size via a MODE SELECT
22879                          * Block Descriptor
22880                          */
22881                         err = sr_read_cd_mode2(dev, (caddr_t)arg, flag);
22882                 } else {
22883                         err = sr_read_mode2(dev, (caddr_t)arg, flag);
22884                 }
22885                 break;
22886 
22887         case CDROMREADMODE1:
22888                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMREADMODE1\n");
22889                 if (!ISCD(un)) {
22890                         err = ENOTTY;
22891                 } else {
22892                         err = sr_read_mode1(dev, (caddr_t)arg, flag);
22893                 }
22894                 break;
22895 
22896         case CDROMREADOFFSET:
22897                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMREADOFFSET\n");
22898                 if (!ISCD(un)) {
22899                         err = ENOTTY;
22900                 } else {
22901                         err = sr_read_sony_session_offset(dev, (caddr_t)arg,
22902                             flag);
22903                 }
22904                 break;
22905 
22906         case CDROMSBLKMODE:
22907                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMSBLKMODE\n");
22908                 /*
22909                  * There is no means of changing block size in case of atapi
22910                  * drives, thus return ENOTTY if drive type is atapi
22911                  */
22912                 if (!ISCD(un) || (un->un_f_cfg_is_atapi == TRUE)) {
22913                         err = ENOTTY;
22914                 } else if (un->un_f_mmc_cap == TRUE) {
22915 
22916                         /*
22917                          * MMC Devices do not support changing the
22918                          * logical block size
22919                          *
22920                          * Note: EINVAL is being returned instead of ENOTTY to
22921                          * maintain consistancy with the original mmc
22922                          * driver update.
22923                          */
22924                         err = EINVAL;
22925                 } else {
22926                         mutex_enter(SD_MUTEX(un));
22927                         if ((!(un->un_exclopen & (1<<SDPART(dev)))) ||
22928                             (un->un_ncmds_in_transport > 0)) {
22929                                 mutex_exit(SD_MUTEX(un));
22930                                 err = EINVAL;
22931                         } else {
22932                                 mutex_exit(SD_MUTEX(un));
22933                                 err = sr_change_blkmode(dev, cmd, arg, flag);
22934                         }
22935                 }
22936                 break;
22937 
22938         case CDROMGBLKMODE:
22939                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMGBLKMODE\n");
22940                 if (!ISCD(un)) {
22941                         err = ENOTTY;
22942                 } else if ((un->un_f_cfg_is_atapi != FALSE) &&
22943                     (un->un_f_blockcount_is_valid != FALSE)) {
22944                         /*
22945                          * Drive is an ATAPI drive so return target block
22946                          * size for ATAPI drives since we cannot change the
22947                          * blocksize on ATAPI drives. Used primarily to detect
22948                          * if an ATAPI cdrom is present.
22949                          */
22950                         if (ddi_copyout(&un->un_tgt_blocksize, (void *)arg,
22951                             sizeof (int), flag) != 0) {
22952                                 err = EFAULT;
22953                         } else {
22954                                 err = 0;
22955                         }
22956 
22957                 } else {
22958                         /*
22959                          * Drive supports changing block sizes via a Mode
22960                          * Select.
22961                          */
22962                         err = sr_change_blkmode(dev, cmd, arg, flag);
22963                 }
22964                 break;
22965 
22966         case CDROMGDRVSPEED:
22967         case CDROMSDRVSPEED:
22968                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMXDRVSPEED\n");
22969                 if (!ISCD(un)) {
22970                         err = ENOTTY;
22971                 } else if (un->un_f_mmc_cap == TRUE) {
22972                         /*
22973                          * Note: In the future the driver implementation
22974                          * for getting and
22975                          * setting cd speed should entail:
22976                          * 1) If non-mmc try the Toshiba mode page
22977                          *    (sr_change_speed)
22978                          * 2) If mmc but no support for Real Time Streaming try
22979                          *    the SET CD SPEED (0xBB) command
22980                          *   (sr_atapi_change_speed)
22981                          * 3) If mmc and support for Real Time Streaming
22982                          *    try the GET PERFORMANCE and SET STREAMING
22983                          *    commands (not yet implemented, 4380808)
22984                          */
22985                         /*
22986                          * As per recent MMC spec, CD-ROM speed is variable
22987                          * and changes with LBA. Since there is no such
22988                          * things as drive speed now, fail this ioctl.
