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  * Copyright (c) 2000, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright 2012 Nexenta Systems, Inc. All rights reserved.
  24  * Copyright 2012 Garrett D'Amore <garrett@damore.org>.  All rights reserved.
  25  */
  26 
  27 #include <sys/note.h>
  28 #include <sys/t_lock.h>
  29 #include <sys/cmn_err.h>
  30 #include <sys/instance.h>
  31 #include <sys/conf.h>
  32 #include <sys/stat.h>
  33 #include <sys/ddi.h>
  34 #include <sys/hwconf.h>
  35 #include <sys/sunddi.h>
  36 #include <sys/sunndi.h>
  37 #include <sys/ddi_impldefs.h>
  38 #include <sys/ndi_impldefs.h>
  39 #include <sys/modctl.h>
  40 #include <sys/contract/device_impl.h>
  41 #include <sys/dacf.h>
  42 #include <sys/promif.h>
  43 #include <sys/pci.h>
  44 #include <sys/cpuvar.h>
  45 #include <sys/pathname.h>
  46 #include <sys/taskq.h>
  47 #include <sys/sysevent.h>
  48 #include <sys/sunmdi.h>
  49 #include <sys/stream.h>
  50 #include <sys/strsubr.h>
  51 #include <sys/fs/snode.h>
  52 #include <sys/fs/dv_node.h>
  53 #include <sys/reboot.h>
  54 #include <sys/sysmacros.h>
  55 #include <sys/systm.h>
  56 #include <sys/fs/sdev_impl.h>
  57 #include <sys/sunldi.h>
  58 #include <sys/sunldi_impl.h>
  59 #include <sys/bootprops.h>
  60 #include <sys/varargs.h>
  61 #include <sys/modhash.h>
  62 #include <sys/instance.h>
  63 
  64 #if defined(__amd64) && !defined(__xpv)
  65 #include <sys/iommulib.h>
  66 #endif
  67 
  68 #ifdef DEBUG
  69 int ddidebug = DDI_AUDIT;
  70 #else
  71 int ddidebug = 0;
  72 #endif
  73 
  74 #define MT_CONFIG_OP    0
  75 #define MT_UNCONFIG_OP  1
  76 
  77 /* Multi-threaded configuration */
  78 struct mt_config_handle {
  79         kmutex_t mtc_lock;
  80         kcondvar_t mtc_cv;
  81         int mtc_thr_count;
  82         dev_info_t *mtc_pdip;   /* parent dip for mt_config_children */
  83         dev_info_t **mtc_fdip;  /* "a" dip where unconfigure failed */
  84         major_t mtc_parmajor;   /* parent major for mt_config_driver */
  85         major_t mtc_major;
  86         int mtc_flags;
  87         int mtc_op;             /* config or unconfig */
  88         int mtc_error;          /* operation error */
  89         struct brevq_node **mtc_brevqp; /* outstanding branch events queue */
  90 #ifdef DEBUG
  91         int total_time;
  92         timestruc_t start_time;
  93 #endif /* DEBUG */
  94 };
  95 
  96 struct devi_nodeid {
  97         pnode_t nodeid;
  98         dev_info_t *dip;
  99         struct devi_nodeid *next;
 100 };
 101 
 102 struct devi_nodeid_list {
 103         kmutex_t dno_lock;              /* Protects other fields */
 104         struct devi_nodeid *dno_head;   /* list of devi nodeid elements */
 105         struct devi_nodeid *dno_free;   /* Free list */
 106         uint_t dno_list_length;         /* number of dips in list */
 107 };
 108 
 109 /* used to keep track of branch remove events to be generated */
 110 struct brevq_node {
 111         char *brn_deviname;
 112         struct brevq_node *brn_sibling;
 113         struct brevq_node *brn_child;
 114 };
 115 
 116 static struct devi_nodeid_list devi_nodeid_list;
 117 static struct devi_nodeid_list *devimap = &devi_nodeid_list;
 118 
 119 /*
 120  * Well known nodes which are attached first at boot time.
 121  */
 122 dev_info_t *top_devinfo;                /* root of device tree */
 123 dev_info_t *options_dip;
 124 dev_info_t *pseudo_dip;
 125 dev_info_t *clone_dip;
 126 dev_info_t *scsi_vhci_dip;              /* MPXIO dip */
 127 major_t clone_major;
 128 
 129 /*
 130  * A non-global zone's /dev is derived from the device tree.
 131  * This generation number serves to indicate when a zone's
 132  * /dev may need to be updated.
 133  */
 134 volatile ulong_t devtree_gen;           /* generation number */
 135 
 136 /* block all future dev_info state changes */
 137 hrtime_t volatile devinfo_freeze = 0;
 138 
 139 /* number of dev_info attaches/detaches currently in progress */
 140 static ulong_t devinfo_attach_detach = 0;
 141 
 142 extern int      sys_shutdown;
 143 extern kmutex_t global_vhci_lock;
 144 
 145 /* bitset of DS_SYSAVAIL & DS_RECONFIG - no races, no lock */
 146 static int devname_state = 0;
 147 
 148 /*
 149  * The devinfo snapshot cache and related variables.
 150  * The only field in the di_cache structure that needs initialization
 151  * is the mutex (cache_lock). However, since this is an adaptive mutex
 152  * (MUTEX_DEFAULT) - it is automatically initialized by being allocated
 153  * in zeroed memory (static storage class). Therefore no explicit
 154  * initialization of the di_cache structure is needed.
 155  */
 156 struct di_cache di_cache = {1};
 157 int             di_cache_debug = 0;
 158 
 159 /* For ddvis, which needs pseudo children under PCI */
 160 int pci_allow_pseudo_children = 0;
 161 
 162 /* Allow path-oriented alias driver binding on driver.conf enumerated nodes */
 163 int driver_conf_allow_path_alias = 1;
 164 
 165 /*
 166  * The following switch is for service people, in case a
 167  * 3rd party driver depends on identify(9e) being called.
 168  */
 169 int identify_9e = 0;
 170 
 171 /*
 172  * Add flag so behaviour of preventing attach for retired persistant nodes
 173  * can be disabled.
 174  */
 175 int retire_prevents_attach = 1;
 176 
 177 int mtc_off;                                    /* turn off mt config */
 178 
 179 int quiesce_debug = 0;
 180 
 181 boolean_t ddi_aliases_present = B_FALSE;
 182 ddi_alias_t ddi_aliases;
 183 uint_t tsd_ddi_redirect;
 184 
 185 #define DDI_ALIAS_HASH_SIZE     (2700)
 186 
 187 static kmem_cache_t *ddi_node_cache;            /* devinfo node cache */
 188 static devinfo_log_header_t *devinfo_audit_log; /* devinfo log */
 189 static int devinfo_log_size;                    /* size in pages */
 190 
 191 boolean_t ddi_err_panic = B_FALSE;
 192 
 193 static int lookup_compatible(dev_info_t *, uint_t);
 194 static char *encode_composite_string(char **, uint_t, size_t *, uint_t);
 195 static void link_to_driver_list(dev_info_t *);
 196 static void unlink_from_driver_list(dev_info_t *);
 197 static void add_to_dn_list(struct devnames *, dev_info_t *);
 198 static void remove_from_dn_list(struct devnames *, dev_info_t *);
 199 static dev_info_t *find_duplicate_child();
 200 static void add_global_props(dev_info_t *);
 201 static void remove_global_props(dev_info_t *);
 202 static int uninit_node(dev_info_t *);
 203 static void da_log_init(void);
 204 static void da_log_enter(dev_info_t *);
 205 static int walk_devs(dev_info_t *, int (*f)(dev_info_t *, void *), void *, int);
 206 static int reset_nexus_flags(dev_info_t *, void *);
 207 static void ddi_optimize_dtree(dev_info_t *);
 208 static int is_leaf_node(dev_info_t *);
 209 static struct mt_config_handle *mt_config_init(dev_info_t *, dev_info_t **,
 210     int, major_t, int, struct brevq_node **);
 211 static void mt_config_children(struct mt_config_handle *);
 212 static void mt_config_driver(struct mt_config_handle *);
 213 static int mt_config_fini(struct mt_config_handle *);
 214 static int devi_unconfig_common(dev_info_t *, dev_info_t **, int, major_t,
 215     struct brevq_node **);
 216 static int
 217 ndi_devi_config_obp_args(dev_info_t *parent, char *devnm,
 218     dev_info_t **childp, int flags);
 219 static void i_link_vhci_node(dev_info_t *);
 220 static void ndi_devi_exit_and_wait(dev_info_t *dip,
 221     int circular, clock_t end_time);
 222 static int ndi_devi_unbind_driver(dev_info_t *dip);
 223 
 224 static int i_ddi_check_retire(dev_info_t *dip);
 225 
 226 static void quiesce_one_device(dev_info_t *, void *);
 227 
 228 dev_info_t *ddi_alias_redirect(char *alias);
 229 char *ddi_curr_redirect(char *currpath);
 230 
 231 
 232 /*
 233  * dev_info cache and node management
 234  */
 235 
 236 /* initialize dev_info node cache */
 237 void
 238 i_ddi_node_cache_init()
 239 {
 240         ASSERT(ddi_node_cache == NULL);
 241         ddi_node_cache = kmem_cache_create("dev_info_node_cache",
 242             sizeof (struct dev_info), 0, NULL, NULL, NULL, NULL, NULL, 0);
 243 
 244         if (ddidebug & DDI_AUDIT)
 245                 da_log_init();
 246 }
 247 
 248 
 249 /*
 250  * Allocating a dev_info node, callable from interrupt context with KM_NOSLEEP
 251  * The allocated node has a reference count of 0.
 252  */
 253 dev_info_t *
 254 i_ddi_alloc_node(dev_info_t *pdip, char *node_name, pnode_t nodeid,
 255     int instance, ddi_prop_t *sys_prop, int flag)
 256 {
 257         struct dev_info *devi;
 258         struct devi_nodeid *elem;
 259         static char failed[] = "i_ddi_alloc_node: out of memory";
 260 
 261         ASSERT(node_name != NULL);
 262 
 263         if ((devi = kmem_cache_alloc(ddi_node_cache, flag)) == NULL) {
 264                 cmn_err(CE_NOTE, failed);
 265                 return (NULL);
 266         }
 267 
 268         bzero(devi, sizeof (struct dev_info));
 269 
 270         if (devinfo_audit_log) {
 271                 devi->devi_audit = kmem_zalloc(sizeof (devinfo_audit_t), flag);
 272                 if (devi->devi_audit == NULL)
 273                         goto fail;
 274         }
 275 
 276         if ((devi->devi_node_name = i_ddi_strdup(node_name, flag)) == NULL)
 277                 goto fail;
 278 
 279         /* default binding name is node name */
 280         devi->devi_binding_name = devi->devi_node_name;
 281         devi->devi_major = DDI_MAJOR_T_NONE; /* unbound by default */
 282 
 283         /*
 284          * Make a copy of system property
 285          */
 286         if (sys_prop &&
 287             (devi->devi_sys_prop_ptr = i_ddi_prop_list_dup(sys_prop, flag))
 288             == NULL)
 289                 goto fail;
 290 
 291         /*
 292          * Assign devi_nodeid, devi_node_class, devi_node_attributes
 293          * according to the following algorithm:
 294          *
 295          * nodeid arg                   node class              node attributes
 296          *
 297          * DEVI_PSEUDO_NODEID           DDI_NC_PSEUDO           A
 298          * DEVI_SID_NODEID              DDI_NC_PSEUDO           A,P
 299          * DEVI_SID_HIDDEN_NODEID       DDI_NC_PSEUDO           A,P,H
 300          * DEVI_SID_HP_NODEID           DDI_NC_PSEUDO           A,P,h
 301          * DEVI_SID_HP_HIDDEN_NODEID    DDI_NC_PSEUDO           A,P,H,h
 302          * other                        DDI_NC_PROM             P
 303          *
 304          * Where A = DDI_AUTO_ASSIGNED_NODEID (auto-assign a nodeid)
 305          * and   P = DDI_PERSISTENT
 306          * and   H = DDI_HIDDEN_NODE
 307          * and   h = DDI_HOTPLUG_NODE
 308          *
 309          * auto-assigned nodeids are also auto-freed.
 310          */
 311         devi->devi_node_attributes = 0;
 312         switch (nodeid) {
 313         case DEVI_SID_HIDDEN_NODEID:
 314                 devi->devi_node_attributes |= DDI_HIDDEN_NODE;
 315                 goto sid;
 316 
 317         case DEVI_SID_HP_NODEID:
 318                 devi->devi_node_attributes |= DDI_HOTPLUG_NODE;
 319                 goto sid;
 320 
 321         case DEVI_SID_HP_HIDDEN_NODEID:
 322                 devi->devi_node_attributes |= DDI_HIDDEN_NODE;
 323                 devi->devi_node_attributes |= DDI_HOTPLUG_NODE;
 324                 goto sid;
 325 
 326         case DEVI_SID_NODEID:
 327 sid:            devi->devi_node_attributes |= DDI_PERSISTENT;
 328                 if ((elem = kmem_zalloc(sizeof (*elem), flag)) == NULL)
 329                         goto fail;
 330                 /*FALLTHROUGH*/
 331 
 332         case DEVI_PSEUDO_NODEID:
 333                 devi->devi_node_attributes |= DDI_AUTO_ASSIGNED_NODEID;
 334                 devi->devi_node_class = DDI_NC_PSEUDO;
 335                 if (impl_ddi_alloc_nodeid(&devi->devi_nodeid)) {
 336                         panic("i_ddi_alloc_node: out of nodeids");
 337                         /*NOTREACHED*/
 338                 }
 339                 break;
 340 
 341         default:
 342                 if ((elem = kmem_zalloc(sizeof (*elem), flag)) == NULL)
 343                         goto fail;
 344 
 345                 /*
 346                  * the nodetype is 'prom', try to 'take' the nodeid now.
 347                  * This requires memory allocation, so check for failure.
 348                  */
 349                 if (impl_ddi_take_nodeid(nodeid, flag) != 0) {
 350                         kmem_free(elem, sizeof (*elem));
 351                         goto fail;
 352                 }
 353 
 354                 devi->devi_nodeid = nodeid;
 355                 devi->devi_node_class = DDI_NC_PROM;
 356                 devi->devi_node_attributes = DDI_PERSISTENT;
 357                 break;
 358         }
 359 
 360         if (ndi_dev_is_persistent_node((dev_info_t *)devi)) {
 361                 mutex_enter(&devimap->dno_lock);
 362                 elem->next = devimap->dno_free;
 363                 devimap->dno_free = elem;
 364                 mutex_exit(&devimap->dno_lock);
 365         }
 366 
 367         /*
 368          * Instance is normally initialized to -1. In a few special
 369          * cases, the caller may specify an instance (e.g. CPU nodes).
 370          */
 371         devi->devi_instance = instance;
 372 
 373         /*
 374          * set parent and bus_ctl parent
 375          */
 376         devi->devi_parent = DEVI(pdip);
 377         devi->devi_bus_ctl = DEVI(pdip);
 378 
 379         NDI_CONFIG_DEBUG((CE_CONT,
 380             "i_ddi_alloc_node: name=%s id=%d\n", node_name, devi->devi_nodeid));
 381 
 382         cv_init(&(devi->devi_cv), NULL, CV_DEFAULT, NULL);
 383         mutex_init(&(devi->devi_lock), NULL, MUTEX_DEFAULT, NULL);
 384         mutex_init(&(devi->devi_pm_lock), NULL, MUTEX_DEFAULT, NULL);
 385         mutex_init(&(devi->devi_pm_busy_lock), NULL, MUTEX_DEFAULT, NULL);
 386 
 387         RIO_TRACE((CE_NOTE, "i_ddi_alloc_node: Initing contract fields: "
 388             "dip=%p, name=%s", (void *)devi, node_name));
 389 
 390         mutex_init(&(devi->devi_ct_lock), NULL, MUTEX_DEFAULT, NULL);
 391         cv_init(&(devi->devi_ct_cv), NULL, CV_DEFAULT, NULL);
 392         devi->devi_ct_count = -1;    /* counter not in use if -1 */
 393         list_create(&(devi->devi_ct), sizeof (cont_device_t),
 394             offsetof(cont_device_t, cond_next));
 395 
 396         i_ddi_set_node_state((dev_info_t *)devi, DS_PROTO);
 397         da_log_enter((dev_info_t *)devi);
 398         return ((dev_info_t *)devi);
 399 
 400 fail:
 401         if (devi->devi_sys_prop_ptr)
 402                 i_ddi_prop_list_delete(devi->devi_sys_prop_ptr);
 403         if (devi->devi_node_name)
 404                 kmem_free(devi->devi_node_name, strlen(node_name) + 1);
 405         if (devi->devi_audit)
 406                 kmem_free(devi->devi_audit, sizeof (devinfo_audit_t));
 407         kmem_cache_free(ddi_node_cache, devi);
 408         cmn_err(CE_NOTE, failed);
 409         return (NULL);
 410 }
 411 
 412 /*
 413  * free a dev_info structure.
 414  * NB. Not callable from interrupt since impl_ddi_free_nodeid may block.
 415  */
 416 void
 417 i_ddi_free_node(dev_info_t *dip)
 418 {
 419         struct dev_info *devi = DEVI(dip);
 420         struct devi_nodeid *elem;
 421 
 422         ASSERT(devi->devi_ref == 0);
 423         ASSERT(devi->devi_addr == NULL);
 424         ASSERT(devi->devi_node_state == DS_PROTO);
 425         ASSERT(devi->devi_child == NULL);
 426         ASSERT(devi->devi_hp_hdlp == NULL);
 427 
 428         /* free devi_addr_buf allocated by ddi_set_name_addr() */
 429         if (devi->devi_addr_buf)
 430                 kmem_free(devi->devi_addr_buf, 2 * MAXNAMELEN);
 431 
 432         if (i_ndi_dev_is_auto_assigned_node(dip))
 433                 impl_ddi_free_nodeid(DEVI(dip)->devi_nodeid);
 434 
 435         if (ndi_dev_is_persistent_node(dip)) {
 436                 mutex_enter(&devimap->dno_lock);
 437                 ASSERT(devimap->dno_free);
 438                 elem = devimap->dno_free;
 439                 devimap->dno_free = elem->next;
 440                 mutex_exit(&devimap->dno_lock);
 441                 kmem_free(elem, sizeof (*elem));
 442         }
 443 
 444         if (DEVI(dip)->devi_compat_names)
 445                 kmem_free(DEVI(dip)->devi_compat_names,
 446                     DEVI(dip)->devi_compat_length);
 447         if (DEVI(dip)->devi_rebinding_name)
 448                 kmem_free(DEVI(dip)->devi_rebinding_name,
 449                     strlen(DEVI(dip)->devi_rebinding_name) + 1);
 450 
 451         ddi_prop_remove_all(dip);       /* remove driver properties */
 452         if (devi->devi_sys_prop_ptr)
 453                 i_ddi_prop_list_delete(devi->devi_sys_prop_ptr);
 454         if (devi->devi_hw_prop_ptr)
 455                 i_ddi_prop_list_delete(devi->devi_hw_prop_ptr);
 456 
 457         if (DEVI(dip)->devi_devid_str)
 458                 ddi_devid_str_free(DEVI(dip)->devi_devid_str);
 459 
 460         i_ddi_set_node_state(dip, DS_INVAL);
 461         da_log_enter(dip);
 462         if (devi->devi_audit) {
 463                 kmem_free(devi->devi_audit, sizeof (devinfo_audit_t));
 464         }
 465         if (devi->devi_device_class)
 466                 kmem_free(devi->devi_device_class,
 467                     strlen(devi->devi_device_class) + 1);
 468         cv_destroy(&(devi->devi_cv));
 469         mutex_destroy(&(devi->devi_lock));
 470         mutex_destroy(&(devi->devi_pm_lock));
 471         mutex_destroy(&(devi->devi_pm_busy_lock));
 472 
 473         RIO_TRACE((CE_NOTE, "i_ddi_free_node: destroying contract fields: "
 474             "dip=%p", (void *)dip));
 475         contract_device_remove_dip(dip);
 476         ASSERT(devi->devi_ct_count == -1);
 477         ASSERT(list_is_empty(&(devi->devi_ct)));
 478         cv_destroy(&(devi->devi_ct_cv));
 479         list_destroy(&(devi->devi_ct));
 480         /* free this last since contract_device_remove_dip() uses it */
 481         mutex_destroy(&(devi->devi_ct_lock));
 482         RIO_TRACE((CE_NOTE, "i_ddi_free_node: destroyed all contract fields: "
 483             "dip=%p, name=%s", (void *)dip, devi->devi_node_name));
 484 
 485         kmem_free(devi->devi_node_name, strlen(devi->devi_node_name) + 1);
 486 
 487         /* free event data */
 488         if (devi->devi_ev_path)
 489                 kmem_free(devi->devi_ev_path, MAXPATHLEN);
 490 
 491         kmem_cache_free(ddi_node_cache, devi);
 492 }
 493 
 494 
 495 /*
 496  * Node state transitions
 497  */
 498 
 499 /*
 500  * Change the node name
 501  */
 502 int
 503 ndi_devi_set_nodename(dev_info_t *dip, char *name, int flags)
 504 {
 505         _NOTE(ARGUNUSED(flags))
 506         char *nname, *oname;
 507 
 508         ASSERT(dip && name);
 509 
 510         oname = DEVI(dip)->devi_node_name;
 511         if (strcmp(oname, name) == 0)
 512                 return (DDI_SUCCESS);
 513 
 514         /*
 515          * pcicfg_fix_ethernet requires a name change after node
 516          * is linked into the tree. When pcicfg is fixed, we
 517          * should only allow name change in DS_PROTO state.
 518          */
 519         if (i_ddi_node_state(dip) >= DS_BOUND) {
 520                 /*
 521                  * Don't allow name change once node is bound
 522                  */
 523                 cmn_err(CE_NOTE,
 524                     "ndi_devi_set_nodename: node already bound dip = %p,"
 525                     " %s -> %s", (void *)dip, ddi_node_name(dip), name);
 526                 return (NDI_FAILURE);
 527         }
 528 
 529         nname = i_ddi_strdup(name, KM_SLEEP);
 530         DEVI(dip)->devi_node_name = nname;
 531         i_ddi_set_binding_name(dip, nname);
 532         kmem_free(oname, strlen(oname) + 1);
 533 
 534         da_log_enter(dip);
 535         return (NDI_SUCCESS);
 536 }
 537 
 538 void
 539 i_ddi_add_devimap(dev_info_t *dip)
 540 {
 541         struct devi_nodeid *elem;
 542 
 543         ASSERT(dip);
 544 
 545         if (!ndi_dev_is_persistent_node(dip))
 546                 return;
 547 
 548         ASSERT(ddi_get_parent(dip) == NULL || (DEVI_VHCI_NODE(dip)) ||
 549             DEVI_BUSY_OWNED(ddi_get_parent(dip)));
 550 
 551         mutex_enter(&devimap->dno_lock);
 552 
 553         ASSERT(devimap->dno_free);
 554 
 555         elem = devimap->dno_free;
 556         devimap->dno_free = elem->next;
 557 
 558         elem->nodeid = ddi_get_nodeid(dip);
 559         elem->dip = dip;
 560         elem->next = devimap->dno_head;
 561         devimap->dno_head = elem;
 562 
 563         devimap->dno_list_length++;
 564 
 565         mutex_exit(&devimap->dno_lock);
 566 }
 567 
 568 static int
 569 i_ddi_remove_devimap(dev_info_t *dip)
 570 {
 571         struct devi_nodeid *prev, *elem;
 572         static const char *fcn = "i_ddi_remove_devimap";
 573 
 574         ASSERT(dip);
 575 
 576         if (!ndi_dev_is_persistent_node(dip))
 577                 return (DDI_SUCCESS);
 578 
 579         mutex_enter(&devimap->dno_lock);
 580 
 581         /*
 582          * The following check is done with dno_lock held
 583          * to prevent race between dip removal and
 584          * e_ddi_prom_node_to_dip()
 585          */
 586         if (e_ddi_devi_holdcnt(dip)) {
 587                 mutex_exit(&devimap->dno_lock);
 588                 return (DDI_FAILURE);
 589         }
 590 
 591         ASSERT(devimap->dno_head);
 592         ASSERT(devimap->dno_list_length > 0);
 593 
 594         prev = NULL;
 595         for (elem = devimap->dno_head; elem; elem = elem->next) {
 596                 if (elem->dip == dip) {
 597                         ASSERT(elem->nodeid == ddi_get_nodeid(dip));
 598                         break;
 599                 }
 600                 prev = elem;
 601         }
 602 
 603         if (elem && prev)
 604                 prev->next = elem->next;
 605         else if (elem)
 606                 devimap->dno_head = elem->next;
 607         else
 608                 panic("%s: devinfo node(%p) not found",
 609                     fcn, (void *)dip);
 610 
 611         devimap->dno_list_length--;
 612 
 613         elem->nodeid = 0;
 614         elem->dip = NULL;
 615 
 616         elem->next = devimap->dno_free;
 617         devimap->dno_free = elem;
 618 
 619         mutex_exit(&devimap->dno_lock);
 620 
 621         return (DDI_SUCCESS);
 622 }
 623 
 624 /*
 625  * Link this node into the devinfo tree and add to orphan list
 626  * Not callable from interrupt context
 627  */
 628 static void
 629 link_node(dev_info_t *dip)
 630 {
 631         struct dev_info *devi = DEVI(dip);
 632         struct dev_info *parent = devi->devi_parent;
 633         dev_info_t **dipp;
 634 
 635         ASSERT(parent); /* never called for root node */
 636 
 637         NDI_CONFIG_DEBUG((CE_CONT, "link_node: parent = %s child = %s\n",
 638             parent->devi_node_name, devi->devi_node_name));
 639 
 640         /*
 641          * Hold the global_vhci_lock before linking any direct
 642          * children of rootnex driver. This special lock protects
 643          * linking and unlinking for rootnext direct children.
 644          */
 645         if ((dev_info_t *)parent == ddi_root_node())
 646                 mutex_enter(&global_vhci_lock);
 647 
 648         /*
 649          * attach the node to end of the list unless the node is already there
 650          */
 651         dipp = (dev_info_t **)(&DEVI(parent)->devi_child);
 652         while (*dipp && (*dipp != dip)) {
 653                 dipp = (dev_info_t **)(&DEVI(*dipp)->devi_sibling);
 654         }
 655         ASSERT(*dipp == NULL);  /* node is not linked */
 656 
 657         /*
 658          * Now that we are in the tree, update the devi-nodeid map.
 659          */
 660         i_ddi_add_devimap(dip);
 661 
 662         /*
 663          * This is a temporary workaround for Bug 4618861.
 664          * We keep the scsi_vhci nexus node on the left side of the devinfo
 665          * tree (under the root nexus driver), so that virtual nodes under
 666          * scsi_vhci will be SUSPENDed first and RESUMEd last.  This ensures
 667          * that the pHCI nodes are active during times when their clients
 668          * may be depending on them.  This workaround embodies the knowledge
 669          * that system PM and CPR both traverse the tree left-to-right during
 670          * SUSPEND and right-to-left during RESUME.
 671          * Extending the workaround to IB Nexus/VHCI
 672          * driver also.
 673          */
 674         if (strcmp(devi->devi_binding_name, "scsi_vhci") == 0) {
 675                 /* Add scsi_vhci to beginning of list */
 676                 ASSERT((dev_info_t *)parent == top_devinfo);
 677                 /* scsi_vhci under rootnex */
 678                 devi->devi_sibling = parent->devi_child;
 679                 parent->devi_child = devi;
 680         } else if (strcmp(devi->devi_binding_name, "ib") == 0) {
 681                 i_link_vhci_node(dip);
 682         } else {
 683                 /* Add to end of list */
 684                 *dipp = dip;
 685                 DEVI(dip)->devi_sibling = NULL;
 686         }
 687 
 688         /*
 689          * Release the global_vhci_lock before linking any direct
 690          * children of rootnex driver.
 691          */
 692         if ((dev_info_t *)parent == ddi_root_node())
 693                 mutex_exit(&global_vhci_lock);
 694 
 695         /* persistent nodes go on orphan list */
 696         if (ndi_dev_is_persistent_node(dip))
 697                 add_to_dn_list(&orphanlist, dip);
 698 }
 699 
 700 /*
 701  * Unlink this node from the devinfo tree
 702  */
 703 static int
 704 unlink_node(dev_info_t *dip)
 705 {
 706         struct dev_info *devi = DEVI(dip);
 707         struct dev_info *parent = devi->devi_parent;
 708         dev_info_t **dipp;
 709         ddi_hp_cn_handle_t *hdlp;
 710 
 711         ASSERT(parent != NULL);
 712         ASSERT(devi->devi_node_state == DS_LINKED);
 713 
 714         NDI_CONFIG_DEBUG((CE_CONT, "unlink_node: name = %s\n",
 715             ddi_node_name(dip)));
 716 
 717         /* check references */
 718         if (devi->devi_ref || i_ddi_remove_devimap(dip) != DDI_SUCCESS)
 719                 return (DDI_FAILURE);
 720 
 721         /*
 722          * Hold the global_vhci_lock before linking any direct
 723          * children of rootnex driver.
 724          */
 725         if ((dev_info_t *)parent == ddi_root_node())
 726                 mutex_enter(&global_vhci_lock);
 727 
 728         dipp = (dev_info_t **)(&DEVI(parent)->devi_child);
 729         while (*dipp && (*dipp != dip)) {
 730                 dipp = (dev_info_t **)(&DEVI(*dipp)->devi_sibling);
 731         }
 732         if (*dipp) {
 733                 *dipp = (dev_info_t *)(devi->devi_sibling);
 734                 devi->devi_sibling = NULL;
 735         } else {
 736                 NDI_CONFIG_DEBUG((CE_NOTE, "unlink_node: %s not linked",
 737                     devi->devi_node_name));
 738         }
 739 
 740         /*
 741          * Release the global_vhci_lock before linking any direct
 742          * children of rootnex driver.
 743          */
 744         if ((dev_info_t *)parent == ddi_root_node())
 745                 mutex_exit(&global_vhci_lock);
 746 
 747         /* Remove node from orphan list */
 748         if (ndi_dev_is_persistent_node(dip)) {
 749                 remove_from_dn_list(&orphanlist, dip);
 750         }
 751 
 752         /* Update parent's hotplug handle list */
 753         for (hdlp = DEVI(parent)->devi_hp_hdlp; hdlp; hdlp = hdlp->next) {
 754                 if (hdlp->cn_info.cn_child == dip)
 755                         hdlp->cn_info.cn_child = NULL;
 756         }
 757         return (DDI_SUCCESS);
 758 }
 759 
 760 /*
 761  * Bind this devinfo node to a driver. If compat is NON-NULL, try that first.
 762  * Else, use the node-name.
 763  *
 764  * NOTE: IEEE1275 specifies that nodename should be tried before compatible.
 765  *      Solaris implementation binds nodename after compatible.
 766  *
 767  * If we find a binding,
 768  * - set the binding name to the string,
 769  * - set major number to driver major
 770  *
 771  * If we don't find a binding,
 772  * - return failure
 773  */
 774 static int
 775 bind_node(dev_info_t *dip)
 776 {
 777         char *p = NULL;
 778         major_t major = DDI_MAJOR_T_NONE;
 779         struct dev_info *devi = DEVI(dip);
 780         dev_info_t *parent = ddi_get_parent(dip);
 781 
 782         ASSERT(devi->devi_node_state == DS_LINKED);
 783 
 784         NDI_CONFIG_DEBUG((CE_CONT, "bind_node: 0x%p(name = %s)\n",
 785             (void *)dip, ddi_node_name(dip)));
 786 
 787         mutex_enter(&DEVI(dip)->devi_lock);
 788         if (DEVI(dip)->devi_flags & DEVI_NO_BIND) {
 789                 mutex_exit(&DEVI(dip)->devi_lock);
 790                 return (DDI_FAILURE);
 791         }
 792         mutex_exit(&DEVI(dip)->devi_lock);
 793 
 794         /* find the driver with most specific binding using compatible */
 795         major = ddi_compatible_driver_major(dip, &p);
 796         if (major == DDI_MAJOR_T_NONE)
 797                 return (DDI_FAILURE);
 798 
 799         devi->devi_major = major;
 800         if (p != NULL) {
 801                 i_ddi_set_binding_name(dip, p);
 802                 NDI_CONFIG_DEBUG((CE_CONT, "bind_node: %s bound to %s\n",
 803                     devi->devi_node_name, p));
 804         }
 805 
 806         /* Link node to per-driver list */
 807         link_to_driver_list(dip);
 808 
 809         /*
 810          * reset parent flag so that nexus will merge .conf props
 811          */
 812         if (ndi_dev_is_persistent_node(dip)) {
 813                 mutex_enter(&DEVI(parent)->devi_lock);
 814                 DEVI(parent)->devi_flags &=
 815                     ~(DEVI_ATTACHED_CHILDREN|DEVI_MADE_CHILDREN);
 816                 mutex_exit(&DEVI(parent)->devi_lock);
 817         }
 818         return (DDI_SUCCESS);
 819 }
 820 
 821 /*
 822  * Unbind this devinfo node
 823  * Called before the node is destroyed or driver is removed from system
 824  */
 825 static int
 826 unbind_node(dev_info_t *dip)
 827 {
 828         ASSERT(DEVI(dip)->devi_node_state == DS_BOUND);
 829         ASSERT(DEVI(dip)->devi_major != DDI_MAJOR_T_NONE);
 830 
 831         /* check references */
 832         if (DEVI(dip)->devi_ref)
 833                 return (DDI_FAILURE);
 834 
 835         NDI_CONFIG_DEBUG((CE_CONT, "unbind_node: 0x%p(name = %s)\n",
 836             (void *)dip, ddi_node_name(dip)));
 837 
 838         unlink_from_driver_list(dip);
 839 
 840         DEVI(dip)->devi_major = DDI_MAJOR_T_NONE;
 841         DEVI(dip)->devi_binding_name = DEVI(dip)->devi_node_name;
 842         return (DDI_SUCCESS);
 843 }
 844 
 845 /*
 846  * Initialize a node: calls the parent nexus' bus_ctl ops to do the operation.
 847  * Must hold parent and per-driver list while calling this function.
 848  * A successful init_node() returns with an active ndi_hold_devi() hold on
 849  * the parent.
 850  */
 851 static int
 852 init_node(dev_info_t *dip)
 853 {
 854         int error;
 855         dev_info_t *pdip = ddi_get_parent(dip);
 856         int (*f)(dev_info_t *, dev_info_t *, ddi_ctl_enum_t, void *, void *);
 857         char *path;
 858         major_t major;
 859         ddi_devid_t devid = NULL;
 860 
 861         ASSERT(i_ddi_node_state(dip) == DS_BOUND);
 862 
 863         /* should be DS_READY except for pcmcia ... */
 864         ASSERT(i_ddi_node_state(pdip) >= DS_PROBED);
 865 
 866         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
 867         (void) ddi_pathname(dip, path);
 868         NDI_CONFIG_DEBUG((CE_CONT, "init_node: entry: path %s 0x%p\n",
 869             path, (void *)dip));
 870 
 871         /*
 872          * The parent must have a bus_ctl operation.
 873          */
 874         if ((DEVI(pdip)->devi_ops->devo_bus_ops == NULL) ||
 875             (f = DEVI(pdip)->devi_ops->devo_bus_ops->bus_ctl) == NULL) {
 876                 error = DDI_FAILURE;
 877                 goto out;
 878         }
 879 
 880         add_global_props(dip);
 881 
 882         /*
 883          * Invoke the parent's bus_ctl operation with the DDI_CTLOPS_INITCHILD
 884          * command to transform the child to canonical form 1. If there
 885          * is an error, ddi_remove_child should be called, to clean up.
 886          */
 887         error = (*f)(pdip, pdip, DDI_CTLOPS_INITCHILD, dip, NULL);
 888         if (error != DDI_SUCCESS) {
 889                 NDI_CONFIG_DEBUG((CE_CONT, "init_node: %s 0x%p failed\n",
 890                     path, (void *)dip));
 891                 remove_global_props(dip);
 892 
 893                 /*
 894                  * If a nexus INITCHILD implementation calls ddi_devid_regster()
 895                  * prior to setting devi_addr, the devid is not recorded in
 896                  * the devid cache (i.e. DEVI_CACHED_DEVID is not set).
 897                  * With mpxio, while the vhci client path may be missing
 898                  * from the cache, phci pathinfo paths may have already be
 899                  * added to the cache, against the client dip, by use of
 900                  * e_devid_cache_pathinfo().  Because of this, when INITCHILD
 901                  * of the client fails, we need to purge the client dip from
 902                  * the cache even if DEVI_CACHED_DEVID is not set - if only
 903                  * devi_devid_str is set.
 904                  */
 905                 mutex_enter(&DEVI(dip)->devi_lock);
 906                 if ((DEVI(dip)->devi_flags & DEVI_CACHED_DEVID) ||
 907                     DEVI(dip)->devi_devid_str) {
 908                         DEVI(dip)->devi_flags &= ~DEVI_CACHED_DEVID;
 909                         mutex_exit(&DEVI(dip)->devi_lock);
 910                         ddi_devid_unregister(dip);
 911                 } else
 912                         mutex_exit(&DEVI(dip)->devi_lock);
 913 
 914                 /* in case nexus driver didn't clear this field */
 915                 ddi_set_name_addr(dip, NULL);
 916                 error = DDI_FAILURE;
 917                 goto out;
 918         }
 919 
 920         ndi_hold_devi(pdip);                    /* initial hold of parent */
 921 
 922         /* recompute path after initchild for @addr information */
 923         (void) ddi_pathname(dip, path);
 924 
 925         /* Check for duplicate nodes */
 926         if (find_duplicate_child(pdip, dip) != NULL) {
 927                 /*
 928                  * uninit_node() the duplicate - a successful uninit_node()
 929                  * will release inital hold of parent using ndi_rele_devi().
 930                  */
 931                 if ((error = uninit_node(dip)) != DDI_SUCCESS) {
 932                         ndi_rele_devi(pdip);    /* release initial hold */
 933                         cmn_err(CE_WARN, "init_node: uninit of duplicate "
 934                             "node %s failed", path);
 935                 }
 936                 NDI_CONFIG_DEBUG((CE_CONT, "init_node: duplicate uninit "
 937                     "%s 0x%p%s\n", path, (void *)dip,
 938                     (error == DDI_SUCCESS) ? "" : " failed"));
 939                 error = DDI_FAILURE;
 940                 goto out;
 941         }
 942 
 943         /*
 944          * If a devid was registered for a DS_BOUND node then the devid_cache
 945          * may not have captured the path. Detect this situation and ensure that
 946          * the path enters the cache now that devi_addr is established.
 947          */
 948         if (!(DEVI(dip)->devi_flags & DEVI_CACHED_DEVID) &&
 949             (ddi_devid_get(dip, &devid) == DDI_SUCCESS)) {
 950                 if (e_devid_cache_register(dip, devid) == DDI_SUCCESS) {
 951                         mutex_enter(&DEVI(dip)->devi_lock);
 952                         DEVI(dip)->devi_flags |= DEVI_CACHED_DEVID;
 953                         mutex_exit(&DEVI(dip)->devi_lock);
 954                 }
 955 
 956                 ddi_devid_free(devid);
 957         }
 958 
 959         /*
 960          * Check to see if we have a path-oriented driver alias that overrides
 961          * the current driver binding. If so, we need to rebind. This check
 962          * needs to be delayed until after a successful DDI_CTLOPS_INITCHILD,
 963          * so the unit-address is established on the last component of the path.
 964          *
 965          * NOTE: Allowing a path-oriented alias to change the driver binding
 966          * of a driver.conf node results in non-intuitive property behavior.
 967          * We provide a tunable (driver_conf_allow_path_alias) to control
 968          * this behavior. See uninit_node() for more details.
 969          *
 970          * NOTE: If you are adding a path-oriented alias for the boot device,
 971          * and there is mismatch between OBP and the kernel in regard to
 972          * generic name use, like "disk" .vs. "ssd", then you will need
 973          * to add a path-oriented alias for both paths.
 974          */
 975         major = ddi_name_to_major(path);
 976         if (driver_active(major) && (major != DEVI(dip)->devi_major) &&
 977             (ndi_dev_is_persistent_node(dip) || driver_conf_allow_path_alias)) {
 978 
 979                 /* Mark node for rebind processing. */
 980                 mutex_enter(&DEVI(dip)->devi_lock);
 981                 DEVI(dip)->devi_flags |= DEVI_REBIND;
 982                 mutex_exit(&DEVI(dip)->devi_lock);
 983 
 984                 /*
 985                  * Add an extra hold on the parent to prevent it from ever
 986                  * having a zero devi_ref during the child rebind process.
 987                  * This is necessary to ensure that the parent will never
 988                  * detach(9E) during the rebind.
 989                  */
 990                 ndi_hold_devi(pdip);            /* extra hold of parent */
 991 
 992                 /*
 993                  * uninit_node() current binding - a successful uninit_node()
 994                  * will release extra hold of parent using ndi_rele_devi().
 995                  */
 996                 if ((error = uninit_node(dip)) != DDI_SUCCESS) {
 997                         ndi_rele_devi(pdip);    /* release extra hold */
 998                         ndi_rele_devi(pdip);    /* release initial hold */
 999                         cmn_err(CE_WARN, "init_node: uninit for rebind "
1000                             "of node %s failed", path);
1001                         goto out;
1002                 }
1003 
1004                 /* Unbind: demote the node back to DS_LINKED.  */
1005                 if ((error = ndi_devi_unbind_driver(dip)) != DDI_SUCCESS) {
1006                         ndi_rele_devi(pdip);    /* release initial hold */
1007                         cmn_err(CE_WARN, "init_node: unbind for rebind "
1008                             "of node %s failed", path);
1009                         goto out;
1010                 }
1011 
1012                 /* establish rebinding name */
1013                 if (DEVI(dip)->devi_rebinding_name == NULL)
1014                         DEVI(dip)->devi_rebinding_name =
1015                             i_ddi_strdup(path, KM_SLEEP);
1016 
1017                 /*
1018                  * Now that we are demoted and marked for rebind, repromote.
1019                  * We need to do this in steps, instead of just calling
1020                  * ddi_initchild, so that we can redo the merge operation
1021                  * after we are rebound to the path-bound driver.
1022                  *
1023                  * Start by rebinding node to the path-bound driver.
1024                  */
1025                 if ((error = ndi_devi_bind_driver(dip, 0)) != DDI_SUCCESS) {
1026                         ndi_rele_devi(pdip);    /* release initial hold */
1027                         cmn_err(CE_WARN, "init_node: rebind "
1028                             "of node %s failed", path);
1029                         goto out;
1030                 }
1031 
1032                 /*
1033                  * If the node is not a driver.conf node then merge
1034                  * driver.conf properties from new path-bound driver.conf.
1035                  */
1036                 if (ndi_dev_is_persistent_node(dip))
1037                         (void) i_ndi_make_spec_children(pdip, 0);
1038 
1039                 /*
1040                  * Now that we have taken care of merge, repromote back
1041                  * to DS_INITIALIZED.
1042                  */
1043                 error = ddi_initchild(pdip, dip);
1044                 NDI_CONFIG_DEBUG((CE_CONT, "init_node: rebind "
1045                     "%s 0x%p\n", path, (void *)dip));
1046 
1047                 /*
1048                  * Release our initial hold. If ddi_initchild() was
1049                  * successful then it will return with the active hold.
1050                  */
1051                 ndi_rele_devi(pdip);
1052                 goto out;
1053         }
1054 
1055         /*
1056          * Apply multi-parent/deep-nexus optimization to the new node
1057          */
1058         DEVI(dip)->devi_instance = e_ddi_assign_instance(dip);
1059         ddi_optimize_dtree(dip);
1060         error = DDI_SUCCESS;            /* return with active hold */
1061 
1062 out:    if (error != DDI_SUCCESS) {
1063                 /* On failure ensure that DEVI_REBIND is cleared */
1064                 mutex_enter(&DEVI(dip)->devi_lock);
1065                 DEVI(dip)->devi_flags &= ~DEVI_REBIND;
1066                 mutex_exit(&DEVI(dip)->devi_lock);
1067         }
1068         kmem_free(path, MAXPATHLEN);
1069         return (error);
1070 }
1071 
1072 /*
1073  * Uninitialize node
1074  * The per-driver list must be held busy during the call.
1075  * A successful uninit_node() releases the init_node() hold on
1076  * the parent by calling ndi_rele_devi().
1077  */
1078 static int
1079 uninit_node(dev_info_t *dip)
1080 {
1081         int node_state_entry;
1082         dev_info_t *pdip;
1083         struct dev_ops *ops;
1084         int (*f)();
1085         int error;
1086         char *addr;
1087 
1088         /*
1089          * Don't check for references here or else a ref-counted
1090          * dip cannot be downgraded by the framework.
1091          */
1092         node_state_entry = i_ddi_node_state(dip);
1093         ASSERT((node_state_entry == DS_BOUND) ||
1094             (node_state_entry == DS_INITIALIZED));
1095         pdip = ddi_get_parent(dip);
1096         ASSERT(pdip);
1097 
1098         NDI_CONFIG_DEBUG((CE_CONT, "uninit_node: 0x%p(%s%d)\n",
1099             (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1100 
1101         if (((ops = ddi_get_driver(pdip)) == NULL) ||
1102             (ops->devo_bus_ops == NULL) ||
1103             ((f = ops->devo_bus_ops->bus_ctl) == NULL)) {
1104                 return (DDI_FAILURE);
1105         }
1106 
1107         /*
1108          * save the @addr prior to DDI_CTLOPS_UNINITCHILD for use in
1109          * freeing the instance if it succeeds.
1110          */
1111         if (node_state_entry == DS_INITIALIZED) {
1112                 addr = ddi_get_name_addr(dip);
1113                 if (addr)
1114                         addr = i_ddi_strdup(addr, KM_SLEEP);
1115         } else {
1116                 addr = NULL;
1117         }
1118 
1119         error = (*f)(pdip, pdip, DDI_CTLOPS_UNINITCHILD, dip, (void *)NULL);
1120         if (error == DDI_SUCCESS) {
1121                 /* ensure that devids are unregistered */
1122                 mutex_enter(&DEVI(dip)->devi_lock);
1123                 if ((DEVI(dip)->devi_flags & DEVI_CACHED_DEVID)) {
1124                         DEVI(dip)->devi_flags &= ~DEVI_CACHED_DEVID;
1125                         mutex_exit(&DEVI(dip)->devi_lock);
1126                         ddi_devid_unregister(dip);
1127                 } else
1128                         mutex_exit(&DEVI(dip)->devi_lock);
1129 
1130                 /* if uninitchild forgot to set devi_addr to NULL do it now */
1131                 ddi_set_name_addr(dip, NULL);
1132 
1133                 /*
1134                  * Free instance number. This is a no-op if instance has
1135                  * been kept by probe_node().  Avoid free when we are called
1136                  * from init_node (DS_BOUND) because the instance has not yet
1137                  * been assigned.
1138                  */
1139                 if (node_state_entry == DS_INITIALIZED) {
1140                         e_ddi_free_instance(dip, addr);
1141                         DEVI(dip)->devi_instance = -1;
1142                 }
1143 
1144                 /* release the init_node hold */
1145                 ndi_rele_devi(pdip);
1146 
1147                 remove_global_props(dip);
1148 
1149                 /*
1150                  * NOTE: The decision on whether to allow a path-oriented
1151                  * rebind of a driver.conf enumerated node is made by
1152                  * init_node() based on driver_conf_allow_path_alias. The
1153                  * rebind code below prevents deletion of system properties
1154                  * on driver.conf nodes.
1155                  *
1156                  * When driver_conf_allow_path_alias is set, property behavior
1157                  * on rebound driver.conf file is non-intuitive. For a
1158                  * driver.conf node, the unit-address properties come from
1159                  * the driver.conf file as system properties. Removing system
1160                  * properties from a driver.conf node makes the node
1161                  * useless (we get node without unit-address properties) - so
1162                  * we leave system properties in place. The result is a node
1163                  * where system properties come from the node being rebound,
1164                  * and global properties come from the driver.conf file
1165                  * of the driver we are rebinding to.  If we could determine
1166                  * that the path-oriented alias driver.conf file defined a
1167                  * node at the same unit address, it would be best to use
1168                  * that node and avoid the non-intuitive property behavior.
1169                  * Unfortunately, the current "merge" code does not support
1170                  * this, so we live with the non-intuitive property behavior.
1171                  */
1172                 if (!((ndi_dev_is_persistent_node(dip) == 0) &&
1173                     (DEVI(dip)->devi_flags & DEVI_REBIND)))
1174                         e_ddi_prop_remove_all(dip);
1175         } else {
1176                 NDI_CONFIG_DEBUG((CE_CONT, "uninit_node failed: 0x%p(%s%d)\n",
1177                     (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1178         }
1179 
1180         if (addr)
1181                 kmem_free(addr, strlen(addr) + 1);
1182         return (error);
1183 }
1184 
1185 /*
1186  * Invoke driver's probe entry point to probe for existence of hardware.
1187  * Keep instance permanent for successful probe and leaf nodes.
1188  *
1189  * Per-driver list must be held busy while calling this function.
1190  */
1191 static int
1192 probe_node(dev_info_t *dip)
1193 {
1194         int rv;
1195 
1196         ASSERT(i_ddi_node_state(dip) == DS_INITIALIZED);
1197 
1198         NDI_CONFIG_DEBUG((CE_CONT, "probe_node: 0x%p(%s%d)\n",
1199             (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1200 
1201         /* temporarily hold the driver while we probe */
1202         DEVI(dip)->devi_ops = ndi_hold_driver(dip);
1203         if (DEVI(dip)->devi_ops == NULL) {
1204                 NDI_CONFIG_DEBUG((CE_CONT,
1205                     "probe_node: 0x%p(%s%d) cannot load driver\n",
1206                     (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1207                 return (DDI_FAILURE);
1208         }
1209 
1210         if (identify_9e != 0)
1211                 (void) devi_identify(dip);
1212 
1213         rv = devi_probe(dip);
1214 
1215         /* release the driver now that probe is complete */
1216         ndi_rele_driver(dip);
1217         DEVI(dip)->devi_ops = NULL;
1218 
1219         switch (rv) {
1220         case DDI_PROBE_SUCCESS:                 /* found */
1221         case DDI_PROBE_DONTCARE:                /* ddi_dev_is_sid */
1222                 e_ddi_keep_instance(dip);       /* persist instance */
1223                 rv = DDI_SUCCESS;
1224                 break;
1225 
1226         case DDI_PROBE_PARTIAL:                 /* maybe later */
1227         case DDI_PROBE_FAILURE:                 /* not found */
1228                 NDI_CONFIG_DEBUG((CE_CONT,
1229                     "probe_node: 0x%p(%s%d) no hardware found%s\n",
1230                     (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip),
1231                     (rv == DDI_PROBE_PARTIAL) ? " yet" : ""));
1232                 rv = DDI_FAILURE;
1233                 break;
1234 
1235         default:
1236 #ifdef  DEBUG
1237                 cmn_err(CE_WARN, "probe_node: %s%d: illegal probe(9E) value",
1238                     ddi_driver_name(dip), ddi_get_instance(dip));
1239 #endif  /* DEBUG */
1240                 rv = DDI_FAILURE;
1241                 break;
1242         }
1243         return (rv);
1244 }
1245 
1246 /*
1247  * Unprobe a node. Simply reset the node state.
1248  * Per-driver list must be held busy while calling this function.
1249  */
1250 static int
1251 unprobe_node(dev_info_t *dip)
1252 {
1253         ASSERT(i_ddi_node_state(dip) == DS_PROBED);
1254 
1255         /*
1256          * Don't check for references here or else a ref-counted
1257          * dip cannot be downgraded by the framework.
1258          */
1259 
1260         NDI_CONFIG_DEBUG((CE_CONT, "unprobe_node: 0x%p(name = %s)\n",
1261             (void *)dip, ddi_node_name(dip)));
1262         return (DDI_SUCCESS);
1263 }
1264 
1265 /*
1266  * Attach devinfo node.
1267  * Per-driver list must be held busy.
1268  */
1269 static int
1270 attach_node(dev_info_t *dip)
1271 {
1272         int rv;
1273 
1274         ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
1275         ASSERT(i_ddi_node_state(dip) == DS_PROBED);
1276 
1277         NDI_CONFIG_DEBUG((CE_CONT, "attach_node: 0x%p(%s%d)\n",
1278             (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1279 
1280         /*
1281          * Tell mpxio framework that a node is about to online.
1282          */
1283         if ((rv = mdi_devi_online(dip, 0)) != NDI_SUCCESS) {
1284                 return (DDI_FAILURE);
1285         }
1286 
1287         /* no recursive attachment */
1288         ASSERT(DEVI(dip)->devi_ops == NULL);
1289 
1290         /*
1291          * Hold driver the node is bound to.
1292          */
1293         DEVI(dip)->devi_ops = ndi_hold_driver(dip);
1294         if (DEVI(dip)->devi_ops == NULL) {
1295                 /*
1296                  * We were able to load driver for probing, so we should
1297                  * not get here unless something really bad happened.
1298                  */
1299                 cmn_err(CE_WARN, "attach_node: no driver for major %d",
1300                     DEVI(dip)->devi_major);
1301                 return (DDI_FAILURE);
1302         }
1303 
1304         if (NEXUS_DRV(DEVI(dip)->devi_ops))
1305                 DEVI(dip)->devi_taskq = ddi_taskq_create(dip,
1306                     "nexus_enum_tq", 1,
1307                     TASKQ_DEFAULTPRI, 0);
1308 
1309         mutex_enter(&(DEVI(dip)->devi_lock));
1310         DEVI_SET_ATTACHING(dip);
1311         DEVI_SET_NEED_RESET(dip);
1312         mutex_exit(&(DEVI(dip)->devi_lock));
1313 
1314         rv = devi_attach(dip, DDI_ATTACH);
1315 
1316         mutex_enter(&(DEVI(dip)->devi_lock));
1317         DEVI_CLR_ATTACHING(dip);
1318 
1319         if (rv != DDI_SUCCESS) {
1320                 DEVI_CLR_NEED_RESET(dip);
1321                 mutex_exit(&DEVI(dip)->devi_lock);
1322 
1323                 /*
1324                  * Cleanup dacf reservations
1325                  */
1326                 mutex_enter(&dacf_lock);
1327                 dacf_clr_rsrvs(dip, DACF_OPID_POSTATTACH);
1328                 dacf_clr_rsrvs(dip, DACF_OPID_PREDETACH);
1329                 mutex_exit(&dacf_lock);
1330                 if (DEVI(dip)->devi_taskq)
1331                         ddi_taskq_destroy(DEVI(dip)->devi_taskq);
1332                 ddi_remove_minor_node(dip, NULL);
1333 
1334                 /* release the driver if attach failed */
1335                 ndi_rele_driver(dip);
1336                 DEVI(dip)->devi_ops = NULL;
1337                 NDI_CONFIG_DEBUG((CE_CONT, "attach_node: 0x%p(%s%d) failed\n",
1338                     (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1339                 return (DDI_FAILURE);
1340         } else
1341                 mutex_exit(&DEVI(dip)->devi_lock);
1342 
1343         /* successful attach, return with driver held */
1344 
1345         return (DDI_SUCCESS);
1346 }
1347 
1348 /*
1349  * Detach devinfo node.
1350  * Per-driver list must be held busy.
1351  */
1352 static int
1353 detach_node(dev_info_t *dip, uint_t flag)
1354 {
1355         struct devnames *dnp;
1356         int             rv;
1357 
1358         ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
1359         ASSERT(i_ddi_node_state(dip) == DS_ATTACHED);
1360 
1361         /* check references */
1362         if (DEVI(dip)->devi_ref)
1363                 return (DDI_FAILURE);
1364 
1365         NDI_CONFIG_DEBUG((CE_CONT, "detach_node: 0x%p(%s%d)\n",
1366             (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1367 
1368         /*
1369          * NOTE: If we are processing a pHCI node then the calling code
1370          * must detect this and ndi_devi_enter() in (vHCI, parent(pHCI))
1371          * order unless pHCI and vHCI are siblings.  Code paths leading
1372          * here that must ensure this ordering include:
1373          * unconfig_immediate_children(), devi_unconfig_one(),
1374          * ndi_devi_unconfig_one(), ndi_devi_offline().