22989                          *
22990                          * Note: EINVAL is returned for consistancy of original
22991                          * implementation which included support for getting
22992                          * the drive speed of mmc devices but not setting
22993                          * the drive speed. Thus EINVAL would be returned
22994                          * if a set request was made for an mmc device.
22995                          * We no longer support get or set speed for
22996                          * mmc but need to remain consistent with regard
22997                          * to the error code returned.
22998                          */
22999                         err = EINVAL;
23000                 } else if (un->un_f_cfg_is_atapi == TRUE) {
23001                         err = sr_atapi_change_speed(dev, cmd, arg, flag);
23002                 } else {
23003                         err = sr_change_speed(dev, cmd, arg, flag);
23004                 }
23005                 break;
23006 
23007         case CDROMCDDA:
23008                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMCDDA\n");
23009                 if (!ISCD(un)) {
23010                         err = ENOTTY;
23011                 } else {
23012                         err = sr_read_cdda(dev, (void *)arg, flag);
23013                 }
23014                 break;
23015 
23016         case CDROMCDXA:
23017                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMCDXA\n");
23018                 if (!ISCD(un)) {
23019                         err = ENOTTY;
23020                 } else {
23021                         err = sr_read_cdxa(dev, (caddr_t)arg, flag);
23022                 }
23023                 break;
23024 
23025         case CDROMSUBCODE:
23026                 SD_TRACE(SD_LOG_IOCTL, un, "CDROMSUBCODE\n");
23027                 if (!ISCD(un)) {
23028                         err = ENOTTY;
23029                 } else {
23030                         err = sr_read_all_subcodes(dev, (caddr_t)arg, flag);
23031                 }
23032                 break;
23033 
23034 
23035 #ifdef SDDEBUG
23036 /* RESET/ABORTS testing ioctls */
23037         case DKIOCRESET: {
23038                 int     reset_level;
23039 
23040                 if (ddi_copyin((void *)arg, &reset_level, sizeof (int), flag)) {
23041                         err = EFAULT;
23042                 } else {
23043                         SD_INFO(SD_LOG_IOCTL, un, "sdioctl: DKIOCRESET: "
23044                             "reset_level = 0x%lx\n", reset_level);
23045                         if (scsi_reset(SD_ADDRESS(un), reset_level)) {
23046                                 err = 0;
23047                         } else {
23048                                 err = EIO;
23049                         }
23050                 }
23051                 break;
23052         }
23053 
23054         case DKIOCABORT:
23055                 SD_INFO(SD_LOG_IOCTL, un, "sdioctl: DKIOCABORT:\n");
23056                 if (scsi_abort(SD_ADDRESS(un), NULL)) {
23057                         err = 0;
23058                 } else {
23059                         err = EIO;
23060                 }
23061                 break;
23062 #endif
23063 
23064 #ifdef SD_FAULT_INJECTION
23065 /* SDIOC FaultInjection testing ioctls */
23066         case SDIOCSTART:
23067         case SDIOCSTOP:
23068         case SDIOCINSERTPKT:
23069         case SDIOCINSERTXB:
23070         case SDIOCINSERTUN:
23071         case SDIOCINSERTARQ:
23072         case SDIOCPUSH:
23073         case SDIOCRETRIEVE:
23074         case SDIOCRUN:
23075                 SD_INFO(SD_LOG_SDTEST, un, "sdioctl:"
23076                     "SDIOC detected cmd:0x%X:\n", cmd);
23077                 /* call error generator */
23078                 sd_faultinjection_ioctl(cmd, arg, un);
23079                 err = 0;
23080                 break;
23081 
23082 #endif /* SD_FAULT_INJECTION */
23083 
23084         case DKIOCFLUSHWRITECACHE:
23085                 {
23086                         struct dk_callback *dkc = (struct dk_callback *)arg;
23087 
23088                         mutex_enter(SD_MUTEX(un));
23089                         if (!un->un_f_sync_cache_supported ||
23090                             !un->un_f_write_cache_enabled) {
23091                                 err = un->un_f_sync_cache_supported ?