1375          */
1376         ASSERT(!MDI_PHCI(dip) ||
1377             (ddi_get_parent(mdi_devi_get_vdip(dip)) == ddi_get_parent(dip)) ||
1378             DEVI_BUSY_OWNED(mdi_devi_get_vdip(dip)));
1379 
1380         /* Offline the device node with the mpxio framework. */
1381         if (mdi_devi_offline(dip, flag) != NDI_SUCCESS) {
1382                 return (DDI_FAILURE);
1383         }
1384 
1385         /* drain the taskq */
1386         if (DEVI(dip)->devi_taskq)
1387                 ddi_taskq_wait(DEVI(dip)->devi_taskq);
1388 
1389         rv = devi_detach(dip, DDI_DETACH);
1390 
1391         if (rv != DDI_SUCCESS) {
1392                 NDI_CONFIG_DEBUG((CE_CONT,
1393                     "detach_node: 0x%p(%s%d) failed\n",
1394                     (void *)dip, ddi_driver_name(dip), ddi_get_instance(dip)));
1395                 return (DDI_FAILURE);
1396         }
1397 
1398         mutex_enter(&(DEVI(dip)->devi_lock));
1399         DEVI_CLR_NEED_RESET(dip);
1400         mutex_exit(&(DEVI(dip)->devi_lock));
1401 
1402 #if defined(__amd64) && !defined(__xpv)
1403         /*
1404          * Close any iommulib mediated linkage to an IOMMU
1405          */
1406         if (IOMMU_USED(dip))
1407                 iommulib_nex_close(dip);
1408 #endif
1409 
1410         /* destroy the taskq */
1411         if (DEVI(dip)->devi_taskq) {
1412                 ddi_taskq_destroy(DEVI(dip)->devi_taskq);
1413                 DEVI(dip)->devi_taskq = NULL;
1414         }
1415 
1416         /* Cleanup dacf reservations */
1417         mutex_enter(&dacf_lock);
1418         dacf_clr_rsrvs(dip, DACF_OPID_POSTATTACH);
1419         dacf_clr_rsrvs(dip, DACF_OPID_PREDETACH);
1420         mutex_exit(&dacf_lock);
1421 
1422         /* remove any additional flavors that were added */
1423         if (DEVI(dip)->devi_flavorv_n > 1 && DEVI(dip)->devi_flavorv != NULL) {
1424                 kmem_free(DEVI(dip)->devi_flavorv,
1425                     (DEVI(dip)->devi_flavorv_n - 1) * sizeof (void *));
1426                 DEVI(dip)->devi_flavorv = NULL;
1427         }
1428 
1429         /* Remove properties and minor nodes in case driver forgots */
1430         ddi_remove_minor_node(dip, NULL);
1431         ddi_prop_remove_all(dip);
1432 
1433         /* a detached node can't have attached or .conf children */
1434         mutex_enter(&DEVI(dip)->devi_lock);
1435         DEVI(dip)->devi_flags &= ~(DEVI_MADE_CHILDREN|DEVI_ATTACHED_CHILDREN);
1436         mutex_exit(&DEVI(dip)->devi_lock);
1437 
1438         /*
1439          * If the instance has successfully detached in detach_driver() context,
1440          * clear DN_DRIVER_HELD for correct ddi_hold_installed_driver()
1441          * behavior. Consumers like qassociate() depend on this (via clnopen()).
1442          */
1443         if (flag & NDI_DETACH_DRIVER) {
1444                 dnp = &(devnamesp[DEVI(dip)->devi_major]);
1445                 LOCK_DEV_OPS(&dnp->dn_lock);
1446                 dnp->dn_flags &= ~DN_DRIVER_HELD;
1447                 UNLOCK_DEV_OPS(&dnp->dn_lock);
1448         }
1449 
1450         /* successful detach, release the driver */
1451         ndi_rele_driver(dip);
1452         DEVI(dip)->devi_ops = NULL;
1453         return (DDI_SUCCESS);
1454 }
1455 
1456 /*
1457  * Run dacf post_attach routines
1458  */
1459 static int
1460 postattach_node(dev_info_t *dip)
1461 {
1462         int rval;
1463 
1464         /*
1465          * For hotplug busses like USB, it's possible that devices
1466          * are removed but dip is still around. We don't want to
1467          * run dacf routines as part of detach failure recovery.
1468          *
1469          * Pretend success until we figure out how to prevent
1470          * access to such devinfo nodes.
1471          */
1472         if (DEVI_IS_DEVICE_REMOVED(dip))
1473                 return (DDI_SUCCESS);
1474 
1475         /*
1476          * if dacf_postattach failed, report it to the framework
1477          * so that it can be retried later at the open time.
1478          */
1479         mutex_enter(&dacf_lock);
1480         rval = dacfc_postattach(dip);
1481         mutex_exit(&dacf_lock);
1482 
1483         /*
1484          * Plumbing during postattach may fail because of the
1485          * underlying device is not ready. This will fail ndi_devi_config()
1486          * in dv_filldir() and a warning message is issued. The message
1487          * from here will explain what happened
1488          */
1489         if (rval != DACF_SUCCESS) {
1490                 cmn_err(CE_WARN, "Postattach failed for %s%d\n",
1491                     ddi_driver_name(dip), ddi_get_instance(dip));
1492                 return (DDI_FAILURE);
1493         }
1494 
1495         return (DDI_SUCCESS);
1496 }
1497 
1498 /*
1499  * Run dacf pre-detach routines
1500  */
1501 static int
1502 predetach_node(dev_info_t *dip, uint_t flag)
1503 {
1504         int ret;
1505 
1506         /*
1507          * Don't auto-detach if DDI_FORCEATTACH or DDI_NO_AUTODETACH
1508          * properties are set.
1509          */
1510         if (flag & NDI_AUTODETACH) {
1511                 struct devnames *dnp;
1512                 int pflag = DDI_PROP_NOTPROM | DDI_PROP_DONTPASS;
1513 
1514                 if ((ddi_prop_get_int(DDI_DEV_T_ANY, dip,
1515                     pflag, DDI_FORCEATTACH, 0) == 1) ||
1516                     (ddi_prop_get_int(DDI_DEV_T_ANY, dip,
1517                     pflag, DDI_NO_AUTODETACH, 0) == 1))
1518                         return (DDI_FAILURE);
1519 
1520                 /* check for driver global version of DDI_NO_AUTODETACH */
1521                 dnp = &devnamesp[DEVI(dip)->devi_major];
1522                 LOCK_DEV_OPS(&dnp->dn_lock);
1523                 if (dnp->dn_flags & DN_NO_AUTODETACH) {
1524                         UNLOCK_DEV_OPS(&dnp->dn_lock);
1525                         return (DDI_FAILURE);
1526                 }
1527                 UNLOCK_DEV_OPS(&dnp->dn_lock);
1528         }
1529 
1530         mutex_enter(&dacf_lock);
1531         ret = dacfc_predetach(dip);
1532         mutex_exit(&dacf_lock);
1533 
1534         return (ret);
1535 }
1536 
1537 /*
1538  * Wrapper for making multiple state transitions
1539  */
1540 
1541 /*
1542  * i_ndi_config_node: upgrade dev_info node into a specified state.
1543  * It is a bit tricky because the locking protocol changes before and
1544  * after a node is bound to a driver. All locks are held external to
1545  * this function.
1546  */
1547 int
1548 i_ndi_config_node(dev_info_t *dip, ddi_node_state_t state, uint_t flag)
1549 {
1550         _NOTE(ARGUNUSED(flag))
1551         int rv = DDI_SUCCESS;
1552 
1553         ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
1554 
1555         while ((i_ddi_node_state(dip) < state) && (rv == DDI_SUCCESS)) {
1556 
1557                 /* don't allow any more changes to the device tree */
1558                 if (devinfo_freeze) {
1559                         rv = DDI_FAILURE;
1560                         break;
1561                 }
1562 
1563                 switch (i_ddi_node_state(dip)) {
1564                 case DS_PROTO:
1565                         /*
1566                          * only caller can reference this node, no external
1567                          * locking needed.
1568                          */
1569                         link_node(dip);
1570                         translate_devid((dev_info_t *)dip);
1571                         i_ddi_set_node_state(dip, DS_LINKED);
1572                         break;
1573                 case DS_LINKED:
1574                         /*
1575                          * Three code path may attempt to bind a node:
1576                          * - boot code
1577                          * - add_drv
1578                          * - hotplug thread
1579                          * Boot code is single threaded, add_drv synchronize
1580                          * on a userland lock, and hotplug synchronize on
1581                          * hotplug_lk. There could be a race between add_drv
1582                          * and hotplug thread. We'll live with this until the
1583                          * conversion to top-down loading.
1584                          */
1585                         if ((rv = bind_node(dip)) == DDI_SUCCESS)
1586                                 i_ddi_set_node_state(dip, DS_BOUND);
1587 
1588                         break;
1589                 case DS_BOUND:
1590                         /*
1591                          * The following transitions synchronizes on the
1592                          * per-driver busy changing flag, since we already
1593                          * have a driver.
1594                          */
1595                         if ((rv = init_node(dip)) == DDI_SUCCESS)
1596                                 i_ddi_set_node_state(dip, DS_INITIALIZED);
1597                         break;
1598                 case DS_INITIALIZED:
1599                         if ((rv = probe_node(dip)) == DDI_SUCCESS)
1600                                 i_ddi_set_node_state(dip, DS_PROBED);
1601                         break;
1602                 case DS_PROBED:
1603                         /*
1604                          * If node is retired and persistent, then prevent
1605                          * attach. We can't do this for non-persistent nodes
1606                          * as we would lose evidence that the node existed.
1607                          */
1608                         if (i_ddi_check_retire(dip) == 1 &&
1609                             ndi_dev_is_persistent_node(dip) &&
1610                             retire_prevents_attach == 1) {
1611                                 rv = DDI_FAILURE;
1612                                 break;
1613                         }
1614                         atomic_add_long(&devinfo_attach_detach, 1);
1615                         if ((rv = attach_node(dip)) == DDI_SUCCESS)
1616                                 i_ddi_set_node_state(dip, DS_ATTACHED);
1617                         atomic_add_long(&devinfo_attach_detach, -1);
1618                         break;
1619                 case DS_ATTACHED:
1620                         if ((rv = postattach_node(dip)) == DDI_SUCCESS)
1621                                 i_ddi_set_node_state(dip, DS_READY);
1622                         break;
1623                 case DS_READY:
1624                         break;
1625                 default:
1626                         /* should never reach here */
1627                         ASSERT("unknown devinfo state");
1628                 }
1629         }
1630 
1631         if (ddidebug & DDI_AUDIT)
1632                 da_log_enter(dip);
1633         return (rv);
1634 }
1635 
1636 /*
1637  * i_ndi_unconfig_node: downgrade dev_info node into a specified state.
1638  */
1639 int
1640 i_ndi_unconfig_node(dev_info_t *dip, ddi_node_state_t state, uint_t flag)
1641 {
1642         int     rv = DDI_SUCCESS;
1643 
1644         ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
1645 
1646         while ((i_ddi_node_state(dip) > state) && (rv == DDI_SUCCESS)) {
1647 
1648                 /* don't allow any more changes to the device tree */
1649                 if (devinfo_freeze) {
1650                         rv = DDI_FAILURE;
1651                         break;
1652                 }
1653 
1654                 switch (i_ddi_node_state(dip)) {
1655                 case DS_PROTO:
1656                         break;
1657                 case DS_LINKED:
1658                         /*
1659                          * Persistent nodes are only removed by hotplug code
1660                          * .conf nodes synchronizes on per-driver list.
1661                          */
1662                         if ((rv = unlink_node(dip)) == DDI_SUCCESS)
1663                                 i_ddi_set_node_state(dip, DS_PROTO);
1664                         break;
1665                 case DS_BOUND:
1666                         /*
1667                          * The following transitions synchronizes on the
1668                          * per-driver busy changing flag, since we already
1669                          * have a driver.
1670                          */
1671                         if ((rv = unbind_node(dip)) == DDI_SUCCESS)
1672                                 i_ddi_set_node_state(dip, DS_LINKED);
1673                         break;
1674                 case DS_INITIALIZED:
1675                         if ((rv = uninit_node(dip)) == DDI_SUCCESS)
1676                                 i_ddi_set_node_state(dip, DS_BOUND);
1677                         break;
1678                 case DS_PROBED:
1679                         if ((rv = unprobe_node(dip)) == DDI_SUCCESS)
1680                                 i_ddi_set_node_state(dip, DS_INITIALIZED);
1681                         break;
1682                 case DS_ATTACHED:
1683                         atomic_add_long(&devinfo_attach_detach, 1);
1684 
1685                         mutex_enter(&(DEVI(dip)->devi_lock));
1686                         DEVI_SET_DETACHING(dip);
1687                         mutex_exit(&(DEVI(dip)->devi_lock));
1688 
1689                         membar_enter(); /* ensure visibility for hold_devi */
1690 
1691                         if ((rv = detach_node(dip, flag)) == DDI_SUCCESS)
1692                                 i_ddi_set_node_state(dip, DS_PROBED);
1693 
1694                         mutex_enter(&(DEVI(dip)->devi_lock));
1695                         DEVI_CLR_DETACHING(dip);
1696                         mutex_exit(&(DEVI(dip)->devi_lock));
1697 
1698                         atomic_add_long(&devinfo_attach_detach, -1);
1699                         break;
1700                 case DS_READY:
1701                         if ((rv = predetach_node(dip, flag)) == DDI_SUCCESS)
1702                                 i_ddi_set_node_state(dip, DS_ATTACHED);
1703                         break;
1704                 default:
1705                         ASSERT("unknown devinfo state");
1706                 }
1707         }
1708         da_log_enter(dip);
1709         return (rv);
1710 }
1711 
1712 /*
1713  * ddi_initchild: transform node to DS_INITIALIZED state
1714  */
1715 int
1716 ddi_initchild(dev_info_t *parent, dev_info_t *proto)
1717 {
1718         int ret, circ;
1719 
1720         ndi_devi_enter(parent, &circ);
1721         ret = i_ndi_config_node(proto, DS_INITIALIZED, 0);
1722         ndi_devi_exit(parent, circ);
1723 
1724         return (ret);
1725 }
1726 
1727 /*
1728  * ddi_uninitchild: transform node down to DS_BOUND state
1729  */
1730 int
1731 ddi_uninitchild(dev_info_t *dip)
1732 {
1733         int ret, circ;
1734         dev_info_t *parent = ddi_get_parent(dip);
1735         ASSERT(parent);
1736 
1737         ndi_devi_enter(parent, &circ);
1738         ret = i_ndi_unconfig_node(dip, DS_BOUND, 0);
1739         ndi_devi_exit(parent, circ);
1740 
1741         return (ret);
1742 }
1743 
1744 /*
1745  * i_ddi_attachchild: transform node to DS_READY/i_ddi_devi_attached() state
1746  */
1747 static int
1748 i_ddi_attachchild(dev_info_t *dip)
1749 {
1750         dev_info_t      *parent = ddi_get_parent(dip);
1751         int             ret;
1752 
1753         ASSERT(parent && DEVI_BUSY_OWNED(parent));
1754 
1755         if ((i_ddi_node_state(dip) < DS_BOUND) || DEVI_IS_DEVICE_OFFLINE(dip))
1756                 return (DDI_FAILURE);
1757 
1758         ret = i_ndi_config_node(dip, DS_READY, 0);
1759         if (ret == NDI_SUCCESS) {
1760                 ret = DDI_SUCCESS;
1761         } else {
1762                 /*
1763                  * Take it down to DS_INITIALIZED so pm_pre_probe is run
1764                  * on the next attach
1765                  */
1766                 (void) i_ndi_unconfig_node(dip, DS_INITIALIZED, 0);
1767                 ret = DDI_FAILURE;
1768         }
1769 
1770         return (ret);
1771 }
1772 
1773 /*
1774  * i_ddi_detachchild: transform node down to DS_PROBED state
1775  *      If it fails, put it back to DS_READY state.
1776  * NOTE: A node that fails detach may be at DS_ATTACHED instead
1777  * of DS_READY for a small amount of time - this is the source of
1778  * transient DS_READY->DS_ATTACHED->DS_READY state changes.
1779  */
1780 static int
1781 i_ddi_detachchild(dev_info_t *dip, uint_t flags)
1782 {
1783         dev_info_t      *parent = ddi_get_parent(dip);
1784         int             ret;
1785 
1786         ASSERT(parent && DEVI_BUSY_OWNED(parent));
1787 
1788         ret = i_ndi_unconfig_node(dip, DS_PROBED, flags);
1789         if (ret != DDI_SUCCESS)
1790                 (void) i_ndi_config_node(dip, DS_READY, 0);
1791         else
1792                 /* allow pm_pre_probe to reestablish pm state */
1793                 (void) i_ndi_unconfig_node(dip, DS_INITIALIZED, 0);
1794         return (ret);
1795 }
1796 
1797 /*
1798  * Add a child and bind to driver
1799  */
1800 dev_info_t *
1801 ddi_add_child(dev_info_t *pdip, char *name, uint_t nodeid, uint_t unit)
1802 {
1803         int circ;
1804         dev_info_t *dip;
1805 
1806         /* allocate a new node */
1807         dip = i_ddi_alloc_node(pdip, name, nodeid, (int)unit, NULL, KM_SLEEP);
1808 
1809         ndi_devi_enter(pdip, &circ);
1810         (void) i_ndi_config_node(dip, DS_BOUND, 0);
1811         ndi_devi_exit(pdip, circ);
1812         return (dip);
1813 }
1814 
1815 /*
1816  * ddi_remove_child: remove the dip. The parent must be attached and held
1817  */
1818 int
1819 ddi_remove_child(dev_info_t *dip, int dummy)
1820 {
1821         _NOTE(ARGUNUSED(dummy))
1822         int circ, ret;
1823         dev_info_t *parent = ddi_get_parent(dip);
1824         ASSERT(parent);
1825 
1826         ndi_devi_enter(parent, &circ);
1827 
1828         /*
1829          * If we still have children, for example SID nodes marked
1830          * as persistent but not attached, attempt to remove them.
1831          */
1832         if (DEVI(dip)->devi_child) {
1833                 ret = ndi_devi_unconfig(dip, NDI_DEVI_REMOVE);
1834                 if (ret != NDI_SUCCESS) {
1835                         ndi_devi_exit(parent, circ);
1836                         return (DDI_FAILURE);
1837                 }
1838                 ASSERT(DEVI(dip)->devi_child == NULL);
1839         }
1840 
1841         ret = i_ndi_unconfig_node(dip, DS_PROTO, 0);
1842         ndi_devi_exit(parent, circ);
1843 
1844         if (ret != DDI_SUCCESS)
1845                 return (ret);
1846 
1847         ASSERT(i_ddi_node_state(dip) == DS_PROTO);
1848         i_ddi_free_node(dip);
1849         return (DDI_SUCCESS);
1850 }
1851 
1852 /*
1853  * NDI wrappers for ref counting, node allocation, and transitions
1854  */
1855 
1856 /*
1857  * Hold/release the devinfo node itself.
1858  * Caller is assumed to prevent the devi from detaching during this call
1859  */
1860 void
1861 ndi_hold_devi(dev_info_t *dip)
1862 {
1863         mutex_enter(&DEVI(dip)->devi_lock);
1864         ASSERT(DEVI(dip)->devi_ref >= 0);
1865         DEVI(dip)->devi_ref++;
1866         membar_enter();                 /* make sure stores are flushed */
1867         mutex_exit(&DEVI(dip)->devi_lock);
1868 }
1869 
1870 void
1871 ndi_rele_devi(dev_info_t *dip)
1872 {
1873         ASSERT(DEVI(dip)->devi_ref > 0);
1874 
1875         mutex_enter(&DEVI(dip)->devi_lock);
1876         DEVI(dip)->devi_ref--;
1877         membar_enter();                 /* make sure stores are flushed */
1878         mutex_exit(&DEVI(dip)->devi_lock);
1879 }
1880 
1881 int
1882 e_ddi_devi_holdcnt(dev_info_t *dip)
1883 {
1884         return (DEVI(dip)->devi_ref);
1885 }
1886 
1887 /*
1888  * Hold/release the driver the devinfo node is bound to.
1889  */
1890 struct dev_ops *
1891 ndi_hold_driver(dev_info_t *dip)
1892 {
1893         if (i_ddi_node_state(dip) < DS_BOUND)
1894                 return (NULL);
1895 
1896         ASSERT(DEVI(dip)->devi_major != -1);
1897         return (mod_hold_dev_by_major(DEVI(dip)->devi_major));
1898 }
1899 
1900 void
1901 ndi_rele_driver(dev_info_t *dip)
1902 {
1903         ASSERT(i_ddi_node_state(dip) >= DS_BOUND);
1904         mod_rele_dev_by_major(DEVI(dip)->devi_major);
1905 }
1906 
1907 /*
1908  * Single thread entry into devinfo node for modifying its children (devinfo,
1909  * pathinfo, and minor). To verify in ASSERTS use DEVI_BUSY_OWNED macro.
1910  */
1911 void
1912 ndi_devi_enter(dev_info_t *dip, int *circular)
1913 {
1914         struct dev_info *devi = DEVI(dip);
1915         ASSERT(dip != NULL);
1916 
1917         /* for vHCI, enforce (vHCI, pHCI) ndi_deve_enter() order */
1918         ASSERT(!MDI_VHCI(dip) || (mdi_devi_pdip_entered(dip) == 0) ||
1919             DEVI_BUSY_OWNED(dip));
1920 
1921         mutex_enter(&devi->devi_lock);
1922         if (devi->devi_busy_thread == curthread) {
1923                 devi->devi_circular++;
1924         } else {
1925                 while (DEVI_BUSY_CHANGING(devi) && !panicstr)
1926                         cv_wait(&(devi->devi_cv), &(devi->devi_lock));
1927                 if (panicstr) {
1928                         mutex_exit(&devi->devi_lock);
1929                         return;
1930                 }
1931                 devi->devi_flags |= DEVI_BUSY;
1932                 devi->devi_busy_thread = curthread;
1933         }
1934         *circular = devi->devi_circular;
1935         mutex_exit(&devi->devi_lock);
1936 }
1937 
1938 /*
1939  * Release ndi_devi_enter or successful ndi_devi_tryenter.
1940  */
1941 void
1942 ndi_devi_exit(dev_info_t *dip, int circular)
1943 {
1944         struct dev_info *devi = DEVI(dip);
1945         struct dev_info *vdevi;
1946         ASSERT(dip != NULL);
1947 
1948         if (panicstr)
1949                 return;
1950 
1951         mutex_enter(&(devi->devi_lock));
1952         if (circular != 0) {
1953                 devi->devi_circular--;
1954         } else {
1955                 devi->devi_flags &= ~DEVI_BUSY;
1956                 ASSERT(devi->devi_busy_thread == curthread);
1957                 devi->devi_busy_thread = NULL;
1958                 cv_broadcast(&(devi->devi_cv));
1959         }
1960         mutex_exit(&(devi->devi_lock));
1961 
1962         /*
1963          * For pHCI exit we issue a broadcast to vHCI for ndi_devi_config_one()
1964          * doing cv_wait on vHCI.
1965          */
1966         if (MDI_PHCI(dip)) {
1967                 vdevi = DEVI(mdi_devi_get_vdip(dip));
1968                 if (vdevi) {
1969                         mutex_enter(&(vdevi->devi_lock));
1970                         if (vdevi->devi_flags & DEVI_PHCI_SIGNALS_VHCI) {
1971                                 vdevi->devi_flags &= ~DEVI_PHCI_SIGNALS_VHCI;
1972                                 cv_broadcast(&(vdevi->devi_cv));
1973                         }
1974                         mutex_exit(&(vdevi->devi_lock));
1975                 }
1976         }
1977 }
1978 
1979 /*
1980  * Release ndi_devi_enter and wait for possibility of new children, avoiding
1981  * possibility of missing broadcast before getting to cv_timedwait().
1982  */
1983 static void
1984 ndi_devi_exit_and_wait(dev_info_t *dip, int circular, clock_t end_time)
1985 {
1986         struct dev_info *devi = DEVI(dip);
1987         ASSERT(dip != NULL);
1988 
1989         if (panicstr)
1990                 return;
1991 
1992         /*
1993          * We are called to wait for of a new child, and new child can
1994          * only be added if circular is zero.
1995          */
1996         ASSERT(circular == 0);
1997 
1998         /* like ndi_devi_exit with circular of zero */
1999         mutex_enter(&(devi->devi_lock));
2000         devi->devi_flags &= ~DEVI_BUSY;
2001         ASSERT(devi->devi_busy_thread == curthread);
2002         devi->devi_busy_thread = NULL;
2003         cv_broadcast(&(devi->devi_cv));
2004 
2005         /* now wait for new children while still holding devi_lock */
2006         (void) cv_timedwait(&devi->devi_cv, &(devi->devi_lock), end_time);
2007         mutex_exit(&(devi->devi_lock));
2008 }
2009 
2010 /*
2011  * Attempt to single thread entry into devinfo node for modifying its children.
2012  */
2013 int
2014 ndi_devi_tryenter(dev_info_t *dip, int *circular)
2015 {
2016         int rval = 1;              /* assume we enter */
2017         struct dev_info *devi = DEVI(dip);
2018         ASSERT(dip != NULL);
2019 
2020         mutex_enter(&devi->devi_lock);
2021         if (devi->devi_busy_thread == (void *)curthread) {
2022                 devi->devi_circular++;
2023         } else {
2024                 if (!DEVI_BUSY_CHANGING(devi)) {
2025                         devi->devi_flags |= DEVI_BUSY;
2026                         devi->devi_busy_thread = (void *)curthread;
2027                 } else {
2028                         rval = 0;       /* devi is busy */
2029                 }
2030         }
2031         *circular = devi->devi_circular;
2032         mutex_exit(&devi->devi_lock);
2033         return (rval);
2034 }
2035 
2036 /*
2037  * Allocate and initialize a new dev_info structure.
2038  *
2039  * This routine may be called at interrupt time by a nexus in
2040  * response to a hotplug event, therefore memory allocations are
2041  * not allowed to sleep.
2042  */
2043 int
2044 ndi_devi_alloc(dev_info_t *parent, char *node_name, pnode_t nodeid,
2045     dev_info_t **ret_dip)
2046 {
2047         ASSERT(node_name != NULL);
2048         ASSERT(ret_dip != NULL);
2049 
2050         *ret_dip = i_ddi_alloc_node(parent, node_name, nodeid, -1, NULL,
2051             KM_NOSLEEP);
2052         if (*ret_dip == NULL) {
2053                 return (NDI_NOMEM);
2054         }
2055 
2056         return (NDI_SUCCESS);
2057 }
2058 
2059 /*
2060  * Allocate and initialize a new dev_info structure
2061  * This routine may sleep and should not be called at interrupt time
2062  */
2063 void
2064 ndi_devi_alloc_sleep(dev_info_t *parent, char *node_name, pnode_t nodeid,
2065     dev_info_t **ret_dip)
2066 {
2067         ASSERT(node_name != NULL);
2068         ASSERT(ret_dip != NULL);
2069 
2070         *ret_dip = i_ddi_alloc_node(parent, node_name, nodeid, -1, NULL,
2071             KM_SLEEP);
2072         ASSERT(*ret_dip);
2073 }
2074 
2075 /*
2076  * Remove an initialized (but not yet attached) dev_info
2077  * node from it's parent.
2078  */
2079 int
2080 ndi_devi_free(dev_info_t *dip)
2081 {
2082         ASSERT(dip != NULL);
2083 
2084         if (i_ddi_node_state(dip) >= DS_INITIALIZED)
2085                 return (DDI_FAILURE);
2086 
2087         NDI_CONFIG_DEBUG((CE_CONT, "ndi_devi_free: %s%d (%p)\n",
2088             ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip));
2089 
2090         (void) ddi_remove_child(dip, 0);
2091 
2092         return (NDI_SUCCESS);
2093 }
2094 
2095 /*
2096  * ndi_devi_bind_driver() binds a driver to a given device. If it fails
2097  * to bind the driver, it returns an appropriate error back. Some drivers
2098  * may want to know if the actually failed to bind.
2099  */
2100 int
2101 ndi_devi_bind_driver(dev_info_t *dip, uint_t flags)
2102 {
2103         int ret = NDI_FAILURE;
2104         int circ;
2105         dev_info_t *pdip = ddi_get_parent(dip);
2106         ASSERT(pdip);
2107 
2108         NDI_CONFIG_DEBUG((CE_CONT,
2109             "ndi_devi_bind_driver: %s%d (%p) flags: %x\n",
2110             ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
2111 
2112         ndi_devi_enter(pdip, &circ);
2113         if (i_ndi_config_node(dip, DS_BOUND, flags) == DDI_SUCCESS)
2114                 ret = NDI_SUCCESS;
2115         ndi_devi_exit(pdip, circ);
2116 
2117         return (ret);
2118 }
2119 
2120 /*
2121  * ndi_devi_unbind_driver: unbind the dip
2122  */
2123 static int
2124 ndi_devi_unbind_driver(dev_info_t *dip)
2125 {
2126         ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
2127 
2128         return (i_ndi_unconfig_node(dip, DS_LINKED, 0));
2129 }
2130 
2131 /*
2132  * Misc. help routines called by framework only
2133  */
2134 
2135 /*
2136  * Get the state of node
2137  */
2138 ddi_node_state_t
2139 i_ddi_node_state(dev_info_t *dip)
2140 {
2141         return (DEVI(dip)->devi_node_state);
2142 }
2143 
2144 /*
2145  * Set the state of node
2146  */
2147 void
2148 i_ddi_set_node_state(dev_info_t *dip, ddi_node_state_t state)
2149 {
2150         DEVI(dip)->devi_node_state = state;
2151         membar_enter();                 /* make sure stores are flushed */
2152 }
2153 
2154 /*
2155  * Determine if node is attached. The implementation accommodates transient
2156  * DS_READY->DS_ATTACHED->DS_READY state changes.  Outside this file, this
2157  * function should be instead of i_ddi_node_state() DS_ATTACHED/DS_READY
2158  * state checks.
2159  */
2160 int
2161 i_ddi_devi_attached(dev_info_t *dip)
2162 {
2163         return (DEVI(dip)->devi_node_state >= DS_ATTACHED);
2164 }
2165 
2166 /*
2167  * Common function for finding a node in a sibling list given name and addr.
2168  *
2169  * By default, name is matched with devi_node_name. The following
2170  * alternative match strategies are supported:
2171  *
2172  *      FIND_NODE_BY_NODENAME: Match on node name - typical use.
2173  *
2174  *      FIND_NODE_BY_DRIVER: A match on driver name bound to node is conducted.
2175  *              This support is used for support of OBP generic names and
2176  *              for the conversion from driver names to generic names. When
2177  *              more consistency in the generic name environment is achieved
2178  *              (and not needed for upgrade) this support can be removed.
2179  *
2180  *      FIND_NODE_BY_ADDR: Match on just the addr.
2181  *              This support is only used/needed during boot to match
2182  *              a node bound via a path-based driver alias.
2183  *
2184  * If a child is not named (dev_addr == NULL), there are three
2185  * possible actions:
2186  *
2187  *      (1) skip it
2188  *      (2) FIND_ADDR_BY_INIT: bring child to DS_INITIALIZED state
2189  *      (3) FIND_ADDR_BY_CALLBACK: use a caller-supplied callback function
2190  */
2191 #define FIND_NODE_BY_NODENAME   0x01
2192 #define FIND_NODE_BY_DRIVER     0x02
2193 #define FIND_NODE_BY_ADDR       0x04
2194 #define FIND_ADDR_BY_INIT       0x10
2195 #define FIND_ADDR_BY_CALLBACK   0x20
2196 
2197 static dev_info_t *
2198 find_sibling(dev_info_t *head, char *cname, char *caddr, uint_t flag,
2199     int (*callback)(dev_info_t *, char *, int))
2200 {
2201         dev_info_t      *dip;
2202         char            *addr, *buf;
2203         major_t         major;
2204         uint_t          by;
2205 
2206         /* only one way to find a node */
2207         by = flag &
2208             (FIND_NODE_BY_DRIVER | FIND_NODE_BY_NODENAME | FIND_NODE_BY_ADDR);
2209         ASSERT(by && BIT_ONLYONESET(by));
2210 
2211         /* only one way to name a node */
2212         ASSERT(((flag & FIND_ADDR_BY_INIT) == 0) ||
2213             ((flag & FIND_ADDR_BY_CALLBACK) == 0));
2214 
2215         if (by == FIND_NODE_BY_DRIVER) {
2216                 major = ddi_name_to_major(cname);
2217                 if (major == DDI_MAJOR_T_NONE)
2218                         return (NULL);
2219         }
2220 
2221         /* preallocate buffer of naming node by callback */
2222         if (flag & FIND_ADDR_BY_CALLBACK)
2223                 buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2224 
2225         /*
2226          * Walk the child list to find a match
2227          */
2228         if (head == NULL)
2229                 return (NULL);
2230         ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(head)));
2231         for (dip = head; dip; dip = ddi_get_next_sibling(dip)) {
2232                 if (by == FIND_NODE_BY_NODENAME) {
2233                         /* match node name */
2234                         if (strcmp(cname, DEVI(dip)->devi_node_name) != 0)
2235                                 continue;
2236                 } else if (by == FIND_NODE_BY_DRIVER) {
2237                         /* match driver major */
2238                         if (DEVI(dip)->devi_major != major)
2239                                 continue;
2240                 }
2241 
2242                 if ((addr = DEVI(dip)->devi_addr) == NULL) {
2243                         /* name the child based on the flag */
2244                         if (flag & FIND_ADDR_BY_INIT) {
2245                                 if (ddi_initchild(ddi_get_parent(dip), dip)
2246                                     != DDI_SUCCESS)
2247                                         continue;
2248                                 addr = DEVI(dip)->devi_addr;
2249                         } else if (flag & FIND_ADDR_BY_CALLBACK) {
2250                                 if ((callback == NULL) || (callback(
2251                                     dip, buf, MAXNAMELEN) != DDI_SUCCESS))
2252                                         continue;
2253                                 addr = buf;
2254                         } else {
2255                                 continue;       /* skip */
2256                         }
2257                 }
2258 
2259                 /* match addr */
2260                 ASSERT(addr != NULL);
2261                 if (strcmp(caddr, addr) == 0)
2262                         break;  /* node found */
2263 
2264         }
2265         if (flag & FIND_ADDR_BY_CALLBACK)
2266                 kmem_free(buf, MAXNAMELEN);
2267         return (dip);
2268 }
2269 
2270 /*
2271  * Find child of pdip with name: cname@caddr
2272  * Called by init_node() to look for duplicate nodes
2273  */
2274 static dev_info_t *
2275 find_duplicate_child(dev_info_t *pdip, dev_info_t *dip)
2276 {
2277         dev_info_t *dup;
2278         char *cname = DEVI(dip)->devi_node_name;
2279         char *caddr = DEVI(dip)->devi_addr;
2280 
2281         /* search nodes before dip */
2282         dup = find_sibling(ddi_get_child(pdip), cname, caddr,
2283             FIND_NODE_BY_NODENAME, NULL);
2284         if (dup != dip)
2285                 return (dup);
2286 
2287         /*
2288          * search nodes after dip; normally this is not needed,
2289          */
2290         return (find_sibling(ddi_get_next_sibling(dip), cname, caddr,
2291             FIND_NODE_BY_NODENAME, NULL));
2292 }
2293 
2294 /*
2295  * Find a child of a given name and address, using a callback to name
2296  * unnamed children. cname is the binding name.
2297  */
2298 dev_info_t *
2299 ndi_devi_findchild_by_callback(dev_info_t *pdip, char *dname, char *ua,
2300     int (*make_ua)(dev_info_t *, char *, int))
2301 {
2302         int     by = FIND_ADDR_BY_CALLBACK;
2303 
2304         ASSERT(DEVI_BUSY_OWNED(pdip));
2305         by |= dname ? FIND_NODE_BY_DRIVER : FIND_NODE_BY_ADDR;
2306         return (find_sibling(ddi_get_child(pdip), dname, ua, by, make_ua));
2307 }
2308 
2309 /*
2310  * Find a child of a given name and address, invoking initchild to name
2311  * unnamed children. cname is the node name.
2312  */
2313 static dev_info_t *
2314 find_child_by_name(dev_info_t *pdip, char *cname, char *caddr)
2315 {
2316         dev_info_t      *dip;
2317 
2318         /* attempt search without changing state of preceding siblings */
2319         dip = find_sibling(ddi_get_child(pdip), cname, caddr,
2320             FIND_NODE_BY_NODENAME, NULL);
2321         if (dip)
2322                 return (dip);
2323 
2324         return (find_sibling(ddi_get_child(pdip), cname, caddr,
2325             FIND_NODE_BY_NODENAME|FIND_ADDR_BY_INIT, NULL));
2326 }
2327 
2328 /*
2329  * Find a child of a given name and address, invoking initchild to name
2330  * unnamed children. cname is the node name.
2331  */
2332 static dev_info_t *
2333 find_child_by_driver(dev_info_t *pdip, char *cname, char *caddr)
2334 {
2335         dev_info_t      *dip;
2336 
2337         /* attempt search without changing state of preceding siblings */
2338         dip = find_sibling(ddi_get_child(pdip), cname, caddr,
2339             FIND_NODE_BY_DRIVER, NULL);
2340         if (dip)
2341                 return (dip);
2342 
2343         return (find_sibling(ddi_get_child(pdip), cname, caddr,
2344             FIND_NODE_BY_DRIVER|FIND_ADDR_BY_INIT, NULL));
2345 }
2346 
2347 /*
2348  * Find a child of a given address, invoking initchild to name
2349  * unnamed children. cname is the node name.
2350  *
2351  * NOTE: This function is only used during boot. One would hope that
2352  * unique sibling unit-addresses on hardware branches of the tree would
2353  * be a requirement to avoid two drivers trying to control the same
2354  * piece of hardware. Unfortunately there are some cases where this
2355  * situation exists (/ssm@0,0/pci@1c,700000 /ssm@0,0/sghsc@1c,700000).
2356  * Until unit-address uniqueness of siblings is guaranteed, use of this
2357  * interface for purposes other than boot should be avoided.
2358  */
2359 static dev_info_t *
2360 find_child_by_addr(dev_info_t *pdip, char *caddr)
2361 {
2362         dev_info_t      *dip;
2363 
2364         /* return NULL if called without a unit-address */
2365         if ((caddr == NULL) || (*caddr == '\0'))
2366                 return (NULL);
2367 
2368         /* attempt search without changing state of preceding siblings */
2369         dip = find_sibling(ddi_get_child(pdip), NULL, caddr,
2370             FIND_NODE_BY_ADDR, NULL);
2371         if (dip)
2372                 return (dip);
2373 
2374         return (find_sibling(ddi_get_child(pdip), NULL, caddr,
2375             FIND_NODE_BY_ADDR|FIND_ADDR_BY_INIT, NULL));
2376 }
2377 
2378 /*
2379  * Deleting a property list. Take care, since some property structures
2380  * may not be fully built.
2381  */
2382 void
2383 i_ddi_prop_list_delete(ddi_prop_t *prop)
2384 {
2385         while (prop) {
2386                 ddi_prop_t *next = prop->prop_next;
2387                 if (prop->prop_name)
2388                         kmem_free(prop->prop_name, strlen(prop->prop_name) + 1);
2389                 if ((prop->prop_len != 0) && prop->prop_val)
2390                         kmem_free(prop->prop_val, prop->prop_len);
2391                 kmem_free(prop, sizeof (struct ddi_prop));
2392                 prop = next;
2393         }
2394 }
2395 
2396 /*
2397  * Duplicate property list
2398  */
2399 ddi_prop_t *
2400 i_ddi_prop_list_dup(ddi_prop_t *prop, uint_t flag)
2401 {
2402         ddi_prop_t *result, *prev, *copy;
2403 
2404         if (prop == NULL)
2405                 return (NULL);
2406 
2407         result = prev = NULL;
2408         for (; prop != NULL; prop = prop->prop_next) {
2409                 ASSERT(prop->prop_name != NULL);
2410                 copy = kmem_zalloc(sizeof (struct ddi_prop), flag);
2411                 if (copy == NULL)
2412                         goto fail;
2413 
2414                 copy->prop_dev = prop->prop_dev;
2415                 copy->prop_flags = prop->prop_flags;
2416                 copy->prop_name = i_ddi_strdup(prop->prop_name, flag);
2417                 if (copy->prop_name == NULL)
2418                         goto fail;
2419 
2420                 if ((copy->prop_len = prop->prop_len) != 0) {
2421                         copy->prop_val = kmem_zalloc(prop->prop_len, flag);
2422                         if (copy->prop_val == NULL)
2423                                 goto fail;
2424 
2425                         bcopy(prop->prop_val, copy->prop_val, prop->prop_len);
2426                 }
2427 
2428                 if (prev == NULL)
2429                         result = prev = copy;
2430                 else
2431                         prev->prop_next = copy;
2432                 prev = copy;
2433         }
2434         return (result);
2435 
2436 fail:
2437         i_ddi_prop_list_delete(result);
2438         return (NULL);
2439 }
2440 
2441 /*
2442  * Create a reference property list, currently used only for
2443  * driver global properties. Created with ref count of 1.
2444  */
2445 ddi_prop_list_t *
2446 i_ddi_prop_list_create(ddi_prop_t *props)
2447 {
2448         ddi_prop_list_t *list = kmem_alloc(sizeof (*list), KM_SLEEP);
2449         list->prop_list = props;
2450         list->prop_ref = 1;
2451         return (list);
2452 }
2453 
2454 /*
2455  * Increment/decrement reference count. The reference is
2456  * protected by dn_lock. The only interfaces modifying
2457  * dn_global_prop_ptr is in impl_make[free]_parlist().
2458  */
2459 void
2460 i_ddi_prop_list_hold(ddi_prop_list_t *prop_list, struct devnames *dnp)
2461 {
2462         ASSERT(prop_list->prop_ref >= 0);
2463         ASSERT(mutex_owned(&dnp->dn_lock));
2464         prop_list->prop_ref++;
2465 }
2466 
2467 void
2468 i_ddi_prop_list_rele(ddi_prop_list_t *prop_list, struct devnames *dnp)
2469 {
2470         ASSERT(prop_list->prop_ref > 0);
2471         ASSERT(mutex_owned(&dnp->dn_lock));
2472         prop_list->prop_ref--;
2473 
2474         if (prop_list->prop_ref == 0) {
2475                 i_ddi_prop_list_delete(prop_list->prop_list);
2476                 kmem_free(prop_list, sizeof (*prop_list));
2477         }
2478 }
2479 
2480 /*
2481  * Free table of classes by drivers
2482  */
2483 void
2484 i_ddi_free_exported_classes(char **classes, int n)
2485 {
2486         if ((n == 0) || (classes == NULL))
2487                 return;
2488 
2489         kmem_free(classes, n * sizeof (char *));
2490 }
2491 
2492 /*
2493  * Get all classes exported by dip
2494  */
2495 int
2496 i_ddi_get_exported_classes(dev_info_t *dip, char ***classes)
2497 {
2498         extern void lock_hw_class_list();
2499         extern void unlock_hw_class_list();
2500         extern int get_class(const char *, char **);
2501 
2502         static char *rootclass = "root";
2503         int n = 0, nclass = 0;
2504         char **buf;
2505 
2506         ASSERT(i_ddi_node_state(dip) >= DS_BOUND);
2507 
2508         if (dip == ddi_root_node())     /* rootnode exports class "root" */
2509                 nclass = 1;
2510         lock_hw_class_list();
2511         nclass += get_class(ddi_driver_name(dip), NULL);
2512         if (nclass == 0) {
2513                 unlock_hw_class_list();
2514                 return (0);             /* no class exported */
2515         }
2516 
2517         *classes = buf = kmem_alloc(nclass * sizeof (char *), KM_SLEEP);
2518         if (dip == ddi_root_node()) {
2519                 *buf++ = rootclass;
2520                 n = 1;
2521         }
2522         n += get_class(ddi_driver_name(dip), buf);
2523         unlock_hw_class_list();
2524 
2525         ASSERT(n == nclass);    /* make sure buf wasn't overrun */
2526         return (nclass);
2527 }
2528 
2529 /*
2530  * Helper functions, returns NULL if no memory.
2531  */
2532 char *
2533 i_ddi_strdup(char *str, uint_t flag)
2534 {
2535         char *copy;
2536 
2537         if (str == NULL)
2538                 return (NULL);
2539 
2540         copy = kmem_alloc(strlen(str) + 1, flag);
2541         if (copy == NULL)
2542                 return (NULL);
2543 
2544         (void) strcpy(copy, str);
2545         return (copy);
2546 }
2547 
2548 /*
2549  * Load driver.conf file for major. Load all if major == -1.
2550  *
2551  * This is called
2552  * - early in boot after devnames array is initialized
2553  * - from vfs code when certain file systems are mounted
2554  * - from add_drv when a new driver is added
2555  */
2556 int
2557 i_ddi_load_drvconf(major_t major)
2558 {
2559         extern int modrootloaded;
2560 
2561         major_t low, high, m;
2562 
2563         if (major == DDI_MAJOR_T_NONE) {
2564                 low = 0;
2565                 high = devcnt - 1;
2566         } else {
2567                 if (major >= devcnt)
2568                         return (EINVAL);
2569                 low = high = major;
2570         }
2571 
2572         for (m = low; m <= high; m++) {
2573                 struct devnames *dnp = &devnamesp[m];
2574                 LOCK_DEV_OPS(&dnp->dn_lock);
2575                 dnp->dn_flags &= ~(DN_DRIVER_HELD|DN_DRIVER_INACTIVE);
2576                 (void) impl_make_parlist(m);
2577                 UNLOCK_DEV_OPS(&dnp->dn_lock);
2578         }
2579 
2580         if (modrootloaded) {
2581                 ddi_walk_devs(ddi_root_node(), reset_nexus_flags,
2582                     (void *)(uintptr_t)major);
2583         }
2584 
2585         /* build dn_list from old entries in path_to_inst */
2586         e_ddi_unorphan_instance_nos();
2587         return (0);
2588 }
2589 
2590 /*
2591  * Unload a specific driver.conf.
2592  * Don't support unload all because it doesn't make any sense
2593  */
2594 int
2595 i_ddi_unload_drvconf(major_t major)
2596 {
2597         int error;
2598         struct devnames *dnp;
2599 
2600         if (major >= devcnt)
2601                 return (EINVAL);
2602 
2603         /*
2604          * Take the per-driver lock while unloading driver.conf
2605          */
2606         dnp = &devnamesp[major];
2607         LOCK_DEV_OPS(&dnp->dn_lock);
2608         error = impl_free_parlist(major);
2609         UNLOCK_DEV_OPS(&dnp->dn_lock);
2610         return (error);
2611 }
2612 
2613 /*
2614  * Merge a .conf node. This is called by nexus drivers to augment
2615  * hw node with properties specified in driver.conf file. This function
2616  * takes a callback routine to name nexus children.
2617  * The parent node must be held busy.
2618  *
2619  * It returns DDI_SUCCESS if the node is merged and DDI_FAILURE otherwise.
2620  */
2621 int
2622 ndi_merge_node(dev_info_t *dip, int (*make_ua)(dev_info_t *, char *, int))
2623 {
2624         dev_info_t *hwdip;
2625 
2626         ASSERT(ndi_dev_is_persistent_node(dip) == 0);
2627         ASSERT(ddi_get_name_addr(dip) != NULL);
2628 
2629         hwdip = ndi_devi_findchild_by_callback(ddi_get_parent(dip),
2630             ddi_binding_name(dip), ddi_get_name_addr(dip), make_ua);
2631 
2632         /*
2633          * Look for the hardware node that is the target of the merge;
2634          * return failure if not found.
2635          */
2636         if ((hwdip == NULL) || (hwdip == dip)) {
2637                 char *buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2638                 NDI_CONFIG_DEBUG((CE_WARN, "No HW node to merge conf node %s",
2639                     ddi_deviname(dip, buf)));
2640                 kmem_free(buf, MAXNAMELEN);
2641                 return (DDI_FAILURE);
2642         }
2643 
2644         /*
2645          * Make sure the hardware node is uninitialized and has no property.
2646          * This may not be the case if new .conf files are load after some
2647          * hardware nodes have already been initialized and attached.
2648          *
2649          * N.B. We return success here because the node was *intended*
2650          *      to be a merge node because there is a hw node with the name.
2651          */
2652         mutex_enter(&DEVI(hwdip)->devi_lock);
2653         if (ndi_dev_is_persistent_node(hwdip) == 0) {
2654                 char *buf;
2655                 mutex_exit(&DEVI(hwdip)->devi_lock);
2656 
2657                 buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2658                 NDI_CONFIG_DEBUG((CE_NOTE, "Duplicate .conf node %s",
2659                     ddi_deviname(dip, buf)));
2660                 kmem_free(buf, MAXNAMELEN);
2661                 return (DDI_SUCCESS);
2662         }
2663 
2664         /*
2665          * If it is possible that the hardware has already been touched
2666          * then don't merge.
2667          */
2668         if (i_ddi_node_state(hwdip) >= DS_INITIALIZED ||
2669             (DEVI(hwdip)->devi_sys_prop_ptr != NULL) ||
2670             (DEVI(hwdip)->devi_drv_prop_ptr != NULL)) {
2671                 char *buf;
2672                 mutex_exit(&DEVI(hwdip)->devi_lock);
2673 
2674                 buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
2675                 NDI_CONFIG_DEBUG((CE_NOTE,
2676                     "!Cannot merge .conf node %s with hw node %p "
2677                     "-- not in proper state",
2678                     ddi_deviname(dip, buf), (void *)hwdip));
2679                 kmem_free(buf, MAXNAMELEN);
2680                 return (DDI_SUCCESS);
2681         }
2682 
2683         mutex_enter(&DEVI(dip)->devi_lock);
2684         DEVI(hwdip)->devi_sys_prop_ptr = DEVI(dip)->devi_sys_prop_ptr;
2685         DEVI(hwdip)->devi_drv_prop_ptr = DEVI(dip)->devi_drv_prop_ptr;
2686         DEVI(dip)->devi_sys_prop_ptr = NULL;
2687         DEVI(dip)->devi_drv_prop_ptr = NULL;
2688         mutex_exit(&DEVI(dip)->devi_lock);
2689         mutex_exit(&DEVI(hwdip)->devi_lock);
2690 
2691         return (DDI_SUCCESS);
2692 }
2693 
2694 /*
2695  * Merge a "wildcard" .conf node. This is called by nexus drivers to
2696  * augment a set of hw node with properties specified in driver.conf file.
2697  * The parent node must be held busy.
2698  *
2699  * There is no failure mode, since the nexus may or may not have child
2700  * node bound the driver specified by the wildcard node.
2701  */
2702 void
2703 ndi_merge_wildcard_node(dev_info_t *dip)
2704 {
2705         dev_info_t *hwdip;
2706         dev_info_t *pdip = ddi_get_parent(dip);
2707         major_t major = ddi_driver_major(dip);
2708 
2709         /* never attempt to merge a hw node */
2710         ASSERT(ndi_dev_is_persistent_node(dip) == 0);
2711         /* must be bound to a driver major number */
2712         ASSERT(major != DDI_MAJOR_T_NONE);
2713 
2714         /*
2715          * Walk the child list to find all nodes bound to major
2716          * and copy properties.
2717          */
2718         mutex_enter(&DEVI(dip)->devi_lock);
2719         ASSERT(DEVI_BUSY_OWNED(pdip));
2720         for (hwdip = ddi_get_child(pdip); hwdip;
2721             hwdip = ddi_get_next_sibling(hwdip)) {
2722                 /*
2723                  * Skip nodes not bound to same driver
2724                  */
2725                 if (ddi_driver_major(hwdip) != major)
2726                         continue;
2727 
2728                 /*
2729                  * Skip .conf nodes
2730                  */
2731                 if (ndi_dev_is_persistent_node(hwdip) == 0)
2732                         continue;
2733 
2734                 /*
2735                  * Make sure the node is uninitialized and has no property.
2736                  */
2737                 mutex_enter(&DEVI(hwdip)->devi_lock);
2738                 if (i_ddi_node_state(hwdip) >= DS_INITIALIZED ||
2739                     (DEVI(hwdip)->devi_sys_prop_ptr != NULL) ||
2740                     (DEVI(hwdip)->devi_drv_prop_ptr != NULL)) {
2741                         mutex_exit(&DEVI(hwdip)->devi_lock);
2742                         NDI_CONFIG_DEBUG((CE_NOTE, "HW node %p state not "
2743                             "suitable for merging wildcard conf node %s",
2744                             (void *)hwdip, ddi_node_name(dip)));
2745                         continue;
2746                 }
2747 
2748                 DEVI(hwdip)->devi_sys_prop_ptr =
2749                     i_ddi_prop_list_dup(DEVI(dip)->devi_sys_prop_ptr, KM_SLEEP);
2750                 DEVI(hwdip)->devi_drv_prop_ptr =
2751                     i_ddi_prop_list_dup(DEVI(dip)->devi_drv_prop_ptr, KM_SLEEP);
2752                 mutex_exit(&DEVI(hwdip)->devi_lock);
2753         }
2754         mutex_exit(&DEVI(dip)->devi_lock);
2755 }
2756 
2757 /*
2758  * Return the major number based on the compatible property. This interface
2759  * may be used in situations where we are trying to detect if a better driver
2760  * now exists for a device, so it must use the 'compatible' property.  If
2761  * a non-NULL formp is specified and the binding was based on compatible then
2762  * return the pointer to the form used in *formp.
2763  */
2764 major_t
2765 ddi_compatible_driver_major(dev_info_t *dip, char **formp)
2766 {
2767         struct dev_info *devi = DEVI(dip);
2768         void            *compat;
2769         size_t          len;
2770         char            *p = NULL;
2771         major_t         major = DDI_MAJOR_T_NONE;
2772 
2773         if (formp)
2774                 *formp = NULL;
2775 
2776         if (ddi_prop_exists(DDI_DEV_T_NONE, dip, DDI_PROP_DONTPASS,
2777             "ddi-assigned")) {
2778                 major = ddi_name_to_major("nulldriver");
2779                 return (major);
2780         }
2781 
2782         /*
2783          * Highest precedence binding is a path-oriented alias. Since this
2784          * requires a 'path', this type of binding occurs via more obtuse
2785          * 'rebind'. The need for a path-oriented alias 'rebind' is detected
2786          * after a successful DDI_CTLOPS_INITCHILD to another driver: this is
2787          * is the first point at which the unit-address (or instance) of the
2788          * last component of the path is available (even though the path is
2789          * bound to the wrong driver at this point).
2790          */
2791         if (devi->devi_flags & DEVI_REBIND) {
2792                 p = devi->devi_rebinding_name;
2793                 major = ddi_name_to_major(p);
2794                 if (driver_active(major)) {
2795                         if (formp)
2796                                 *formp = p;
2797                         return (major);
2798                 }
2799 
2800                 /*
2801                  * If for some reason devi_rebinding_name no longer resolves
2802                  * to a proper driver then clear DEVI_REBIND.
2803                  */
2804                 mutex_enter(&devi->devi_lock);
2805                 devi->devi_flags &= ~DEVI_REBIND;
2806                 mutex_exit(&devi->devi_lock);
2807         }
2808 
2809         /* look up compatible property */
2810         (void) lookup_compatible(dip, KM_SLEEP);
2811         compat = (void *)(devi->devi_compat_names);
2812         len = devi->devi_compat_length;
2813 
2814         /* find the highest precedence compatible form with a driver binding */
2815         while ((p = prom_decode_composite_string(compat, len, p)) != NULL) {
2816                 major = ddi_name_to_major(p);
2817                 if (driver_active(major)) {
2818                         if (formp)
2819                                 *formp = p;
2820                         return (major);
2821                 }
2822         }
2823 
2824         /*
2825          * none of the compatible forms have a driver binding, see if
2826          * the node name has a driver binding.
2827          */
2828         major = ddi_name_to_major(ddi_node_name(dip));
2829         if (driver_active(major))
2830                 return (major);
2831 
2832         /* no driver */
2833         return (DDI_MAJOR_T_NONE);
2834 }
2835 
2836 /*
2837  * Static help functions
2838  */
2839 
2840 /*
2841  * lookup the "compatible" property and cache it's contents in the
2842  * device node.
2843  */
2844 static int
2845 lookup_compatible(dev_info_t *dip, uint_t flag)
2846 {
2847         int rv;
2848         int prop_flags;
2849         uint_t ncompatstrs;
2850         char **compatstrpp;
2851         char *di_compat_strp;
2852         size_t di_compat_strlen;
2853 
2854         if (DEVI(dip)->devi_compat_names) {
2855                 return (DDI_SUCCESS);
2856         }
2857 
2858         prop_flags = DDI_PROP_TYPE_STRING | DDI_PROP_DONTPASS;
2859 
2860         if (flag & KM_NOSLEEP) {
2861                 prop_flags |= DDI_PROP_DONTSLEEP;
2862         }
2863 
2864         if (ndi_dev_is_prom_node(dip) == 0) {
2865                 prop_flags |= DDI_PROP_NOTPROM;
2866         }
2867 
2868         rv = ddi_prop_lookup_common(DDI_DEV_T_ANY, dip, prop_flags,
2869             "compatible", &compatstrpp, &ncompatstrs,
2870             ddi_prop_fm_decode_strings);
2871 
2872         if (rv == DDI_PROP_NOT_FOUND) {
2873                 return (DDI_SUCCESS);
2874         }
2875 
2876         if (rv != DDI_PROP_SUCCESS) {
2877                 return (DDI_FAILURE);
2878         }
2879 
2880         /*
2881          * encode the compatible property data in the dev_info node
2882          */
2883         rv = DDI_SUCCESS;
2884         if (ncompatstrs != 0) {
2885                 di_compat_strp = encode_composite_string(compatstrpp,
2886                     ncompatstrs, &di_compat_strlen, flag);
2887                 if (di_compat_strp != NULL) {
2888                         DEVI(dip)->devi_compat_names = di_compat_strp;
2889                         DEVI(dip)->devi_compat_length = di_compat_strlen;
2890                 } else {
2891                         rv = DDI_FAILURE;
2892                 }
2893         }
2894         ddi_prop_free(compatstrpp);
2895         return (rv);
2896 }
2897 
2898 /*
2899  * Create a composite string from a list of strings.