23092                                     0 : ENOTSUP;
23093                                 mutex_exit(SD_MUTEX(un));
23094                                 if ((flag & FKIOCTL) && dkc != NULL &&
23095                                     dkc->dkc_callback != NULL) {
23096                                         (*dkc->dkc_callback)(dkc->dkc_cookie,
23097                                             err);
23098                                         /*
23099                                          * Did callback and reported error.
23100                                          * Since we did a callback, ioctl
23101                                          * should return 0.
23102                                          */
23103                                         err = 0;
23104                                 }
23105                                 break;
23106                         }
23107                         mutex_exit(SD_MUTEX(un));
23108 
23109                         if ((flag & FKIOCTL) && dkc != NULL &&
23110                             dkc->dkc_callback != NULL) {
23111                                 /* async SYNC CACHE request */
23112                                 err = sd_send_scsi_SYNCHRONIZE_CACHE(un, dkc);
23113                         } else {
23114                                 /* synchronous SYNC CACHE request */
23115                                 err = sd_send_scsi_SYNCHRONIZE_CACHE(un, NULL);
23116                         }
23117                 }
23118                 break;
23119 
23120         case DKIOCGETWCE: {
23121 
23122                 int wce;
23123 
23124                 if ((err = sd_get_write_cache_enabled(ssc, &wce)) != 0) {
23125                         break;
23126                 }
23127 
23128                 if (ddi_copyout(&wce, (void *)arg, sizeof (wce), flag)) {
23129                         err = EFAULT;
23130                 }
23131                 break;
23132         }
23133 
23134         case DKIOCSETWCE: {
23135 
23136                 int wce, sync_supported;
23137                 int cur_wce = 0;
23138 
23139                 if (ddi_copyin((void *)arg, &wce, sizeof (wce), flag)) {
23140                         err = EFAULT;
23141                         break;
23142                 }
23143 
23144                 /*
23145                  * Synchronize multiple threads trying to enable
23146                  * or disable the cache via the un_f_wcc_cv
23147                  * condition variable.
23148                  */
23149                 mutex_enter(SD_MUTEX(un));
23150 
23151                 /*
23152                  * Don't allow the cache to be enabled if the
23153                  * config file has it disabled.
23154                  */
23155                 if (un->un_f_opt_disable_cache && wce) {
23156                         mutex_exit(SD_MUTEX(un));
23157                         err = EINVAL;
23158                         break;
23159                 }
23160 
23161                 /*
23162                  * Wait for write cache change in progress
23163                  * bit to be clear before proceeding.
23164                  */
23165                 while (un->un_f_wcc_inprog)
23166                         cv_wait(&un->un_wcc_cv, SD_MUTEX(un));
23167 
23168                 un->un_f_wcc_inprog = 1;
23169 
23170                 mutex_exit(SD_MUTEX(un));
23171 
23172                 /*
23173                  * Get the current write cache state
23174                  */
23175                 if ((err = sd_get_write_cache_enabled(ssc, &cur_wce)) != 0) {
23176                         mutex_enter(SD_MUTEX(un));
23177                         un->un_f_wcc_inprog = 0;
23178                         cv_broadcast(&un->un_wcc_cv);
23179                         mutex_exit(SD_MUTEX(un));
23180                         break;
23181                 }
23182 
23183                 mutex_enter(SD_MUTEX(un));
23184                 un->un_f_write_cache_enabled = (cur_wce != 0);
23185 
23186                 if (un->un_f_write_cache_enabled && wce == 0) {
23187                         /*
23188                          * Disable the write cache.  Don't clear
23189                          * un_f_write_cache_enabled until after
23190                          * the mode select and flush are complete.
23191                          */
23192                         sync_supported = un->un_f_sync_cache_supported;
23193 
23194                         /*
23195                          * If cache flush is suppressed, we assume that the
23196                          * controller firmware will take care of managing the
23197                          * write cache for us: no need to explicitly
23198                          * disable it.