2900  *
2901  * A composite string consists of a single buffer containing one
2902  * or more NULL terminated strings.
2903  */
2904 static char *
2905 encode_composite_string(char **strings, uint_t nstrings, size_t *retsz,
2906     uint_t flag)
2907 {
2908         uint_t index;
2909         char  **strpp;
2910         uint_t slen;
2911         size_t cbuf_sz = 0;
2912         char *cbuf_p;
2913         char *cbuf_ip;
2914 
2915         if (strings == NULL || nstrings == 0 || retsz == NULL) {
2916                 return (NULL);
2917         }
2918 
2919         for (index = 0, strpp = strings; index < nstrings; index++)
2920                 cbuf_sz += strlen(*(strpp++)) + 1;
2921 
2922         if ((cbuf_p = kmem_alloc(cbuf_sz, flag)) == NULL) {
2923                 cmn_err(CE_NOTE,
2924                     "?failed to allocate device node compatstr");
2925                 return (NULL);
2926         }
2927 
2928         cbuf_ip = cbuf_p;
2929         for (index = 0, strpp = strings; index < nstrings; index++) {
2930                 slen = strlen(*strpp);
2931                 bcopy(*(strpp++), cbuf_ip, slen);
2932                 cbuf_ip += slen;
2933                 *(cbuf_ip++) = '\0';
2934         }
2935 
2936         *retsz = cbuf_sz;
2937         return (cbuf_p);
2938 }
2939 
2940 static void
2941 link_to_driver_list(dev_info_t *dip)
2942 {
2943         major_t major = DEVI(dip)->devi_major;
2944         struct devnames *dnp;
2945 
2946         ASSERT(major != DDI_MAJOR_T_NONE);
2947 
2948         /*
2949          * Remove from orphan list
2950          */
2951         if (ndi_dev_is_persistent_node(dip)) {
2952                 dnp = &orphanlist;
2953                 remove_from_dn_list(dnp, dip);
2954         }
2955 
2956         /*
2957          * Add to per driver list
2958          */
2959         dnp = &devnamesp[major];
2960         add_to_dn_list(dnp, dip);
2961 }
2962 
2963 static void
2964 unlink_from_driver_list(dev_info_t *dip)
2965 {
2966         major_t major = DEVI(dip)->devi_major;
2967         struct devnames *dnp;
2968 
2969         ASSERT(major != DDI_MAJOR_T_NONE);
2970 
2971         /*
2972          * Remove from per-driver list
2973          */
2974         dnp = &devnamesp[major];
2975         remove_from_dn_list(dnp, dip);
2976 
2977         /*
2978          * Add to orphan list
2979          */
2980         if (ndi_dev_is_persistent_node(dip)) {
2981                 dnp = &orphanlist;
2982                 add_to_dn_list(dnp, dip);
2983         }
2984 }
2985 
2986 /*
2987  * scan the per-driver list looking for dev_info "dip"
2988  */
2989 static dev_info_t *
2990 in_dn_list(struct devnames *dnp, dev_info_t *dip)
2991 {
2992         struct dev_info *idevi;
2993 
2994         if ((idevi = DEVI(dnp->dn_head)) == NULL)
2995                 return (NULL);
2996 
2997         while (idevi) {
2998                 if (idevi == DEVI(dip))
2999                         return (dip);
3000                 idevi = idevi->devi_next;
3001         }
3002         return (NULL);
3003 }
3004 
3005 /*
3006  * insert devinfo node 'dip' into the per-driver instance list
3007  * headed by 'dnp'
3008  *
3009  * Nodes on the per-driver list are ordered: HW - SID - PSEUDO.  The order is
3010  * required for merging of .conf file data to work properly.
3011  */
3012 static void
3013 add_to_ordered_dn_list(struct devnames *dnp, dev_info_t *dip)
3014 {
3015         dev_info_t **dipp;
3016 
3017         ASSERT(mutex_owned(&(dnp->dn_lock)));
3018 
3019         dipp = &dnp->dn_head;
3020         if (ndi_dev_is_prom_node(dip)) {
3021                 /*
3022                  * Find the first non-prom node or end of list
3023                  */
3024                 while (*dipp && (ndi_dev_is_prom_node(*dipp) != 0)) {
3025                         dipp = (dev_info_t **)&DEVI(*dipp)->devi_next;
3026                 }
3027         } else if (ndi_dev_is_persistent_node(dip)) {
3028                 /*
3029                  * Find the first non-persistent node
3030                  */
3031                 while (*dipp && (ndi_dev_is_persistent_node(*dipp) != 0)) {
3032                         dipp = (dev_info_t **)&DEVI(*dipp)->devi_next;
3033                 }
3034         } else {
3035                 /*
3036                  * Find the end of the list
3037                  */
3038                 while (*dipp) {
3039                         dipp = (dev_info_t **)&DEVI(*dipp)->devi_next;
3040                 }
3041         }
3042 
3043         DEVI(dip)->devi_next = DEVI(*dipp);
3044         *dipp = dip;
3045 }
3046 
3047 /*
3048  * add a list of device nodes to the device node list in the
3049  * devnames structure
3050  */
3051 static void
3052 add_to_dn_list(struct devnames *dnp, dev_info_t *dip)
3053 {
3054         /*
3055          * Look to see if node already exists
3056          */
3057         LOCK_DEV_OPS(&(dnp->dn_lock));
3058         if (in_dn_list(dnp, dip)) {
3059                 cmn_err(CE_NOTE, "add_to_dn_list: node %s already in list",
3060                     DEVI(dip)->devi_node_name);
3061         } else {
3062                 add_to_ordered_dn_list(dnp, dip);
3063         }
3064         UNLOCK_DEV_OPS(&(dnp->dn_lock));
3065 }
3066 
3067 static void
3068 remove_from_dn_list(struct devnames *dnp, dev_info_t *dip)
3069 {
3070         dev_info_t **plist;
3071 
3072         LOCK_DEV_OPS(&(dnp->dn_lock));
3073 
3074         plist = (dev_info_t **)&dnp->dn_head;
3075         while (*plist && (*plist != dip)) {
3076                 plist = (dev_info_t **)&DEVI(*plist)->devi_next;
3077         }
3078 
3079         if (*plist != NULL) {
3080                 ASSERT(*plist == dip);
3081                 *plist = (dev_info_t *)(DEVI(dip)->devi_next);
3082                 DEVI(dip)->devi_next = NULL;
3083         } else {
3084                 NDI_CONFIG_DEBUG((CE_NOTE,
3085                     "remove_from_dn_list: node %s not found in list",
3086                     DEVI(dip)->devi_node_name));
3087         }
3088 
3089         UNLOCK_DEV_OPS(&(dnp->dn_lock));
3090 }
3091 
3092 /*
3093  * Add and remove reference driver global property list
3094  */
3095 static void
3096 add_global_props(dev_info_t *dip)
3097 {
3098         struct devnames *dnp;
3099         ddi_prop_list_t *plist;
3100 
3101         ASSERT(DEVI(dip)->devi_global_prop_list == NULL);
3102         ASSERT(DEVI(dip)->devi_major != DDI_MAJOR_T_NONE);
3103 
3104         dnp = &devnamesp[DEVI(dip)->devi_major];
3105         LOCK_DEV_OPS(&dnp->dn_lock);
3106         plist = dnp->dn_global_prop_ptr;
3107         if (plist == NULL) {
3108                 UNLOCK_DEV_OPS(&dnp->dn_lock);
3109                 return;
3110         }
3111         i_ddi_prop_list_hold(plist, dnp);
3112         UNLOCK_DEV_OPS(&dnp->dn_lock);
3113 
3114         mutex_enter(&DEVI(dip)->devi_lock);
3115         DEVI(dip)->devi_global_prop_list = plist;
3116         mutex_exit(&DEVI(dip)->devi_lock);
3117 }
3118 
3119 static void
3120 remove_global_props(dev_info_t *dip)
3121 {
3122         ddi_prop_list_t *proplist;
3123 
3124         mutex_enter(&DEVI(dip)->devi_lock);
3125         proplist = DEVI(dip)->devi_global_prop_list;
3126         DEVI(dip)->devi_global_prop_list = NULL;
3127         mutex_exit(&DEVI(dip)->devi_lock);
3128 
3129         if (proplist) {
3130                 major_t major;
3131                 struct devnames *dnp;
3132 
3133                 major = ddi_driver_major(dip);
3134                 ASSERT(major != DDI_MAJOR_T_NONE);
3135                 dnp = &devnamesp[major];
3136                 LOCK_DEV_OPS(&dnp->dn_lock);
3137                 i_ddi_prop_list_rele(proplist, dnp);
3138                 UNLOCK_DEV_OPS(&dnp->dn_lock);
3139         }
3140 }
3141 
3142 #ifdef DEBUG
3143 /*
3144  * Set this variable to '0' to disable the optimization,
3145  * and to 2 to print debug message.
3146  */
3147 static int optimize_dtree = 1;
3148 
3149 static void
3150 debug_dtree(dev_info_t *devi, struct dev_info *adevi, char *service)
3151 {
3152         char *adeviname, *buf;
3153 
3154         /*
3155          * Don't print unless optimize dtree is set to 2+
3156          */
3157         if (optimize_dtree <= 1)
3158                 return;
3159 
3160         buf = kmem_alloc(MAXNAMELEN, KM_SLEEP);
3161         adeviname = ddi_deviname((dev_info_t *)adevi, buf);
3162         if (*adeviname == '\0')
3163                 adeviname = "root";
3164 
3165         cmn_err(CE_CONT, "%s %s -> %s\n",
3166             ddi_deviname(devi, buf), service, adeviname);
3167 
3168         kmem_free(buf, MAXNAMELEN);
3169 }
3170 #else /* DEBUG */
3171 #define debug_dtree(a1, a2, a3)  /* nothing */
3172 #endif  /* DEBUG */
3173 
3174 static void
3175 ddi_optimize_dtree(dev_info_t *devi)
3176 {
3177         struct dev_info *pdevi;
3178         struct bus_ops *b;
3179 
3180         pdevi = DEVI(devi)->devi_parent;
3181         ASSERT(pdevi);
3182 
3183         /*
3184          * Set the unoptimized values
3185          */
3186         DEVI(devi)->devi_bus_map_fault = pdevi;
3187         DEVI(devi)->devi_bus_dma_allochdl = pdevi;
3188         DEVI(devi)->devi_bus_dma_freehdl = pdevi;
3189         DEVI(devi)->devi_bus_dma_bindhdl = pdevi;
3190         DEVI(devi)->devi_bus_dma_bindfunc =
3191             pdevi->devi_ops->devo_bus_ops->bus_dma_bindhdl;
3192         DEVI(devi)->devi_bus_dma_unbindhdl = pdevi;
3193         DEVI(devi)->devi_bus_dma_unbindfunc =
3194             pdevi->devi_ops->devo_bus_ops->bus_dma_unbindhdl;
3195         DEVI(devi)->devi_bus_dma_flush = pdevi;
3196         DEVI(devi)->devi_bus_dma_win = pdevi;
3197         DEVI(devi)->devi_bus_dma_ctl = pdevi;
3198         DEVI(devi)->devi_bus_ctl = pdevi;
3199 
3200 #ifdef DEBUG
3201         if (optimize_dtree == 0)
3202                 return;
3203 #endif /* DEBUG */
3204 
3205         b = pdevi->devi_ops->devo_bus_ops;
3206 
3207         if (i_ddi_map_fault == b->bus_map_fault) {
3208                 DEVI(devi)->devi_bus_map_fault = pdevi->devi_bus_map_fault;
3209                 debug_dtree(devi, DEVI(devi)->devi_bus_map_fault,
3210                     "bus_map_fault");
3211         }
3212 
3213         if (ddi_dma_allochdl == b->bus_dma_allochdl) {
3214                 DEVI(devi)->devi_bus_dma_allochdl =
3215                     pdevi->devi_bus_dma_allochdl;
3216                 debug_dtree(devi, DEVI(devi)->devi_bus_dma_allochdl,
3217                     "bus_dma_allochdl");
3218         }
3219 
3220         if (ddi_dma_freehdl == b->bus_dma_freehdl) {
3221                 DEVI(devi)->devi_bus_dma_freehdl = pdevi->devi_bus_dma_freehdl;
3222                 debug_dtree(devi, DEVI(devi)->devi_bus_dma_freehdl,
3223                     "bus_dma_freehdl");
3224         }
3225 
3226         if (ddi_dma_bindhdl == b->bus_dma_bindhdl) {
3227                 DEVI(devi)->devi_bus_dma_bindhdl = pdevi->devi_bus_dma_bindhdl;
3228                 DEVI(devi)->devi_bus_dma_bindfunc =
3229                     pdevi->devi_bus_dma_bindhdl->devi_ops->
3230                     devo_bus_ops->bus_dma_bindhdl;
3231                 debug_dtree(devi, DEVI(devi)->devi_bus_dma_bindhdl,
3232                     "bus_dma_bindhdl");
3233         }
3234 
3235         if (ddi_dma_unbindhdl == b->bus_dma_unbindhdl) {
3236                 DEVI(devi)->devi_bus_dma_unbindhdl =
3237                     pdevi->devi_bus_dma_unbindhdl;
3238                 DEVI(devi)->devi_bus_dma_unbindfunc =
3239                     pdevi->devi_bus_dma_unbindhdl->devi_ops->
3240                     devo_bus_ops->bus_dma_unbindhdl;
3241                 debug_dtree(devi, DEVI(devi)->devi_bus_dma_unbindhdl,
3242                     "bus_dma_unbindhdl");
3243         }
3244 
3245         if (ddi_dma_flush == b->bus_dma_flush) {
3246                 DEVI(devi)->devi_bus_dma_flush = pdevi->devi_bus_dma_flush;
3247                 debug_dtree(devi, DEVI(devi)->devi_bus_dma_flush,
3248                     "bus_dma_flush");
3249         }
3250 
3251         if (ddi_dma_win == b->bus_dma_win) {
3252                 DEVI(devi)->devi_bus_dma_win = pdevi->devi_bus_dma_win;
3253                 debug_dtree(devi, DEVI(devi)->devi_bus_dma_win,
3254                     "bus_dma_win");
3255         }
3256 
3257         if (ddi_dma_mctl == b->bus_dma_ctl) {
3258                 DEVI(devi)->devi_bus_dma_ctl = pdevi->devi_bus_dma_ctl;
3259                 debug_dtree(devi, DEVI(devi)->devi_bus_dma_ctl, "bus_dma_ctl");
3260         }
3261 
3262         if (ddi_ctlops == b->bus_ctl) {
3263                 DEVI(devi)->devi_bus_ctl = pdevi->devi_bus_ctl;
3264                 debug_dtree(devi, DEVI(devi)->devi_bus_ctl, "bus_ctl");
3265         }
3266 }
3267 
3268 #define MIN_DEVINFO_LOG_SIZE    max_ncpus
3269 #define MAX_DEVINFO_LOG_SIZE    max_ncpus * 10
3270 
3271 static void
3272 da_log_init()
3273 {
3274         devinfo_log_header_t *dh;
3275         int logsize = devinfo_log_size;
3276 
3277         if (logsize == 0)
3278                 logsize = MIN_DEVINFO_LOG_SIZE;
3279         else if (logsize > MAX_DEVINFO_LOG_SIZE)
3280                 logsize = MAX_DEVINFO_LOG_SIZE;
3281 
3282         dh = kmem_alloc(logsize * PAGESIZE, KM_SLEEP);
3283         mutex_init(&dh->dh_lock, NULL, MUTEX_DEFAULT, NULL);
3284         dh->dh_max = ((logsize * PAGESIZE) - sizeof (*dh)) /
3285             sizeof (devinfo_audit_t) + 1;
3286         dh->dh_curr = -1;
3287         dh->dh_hits = 0;
3288 
3289         devinfo_audit_log = dh;
3290 }
3291 
3292 /*
3293  * Log the stack trace in per-devinfo audit structure and also enter
3294  * it into a system wide log for recording the time history.
3295  */
3296 static void
3297 da_log_enter(dev_info_t *dip)
3298 {
3299         devinfo_audit_t *da_log, *da = DEVI(dip)->devi_audit;
3300         devinfo_log_header_t *dh = devinfo_audit_log;
3301 
3302         if (devinfo_audit_log == NULL)
3303                 return;
3304 
3305         ASSERT(da != NULL);
3306 
3307         da->da_devinfo = dip;
3308         da->da_timestamp = gethrtime();
3309         da->da_thread = curthread;
3310         da->da_node_state = DEVI(dip)->devi_node_state;
3311         da->da_device_state = DEVI(dip)->devi_state;
3312         da->da_depth = getpcstack(da->da_stack, DDI_STACK_DEPTH);
3313 
3314         /*
3315          * Copy into common log and note the location for tracing history
3316          */
3317         mutex_enter(&dh->dh_lock);
3318         dh->dh_hits++;
3319         dh->dh_curr++;
3320         if (dh->dh_curr >= dh->dh_max)
3321                 dh->dh_curr -= dh->dh_max;
3322         da_log = &dh->dh_entry[dh->dh_curr];
3323         mutex_exit(&dh->dh_lock);
3324 
3325         bcopy(da, da_log, sizeof (devinfo_audit_t));
3326         da->da_lastlog = da_log;
3327 }
3328 
3329 static void
3330 attach_drivers()
3331 {
3332         int i;
3333         for (i = 0; i < devcnt; i++) {
3334                 struct devnames *dnp = &devnamesp[i];
3335                 if ((dnp->dn_flags & DN_FORCE_ATTACH) &&
3336                     (ddi_hold_installed_driver((major_t)i) != NULL))
3337                         ddi_rele_driver((major_t)i);
3338         }
3339 }
3340 
3341 /*
3342  * Launch a thread to force attach drivers. This avoids penalty on boot time.
3343  */
3344 void
3345 i_ddi_forceattach_drivers()
3346 {
3347 
3348         /*
3349          * Attach IB VHCI driver before the force-attach thread attaches the
3350          * IB HCA driver. IB HCA driver will fail if IB Nexus has not yet
3351          * been attached.
3352          */
3353         (void) ddi_hold_installed_driver(ddi_name_to_major("ib"));
3354 
3355         (void) thread_create(NULL, 0, (void (*)())attach_drivers, NULL, 0, &p0,
3356             TS_RUN, minclsyspri);
3357 }
3358 
3359 /*
3360  * This is a private DDI interface for optimizing boot performance.
3361  * I/O subsystem initialization is considered complete when devfsadm
3362  * is executed.
3363  *
3364  * NOTE: The start of syseventd happens to be a convenient indicator
3365  *      of the completion of I/O initialization during boot.
3366  *      The implementation should be replaced by something more robust.
3367  */
3368 int
3369 i_ddi_io_initialized()
3370 {
3371         extern int sysevent_daemon_init;
3372         return (sysevent_daemon_init);
3373 }
3374 
3375 /*
3376  * May be used to determine system boot state
3377  * "Available" means the system is for the most part up
3378  * and initialized, with all system services either up or
3379  * capable of being started.  This state is set by devfsadm
3380  * during the boot process.  The /dev filesystem infers
3381  * from this when implicit reconfig can be performed,
3382  * ie, devfsadm can be invoked.  Please avoid making
3383  * further use of this unless it's really necessary.
3384  */
3385 int
3386 i_ddi_sysavail()
3387 {
3388         return (devname_state & DS_SYSAVAIL);
3389 }
3390 
3391 /*
3392  * May be used to determine if boot is a reconfigure boot.
3393  */
3394 int
3395 i_ddi_reconfig()
3396 {
3397         return (devname_state & DS_RECONFIG);
3398 }
3399 
3400 /*
3401  * Note system services are up, inform /dev.
3402  */
3403 void
3404 i_ddi_set_sysavail()
3405 {
3406         if ((devname_state & DS_SYSAVAIL) == 0) {
3407                 devname_state |= DS_SYSAVAIL;
3408                 sdev_devstate_change();
3409         }
3410 }
3411 
3412 /*
3413  * Note reconfiguration boot, inform /dev.
3414  */
3415 void
3416 i_ddi_set_reconfig()
3417 {
3418         if ((devname_state & DS_RECONFIG) == 0) {
3419                 devname_state |= DS_RECONFIG;
3420                 sdev_devstate_change();
3421         }
3422 }
3423 
3424 
3425 /*
3426  * device tree walking
3427  */
3428 
3429 struct walk_elem {
3430         struct walk_elem *next;
3431         dev_info_t *dip;
3432 };
3433 
3434 static void
3435 free_list(struct walk_elem *list)
3436 {
3437         while (list) {
3438                 struct walk_elem *next = list->next;
3439                 kmem_free(list, sizeof (*list));
3440                 list = next;
3441         }
3442 }
3443 
3444 static void
3445 append_node(struct walk_elem **list, dev_info_t *dip)
3446 {
3447         struct walk_elem *tail;
3448         struct walk_elem *elem = kmem_alloc(sizeof (*elem), KM_SLEEP);
3449 
3450         elem->next = NULL;
3451         elem->dip = dip;
3452 
3453         if (*list == NULL) {
3454                 *list = elem;
3455                 return;
3456         }
3457 
3458         tail = *list;
3459         while (tail->next)
3460                 tail = tail->next;
3461 
3462         tail->next = elem;
3463 }
3464 
3465 /*
3466  * The implementation of ddi_walk_devs().
3467  */
3468 static int
3469 walk_devs(dev_info_t *dip, int (*f)(dev_info_t *, void *), void *arg,
3470     int do_locking)
3471 {
3472         struct walk_elem *head = NULL;
3473 
3474         /*
3475          * Do it in two passes. First pass invoke callback on each
3476          * dip on the sibling list. Second pass invoke callback on
3477          * children of each dip.
3478          */
3479         while (dip) {
3480                 switch ((*f)(dip, arg)) {
3481                 case DDI_WALK_TERMINATE:
3482                         free_list(head);
3483                         return (DDI_WALK_TERMINATE);
3484 
3485                 case DDI_WALK_PRUNESIB:
3486                         /* ignore sibling by setting dip to NULL */
3487                         append_node(&head, dip);
3488                         dip = NULL;
3489                         break;
3490 
3491                 case DDI_WALK_PRUNECHILD:
3492                         /* don't worry about children */
3493                         dip = ddi_get_next_sibling(dip);
3494                         break;
3495 
3496                 case DDI_WALK_CONTINUE:
3497                 default:
3498                         append_node(&head, dip);
3499                         dip = ddi_get_next_sibling(dip);
3500                         break;
3501                 }
3502 
3503         }
3504 
3505         /* second pass */
3506         while (head) {
3507                 int circ;
3508                 struct walk_elem *next = head->next;
3509 
3510                 if (do_locking)
3511                         ndi_devi_enter(head->dip, &circ);
3512                 if (walk_devs(ddi_get_child(head->dip), f, arg, do_locking) ==
3513                     DDI_WALK_TERMINATE) {
3514                         if (do_locking)
3515                                 ndi_devi_exit(head->dip, circ);
3516                         free_list(head);
3517                         return (DDI_WALK_TERMINATE);
3518                 }
3519                 if (do_locking)
3520                         ndi_devi_exit(head->dip, circ);
3521                 kmem_free(head, sizeof (*head));
3522                 head = next;
3523         }
3524 
3525         return (DDI_WALK_CONTINUE);
3526 }
3527 
3528 /*
3529  * This general-purpose routine traverses the tree of dev_info nodes,
3530  * starting from the given node, and calls the given function for each
3531  * node that it finds with the current node and the pointer arg (which
3532  * can point to a structure of information that the function
3533  * needs) as arguments.
3534  *
3535  * It does the walk a layer at a time, not depth-first. The given function
3536  * must return one of the following values:
3537  *      DDI_WALK_CONTINUE
3538  *      DDI_WALK_PRUNESIB
3539  *      DDI_WALK_PRUNECHILD
3540  *      DDI_WALK_TERMINATE
3541  *
3542  * N.B. Since we walk the sibling list, the caller must ensure that
3543  *      the parent of dip is held against changes, unless the parent
3544  *      is rootnode.  ndi_devi_enter() on the parent is sufficient.
3545  *
3546  *      To avoid deadlock situations, caller must not attempt to
3547  *      configure/unconfigure/remove device node in (*f)(), nor should
3548  *      it attempt to recurse on other nodes in the system. Any
3549  *      ndi_devi_enter() done by (*f)() must occur 'at-or-below' the
3550  *      node entered prior to ddi_walk_devs(). Furthermore, if (*f)()
3551  *      does any multi-threading (in framework *or* in driver) then the
3552  *      ndi_devi_enter() calls done by dependent threads must be
3553  *      'strictly-below'.
3554  *
3555  *      This is not callable from device autoconfiguration routines.
3556  *      They include, but not limited to, _init(9e), _fini(9e), probe(9e),
3557  *      attach(9e), and detach(9e).
3558  */
3559 
3560 void
3561 ddi_walk_devs(dev_info_t *dip, int (*f)(dev_info_t *, void *), void *arg)
3562 {
3563 
3564         ASSERT(dip == NULL || ddi_get_parent(dip) == NULL ||
3565             DEVI_BUSY_OWNED(ddi_get_parent(dip)));
3566 
3567         (void) walk_devs(dip, f, arg, 1);
3568 }
3569 
3570 /*
3571  * This is a general-purpose routine traverses the per-driver list
3572  * and calls the given function for each node. must return one of
3573  * the following values:
3574  *      DDI_WALK_CONTINUE
3575  *      DDI_WALK_TERMINATE
3576  *
3577  * N.B. The same restrictions from ddi_walk_devs() apply.
3578  */
3579 
3580 void
3581 e_ddi_walk_driver(char *drv, int (*f)(dev_info_t *, void *), void *arg)
3582 {
3583         major_t major;
3584         struct devnames *dnp;
3585         dev_info_t *dip;
3586 
3587         major = ddi_name_to_major(drv);
3588         if (major == DDI_MAJOR_T_NONE)
3589                 return;
3590 
3591         dnp = &devnamesp[major];
3592         LOCK_DEV_OPS(&dnp->dn_lock);
3593         dip = dnp->dn_head;
3594         while (dip) {
3595                 ndi_hold_devi(dip);
3596                 UNLOCK_DEV_OPS(&dnp->dn_lock);
3597                 if ((*f)(dip, arg) == DDI_WALK_TERMINATE) {
3598                         ndi_rele_devi(dip);
3599                         return;
3600                 }
3601                 LOCK_DEV_OPS(&dnp->dn_lock);
3602                 ndi_rele_devi(dip);
3603                 dip = ddi_get_next(dip);
3604         }
3605         UNLOCK_DEV_OPS(&dnp->dn_lock);
3606 }
3607 
3608 /*
3609  * argument to i_find_devi, a devinfo node search callback function.
3610  */
3611 struct match_info {
3612         dev_info_t      *dip;           /* result */
3613         char            *nodename;      /* if non-null, nodename must match */
3614         int             instance;       /* if != -1, instance must match */
3615         int             attached;       /* if != 0, i_ddi_devi_attached() */
3616 };
3617 
3618 static int
3619 i_find_devi(dev_info_t *dip, void *arg)
3620 {
3621         struct match_info *info = (struct match_info *)arg;
3622 
3623         if (((info->nodename == NULL) ||
3624             (strcmp(ddi_node_name(dip), info->nodename) == 0)) &&
3625             ((info->instance == -1) ||
3626             (ddi_get_instance(dip) == info->instance)) &&
3627             ((info->attached == 0) || i_ddi_devi_attached(dip))) {
3628                 info->dip = dip;
3629                 ndi_hold_devi(dip);
3630                 return (DDI_WALK_TERMINATE);
3631         }
3632 
3633         return (DDI_WALK_CONTINUE);
3634 }
3635 
3636 /*
3637  * Find dip with a known node name and instance and return with it held
3638  */
3639 dev_info_t *
3640 ddi_find_devinfo(char *nodename, int instance, int attached)
3641 {
3642         struct match_info       info;
3643 
3644         info.nodename = nodename;
3645         info.instance = instance;
3646         info.attached = attached;
3647         info.dip = NULL;
3648 
3649         ddi_walk_devs(ddi_root_node(), i_find_devi, &info);
3650         return (info.dip);
3651 }
3652 
3653 extern ib_boot_prop_t *iscsiboot_prop;
3654 static void
3655 i_ddi_parse_iscsi_name(char *name, char **nodename, char **addrname,
3656     char **minorname)
3657 {
3658         char *cp, *colon;
3659         static char nulladdrname[] = "";
3660 
3661         /* default values */
3662         if (nodename)
3663                 *nodename = name;
3664         if (addrname)
3665                 *addrname = nulladdrname;
3666         if (minorname)
3667                 *minorname = NULL;
3668 
3669         cp = colon = name;
3670         while (*cp != '\0') {
3671                 if (addrname && *cp == '@') {
3672                         *addrname = cp + 1;
3673                         *cp = '\0';
3674                 } else if (minorname && *cp == ':') {
3675                         *minorname = cp + 1;
3676                         colon = cp;
3677                 }
3678                 ++cp;
3679         }
3680         if (colon != name) {
3681                 *colon = '\0';
3682         }
3683 }
3684 
3685 /*
3686  * Parse for name, addr, and minor names. Some args may be NULL.