23199                          */
23200                         if (!un->un_f_suppress_cache_flush) {
23201                                 mutex_exit(SD_MUTEX(un));
23202                                 if ((err = sd_cache_control(ssc,
23203                                     SD_CACHE_NOCHANGE,
23204                                     SD_CACHE_DISABLE)) == 0 &&
23205                                     sync_supported) {
23206                                         err = sd_send_scsi_SYNCHRONIZE_CACHE(un,
23207                                             NULL);
23208                                 }
23209                         } else {
23210                                 mutex_exit(SD_MUTEX(un));
23211                         }
23212 
23213                         mutex_enter(SD_MUTEX(un));
23214                         if (err == 0) {
23215                                 un->un_f_write_cache_enabled = 0;
23216                         }
23217 
23218                 } else if (!un->un_f_write_cache_enabled && wce != 0) {
23219                         /*
23220                          * Set un_f_write_cache_enabled first, so there is
23221                          * no window where the cache is enabled, but the
23222                          * bit says it isn't.
23223                          */
23224                         un->un_f_write_cache_enabled = 1;
23225 
23226                         /*
23227                          * If cache flush is suppressed, we assume that the
23228                          * controller firmware will take care of managing the
23229                          * write cache for us: no need to explicitly
23230                          * enable it.
23231                          */
23232                         if (!un->un_f_suppress_cache_flush) {
23233                                 mutex_exit(SD_MUTEX(un));
23234                                 err = sd_cache_control(ssc, SD_CACHE_NOCHANGE,
23235                                     SD_CACHE_ENABLE);
23236                         } else {
23237                                 mutex_exit(SD_MUTEX(un));
23238                         }
23239 
23240                         mutex_enter(SD_MUTEX(un));
23241 
23242                         if (err) {
23243                                 un->un_f_write_cache_enabled = 0;
23244                         }
23245                 }
23246 
23247                 un->un_f_wcc_inprog = 0;
23248                 cv_broadcast(&un->un_wcc_cv);
23249                 mutex_exit(SD_MUTEX(un));
23250                 break;
23251         }
23252 
23253         default:
23254                 err = ENOTTY;
23255                 break;
23256         }
23257         mutex_enter(SD_MUTEX(un));
23258         un->un_ncmds_in_driver--;
23259         ASSERT(un->un_ncmds_in_driver >= 0);
23260         mutex_exit(SD_MUTEX(un));
23261 
23262 
23263 done_without_assess:
23264         sd_ssc_fini(ssc);
23265 
23266         SD_TRACE(SD_LOG_IOCTL, un, "sdioctl: exit: %d\n", err);
23267         return (err);
23268 
23269 done_with_assess:
23270         mutex_enter(SD_MUTEX(un));
23271         un->un_ncmds_in_driver--;
23272         ASSERT(un->un_ncmds_in_driver >= 0);
23273         mutex_exit(SD_MUTEX(un));
23274 
23275 done_quick_assess:
23276         if (err != 0)
23277                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
23278         /* Uninitialize sd_ssc_t pointer */
23279         sd_ssc_fini(ssc);
23280 
23281         SD_TRACE(SD_LOG_IOCTL, un, "sdioctl: exit: %d\n", err);
23282         return (err);
23283 }
23284 
23285 
23286 /*
23287  *    Function: sd_dkio_ctrl_info
23288  *
23289  * Description: This routine is the driver entry point for handling controller
23290  *              information ioctl requests (DKIOCINFO).
23291  *
23292  *   Arguments: dev  - the device number
23293  *              arg  - pointer to user provided dk_cinfo structure
23294  *                     specifying the controller type and attributes.
23295  *              flag - this argument is a pass through to ddi_copyxxx()
23296  *                     directly from the mode argument of ioctl().
23297  *
23298  * Return Code: 0
23299  *              EFAULT
23300  *              ENXIO
23301  */
23302 
23303 static int
23304 sd_dkio_ctrl_info(dev_t dev, caddr_t arg, int flag)
23305 {
23306         struct sd_lun   *un = NULL;
23307         struct dk_cinfo *info;
23308         dev_info_t      *pdip;
23309         int             lun, tgt;
23310 
23311         if ((un = ddi_get_soft_state(sd_state, SDUNIT(dev))) == NULL) {
23312                 return (ENXIO);
23313         }
23314 
23315         info = (struct dk_cinfo *)
23316             kmem_zalloc(sizeof (struct dk_cinfo), KM_SLEEP);
23317 
23318         switch (un->un_ctype) {
23319         case CTYPE_CDROM:
23320                 info->dki_ctype = DKC_CDROM;
23321                 break;
23322         default:
23323                 info->dki_ctype = DKC_SCSI_CCS;
23324                 break;
23325         }
23326         pdip = ddi_get_parent(SD_DEVINFO(un));
23327         info->dki_cnum = ddi_get_instance(pdip);
23328         if (strlen(ddi_get_name(pdip)) < DK_DEVLEN) {
23329                 (void) strcpy(info->dki_cname, ddi_get_name(pdip));
23330         } else {
23331                 (void) strncpy(info->dki_cname, ddi_node_name(pdip),
23332                     DK_DEVLEN - 1);
23333         }
23334 
23335         lun = ddi_prop_get_int(DDI_DEV_T_ANY, SD_DEVINFO(un),
23336             DDI_PROP_DONTPASS, SCSI_ADDR_PROP_LUN, 0);
23337         tgt = ddi_prop_get_int(DDI_DEV_T_ANY, SD_DEVINFO(un),
23338             DDI_PROP_DONTPASS, SCSI_ADDR_PROP_TARGET, 0);
23339 
23340         /* Unit Information */
23341         info->dki_unit = ddi_get_instance(SD_DEVINFO(un));
23342         info->dki_slave = ((tgt << 3) | lun);
23343         (void) strncpy(info->dki_dname, ddi_driver_name(SD_DEVINFO(un)),
23344             DK_DEVLEN - 1);
23345         info->dki_flags = DKI_FMTVOL;
23346         info->dki_partition = SDPART(dev);
23347 
23348         /* Max Transfer size of this device in blocks */
23349         info->dki_maxtransfer = un->un_max_xfer_size / un->un_sys_blocksize;
23350         info->dki_addr = 0;
23351         info->dki_space = 0;
23352         info->dki_prio = 0;
23353         info->dki_vec = 0;
23354 
23355         if (ddi_copyout(info, arg, sizeof (struct dk_cinfo), flag) != 0) {
23356                 kmem_free(info, sizeof (struct dk_cinfo));
23357                 return (EFAULT);
23358         } else {
23359                 kmem_free(info, sizeof (struct dk_cinfo));
23360                 return (0);
23361         }
23362 }
23363 
23364 /*
23365  *    Function: sd_get_media_info_com
23366  *
23367  * Description: This routine returns the information required to populate
23368  *              the fields for the dk_minfo/dk_minfo_ext structures.
23369  *
23370  *   Arguments: dev             - the device number
23371  *              dki_media_type  - media_type
23372  *              dki_lbsize      - logical block size
23373  *              dki_capacity    - capacity in blocks
23374  *              dki_pbsize      - physical block size (if requested)
23375  *
23376  * Return Code: 0
23377  *              EACCESS
23378  *              EFAULT
23379  *              ENXIO
23380  *              EIO
23381  */
23382 static int
23383 sd_get_media_info_com(dev_t dev, uint_t *dki_media_type, uint_t *dki_lbsize,
23384         diskaddr_t *dki_capacity, uint_t *dki_pbsize)
23385 {
23386         struct sd_lun           *un = NULL;
23387         struct uscsi_cmd        com;
23388         struct scsi_inquiry     *sinq;
23389         u_longlong_t            media_capacity;
23390         uint64_t                capacity;
23391         uint_t                  lbasize;
23392         uint_t                  pbsize;
23393         uchar_t                 *out_data;
23394         uchar_t                 *rqbuf;
23395         int                     rval = 0;
23396         int                     rtn;
23397         sd_ssc_t                *ssc;
23398 
23399         if ((un = ddi_get_soft_state(sd_state, SDUNIT(dev))) == NULL ||
23400             (un->un_state == SD_STATE_OFFLINE)) {
23401                 return (ENXIO);
23402         }
23403 
23404         SD_TRACE(SD_LOG_IOCTL_DKIO, un, "sd_get_media_info_com: entry\n");
23405 
23406         out_data = kmem_zalloc(SD_PROFILE_HEADER_LEN, KM_SLEEP);
23407         rqbuf = kmem_zalloc(SENSE_LENGTH, KM_SLEEP);
23408         ssc = sd_ssc_init(un);
23409 
23410         /* Issue a TUR to determine if the drive is ready with media present */
23411         rval = sd_send_scsi_TEST_UNIT_READY(ssc, SD_CHECK_FOR_MEDIA);
23412         if (rval == ENXIO) {
23413                 goto done;
23414         } else if (rval != 0) {
23415                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
23416         }
23417 
23418         /* Now get configuration data */
23419         if (ISCD(un)) {
23420                 *dki_media_type = DK_CDROM;
23421 
23422                 /* Allow SCMD_GET_CONFIGURATION to MMC devices only */
23423                 if (un->un_f_mmc_cap == TRUE) {
23424                         rtn = sd_send_scsi_GET_CONFIGURATION(ssc, &com, rqbuf,
23425                             SENSE_LENGTH, out_data, SD_PROFILE_HEADER_LEN,
23426                             SD_PATH_STANDARD);
23427 
23428                         if (rtn) {
23429                                 /*
23430                                  * We ignore all failures for CD and need to
23431                                  * put the assessment before processing code
23432                                  * to avoid missing assessment for FMA.