3687  */
3688 void
3689 i_ddi_parse_name(char *name, char **nodename, char **addrname, char **minorname)
3690 {
3691         char *cp;
3692         static char nulladdrname[] = "";
3693 
3694         /* default values */
3695         if (nodename)
3696                 *nodename = name;
3697         if (addrname)
3698                 *addrname = nulladdrname;
3699         if (minorname)
3700                 *minorname = NULL;
3701 
3702         cp = name;
3703         while (*cp != '\0') {
3704                 if (addrname && *cp == '@') {
3705                         *addrname = cp + 1;
3706                         *cp = '\0';
3707                 } else if (minorname && *cp == ':') {
3708                         *minorname = cp + 1;
3709                         *cp = '\0';
3710                 }
3711                 ++cp;
3712         }
3713 }
3714 
3715 static char *
3716 child_path_to_driver(dev_info_t *parent, char *child_name, char *unit_address)
3717 {
3718         char *p, *drvname = NULL;
3719         major_t maj;
3720 
3721         /*
3722          * Construct the pathname and ask the implementation
3723          * if it can do a driver = f(pathname) for us, if not
3724          * we'll just default to using the node-name that
3725          * was given to us.  We want to do this first to
3726          * allow the platform to use 'generic' names for
3727          * legacy device drivers.
3728          */
3729         p = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
3730         (void) ddi_pathname(parent, p);
3731         (void) strcat(p, "/");
3732         (void) strcat(p, child_name);
3733         if (unit_address && *unit_address) {
3734                 (void) strcat(p, "@");
3735                 (void) strcat(p, unit_address);
3736         }
3737 
3738         /*
3739          * Get the binding. If there is none, return the child_name
3740          * and let the caller deal with it.
3741          */
3742         maj = path_to_major(p);
3743 
3744         kmem_free(p, MAXPATHLEN);
3745 
3746         if (maj != DDI_MAJOR_T_NONE)
3747                 drvname = ddi_major_to_name(maj);
3748         if (drvname == NULL)
3749                 drvname = child_name;
3750 
3751         return (drvname);
3752 }
3753 
3754 
3755 #define PCI_EX_CLASS    "pciexclass"
3756 #define PCI_EX          "pciex"
3757 #define PCI_CLASS       "pciclass"
3758 #define PCI             "pci"
3759 
3760 int
3761 ddi_is_pci_dip(dev_info_t *dip)
3762 {
3763         char    *prop = NULL;
3764 
3765         if (ddi_prop_lookup_string(DDI_DEV_T_ANY, dip, DDI_PROP_DONTPASS,
3766             "compatible", &prop) == DDI_PROP_SUCCESS) {
3767                 ASSERT(prop);
3768                 if (strncmp(prop, PCI_EX_CLASS, sizeof (PCI_EX_CLASS) - 1)
3769                     == 0 ||
3770                     strncmp(prop, PCI_EX, sizeof (PCI_EX)- 1)
3771                     == 0 ||
3772                     strncmp(prop, PCI_CLASS, sizeof (PCI_CLASS) - 1)
3773                     == 0 ||
3774                     strncmp(prop, PCI, sizeof (PCI) - 1)
3775                     == 0) {
3776                         ddi_prop_free(prop);
3777                         return (1);
3778                 }
3779         }
3780 
3781         if (prop != NULL) {
3782                 ddi_prop_free(prop);
3783         }
3784 
3785         return (0);
3786 }
3787 
3788 /*
3789  * Given the pathname of a device, fill in the dev_info_t value and/or the
3790  * dev_t value and/or the spectype, depending on which parameters are non-NULL.
3791  * If there is an error, this function returns -1.
3792  *
3793  * NOTE: If this function returns the dev_info_t structure, then it
3794  * does so with a hold on the devi. Caller should ensure that they get
3795  * decremented via ddi_release_devi() or ndi_rele_devi();
3796  *
3797  * This function can be invoked in the boot case for a pathname without
3798  * device argument (:xxxx), traditionally treated as a minor name.
3799  * In this case, we do the following
3800  * (1) search the minor node of type DDM_DEFAULT.
3801  * (2) if no DDM_DEFAULT minor exists, then the first non-alias minor is chosen.
3802  * (3) if neither exists, a dev_t is faked with minor number = instance.
3803  * As of S9 FCS, no instance of #1 exists. #2 is used by several platforms
3804  * to default the boot partition to :a possibly by other OBP definitions.
3805  * #3 is used for booting off network interfaces, most SPARC network
3806  * drivers support Style-2 only, so only DDM_ALIAS minor exists.
3807  *
3808  * It is possible for OBP to present device args at the end of the path as
3809  * well as in the middle. For example, with IB the following strings are
3810  * valid boot paths.
3811  *      a /pci@8,700000/ib@1,2:port=1,pkey=ff,dhcp,...
3812  *      b /pci@8,700000/ib@1,1:port=1/ioc@xxxxxx,yyyyyyy:dhcp
3813  * Case (a), we first look for minor node "port=1,pkey...".
3814  * Failing that, we will pass "port=1,pkey..." to the bus_config
3815  * entry point of ib (HCA) driver.
3816  * Case (b), configure ib@1,1 as usual. Then invoke ib's bus_config
3817  * with argument "ioc@xxxxxxx,yyyyyyy:port=1". After configuring
3818  * the ioc, look for minor node dhcp. If not found, pass ":dhcp"
3819  * to ioc's bus_config entry point.
3820  */
3821 int
3822 resolve_pathname(char *pathname,
3823         dev_info_t **dipp, dev_t *devtp, int *spectypep)
3824 {
3825         int                     error;
3826         dev_info_t              *parent, *child;
3827         struct pathname         pn;
3828         char                    *component, *config_name;
3829         char                    *minorname = NULL;
3830         char                    *prev_minor = NULL;
3831         dev_t                   devt = NODEV;
3832         int                     spectype;
3833         struct ddi_minor_data   *dmn;
3834         int                     circ;
3835 
3836         if (*pathname != '/')
3837                 return (EINVAL);
3838         parent = ddi_root_node();       /* Begin at the top of the tree */
3839 
3840         if (error = pn_get(pathname, UIO_SYSSPACE, &pn))
3841                 return (error);
3842         pn_skipslash(&pn);
3843 
3844         ASSERT(i_ddi_devi_attached(parent));
3845         ndi_hold_devi(parent);
3846 
3847         component = kmem_alloc(MAXNAMELEN, KM_SLEEP);
3848         config_name = kmem_alloc(MAXNAMELEN, KM_SLEEP);
3849 
3850         while (pn_pathleft(&pn)) {
3851                 /* remember prev minor (:xxx) in the middle of path */
3852                 if (minorname)
3853                         prev_minor = i_ddi_strdup(minorname, KM_SLEEP);
3854 
3855                 /* Get component and chop off minorname */
3856                 (void) pn_getcomponent(&pn, component);
3857                 if ((iscsiboot_prop != NULL) &&
3858                     (strcmp((DEVI(parent)->devi_node_name), "iscsi") == 0)) {
3859                         i_ddi_parse_iscsi_name(component, NULL, NULL,
3860                             &minorname);
3861                 } else {
3862                         i_ddi_parse_name(component, NULL, NULL, &minorname);
3863                 }
3864                 if (prev_minor == NULL) {
3865                         (void) snprintf(config_name, MAXNAMELEN, "%s",
3866                             component);
3867                 } else {
3868                         (void) snprintf(config_name, MAXNAMELEN, "%s:%s",
3869                             component, prev_minor);
3870                         kmem_free(prev_minor, strlen(prev_minor) + 1);
3871                         prev_minor = NULL;
3872                 }
3873 
3874                 /*
3875                  * Find and configure the child
3876                  */
3877                 if (ndi_devi_config_one(parent, config_name, &child,
3878                     NDI_PROMNAME | NDI_NO_EVENT) != NDI_SUCCESS) {
3879                         ndi_rele_devi(parent);
3880                         pn_free(&pn);
3881                         kmem_free(component, MAXNAMELEN);
3882                         kmem_free(config_name, MAXNAMELEN);
3883                         return (-1);
3884                 }
3885 
3886                 ASSERT(i_ddi_devi_attached(child));
3887                 ndi_rele_devi(parent);
3888                 parent = child;
3889                 pn_skipslash(&pn);
3890         }
3891 
3892         /*
3893          * First look for a minor node matching minorname.
3894          * Failing that, try to pass minorname to bus_config().
3895          */
3896         if (minorname && i_ddi_minorname_to_devtspectype(parent,
3897             minorname, &devt, &spectype) == DDI_FAILURE) {
3898                 (void) snprintf(config_name, MAXNAMELEN, "%s", minorname);
3899                 if (ndi_devi_config_obp_args(parent,
3900                     config_name, &child, 0) != NDI_SUCCESS) {
3901                         ndi_rele_devi(parent);
3902                         pn_free(&pn);
3903                         kmem_free(component, MAXNAMELEN);
3904                         kmem_free(config_name, MAXNAMELEN);
3905                         NDI_CONFIG_DEBUG((CE_NOTE,
3906                             "%s: minor node not found\n", pathname));
3907                         return (-1);
3908                 }
3909                 minorname = NULL;       /* look for default minor */
3910                 ASSERT(i_ddi_devi_attached(child));
3911                 ndi_rele_devi(parent);
3912                 parent = child;
3913         }
3914 
3915         if (devtp || spectypep) {
3916                 if (minorname == NULL) {
3917                         /*
3918                          * Search for a default entry with an active
3919                          * ndi_devi_enter to protect the devi_minor list.
3920                          */
3921                         ndi_devi_enter(parent, &circ);
3922                         for (dmn = DEVI(parent)->devi_minor; dmn;
3923                             dmn = dmn->next) {
3924                                 if (dmn->type == DDM_DEFAULT) {
3925                                         devt = dmn->ddm_dev;
3926                                         spectype = dmn->ddm_spec_type;
3927                                         break;
3928                                 }
3929                         }
3930 
3931                         if (devt == NODEV) {
3932                                 /*
3933                                  * No default minor node, try the first one;
3934                                  * else, assume 1-1 instance-minor mapping
3935                                  */
3936                                 dmn = DEVI(parent)->devi_minor;
3937                                 if (dmn && ((dmn->type == DDM_MINOR) ||
3938                                     (dmn->type == DDM_INTERNAL_PATH))) {
3939                                         devt = dmn->ddm_dev;
3940                                         spectype = dmn->ddm_spec_type;
3941                                 } else {
3942                                         devt = makedevice(
3943                                             DEVI(parent)->devi_major,
3944                                             ddi_get_instance(parent));
3945                                         spectype = S_IFCHR;
3946                                 }
3947                         }
3948                         ndi_devi_exit(parent, circ);
3949                 }
3950                 if (devtp)
3951                         *devtp = devt;
3952                 if (spectypep)
3953                         *spectypep = spectype;
3954         }
3955 
3956         pn_free(&pn);
3957         kmem_free(component, MAXNAMELEN);
3958         kmem_free(config_name, MAXNAMELEN);
3959 
3960         /*
3961          * If there is no error, return the appropriate parameters
3962          */
3963         if (dipp != NULL)
3964                 *dipp = parent;
3965         else {
3966                 /*
3967                  * We should really keep the ref count to keep the node from
3968                  * detaching but ddi_pathname_to_dev_t() specifies a NULL dipp,
3969                  * so we have no way of passing back the held dip.  Not holding
3970                  * the dip allows detaches to occur - which can cause problems
3971                  * for subsystems which call ddi_pathname_to_dev_t (console).
3972                  *
3973                  * Instead of holding the dip, we place a ddi-no-autodetach
3974                  * property on the node to prevent auto detaching.
3975                  *
3976                  * The right fix is to remove ddi_pathname_to_dev_t and replace
3977                  * it, and all references, with a call that specifies a dipp.
3978                  * In addition, the callers of this new interfaces would then
3979                  * need to call ndi_rele_devi when the reference is complete.
3980                  *
3981                  */
3982                 (void) ddi_prop_update_int(DDI_DEV_T_NONE, parent,
3983                     DDI_NO_AUTODETACH, 1);
3984                 ndi_rele_devi(parent);
3985         }
3986 
3987         return (0);
3988 }
3989 
3990 /*
3991  * Given the pathname of a device, return the dev_t of the corresponding
3992  * device.  Returns NODEV on failure.
3993  *
3994  * Note that this call sets the DDI_NO_AUTODETACH property on the devinfo node.
3995  */
3996 dev_t
3997 ddi_pathname_to_dev_t(char *pathname)
3998 {
3999         dev_t devt;
4000         int error;
4001 
4002         error = resolve_pathname(pathname, NULL, &devt, NULL);
4003 
4004         return (error ? NODEV : devt);
4005 }
4006 
4007 /*
4008  * Translate a prom pathname to kernel devfs pathname.
4009  * Caller is assumed to allocate devfspath memory of
4010  * size at least MAXPATHLEN
4011  *
4012  * The prom pathname may not include minor name, but
4013  * devfs pathname has a minor name portion.
4014  */
4015 int
4016 i_ddi_prompath_to_devfspath(char *prompath, char *devfspath)
4017 {
4018         dev_t           devt = (dev_t)NODEV;
4019         dev_info_t      *dip = NULL;
4020         char            *minor_name = NULL;
4021         int             spectype;
4022         int             error;
4023         int             circ;
4024 
4025         error = resolve_pathname(prompath, &dip, &devt, &spectype);
4026         if (error)
4027                 return (DDI_FAILURE);
4028         ASSERT(dip && devt != NODEV);
4029 
4030         /*
4031          * Get in-kernel devfs pathname
4032          */
4033         (void) ddi_pathname(dip, devfspath);
4034 
4035         ndi_devi_enter(dip, &circ);
4036         minor_name = i_ddi_devtspectype_to_minorname(dip, devt, spectype);
4037         if (minor_name) {
4038                 (void) strcat(devfspath, ":");
4039                 (void) strcat(devfspath, minor_name);
4040         } else {
4041                 /*
4042                  * If minor_name is NULL, we have an alias minor node.
4043                  * So manufacture a path to the corresponding clone minor.
4044                  */
4045                 (void) snprintf(devfspath, MAXPATHLEN, "%s:%s",
4046                     CLONE_PATH, ddi_driver_name(dip));
4047         }
4048         ndi_devi_exit(dip, circ);
4049 
4050         /* release hold from resolve_pathname() */
4051         ndi_rele_devi(dip);
4052         return (0);
4053 }
4054 
4055 /*
4056  * This function is intended to identify drivers that must quiesce for fast
4057  * reboot to succeed.  It does not claim to have more knowledge about the device
4058  * than its driver.  If a driver has implemented quiesce(), it will be invoked;
4059  * if a so identified driver does not manage any device that needs to be
4060  * quiesced, it must explicitly set its devo_quiesce dev_op to
4061  * ddi_quiesce_not_needed.
4062  */
4063 static int skip_pseudo = 1;     /* Skip pseudo devices */
4064 static int skip_non_hw = 1;     /* Skip devices with no hardware property */
4065 static int
4066 should_implement_quiesce(dev_info_t *dip)
4067 {
4068         struct dev_info *devi = DEVI(dip);
4069         dev_info_t *pdip;
4070 
4071         /*
4072          * If dip is pseudo and skip_pseudo is set, driver doesn't have to
4073          * implement quiesce().
4074          */
4075         if (skip_pseudo &&
4076             strncmp(ddi_binding_name(dip), "pseudo", sizeof ("pseudo")) == 0)
4077                 return (0);
4078 
4079         /*
4080          * If parent dip is pseudo and skip_pseudo is set, driver doesn't have
4081          * to implement quiesce().
4082          */
4083         if (skip_pseudo && (pdip = ddi_get_parent(dip)) != NULL &&
4084             strncmp(ddi_binding_name(pdip), "pseudo", sizeof ("pseudo")) == 0)
4085                 return (0);
4086 
4087         /*
4088          * If not attached, driver doesn't have to implement quiesce().
4089          */
4090         if (!i_ddi_devi_attached(dip))
4091                 return (0);
4092 
4093         /*
4094          * If dip has no hardware property and skip_non_hw is set,
4095          * driver doesn't have to implement quiesce().
4096          */
4097         if (skip_non_hw && devi->devi_hw_prop_ptr == NULL)
4098                 return (0);
4099 
4100         return (1);
4101 }
4102 
4103 static int
4104 driver_has_quiesce(struct dev_ops *ops)
4105 {
4106         if ((ops->devo_rev >= 4) && (ops->devo_quiesce != nodev) &&
4107             (ops->devo_quiesce != NULL) && (ops->devo_quiesce != nulldev) &&
4108             (ops->devo_quiesce != ddi_quiesce_not_supported))
4109                 return (1);
4110         else
4111                 return (0);
4112 }
4113 
4114 /*
4115  * Check to see if a driver has implemented the quiesce() DDI function.
4116  */
4117 int
4118 check_driver_quiesce(dev_info_t *dip, void *arg)
4119 {
4120         struct dev_ops *ops;
4121 
4122         if (!should_implement_quiesce(dip))
4123                 return (DDI_WALK_CONTINUE);
4124 
4125         if ((ops = ddi_get_driver(dip)) == NULL)
4126                 return (DDI_WALK_CONTINUE);
4127 
4128         if (driver_has_quiesce(ops)) {
4129                 if ((quiesce_debug & 0x2) == 0x2) {
4130                         if (ops->devo_quiesce == ddi_quiesce_not_needed)
4131                                 cmn_err(CE_CONT, "%s does not need to be "
4132                                     "quiesced", ddi_driver_name(dip));
4133                         else
4134                                 cmn_err(CE_CONT, "%s has quiesce routine",
4135                                     ddi_driver_name(dip));
4136                 }
4137         } else {
4138                 if (arg != NULL)
4139                         *((int *)arg) = -1;
4140                 cmn_err(CE_WARN, "%s has no quiesce()", ddi_driver_name(dip));
4141         }
4142 
4143         return (DDI_WALK_CONTINUE);
4144 }
4145 
4146 /*
4147  * Quiesce device.
4148  */
4149 static void
4150 quiesce_one_device(dev_info_t *dip, void *arg)
4151 {
4152         struct dev_ops *ops;
4153         int should_quiesce = 0;
4154 
4155         /*
4156          * If the device is not attached it doesn't need to be quiesced.
4157          */
4158         if (!i_ddi_devi_attached(dip))
4159                 return;
4160 
4161         if ((ops = ddi_get_driver(dip)) == NULL)
4162                 return;
4163 
4164         should_quiesce = should_implement_quiesce(dip);
4165 
4166         /*
4167          * If there's an implementation of quiesce(), always call it even if
4168          * some of the drivers don't have quiesce() or quiesce() have failed
4169          * so we can do force fast reboot.  The implementation of quiesce()
4170          * should not negatively affect a regular reboot.
4171          */
4172         if (driver_has_quiesce(ops)) {
4173                 int rc = DDI_SUCCESS;
4174 
4175                 if (ops->devo_quiesce == ddi_quiesce_not_needed)
4176                         return;
4177 
4178                 rc = devi_quiesce(dip);
4179 
4180                 if (rc != DDI_SUCCESS && should_quiesce) {
4181 #ifdef DEBUG
4182                         cmn_err(CE_WARN, "quiesce() failed for %s%d",
4183                             ddi_driver_name(dip), ddi_get_instance(dip));
4184 #endif /* DEBUG */
4185                         if (arg != NULL)
4186                                 *((int *)arg) = -1;
4187                 }
4188         } else if (should_quiesce && arg != NULL) {
4189                 *((int *)arg) = -1;
4190         }
4191 }
4192 
4193 /*
4194  * Traverse the dev info tree in a breadth-first manner so that we quiesce
4195  * children first.  All subtrees under the parent of dip will be quiesced.
4196  */
4197 void
4198 quiesce_devices(dev_info_t *dip, void *arg)
4199 {
4200         /*
4201          * if we're reached here, the device tree better not be changing.
4202          * so either devinfo_freeze better be set or we better be panicing.
4203          */
4204         ASSERT(devinfo_freeze || panicstr);
4205 
4206         for (; dip != NULL; dip = ddi_get_next_sibling(dip)) {
4207                 quiesce_devices(ddi_get_child(dip), arg);
4208 
4209                 quiesce_one_device(dip, arg);
4210         }
4211 }
4212 
4213 /*
4214  * Reset all the pure leaf drivers on the system at halt time
4215  */
4216 static int
4217 reset_leaf_device(dev_info_t *dip, void *arg)
4218 {
4219         _NOTE(ARGUNUSED(arg))
4220         struct dev_ops *ops;
4221 
4222         /* if the device doesn't need to be reset then there's nothing to do */
4223         if (!DEVI_NEED_RESET(dip))
4224                 return (DDI_WALK_CONTINUE);
4225 
4226         /*
4227          * if the device isn't a char/block device or doesn't have a
4228          * reset entry point then there's nothing to do.
4229          */
4230         ops = ddi_get_driver(dip);
4231         if ((ops == NULL) || (ops->devo_cb_ops == NULL) ||
4232             (ops->devo_reset == nodev) || (ops->devo_reset == nulldev) ||
4233             (ops->devo_reset == NULL))
4234                 return (DDI_WALK_CONTINUE);
4235 
4236         if (DEVI_IS_ATTACHING(dip) || DEVI_IS_DETACHING(dip)) {
4237                 static char path[MAXPATHLEN];
4238 
4239                 /*
4240                  * bad news, this device has blocked in it's attach or
4241                  * detach routine, which means it not safe to call it's
4242                  * devo_reset() entry point.
4243                  */
4244                 cmn_err(CE_WARN, "unable to reset device: %s",
4245                     ddi_pathname(dip, path));
4246                 return (DDI_WALK_CONTINUE);
4247         }
4248 
4249         NDI_CONFIG_DEBUG((CE_NOTE, "resetting %s%d\n",
4250             ddi_driver_name(dip), ddi_get_instance(dip)));
4251 
4252         (void) devi_reset(dip, DDI_RESET_FORCE);
4253         return (DDI_WALK_CONTINUE);
4254 }
4255 
4256 void
4257 reset_leaves(void)
4258 {
4259         /*
4260          * if we're reached here, the device tree better not be changing.
4261          * so either devinfo_freeze better be set or we better be panicing.
4262          */
4263         ASSERT(devinfo_freeze || panicstr);
4264 
4265         (void) walk_devs(top_devinfo, reset_leaf_device, NULL, 0);
4266 }
4267 
4268 
4269 /*
4270  * devtree_freeze() must be called before quiesce_devices() and reset_leaves()
4271  * during a normal system shutdown.  It attempts to ensure that there are no
4272  * outstanding attach or detach operations in progress when quiesce_devices() or
4273  * reset_leaves()is invoked.  It must be called before the system becomes
4274  * single-threaded because device attach and detach are multi-threaded
4275  * operations.  (note that during system shutdown the system doesn't actually
4276  * become single-thread since other threads still exist, but the shutdown thread
4277  * will disable preemption for itself, raise it's pil, and stop all the other
4278  * cpus in the system there by effectively making the system single-threaded.)
4279  */
4280 void
4281 devtree_freeze(void)
4282 {
4283         int delayed = 0;
4284 
4285         /* if we're panicing then the device tree isn't going to be changing */
4286         if (panicstr)
4287                 return;
4288 
4289         /* stop all dev_info state changes in the device tree */
4290         devinfo_freeze = gethrtime();
4291 
4292         /*
4293          * if we're not panicing and there are on-going attach or detach
4294          * operations, wait for up to 3 seconds for them to finish.  This
4295          * is a randomly chosen interval but this should be ok because:
4296          * - 3 seconds is very small relative to the deadman timer.
4297          * - normal attach and detach operations should be very quick.
4298          * - attach and detach operations are fairly rare.
4299          */
4300         while (!panicstr && atomic_add_long_nv(&devinfo_attach_detach, 0) &&
4301             (delayed < 3)) {
4302                 delayed += 1;
4303 
4304                 /* do a sleeping wait for one second */
4305                 ASSERT(!servicing_interrupt());
4306                 delay(drv_usectohz(MICROSEC));
4307         }
4308 }
4309 
4310 static int
4311 bind_dip(dev_info_t *dip, void *arg)
4312 {
4313         _NOTE(ARGUNUSED(arg))
4314         char    *path;
4315         major_t major, pmajor;
4316 
4317         /*
4318          * If the node is currently bound to the wrong driver, try to unbind
4319          * so that we can rebind to the correct driver.
4320          */
4321         if (i_ddi_node_state(dip) >= DS_BOUND) {
4322                 major = ddi_compatible_driver_major(dip, NULL);
4323                 if ((DEVI(dip)->devi_major == major) &&
4324                     (i_ddi_node_state(dip) >= DS_INITIALIZED)) {
4325                         /*
4326                          * Check for a path-oriented driver alias that
4327                          * takes precedence over current driver binding.
4328                          */
4329                         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4330                         (void) ddi_pathname(dip, path);
4331                         pmajor = ddi_name_to_major(path);
4332                         if (driver_active(pmajor))
4333                                 major = pmajor;
4334                         kmem_free(path, MAXPATHLEN);
4335                 }
4336 
4337                 /* attempt unbind if current driver is incorrect */
4338                 if (driver_active(major) &&
4339                     (major != DEVI(dip)->devi_major))
4340                         (void) ndi_devi_unbind_driver(dip);
4341         }
4342 
4343         /* If unbound, try to bind to a driver */
4344         if (i_ddi_node_state(dip) < DS_BOUND)
4345                 (void) ndi_devi_bind_driver(dip, 0);
4346 
4347         return (DDI_WALK_CONTINUE);
4348 }
4349 
4350 void
4351 i_ddi_bind_devs(void)
4352 {
4353         /* flush devfs so that ndi_devi_unbind_driver will work when possible */
4354         (void) devfs_clean(top_devinfo, NULL, 0);
4355 
4356         ddi_walk_devs(top_devinfo, bind_dip, (void *)NULL);
4357 }
4358 
4359 /* callback data for unbind_children_by_alias() */
4360 typedef struct unbind_data {
4361         major_t drv_major;
4362         char    *drv_alias;
4363         int     ndevs_bound;
4364         int     unbind_errors;
4365 } unbind_data_t;
4366 
4367 /*
4368  * A utility function provided for testing and support convenience
4369  * Called for each device during an upgrade_drv -d bound to the alias
4370  * that cannot be unbound due to device in use.
4371  */
4372 static void
4373 unbind_alias_dev_in_use(dev_info_t *dip, char *alias)
4374 {
4375         if (moddebug & MODDEBUG_BINDING) {
4376                 cmn_err(CE_CONT, "%s%d: state %d: bound to %s\n",
4377                     ddi_driver_name(dip), ddi_get_instance(dip),
4378                     i_ddi_node_state(dip), alias);
4379         }
4380 }
4381 
4382 /*
4383  * walkdevs callback for unbind devices bound to specific driver
4384  * and alias.  Invoked within the context of update_drv -d <alias>.
4385  */
4386 static int
4387 unbind_children_by_alias(dev_info_t *dip, void *arg)
4388 {
4389         int             circ;
4390         dev_info_t      *cdip;
4391         dev_info_t      *next;
4392         unbind_data_t   *ub = (unbind_data_t *)(uintptr_t)arg;
4393         int             rv;
4394 
4395         /*
4396          * We are called from update_drv to try to unbind a specific
4397          * set of aliases for a driver.  Unbind what persistent nodes
4398          * we can, and return the number of nodes which cannot be unbound.
4399          * If not all nodes can be unbound, update_drv leaves the
4400          * state of the driver binding files unchanged, except in
4401          * the case of -f.