23433                                  */
23434                                 sd_ssc_assessment(ssc, SD_FMT_IGNORE);
23435                                 /*
23436                                  * Failed for other than an illegal request
23437                                  * or command not supported
23438                                  */
23439                                 if ((com.uscsi_status == STATUS_CHECK) &&
23440                                     (com.uscsi_rqstatus == STATUS_GOOD)) {
23441                                         if ((rqbuf[2] != KEY_ILLEGAL_REQUEST) ||
23442                                             (rqbuf[12] != 0x20)) {
23443                                                 rval = EIO;
23444                                                 goto no_assessment;
23445                                         }
23446                                 }
23447                         } else {
23448                                 /*
23449                                  * The GET CONFIGURATION command succeeded
23450                                  * so set the media type according to the
23451                                  * returned data
23452                                  */
23453                                 *dki_media_type = out_data[6];
23454                                 *dki_media_type <<= 8;
23455                                 *dki_media_type |= out_data[7];
23456                         }
23457                 }
23458         } else {
23459                 /*
23460                  * The profile list is not available, so we attempt to identify
23461                  * the media type based on the inquiry data
23462                  */
23463                 sinq = un->un_sd->sd_inq;
23464                 if ((sinq->inq_dtype == DTYPE_DIRECT) ||
23465                     (sinq->inq_dtype == DTYPE_OPTICAL)) {
23466                         /* This is a direct access device  or optical disk */
23467                         *dki_media_type = DK_FIXED_DISK;
23468 
23469                         if ((bcmp(sinq->inq_vid, "IOMEGA", 6) == 0) ||
23470                             (bcmp(sinq->inq_vid, "iomega", 6) == 0)) {
23471                                 if ((bcmp(sinq->inq_pid, "ZIP", 3) == 0)) {
23472                                         *dki_media_type = DK_ZIP;
23473                                 } else if (
23474                                     (bcmp(sinq->inq_pid, "jaz", 3) == 0)) {
23475                                         *dki_media_type = DK_JAZ;
23476                                 }
23477                         }
23478                 } else {
23479                         /*
23480                          * Not a CD, direct access or optical disk so return
23481                          * unknown media
23482                          */
23483                         *dki_media_type = DK_UNKNOWN;
23484                 }
23485         }
23486 
23487         /*
23488          * Now read the capacity so we can provide the lbasize,
23489          * pbsize and capacity.
23490          */
23491         if (dki_pbsize && un->un_f_descr_format_supported) {
23492                 rval = sd_send_scsi_READ_CAPACITY_16(ssc, &capacity, &lbasize,
23493                     &pbsize, SD_PATH_DIRECT);
23494 
23495                 /*
23496                  * Override the physical blocksize if the instance already
23497                  * has a larger value.
23498                  */
23499                 pbsize = MAX(pbsize, un->un_phy_blocksize);
23500         }
23501 
23502         if (dki_pbsize == NULL || rval != 0 ||
23503             !un->un_f_descr_format_supported) {
23504                 rval = sd_send_scsi_READ_CAPACITY(ssc, &capacity, &lbasize,
23505                     SD_PATH_DIRECT);
23506 
23507                 switch (rval) {
23508                 case 0:
23509                         if (un->un_f_enable_rmw &&
23510                             un->un_phy_blocksize != 0) {
23511                                 pbsize = un->un_phy_blocksize;
23512                         } else {
23513                                 pbsize = lbasize;
23514                         }
23515                         media_capacity = capacity;
23516 
23517                         /*
23518                          * sd_send_scsi_READ_CAPACITY() reports capacity in
23519                          * un->un_sys_blocksize chunks. So we need to convert
23520                          * it into cap.lbsize chunks.