4402          */
4403         ndi_devi_enter(dip, &circ);
4404         for (cdip = ddi_get_child(dip); cdip; cdip = next) {
4405                 next = ddi_get_next_sibling(cdip);
4406                 if ((ddi_driver_major(cdip) != ub->drv_major) ||
4407                     (strcmp(DEVI(cdip)->devi_node_name, ub->drv_alias) != 0))
4408                         continue;
4409                 if (i_ddi_node_state(cdip) >= DS_BOUND) {
4410                         rv = ndi_devi_unbind_driver(cdip);
4411                         if (rv != DDI_SUCCESS ||
4412                             (i_ddi_node_state(cdip) >= DS_BOUND)) {
4413                                 unbind_alias_dev_in_use(cdip, ub->drv_alias);
4414                                 ub->ndevs_bound++;
4415                                 continue;
4416                         }
4417                         if (ndi_dev_is_persistent_node(cdip) == 0)
4418                                 (void) ddi_remove_child(cdip, 0);
4419                 }
4420         }
4421         ndi_devi_exit(dip, circ);
4422 
4423         return (DDI_WALK_CONTINUE);
4424 }
4425 
4426 /*
4427  * Unbind devices by driver & alias
4428  * Context: update_drv [-f] -d -i <alias> <driver>
4429  */
4430 int
4431 i_ddi_unbind_devs_by_alias(major_t major, char *alias)
4432 {
4433         unbind_data_t   *ub;
4434         int             rv;
4435 
4436         ub = kmem_zalloc(sizeof (*ub), KM_SLEEP);
4437         ub->drv_major = major;
4438         ub->drv_alias = alias;
4439         ub->ndevs_bound = 0;
4440         ub->unbind_errors = 0;
4441 
4442         /* flush devfs so that ndi_devi_unbind_driver will work when possible */
4443         (void) devfs_clean(top_devinfo, NULL, 0);
4444         ddi_walk_devs(top_devinfo, unbind_children_by_alias,
4445             (void *)(uintptr_t)ub);
4446 
4447         /* return the number of devices remaining bound to the alias */
4448         rv = ub->ndevs_bound + ub->unbind_errors;
4449         kmem_free(ub, sizeof (*ub));
4450         return (rv);
4451 }
4452 
4453 /*
4454  * walkdevs callback for unbind devices by driver
4455  */
4456 static int
4457 unbind_children_by_driver(dev_info_t *dip, void *arg)
4458 {
4459         int             circ;
4460         dev_info_t      *cdip;
4461         dev_info_t      *next;
4462         major_t         major = (major_t)(uintptr_t)arg;
4463         int             rv;
4464 
4465         /*
4466          * We are called either from rem_drv or update_drv when reloading
4467          * a driver.conf file. In either case, we unbind persistent nodes
4468          * and destroy .conf nodes. In the case of rem_drv, this will be
4469          * the final state. In the case of update_drv,  i_ddi_bind_devs()
4470          * may be invoked later to re-enumerate (new) driver.conf rebind
4471          * persistent nodes.
4472          */
4473         ndi_devi_enter(dip, &circ);
4474         for (cdip = ddi_get_child(dip); cdip; cdip = next) {
4475                 next = ddi_get_next_sibling(cdip);
4476                 if (ddi_driver_major(cdip) != major)
4477                         continue;
4478                 if (i_ddi_node_state(cdip) >= DS_BOUND) {
4479                         rv = ndi_devi_unbind_driver(cdip);
4480                         if (rv == DDI_FAILURE ||
4481                             (i_ddi_node_state(cdip) >= DS_BOUND))
4482                                 continue;
4483                         if (ndi_dev_is_persistent_node(cdip) == 0)
4484                                 (void) ddi_remove_child(cdip, 0);
4485                 }
4486         }
4487         ndi_devi_exit(dip, circ);
4488 
4489         return (DDI_WALK_CONTINUE);
4490 }
4491 
4492 /*
4493  * Unbind devices by driver
4494  * Context: rem_drv or unload driver.conf
4495  */
4496 void
4497 i_ddi_unbind_devs(major_t major)
4498 {
4499         /* flush devfs so that ndi_devi_unbind_driver will work when possible */
4500         (void) devfs_clean(top_devinfo, NULL, 0);
4501         ddi_walk_devs(top_devinfo, unbind_children_by_driver,
4502             (void *)(uintptr_t)major);
4503 }
4504 
4505 /*
4506  * I/O Hotplug control
4507  */
4508 
4509 /*
4510  * create and attach a dev_info node from a .conf file spec
4511  */
4512 static void
4513 init_spec_child(dev_info_t *pdip, struct hwc_spec *specp, uint_t flags)
4514 {
4515         _NOTE(ARGUNUSED(flags))
4516         dev_info_t *dip;
4517         char *node_name;
4518 
4519         if (((node_name = specp->hwc_devi_name) == NULL) ||
4520             (ddi_name_to_major(node_name) == DDI_MAJOR_T_NONE)) {
4521                 char *tmp = node_name;
4522                 if (tmp == NULL)
4523                         tmp = "<none>";
4524                 cmn_err(CE_CONT,
4525                     "init_spec_child: parent=%s, bad spec (%s)\n",
4526                     ddi_node_name(pdip), tmp);
4527                 return;
4528         }
4529 
4530         dip = i_ddi_alloc_node(pdip, node_name, (pnode_t)DEVI_PSEUDO_NODEID,
4531             -1, specp->hwc_devi_sys_prop_ptr, KM_SLEEP);
4532 
4533         if (dip == NULL)
4534                 return;
4535 
4536         if (ddi_initchild(pdip, dip) != DDI_SUCCESS)
4537                 (void) ddi_remove_child(dip, 0);
4538 }
4539 
4540 /*
4541  * Lookup hwc specs from hash tables and make children from the spec
4542  * Because some .conf children are "merge" nodes, we also initialize
4543  * .conf children to merge properties onto hardware nodes.
4544  *
4545  * The pdip must be held busy.
4546  */
4547 int
4548 i_ndi_make_spec_children(dev_info_t *pdip, uint_t flags)
4549 {
4550         extern struct hwc_spec *hwc_get_child_spec(dev_info_t *, major_t);
4551         int                     circ;
4552         struct hwc_spec         *list, *spec;
4553 
4554         ndi_devi_enter(pdip, &circ);
4555         if (DEVI(pdip)->devi_flags & DEVI_MADE_CHILDREN) {
4556                 ndi_devi_exit(pdip, circ);
4557                 return (DDI_SUCCESS);
4558         }
4559 
4560         list = hwc_get_child_spec(pdip, DDI_MAJOR_T_NONE);
4561         for (spec = list; spec != NULL; spec = spec->hwc_next) {
4562                 init_spec_child(pdip, spec, flags);
4563         }
4564         hwc_free_spec_list(list);
4565 
4566         mutex_enter(&DEVI(pdip)->devi_lock);
4567         DEVI(pdip)->devi_flags |= DEVI_MADE_CHILDREN;
4568         mutex_exit(&DEVI(pdip)->devi_lock);
4569         ndi_devi_exit(pdip, circ);
4570         return (DDI_SUCCESS);
4571 }
4572 
4573 /*
4574  * Run initchild on all child nodes such that instance assignment
4575  * for multiport network cards are contiguous.
4576  *
4577  * The pdip must be held busy.
4578  */
4579 static void
4580 i_ndi_init_hw_children(dev_info_t *pdip, uint_t flags)
4581 {
4582         dev_info_t *dip;
4583 
4584         ASSERT(DEVI(pdip)->devi_flags & DEVI_MADE_CHILDREN);
4585 
4586         /* contiguous instance assignment */
4587         e_ddi_enter_instance();
4588         dip = ddi_get_child(pdip);
4589         while (dip) {
4590                 if (ndi_dev_is_persistent_node(dip))
4591                         (void) i_ndi_config_node(dip, DS_INITIALIZED, flags);
4592                 dip = ddi_get_next_sibling(dip);
4593         }
4594         e_ddi_exit_instance();
4595 }
4596 
4597 /*
4598  * report device status
4599  */
4600 static void
4601 i_ndi_devi_report_status_change(dev_info_t *dip, char *path)
4602 {
4603         char *status;
4604 
4605         if (!DEVI_NEED_REPORT(dip) ||
4606             (i_ddi_node_state(dip) < DS_INITIALIZED) ||
4607             ndi_dev_is_hidden_node(dip)) {
4608                 return;
4609         }
4610 
4611         /* Invalidate the devinfo snapshot cache */
4612         i_ddi_di_cache_invalidate();
4613 
4614         if (DEVI_IS_DEVICE_REMOVED(dip)) {
4615                 status = "removed";
4616         } else if (DEVI_IS_DEVICE_OFFLINE(dip)) {
4617                 status = "offline";
4618         } else if (DEVI_IS_DEVICE_DOWN(dip)) {
4619                 status = "down";
4620         } else if (DEVI_IS_BUS_QUIESCED(dip)) {
4621                 status = "quiesced";
4622         } else if (DEVI_IS_BUS_DOWN(dip)) {
4623                 status = "down";
4624         } else if (i_ddi_devi_attached(dip)) {
4625                 status = "online";
4626         } else {
4627                 status = "unknown";
4628         }
4629 
4630         if (path == NULL) {
4631                 path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4632                 cmn_err(CE_CONT, "?%s (%s%d) %s\n",
4633                     ddi_pathname(dip, path), ddi_driver_name(dip),
4634                     ddi_get_instance(dip), status);
4635                 kmem_free(path, MAXPATHLEN);
4636         } else {
4637                 cmn_err(CE_CONT, "?%s (%s%d) %s\n",
4638                     path, ddi_driver_name(dip),
4639                     ddi_get_instance(dip), status);
4640         }
4641 
4642         mutex_enter(&(DEVI(dip)->devi_lock));
4643         DEVI_REPORT_DONE(dip);
4644         mutex_exit(&(DEVI(dip)->devi_lock));
4645 }
4646 
4647 /*
4648  * log a notification that a dev_info node has been configured.
4649  */
4650 static int
4651 i_log_devfs_add_devinfo(dev_info_t *dip, uint_t flags)
4652 {
4653         int                     se_err;
4654         char                    *pathname;
4655         sysevent_t              *ev;
4656         sysevent_id_t           eid;
4657         sysevent_value_t        se_val;
4658         sysevent_attr_list_t    *ev_attr_list = NULL;
4659         char                    *class_name;
4660         int                     no_transport = 0;
4661 
4662         ASSERT(dip && ddi_get_parent(dip) &&
4663             DEVI_BUSY_OWNED(ddi_get_parent(dip)));
4664 
4665         /* do not generate ESC_DEVFS_DEVI_ADD event during boot */
4666         if (!i_ddi_io_initialized())
4667                 return (DDI_SUCCESS);
4668 
4669         /* Invalidate the devinfo snapshot cache */
4670         i_ddi_di_cache_invalidate();
4671 
4672         ev = sysevent_alloc(EC_DEVFS, ESC_DEVFS_DEVI_ADD, EP_DDI, SE_SLEEP);
4673 
4674         pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4675 
4676         (void) ddi_pathname(dip, pathname);
4677         ASSERT(strlen(pathname));
4678 
4679         se_val.value_type = SE_DATA_TYPE_STRING;
4680         se_val.value.sv_string = pathname;
4681         if (sysevent_add_attr(&ev_attr_list, DEVFS_PATHNAME,
4682             &se_val, SE_SLEEP) != 0) {
4683                 goto fail;
4684         }
4685 
4686         /* add the device class attribute */
4687         if ((class_name = i_ddi_devi_class(dip)) != NULL) {
4688                 se_val.value_type = SE_DATA_TYPE_STRING;
4689                 se_val.value.sv_string = class_name;
4690 
4691                 if (sysevent_add_attr(&ev_attr_list,
4692                     DEVFS_DEVI_CLASS, &se_val, SE_SLEEP) != 0) {
4693                         sysevent_free_attr(ev_attr_list);
4694                         goto fail;
4695                 }
4696         }
4697 
4698         /*
4699          * must log a branch event too unless NDI_BRANCH_EVENT_OP is set,
4700          * in which case the branch event will be logged by the caller
4701          * after the entire branch has been configured.
4702          */
4703         if ((flags & NDI_BRANCH_EVENT_OP) == 0) {
4704                 /*
4705                  * Instead of logging a separate branch event just add
4706                  * DEVFS_BRANCH_EVENT attribute. It indicates devfsadmd to
4707                  * generate a EC_DEV_BRANCH event.
4708                  */
4709                 se_val.value_type = SE_DATA_TYPE_INT32;
4710                 se_val.value.sv_int32 = 1;
4711                 if (sysevent_add_attr(&ev_attr_list,
4712                     DEVFS_BRANCH_EVENT, &se_val, SE_SLEEP) != 0) {
4713                         sysevent_free_attr(ev_attr_list);
4714                         goto fail;
4715                 }
4716         }
4717 
4718         if (sysevent_attach_attributes(ev, ev_attr_list) != 0) {
4719                 sysevent_free_attr(ev_attr_list);
4720                 goto fail;
4721         }
4722 
4723         if ((se_err = log_sysevent(ev, SE_SLEEP, &eid)) != 0) {
4724                 if (se_err == SE_NO_TRANSPORT)
4725                         no_transport = 1;
4726                 goto fail;
4727         }
4728 
4729         sysevent_free(ev);
4730         kmem_free(pathname, MAXPATHLEN);
4731 
4732         return (DDI_SUCCESS);
4733 
4734 fail:
4735         cmn_err(CE_WARN, "failed to log ESC_DEVFS_DEVI_ADD event for %s%s",
4736             pathname, (no_transport) ? " (syseventd not responding)" : "");
4737 
4738         cmn_err(CE_WARN, "/dev may not be current for driver %s. "
4739             "Run devfsadm -i %s",
4740             ddi_driver_name(dip), ddi_driver_name(dip));
4741 
4742         sysevent_free(ev);
4743         kmem_free(pathname, MAXPATHLEN);
4744         return (DDI_SUCCESS);
4745 }
4746 
4747 /*
4748  * log a notification that a dev_info node has been unconfigured.
4749  */
4750 static int
4751 i_log_devfs_remove_devinfo(char *pathname, char *class_name, char *driver_name,
4752     int instance, uint_t flags)
4753 {
4754         sysevent_t              *ev;
4755         sysevent_id_t           eid;
4756         sysevent_value_t        se_val;
4757         sysevent_attr_list_t    *ev_attr_list = NULL;
4758         int                     se_err;
4759         int                     no_transport = 0;
4760 
4761         if (!i_ddi_io_initialized())
4762                 return (DDI_SUCCESS);
4763 
4764         /* Invalidate the devinfo snapshot cache */
4765         i_ddi_di_cache_invalidate();
4766 
4767         ev = sysevent_alloc(EC_DEVFS, ESC_DEVFS_DEVI_REMOVE, EP_DDI, SE_SLEEP);
4768 
4769         se_val.value_type = SE_DATA_TYPE_STRING;
4770         se_val.value.sv_string = pathname;
4771         if (sysevent_add_attr(&ev_attr_list, DEVFS_PATHNAME,
4772             &se_val, SE_SLEEP) != 0) {
4773                 goto fail;
4774         }
4775 
4776         if (class_name) {
4777                 /* add the device class, driver name and instance attributes */
4778 
4779                 se_val.value_type = SE_DATA_TYPE_STRING;
4780                 se_val.value.sv_string = class_name;
4781                 if (sysevent_add_attr(&ev_attr_list,
4782                     DEVFS_DEVI_CLASS, &se_val, SE_SLEEP) != 0) {
4783                         sysevent_free_attr(ev_attr_list);
4784                         goto fail;
4785                 }
4786 
4787                 se_val.value_type = SE_DATA_TYPE_STRING;
4788                 se_val.value.sv_string = driver_name;
4789                 if (sysevent_add_attr(&ev_attr_list,
4790                     DEVFS_DRIVER_NAME, &se_val, SE_SLEEP) != 0) {
4791                         sysevent_free_attr(ev_attr_list);
4792                         goto fail;
4793                 }
4794 
4795                 se_val.value_type = SE_DATA_TYPE_INT32;
4796                 se_val.value.sv_int32 = instance;
4797                 if (sysevent_add_attr(&ev_attr_list,
4798                     DEVFS_INSTANCE, &se_val, SE_SLEEP) != 0) {
4799                         sysevent_free_attr(ev_attr_list);
4800                         goto fail;
4801                 }
4802         }
4803 
4804         /*
4805          * must log a branch event too unless NDI_BRANCH_EVENT_OP is set,
4806          * in which case the branch event will be logged by the caller
4807          * after the entire branch has been unconfigured.
4808          */
4809         if ((flags & NDI_BRANCH_EVENT_OP) == 0) {
4810                 /*
4811                  * Instead of logging a separate branch event just add
4812                  * DEVFS_BRANCH_EVENT attribute. It indicates devfsadmd to
4813                  * generate a EC_DEV_BRANCH event.
4814                  */
4815                 se_val.value_type = SE_DATA_TYPE_INT32;
4816                 se_val.value.sv_int32 = 1;
4817                 if (sysevent_add_attr(&ev_attr_list,
4818                     DEVFS_BRANCH_EVENT, &se_val, SE_SLEEP) != 0) {
4819                         sysevent_free_attr(ev_attr_list);
4820                         goto fail;
4821                 }
4822         }
4823 
4824         if (sysevent_attach_attributes(ev, ev_attr_list) != 0) {
4825                 sysevent_free_attr(ev_attr_list);
4826                 goto fail;
4827         }
4828 
4829         if ((se_err = log_sysevent(ev, SE_SLEEP, &eid)) != 0) {
4830                 if (se_err == SE_NO_TRANSPORT)
4831                         no_transport = 1;
4832                 goto fail;
4833         }
4834 
4835         sysevent_free(ev);
4836         return (DDI_SUCCESS);
4837 
4838 fail:
4839         sysevent_free(ev);
4840         cmn_err(CE_WARN, "failed to log ESC_DEVFS_DEVI_REMOVE event for %s%s",
4841             pathname, (no_transport) ? " (syseventd not responding)" : "");
4842         return (DDI_SUCCESS);
4843 }
4844 
4845 static void
4846 i_ddi_log_devfs_device_remove(dev_info_t *dip)
4847 {
4848         char    *path;
4849 
4850         ASSERT(dip && ddi_get_parent(dip) &&
4851             DEVI_BUSY_OWNED(ddi_get_parent(dip)));
4852         ASSERT(DEVI_IS_DEVICE_REMOVED(dip));
4853 
4854         ASSERT(i_ddi_node_state(dip) >= DS_INITIALIZED);
4855         if (i_ddi_node_state(dip) < DS_INITIALIZED)
4856                 return;
4857 
4858         /* Inform LDI_EV_DEVICE_REMOVE callbacks. */
4859         ldi_invoke_finalize(dip, DDI_DEV_T_ANY, 0, LDI_EV_DEVICE_REMOVE,
4860             LDI_EV_SUCCESS, NULL);
4861 
4862         /* Generate EC_DEVFS_DEVI_REMOVE sysevent. */
4863         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4864         (void) i_log_devfs_remove_devinfo(ddi_pathname(dip, path),
4865             i_ddi_devi_class(dip), (char *)ddi_driver_name(dip),
4866             ddi_get_instance(dip), 0);
4867         kmem_free(path, MAXPATHLEN);
4868 }
4869 
4870 static void
4871 i_ddi_log_devfs_device_insert(dev_info_t *dip)
4872 {
4873         ASSERT(dip && ddi_get_parent(dip) &&
4874             DEVI_BUSY_OWNED(ddi_get_parent(dip)));
4875         ASSERT(!DEVI_IS_DEVICE_REMOVED(dip));
4876 
4877         (void) i_log_devfs_add_devinfo(dip, 0);
4878 }
4879 
4880 
4881 /*
4882  * log an event that a dev_info branch has been configured or unconfigured.
4883  */
4884 static int
4885 i_log_devfs_branch(char *node_path, char *subclass)
4886 {
4887         int se_err;
4888         sysevent_t *ev;
4889         sysevent_id_t eid;
4890         sysevent_value_t se_val;
4891         sysevent_attr_list_t *ev_attr_list = NULL;
4892         int no_transport = 0;
4893 
4894         /* do not generate the event during boot */
4895         if (!i_ddi_io_initialized())
4896                 return (DDI_SUCCESS);
4897 
4898         /* Invalidate the devinfo snapshot cache */
4899         i_ddi_di_cache_invalidate();
4900 
4901         ev = sysevent_alloc(EC_DEVFS, subclass, EP_DDI, SE_SLEEP);
4902 
4903         se_val.value_type = SE_DATA_TYPE_STRING;
4904         se_val.value.sv_string = node_path;
4905 
4906         if (sysevent_add_attr(&ev_attr_list, DEVFS_PATHNAME,
4907             &se_val, SE_SLEEP) != 0) {
4908                 goto fail;
4909         }
4910 
4911         if (sysevent_attach_attributes(ev, ev_attr_list) != 0) {
4912                 sysevent_free_attr(ev_attr_list);
4913                 goto fail;
4914         }
4915 
4916         if ((se_err = log_sysevent(ev, SE_SLEEP, &eid)) != 0) {
4917                 if (se_err == SE_NO_TRANSPORT)
4918                         no_transport = 1;
4919                 goto fail;
4920         }
4921 
4922         sysevent_free(ev);
4923         return (DDI_SUCCESS);
4924 
4925 fail:
4926         cmn_err(CE_WARN, "failed to log %s branch event for %s%s",
4927             subclass, node_path,
4928             (no_transport) ? " (syseventd not responding)" : "");
4929 
4930         sysevent_free(ev);
4931         return (DDI_FAILURE);
4932 }
4933 
4934 /*
4935  * log an event that a dev_info tree branch has been configured.
4936  */
4937 static int
4938 i_log_devfs_branch_add(dev_info_t *dip)
4939 {
4940         char *node_path;
4941         int rv;
4942 
4943         node_path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
4944         (void) ddi_pathname(dip, node_path);
4945         rv = i_log_devfs_branch(node_path, ESC_DEVFS_BRANCH_ADD);
4946         kmem_free(node_path, MAXPATHLEN);
4947 
4948         return (rv);
4949 }
4950 
4951 /*
4952  * log an event that a dev_info tree branch has been unconfigured.
4953  */
4954 static int
4955 i_log_devfs_branch_remove(char *node_path)
4956 {
4957         return (i_log_devfs_branch(node_path, ESC_DEVFS_BRANCH_REMOVE));
4958 }
4959 
4960 /*
4961  * enqueue the dip's deviname on the branch event queue.
4962  */
4963 static struct brevq_node *
4964 brevq_enqueue(struct brevq_node **brevqp, dev_info_t *dip,
4965     struct brevq_node *child)
4966 {
4967         struct brevq_node *brn;
4968         char *deviname;
4969 
4970         deviname = kmem_alloc(MAXNAMELEN, KM_SLEEP);
4971         (void) ddi_deviname(dip, deviname);
4972 
4973         brn = kmem_zalloc(sizeof (*brn), KM_SLEEP);
4974         brn->brn_deviname = i_ddi_strdup(deviname, KM_SLEEP);
4975         kmem_free(deviname, MAXNAMELEN);
4976         brn->brn_child = child;
4977         brn->brn_sibling = *brevqp;
4978         *brevqp = brn;
4979 
4980         return (brn);
4981 }
4982 
4983 /*
4984  * free the memory allocated for the elements on the branch event queue.
4985  */
4986 static void
4987 free_brevq(struct brevq_node *brevq)
4988 {
4989         struct brevq_node *brn, *next_brn;
4990 
4991         for (brn = brevq; brn != NULL; brn = next_brn) {
4992                 next_brn = brn->brn_sibling;
4993                 ASSERT(brn->brn_child == NULL);
4994                 kmem_free(brn->brn_deviname, strlen(brn->brn_deviname) + 1);
4995                 kmem_free(brn, sizeof (*brn));
4996         }
4997 }
4998 
4999 /*
5000  * log the events queued up on the branch event queue and free the
5001  * associated memory.
5002  *
5003  * node_path must have been allocated with at least MAXPATHLEN bytes.
5004  */
5005 static void
5006 log_and_free_brevq(char *node_path, struct brevq_node *brevq)
5007 {
5008         struct brevq_node *brn;
5009         char *p;
5010 
5011         p = node_path + strlen(node_path);
5012         for (brn = brevq; brn != NULL; brn = brn->brn_sibling) {
5013                 (void) strcpy(p, brn->brn_deviname);
5014                 (void) i_log_devfs_branch_remove(node_path);
5015         }
5016         *p = '\0';
5017 
5018         free_brevq(brevq);
5019 }
5020 
5021 /*
5022  * log the events queued up on the branch event queue and free the
5023  * associated memory. Same as the previous function but operates on dip.
5024  */
5025 static void
5026 log_and_free_brevq_dip(dev_info_t *dip, struct brevq_node *brevq)
5027 {
5028         char *path;
5029 
5030         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5031         (void) ddi_pathname(dip, path);
5032         log_and_free_brevq(path, brevq);
5033         kmem_free(path, MAXPATHLEN);
5034 }
5035 
5036 /*
5037  * log the outstanding branch remove events for the grand children of the dip
5038  * and free the associated memory.
5039  */
5040 static void
5041 log_and_free_br_events_on_grand_children(dev_info_t *dip,
5042     struct brevq_node *brevq)
5043 {
5044         struct brevq_node *brn;
5045         char *path;
5046         char *p;
5047 
5048         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5049         (void) ddi_pathname(dip, path);
5050         p = path + strlen(path);
5051         for (brn = brevq; brn != NULL; brn = brn->brn_sibling) {
5052                 if (brn->brn_child) {
5053                         (void) strcpy(p, brn->brn_deviname);
5054                         /* now path contains the node path to the dip's child */
5055                         log_and_free_brevq(path, brn->brn_child);
5056                         brn->brn_child = NULL;
5057                 }
5058         }
5059         kmem_free(path, MAXPATHLEN);
5060 }
5061 
5062 /*
5063  * log and cleanup branch remove events for the grand children of the dip.
5064  */
5065 static void
5066 cleanup_br_events_on_grand_children(dev_info_t *dip, struct brevq_node **brevqp)
5067 {
5068         dev_info_t *child;
5069         struct brevq_node *brevq, *brn, *prev_brn, *next_brn;
5070         char *path;
5071         int circ;
5072 
5073         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5074         prev_brn = NULL;
5075         brevq = *brevqp;
5076 
5077         ndi_devi_enter(dip, &circ);
5078         for (brn = brevq; brn != NULL; brn = next_brn) {
5079                 next_brn = brn->brn_sibling;
5080                 for (child = ddi_get_child(dip); child != NULL;
5081                     child = ddi_get_next_sibling(child)) {
5082                         if (i_ddi_node_state(child) >= DS_INITIALIZED) {
5083                                 (void) ddi_deviname(child, path);
5084                                 if (strcmp(path, brn->brn_deviname) == 0)
5085                                         break;
5086                         }
5087                 }
5088 
5089                 if (child != NULL && !(DEVI_EVREMOVE(child))) {
5090                         /*
5091                          * Event state is not REMOVE. So branch remove event
5092                          * is not going be generated on brn->brn_child.
5093                          * If any branch remove events were queued up on
5094                          * brn->brn_child log them and remove the brn
5095                          * from the queue.
5096                          */
5097                         if (brn->brn_child) {
5098                                 (void) ddi_pathname(dip, path);
5099                                 (void) strcat(path, brn->brn_deviname);
5100                                 log_and_free_brevq(path, brn->brn_child);
5101                         }
5102 
5103                         if (prev_brn)
5104                                 prev_brn->brn_sibling = next_brn;
5105                         else
5106                                 *brevqp = next_brn;
5107 
5108                         kmem_free(brn->brn_deviname,
5109                             strlen(brn->brn_deviname) + 1);
5110                         kmem_free(brn, sizeof (*brn));
5111                 } else {
5112                         /*
5113                          * Free up the outstanding branch remove events
5114                          * queued on brn->brn_child since brn->brn_child
5115                          * itself is eligible for branch remove event.
5116                          */
5117                         if (brn->brn_child) {
5118                                 free_brevq(brn->brn_child);
5119                                 brn->brn_child = NULL;
5120                         }
5121                         prev_brn = brn;
5122                 }
5123         }
5124 
5125         ndi_devi_exit(dip, circ);
5126         kmem_free(path, MAXPATHLEN);
5127 }
5128 
5129 static int
5130 need_remove_event(dev_info_t *dip, int flags)
5131 {
5132         if ((flags & (NDI_NO_EVENT | NDI_AUTODETACH)) == 0 &&
5133             (flags & (NDI_DEVI_OFFLINE | NDI_UNCONFIG | NDI_DEVI_REMOVE)) &&
5134             !(DEVI_EVREMOVE(dip)))
5135                 return (1);
5136         else
5137                 return (0);
5138 }
5139 
5140 /*
5141  * Unconfigure children/descendants of the dip.
5142  *
5143  * If the operation involves a branch event NDI_BRANCH_EVENT_OP is set
5144  * through out the unconfiguration. On successful return *brevqp is set to
5145  * a queue of dip's child devinames for which branch remove events need
5146  * to be generated.