23521                          */
23522                         if (un->un_f_has_removable_media) {
23523                                 media_capacity *= un->un_sys_blocksize;
23524                                 media_capacity /= lbasize;
23525                         }
23526                         break;
23527                 case EACCES:
23528                         rval = EACCES;
23529                         goto done;
23530                 default:
23531                         rval = EIO;
23532                         goto done;
23533                 }
23534         } else {
23535                 if (un->un_f_enable_rmw &&
23536                     !ISP2(pbsize % DEV_BSIZE)) {
23537                         pbsize = SSD_SECSIZE;
23538                 } else if (!ISP2(lbasize % DEV_BSIZE) ||
23539                     !ISP2(pbsize % DEV_BSIZE)) {
23540                         pbsize = lbasize = DEV_BSIZE;
23541                 }
23542                 media_capacity = capacity;
23543         }
23544 
23545         /*
23546          * If lun is expanded dynamically, update the un structure.
23547          */
23548         mutex_enter(SD_MUTEX(un));
23549         if ((un->un_f_blockcount_is_valid == TRUE) &&
23550             (un->un_f_tgt_blocksize_is_valid == TRUE) &&
23551             (capacity > un->un_blockcount)) {
23552                 un->un_f_expnevent = B_FALSE;
23553                 sd_update_block_info(un, lbasize, capacity);
23554         }
23555         mutex_exit(SD_MUTEX(un));
23556 
23557         *dki_lbsize = lbasize;
23558         *dki_capacity = media_capacity;
23559         if (dki_pbsize)
23560                 *dki_pbsize = pbsize;
23561 
23562 done:
23563         if (rval != 0) {
23564                 if (rval == EIO)
23565                         sd_ssc_assessment(ssc, SD_FMT_STATUS_CHECK);
23566                 else
23567                         sd_ssc_assessment(ssc, SD_FMT_IGNORE);
23568         }
23569 no_assessment:
23570         sd_ssc_fini(ssc);
23571         kmem_free(out_data, SD_PROFILE_HEADER_LEN);
23572         kmem_free(rqbuf, SENSE_LENGTH);
23573         return (rval);
23574 }
23575 
23576 /*
23577  *    Function: sd_get_media_info
23578  *
23579  * Description: This routine is the driver entry point for handling ioctl
23580  *              requests for the media type or command set profile used by the
23581  *              drive to operate on the media (DKIOCGMEDIAINFO).
23582  *
23583  *   Arguments: dev     - the device number
23584  *              arg     - pointer to user provided dk_minfo structure
23585  *                        specifying the media type, logical block size and
23586  *                        drive capacity.
23587  *              flag    - this argument is a pass through to ddi_copyxxx()
23588  *                        directly from the mode argument of ioctl().
23589  *
23590  * Return Code: returns the value from sd_get_media_info_com
23591  */
23592 static int
23593 sd_get_media_info(dev_t dev, caddr_t arg, int flag)
23594 {
23595         struct dk_minfo         mi;
23596         int                     rval;
23597 
23598         rval = sd_get_media_info_com(dev, &mi.dki_media_type,
23599             &mi.dki_lbsize, &mi.dki_capacity, NULL);
23600 
23601         if (rval)
23602                 return (rval);
23603         if (ddi_copyout(&mi, arg, sizeof (struct dk_minfo), flag))
23604                 rval = EFAULT;
23605         return (rval);
23606 }
23607 
23608 /*
23609  *    Function: sd_get_media_info_ext
23610  *
23611  * Description: This routine is the driver entry point for handling ioctl
23612  *              requests for the media type or command set profile used by the
23613  *              drive to operate on the media (DKIOCGMEDIAINFOEXT). The
23614  *              difference this ioctl and DKIOCGMEDIAINFO is the return value
23615  *              of this ioctl contains both logical block size and physical
23616  *              block size.
23617  *
23618  *
23619  *   Arguments: dev     - the device number
23620  *              arg     - pointer to user provided dk_minfo_ext structure
23621  *                        specifying the media type, logical block size,
23622  *                        physical block