5147  */
5148 static int
5149 devi_unconfig_branch(dev_info_t *dip, dev_info_t **dipp, int flags,
5150     struct brevq_node **brevqp)
5151 {
5152         int rval;
5153 
5154         *brevqp = NULL;
5155 
5156         if ((!(flags & NDI_BRANCH_EVENT_OP)) && need_remove_event(dip, flags))
5157                 flags |= NDI_BRANCH_EVENT_OP;
5158 
5159         if (flags & NDI_BRANCH_EVENT_OP) {
5160                 rval = devi_unconfig_common(dip, dipp, flags, DDI_MAJOR_T_NONE,
5161                     brevqp);
5162 
5163                 if (rval != NDI_SUCCESS && (*brevqp)) {
5164                         log_and_free_brevq_dip(dip, *brevqp);
5165                         *brevqp = NULL;
5166                 }
5167         } else
5168                 rval = devi_unconfig_common(dip, dipp, flags, DDI_MAJOR_T_NONE,
5169                     NULL);
5170 
5171         return (rval);
5172 }
5173 
5174 /*
5175  * If the dip is already bound to a driver transition to DS_INITIALIZED
5176  * in order to generate an event in the case where the node was left in
5177  * DS_BOUND state since boot (never got attached) and the node is now
5178  * being offlined.
5179  */
5180 static void
5181 init_bound_node_ev(dev_info_t *pdip, dev_info_t *dip, int flags)
5182 {
5183         if (need_remove_event(dip, flags) &&
5184             i_ddi_node_state(dip) == DS_BOUND &&
5185             i_ddi_devi_attached(pdip) && !DEVI_IS_DEVICE_OFFLINE(dip))
5186                 (void) ddi_initchild(pdip, dip);
5187 }
5188 
5189 /*
5190  * attach a node/branch with parent already held busy
5191  */
5192 static int
5193 devi_attach_node(dev_info_t *dip, uint_t flags)
5194 {
5195         dev_info_t *pdip = ddi_get_parent(dip);
5196 
5197         ASSERT(pdip && DEVI_BUSY_OWNED(pdip));
5198 
5199         mutex_enter(&(DEVI(dip)->devi_lock));
5200         if (flags & NDI_DEVI_ONLINE) {
5201                 if (!i_ddi_devi_attached(dip))
5202                         DEVI_SET_REPORT(dip);
5203                 DEVI_SET_DEVICE_ONLINE(dip);
5204         }
5205         if (DEVI_IS_DEVICE_OFFLINE(dip)) {
5206                 mutex_exit(&(DEVI(dip)->devi_lock));
5207                 return (NDI_FAILURE);
5208         }
5209         mutex_exit(&(DEVI(dip)->devi_lock));
5210 
5211         if (i_ddi_attachchild(dip) != DDI_SUCCESS) {
5212                 mutex_enter(&(DEVI(dip)->devi_lock));
5213                 DEVI_SET_EVUNINIT(dip);
5214                 mutex_exit(&(DEVI(dip)->devi_lock));
5215 
5216                 if (ndi_dev_is_persistent_node(dip))
5217                         (void) ddi_uninitchild(dip);
5218                 else {
5219                         /*
5220                          * Delete .conf nodes and nodes that are not
5221                          * well formed.
5222                          */
5223                         (void) ddi_remove_child(dip, 0);
5224                 }
5225                 return (NDI_FAILURE);
5226         }
5227 
5228         i_ndi_devi_report_status_change(dip, NULL);
5229 
5230         /*
5231          * log an event, but not during devfs lookups in which case
5232          * NDI_NO_EVENT is set.
5233          */
5234         if ((flags & NDI_NO_EVENT) == 0 && !(DEVI_EVADD(dip))) {
5235                 (void) i_log_devfs_add_devinfo(dip, flags);
5236 
5237                 mutex_enter(&(DEVI(dip)->devi_lock));
5238                 DEVI_SET_EVADD(dip);
5239                 mutex_exit(&(DEVI(dip)->devi_lock));
5240         } else if (!(flags & NDI_NO_EVENT_STATE_CHNG)) {
5241                 mutex_enter(&(DEVI(dip)->devi_lock));
5242                 DEVI_SET_EVADD(dip);
5243                 mutex_exit(&(DEVI(dip)->devi_lock));
5244         }
5245 
5246         return (NDI_SUCCESS);
5247 }
5248 
5249 /* internal function to config immediate children */
5250 static int
5251 config_immediate_children(dev_info_t *pdip, uint_t flags, major_t major)
5252 {
5253         dev_info_t      *child, *next;
5254         int             circ;
5255 
5256         ASSERT(i_ddi_devi_attached(pdip));
5257 
5258         if (!NEXUS_DRV(ddi_get_driver(pdip)))
5259                 return (NDI_SUCCESS);
5260 
5261         NDI_CONFIG_DEBUG((CE_CONT,
5262             "config_immediate_children: %s%d (%p), flags=%x\n",
5263             ddi_driver_name(pdip), ddi_get_instance(pdip),
5264             (void *)pdip, flags));
5265 
5266         ndi_devi_enter(pdip, &circ);
5267 
5268         if (flags & NDI_CONFIG_REPROBE) {
5269                 mutex_enter(&DEVI(pdip)->devi_lock);
5270                 DEVI(pdip)->devi_flags &= ~DEVI_MADE_CHILDREN;
5271                 mutex_exit(&DEVI(pdip)->devi_lock);
5272         }
5273         (void) i_ndi_make_spec_children(pdip, flags);
5274         i_ndi_init_hw_children(pdip, flags);
5275 
5276         child = ddi_get_child(pdip);
5277         while (child) {
5278                 /* NOTE: devi_attach_node() may remove the dip */
5279                 next = ddi_get_next_sibling(child);
5280 
5281                 /*
5282                  * Configure all nexus nodes or leaf nodes with
5283                  * matching driver major
5284                  */
5285                 if ((major == DDI_MAJOR_T_NONE) ||
5286                     (major == ddi_driver_major(child)) ||
5287                     ((flags & NDI_CONFIG) && (is_leaf_node(child) == 0)))
5288                         (void) devi_attach_node(child, flags);
5289                 child = next;
5290         }
5291 
5292         ndi_devi_exit(pdip, circ);
5293 
5294         return (NDI_SUCCESS);
5295 }
5296 
5297 /* internal function to config grand children */
5298 static int
5299 config_grand_children(dev_info_t *pdip, uint_t flags, major_t major)
5300 {
5301         struct mt_config_handle *hdl;
5302 
5303         /* multi-threaded configuration of child nexus */
5304         hdl = mt_config_init(pdip, NULL, flags, major, MT_CONFIG_OP, NULL);
5305         mt_config_children(hdl);
5306 
5307         return (mt_config_fini(hdl));   /* wait for threads to exit */
5308 }
5309 
5310 /*
5311  * Common function for device tree configuration,
5312  * either BUS_CONFIG_ALL or BUS_CONFIG_DRIVER.
5313  * The NDI_CONFIG flag causes recursive configuration of
5314  * grandchildren, devfs usage should not recurse.
5315  */
5316 static int
5317 devi_config_common(dev_info_t *dip, int flags, major_t major)
5318 {
5319         int error;
5320         int (*f)();
5321 
5322         if (!i_ddi_devi_attached(dip))
5323                 return (NDI_FAILURE);
5324 
5325         if (pm_pre_config(dip, NULL) != DDI_SUCCESS)
5326                 return (NDI_FAILURE);
5327 
5328         if ((DEVI(dip)->devi_ops->devo_bus_ops == NULL) ||
5329             (DEVI(dip)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
5330             (f = DEVI(dip)->devi_ops->devo_bus_ops->bus_config) == NULL) {
5331                 error = config_immediate_children(dip, flags, major);
5332         } else {
5333                 /* call bus_config entry point */
5334                 ddi_bus_config_op_t bus_op = (major == DDI_MAJOR_T_NONE) ?
5335                     BUS_CONFIG_ALL : BUS_CONFIG_DRIVER;
5336                 error = (*f)(dip,
5337                     flags, bus_op, (void *)(uintptr_t)major, NULL, 0);
5338         }
5339 
5340         if (error) {
5341                 pm_post_config(dip, NULL);
5342                 return (error);
5343         }
5344 
5345         /*
5346          * Some callers, notably SCSI, need to mark the devfs cache
5347          * to be rebuilt together with the config operation.
5348          */
5349         if (flags & NDI_DEVFS_CLEAN)
5350                 (void) devfs_clean(dip, NULL, 0);
5351 
5352         if (flags & NDI_CONFIG)
5353                 (void) config_grand_children(dip, flags, major);
5354 
5355         pm_post_config(dip, NULL);
5356 
5357         return (NDI_SUCCESS);
5358 }
5359 
5360 /*
5361  * Framework entry point for BUS_CONFIG_ALL
5362  */
5363 int
5364 ndi_devi_config(dev_info_t *dip, int flags)
5365 {
5366         NDI_CONFIG_DEBUG((CE_CONT,
5367             "ndi_devi_config: par = %s%d (%p), flags = 0x%x\n",
5368             ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
5369 
5370         return (devi_config_common(dip, flags, DDI_MAJOR_T_NONE));
5371 }
5372 
5373 /*
5374  * Framework entry point for BUS_CONFIG_DRIVER, bound to major
5375  */
5376 int
5377 ndi_devi_config_driver(dev_info_t *dip, int flags, major_t major)
5378 {
5379         /* don't abuse this function */
5380         ASSERT(major != DDI_MAJOR_T_NONE);
5381 
5382         NDI_CONFIG_DEBUG((CE_CONT,
5383             "ndi_devi_config_driver: par = %s%d (%p), flags = 0x%x\n",
5384             ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
5385 
5386         return (devi_config_common(dip, flags, major));
5387 }
5388 
5389 /*
5390  * Called by nexus drivers to configure its children.
5391  */
5392 static int
5393 devi_config_one(dev_info_t *pdip, char *devnm, dev_info_t **cdipp,
5394     uint_t flags, clock_t timeout)
5395 {
5396         dev_info_t      *vdip = NULL;
5397         char            *drivername = NULL;
5398         int             find_by_addr = 0;
5399         char            *name, *addr;
5400         int             v_circ, p_circ;
5401         clock_t         end_time;       /* 60 sec */
5402         int             probed;
5403         dev_info_t      *cdip;
5404         mdi_pathinfo_t  *cpip;
5405 
5406         *cdipp = NULL;
5407 
5408         if (!NEXUS_DRV(ddi_get_driver(pdip)))
5409                 return (NDI_FAILURE);
5410 
5411         /* split name into "name@addr" parts */
5412         i_ddi_parse_name(devnm, &name, &addr, NULL);
5413 
5414         /*
5415          * If the nexus is a pHCI and we are not processing a pHCI from
5416          * mdi bus_config code then we need to know the vHCI.
5417          */
5418         if (MDI_PHCI(pdip))
5419                 vdip = mdi_devi_get_vdip(pdip);
5420 
5421         /*
5422          * We may have a genericname on a system that creates drivername
5423          * nodes (from .conf files).  Find the drivername by nodeid. If we
5424          * can't find a node with devnm as the node name then we search by
5425          * drivername.  This allows an implementation to supply a genericly
5426          * named boot path (disk) and locate drivename nodes (sd).  The
5427          * NDI_PROMNAME flag does not apply to /devices/pseudo paths.
5428          */
5429         if ((flags & NDI_PROMNAME) && (pdip != pseudo_dip)) {
5430                 drivername = child_path_to_driver(pdip, name, addr);
5431                 find_by_addr = 1;
5432         }
5433 
5434         /*
5435          * Determine end_time: This routine should *not* be called with a
5436          * constant non-zero timeout argument, the caller should be adjusting
5437          * the timeout argument relative to when it *started* its asynchronous
5438          * enumeration.
5439          */
5440         if (timeout > 0)
5441                 end_time = ddi_get_lbolt() + timeout;
5442 
5443         for (;;) {
5444                 /*
5445                  * For pHCI, enter (vHCI, pHCI) and search for pathinfo/client
5446                  * child - break out of for(;;) loop if child found.
5447                  * NOTE: Lock order for ndi_devi_enter is (vHCI, pHCI).
5448                  */
5449                 if (vdip) {
5450                         /* use mdi_devi_enter ordering */
5451                         ndi_devi_enter(vdip, &v_circ);
5452                         ndi_devi_enter(pdip, &p_circ);
5453                         cpip = mdi_pi_find(pdip, NULL, addr);
5454                         cdip = mdi_pi_get_client(cpip);
5455                         if (cdip)
5456                                 break;
5457                 } else
5458                         ndi_devi_enter(pdip, &p_circ);
5459 
5460                 /*
5461                  * When not a  vHCI or not all pHCI devices are required to
5462                  * enumerated under the vHCI (NDI_MDI_FALLBACK) search for
5463                  * devinfo child.
5464                  */
5465                 if ((vdip == NULL) || (flags & NDI_MDI_FALLBACK)) {
5466                         /* determine if .conf nodes already built */
5467                         probed = (DEVI(pdip)->devi_flags & DEVI_MADE_CHILDREN);
5468 
5469                         /*
5470                          * Search for child by name, if not found then search
5471                          * for a node bound to the drivername driver with the
5472                          * specified "@addr". Break out of for(;;) loop if
5473                          * child found.  To support path-oriented aliases
5474                          * binding on boot-device, we do a search_by_addr too.
5475                          */
5476 again:                  (void) i_ndi_make_spec_children(pdip, flags);
5477                         cdip = find_child_by_name(pdip, name, addr);
5478                         if ((cdip == NULL) && drivername)
5479                                 cdip = find_child_by_driver(pdip,
5480                                     drivername, addr);
5481                         if ((cdip == NULL) && find_by_addr)
5482                                 cdip = find_child_by_addr(pdip, addr);
5483                         if (cdip)
5484                                 break;
5485 
5486                         /*
5487                          * determine if we should reenumerate .conf nodes
5488                          * and look for child again.
5489                          */
5490                         if (probed &&
5491                             i_ddi_io_initialized() &&
5492                             (flags & NDI_CONFIG_REPROBE) &&
5493                             ((timeout <= 0) || (ddi_get_lbolt() >= end_time))) {
5494                                 probed = 0;
5495                                 mutex_enter(&DEVI(pdip)->devi_lock);
5496                                 DEVI(pdip)->devi_flags &= ~DEVI_MADE_CHILDREN;
5497                                 mutex_exit(&DEVI(pdip)->devi_lock);
5498                                 goto again;
5499                         }
5500                 }
5501 
5502                 /* break out of for(;;) if time expired */
5503                 if ((timeout <= 0) || (ddi_get_lbolt() >= end_time))
5504                         break;
5505 
5506                 /*
5507                  * Child not found, exit and wait for asynchronous enumeration
5508                  * to add child (or timeout). The addition of a new child (vhci
5509                  * or phci) requires the asynchronous enumeration thread to
5510                  * ndi_devi_enter/ndi_devi_exit. This exit will signal devi_cv
5511                  * and cause us to return from ndi_devi_exit_and_wait, after
5512                  * which we loop and search for the requested child again.
5513                  */
5514                 NDI_DEBUG(flags, (CE_CONT,
5515                     "%s%d: waiting for child %s@%s, timeout %ld",
5516                     ddi_driver_name(pdip), ddi_get_instance(pdip),
5517                     name, addr, timeout));
5518                 if (vdip) {
5519                         /*
5520                          * Mark vHCI for pHCI ndi_devi_exit broadcast.
5521                          */
5522                         mutex_enter(&DEVI(vdip)->devi_lock);
5523                         DEVI(vdip)->devi_flags |=
5524                             DEVI_PHCI_SIGNALS_VHCI;
5525                         mutex_exit(&DEVI(vdip)->devi_lock);
5526                         ndi_devi_exit(pdip, p_circ);
5527 
5528                         /*
5529                          * NB: There is a small race window from above
5530                          * ndi_devi_exit() of pdip to cv_wait() in
5531                          * ndi_devi_exit_and_wait() which can result in
5532                          * not immediately finding a new pHCI child
5533                          * of a pHCI that uses NDI_MDI_FAILBACK.
5534                          */
5535                         ndi_devi_exit_and_wait(vdip, v_circ, end_time);
5536                 } else {
5537                         ndi_devi_exit_and_wait(pdip, p_circ, end_time);
5538                 }
5539         }
5540 
5541         /* done with paddr, fixup i_ddi_parse_name '@'->'\0' change */
5542         if (addr && *addr != '\0')
5543                 *(addr - 1) = '@';
5544 
5545         /* attach and hold the child, returning pointer to child */
5546         if (cdip && (devi_attach_node(cdip, flags) == NDI_SUCCESS)) {
5547                 ndi_hold_devi(cdip);
5548                 *cdipp = cdip;
5549         }
5550 
5551         ndi_devi_exit(pdip, p_circ);
5552         if (vdip)
5553                 ndi_devi_exit(vdip, v_circ);
5554         return (*cdipp ? NDI_SUCCESS : NDI_FAILURE);
5555 }
5556 
5557 /*
5558  * Enumerate and attach a child specified by name 'devnm'.
5559  * Called by devfs lookup and DR to perform a BUS_CONFIG_ONE.
5560  * Note: devfs does not make use of NDI_CONFIG to configure
5561  * an entire branch.
5562  */
5563 int
5564 ndi_devi_config_one(dev_info_t *pdip, char *devnm, dev_info_t **dipp, int flags)
5565 {
5566         int error;
5567         int (*f)();
5568         char *nmdup;
5569         int duplen;
5570         int branch_event = 0;
5571 
5572         ASSERT(pdip);
5573         ASSERT(devnm);
5574         ASSERT(dipp);
5575         ASSERT(i_ddi_devi_attached(pdip));
5576 
5577         NDI_CONFIG_DEBUG((CE_CONT,
5578             "ndi_devi_config_one: par = %s%d (%p), child = %s\n",
5579             ddi_driver_name(pdip), ddi_get_instance(pdip),
5580             (void *)pdip, devnm));
5581 
5582         *dipp = NULL;
5583 
5584         if (pm_pre_config(pdip, devnm) != DDI_SUCCESS) {
5585                 cmn_err(CE_WARN, "preconfig failed: %s", devnm);
5586                 return (NDI_FAILURE);
5587         }
5588 
5589         if ((flags & (NDI_NO_EVENT | NDI_BRANCH_EVENT_OP)) == 0 &&
5590             (flags & NDI_CONFIG)) {
5591                 flags |= NDI_BRANCH_EVENT_OP;
5592                 branch_event = 1;
5593         }
5594 
5595         nmdup = strdup(devnm);
5596         duplen = strlen(devnm) + 1;
5597 
5598         if ((DEVI(pdip)->devi_ops->devo_bus_ops == NULL) ||
5599             (DEVI(pdip)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
5600             (f = DEVI(pdip)->devi_ops->devo_bus_ops->bus_config) == NULL) {
5601                 error = devi_config_one(pdip, devnm, dipp, flags, 0);
5602         } else {
5603                 /* call bus_config entry point */
5604                 error = (*f)(pdip, flags, BUS_CONFIG_ONE, (void *)devnm, dipp);
5605         }
5606 
5607         if (error) {
5608                 *dipp = NULL;
5609         }
5610 
5611         /*
5612          * if we fail to lookup and this could be an alias, lookup currdip
5613          * To prevent recursive lookups into the same hash table, only
5614          * do the currdip lookups once the hash table init is complete.
5615          * Use tsd so that redirection doesn't recurse
5616          */
5617         if (error) {
5618                 char *alias = kmem_alloc(MAXPATHLEN, KM_NOSLEEP);
5619                 if (alias == NULL) {
5620                         ddi_err(DER_PANIC, pdip, "alias alloc failed: %s",
5621                             nmdup);
5622                 }
5623                 (void) ddi_pathname(pdip, alias);
5624                 (void) strlcat(alias, "/", MAXPATHLEN);
5625                 (void) strlcat(alias, nmdup, MAXPATHLEN);
5626 
5627                 *dipp = ddi_alias_redirect(alias);
5628                 error = (*dipp ? NDI_SUCCESS : NDI_FAILURE);
5629 
5630                 kmem_free(alias, MAXPATHLEN);
5631         }
5632         kmem_free(nmdup, duplen);
5633 
5634         if (error || !(flags & NDI_CONFIG)) {
5635                 pm_post_config(pdip, devnm);
5636                 return (error);
5637         }
5638 
5639         /*
5640          * DR usage (i.e. call with NDI_CONFIG) recursively configures
5641          * grandchildren, performing a BUS_CONFIG_ALL from the node attached
5642          * by the BUS_CONFIG_ONE.
5643          */
5644         ASSERT(*dipp);
5645         error = devi_config_common(*dipp, flags, DDI_MAJOR_T_NONE);
5646 
5647         pm_post_config(pdip, devnm);
5648 
5649         if (branch_event)
5650                 (void) i_log_devfs_branch_add(*dipp);
5651 
5652         return (error);
5653 }
5654 
5655 /*
5656  * Enumerate and attach a child specified by name 'devnm'.
5657  * Called during configure the OBP options. This configures
5658  * only one node.
5659  */
5660 static int
5661 ndi_devi_config_obp_args(dev_info_t *parent, char *devnm,
5662     dev_info_t **childp, int flags)
5663 {
5664         int error;
5665         int (*f)();
5666 
5667         ASSERT(childp);
5668         ASSERT(i_ddi_devi_attached(parent));
5669 
5670         NDI_CONFIG_DEBUG((CE_CONT, "ndi_devi_config_obp_args: "
5671             "par = %s%d (%p), child = %s\n", ddi_driver_name(parent),
5672             ddi_get_instance(parent), (void *)parent, devnm));
5673 
5674         if ((DEVI(parent)->devi_ops->devo_bus_ops == NULL) ||
5675             (DEVI(parent)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
5676             (f = DEVI(parent)->devi_ops->devo_bus_ops->bus_config) == NULL) {
5677                 error = NDI_FAILURE;
5678         } else {
5679                 /* call bus_config entry point */
5680                 error = (*f)(parent, flags,
5681                     BUS_CONFIG_OBP_ARGS, (void *)devnm, childp);
5682         }
5683         return (error);
5684 }
5685 
5686 /*
5687  * Pay attention, the following is a bit tricky:
5688  * There are three possible cases when constraints are applied
5689  *
5690  *      - A constraint is applied and the offline is disallowed.
5691  *        Simply return failure and block the offline
5692  *
5693  *      - A constraint is applied and the offline is allowed.
5694  *        Mark the dip as having passed the constraint and allow
5695  *        offline to proceed.
5696  *
5697  *      - A constraint is not applied. Allow the offline to proceed for now.
5698  *
5699  * In the latter two cases we allow the offline to proceed. If the
5700  * offline succeeds (no users) everything is fine. It is ok for an unused
5701  * device to be offlined even if no constraints were imposed on the offline.
5702  * If the offline fails because there are users, we look at the constraint
5703  * flag on the dip. If the constraint flag is set (implying that it passed
5704  * a constraint) we allow the dip to be retired. If not, we don't allow
5705  * the retire. This ensures that we don't allow unconstrained retire.
5706  */
5707 int
5708 e_ddi_offline_notify(dev_info_t *dip)
5709 {
5710         int retval;
5711         int constraint;
5712         int failure;
5713 
5714         RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): entered: dip=%p",
5715             (void *) dip));
5716 
5717         constraint = 0;
5718         failure = 0;
5719 
5720         /*
5721          * Start with userland constraints first - applied via device contracts
5722          */
5723         retval = contract_device_offline(dip, DDI_DEV_T_ANY, 0);
5724         switch (retval) {
5725         case CT_NACK:
5726                 RIO_DEBUG((CE_NOTE, "Received NACK for dip=%p", (void *)dip));
5727                 failure = 1;
5728                 goto out;
5729         case CT_ACK:
5730                 constraint = 1;
5731                 RIO_DEBUG((CE_NOTE, "Received ACK for dip=%p", (void *)dip));
5732                 break;
5733         case CT_NONE:
5734                 /* no contracts */
5735                 RIO_DEBUG((CE_NOTE, "No contracts on dip=%p", (void *)dip));
5736                 break;
5737         default:
5738                 ASSERT(retval == CT_NONE);
5739         }
5740 
5741         /*
5742          * Next, use LDI to impose kernel constraints
5743          */
5744         retval = ldi_invoke_notify(dip, DDI_DEV_T_ANY, 0, LDI_EV_OFFLINE, NULL);
5745         switch (retval) {
5746         case LDI_EV_FAILURE:
5747                 contract_device_negend(dip, DDI_DEV_T_ANY, 0, CT_EV_FAILURE);
5748                 RIO_DEBUG((CE_NOTE, "LDI callback failed on dip=%p",
5749                     (void *)dip));
5750                 failure = 1;
5751                 goto out;
5752         case LDI_EV_SUCCESS:
5753                 constraint = 1;
5754                 RIO_DEBUG((CE_NOTE, "LDI callback success on dip=%p",
5755                     (void *)dip));
5756                 break;
5757         case LDI_EV_NONE:
5758                 /* no matching LDI callbacks */
5759                 RIO_DEBUG((CE_NOTE, "No LDI callbacks for dip=%p",
5760                     (void *)dip));
5761                 break;
5762         default:
5763                 ASSERT(retval == LDI_EV_NONE);
5764         }
5765 
5766 out:
5767         mutex_enter(&(DEVI(dip)->devi_lock));
5768         if ((DEVI(dip)->devi_flags & DEVI_RETIRING) && failure) {
5769                 RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): setting "
5770                     "BLOCKED flag. dip=%p", (void *)dip));
5771                 DEVI(dip)->devi_flags |= DEVI_R_BLOCKED;
5772                 if (DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT) {
5773                         RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): "
5774                             "blocked. clearing RCM CONSTRAINT flag. dip=%p",
5775                             (void *)dip));
5776                         DEVI(dip)->devi_flags &= ~DEVI_R_CONSTRAINT;
5777                 }
5778         } else if ((DEVI(dip)->devi_flags & DEVI_RETIRING) && constraint) {
5779                 RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): setting "
5780                     "CONSTRAINT flag. dip=%p", (void *)dip));
5781                 DEVI(dip)->devi_flags |= DEVI_R_CONSTRAINT;
5782         } else if ((DEVI(dip)->devi_flags & DEVI_RETIRING) &&
5783             ((DEVI(dip)->devi_ops != NULL &&
5784             DEVI(dip)->devi_ops->devo_bus_ops != NULL) ||
5785             DEVI(dip)->devi_ref == 0)) {
5786                 /* also allow retire if nexus or if device is not in use */
5787                 RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): device not in "
5788                     "use. Setting CONSTRAINT flag. dip=%p", (void *)dip));
5789                 DEVI(dip)->devi_flags |= DEVI_R_CONSTRAINT;
5790         } else {
5791                 /*
5792                  * Note: We cannot ASSERT here that DEVI_R_CONSTRAINT is
5793                  * not set, since other sources (such as RCM) may have
5794                  * set the flag.
5795                  */
5796                 RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): not setting "
5797                     "constraint flag. dip=%p", (void *)dip));
5798         }
5799         mutex_exit(&(DEVI(dip)->devi_lock));
5800 
5801 
5802         RIO_VERBOSE((CE_NOTE, "e_ddi_offline_notify(): exit: dip=%p",
5803             (void *) dip));
5804 
5805         return (failure ? DDI_FAILURE : DDI_SUCCESS);
5806 }
5807 
5808 void
5809 e_ddi_offline_finalize(dev_info_t *dip, int result)
5810 {
5811         RIO_DEBUG((CE_NOTE, "e_ddi_offline_finalize(): entry: result=%s, "
5812             "dip=%p", result == DDI_SUCCESS ? "SUCCESS" : "FAILURE",
5813             (void *)dip));
5814 
5815         contract_device_negend(dip, DDI_DEV_T_ANY, 0,  result == DDI_SUCCESS ?
5816             CT_EV_SUCCESS : CT_EV_FAILURE);
5817 
5818         ldi_invoke_finalize(dip, DDI_DEV_T_ANY, 0,
5819             LDI_EV_OFFLINE, result == DDI_SUCCESS ?
5820             LDI_EV_SUCCESS : LDI_EV_FAILURE, NULL);
5821 
5822         RIO_VERBOSE((CE_NOTE, "e_ddi_offline_finalize(): exit: dip=%p",
5823             (void *)dip));
5824 }
5825 
5826 void
5827 e_ddi_degrade_finalize(dev_info_t *dip)
5828 {
5829         RIO_DEBUG((CE_NOTE, "e_ddi_degrade_finalize(): entry: "
5830             "result always = DDI_SUCCESS, dip=%p", (void *)dip));
5831 
5832         contract_device_degrade(dip, DDI_DEV_T_ANY, 0);
5833         contract_device_negend(dip, DDI_DEV_T_ANY, 0, CT_EV_SUCCESS);
5834 
5835         ldi_invoke_finalize(dip, DDI_DEV_T_ANY, 0, LDI_EV_DEGRADE,
5836             LDI_EV_SUCCESS, NULL);
5837 
5838         RIO_VERBOSE((CE_NOTE, "e_ddi_degrade_finalize(): exit: dip=%p",
5839             (void *)dip));
5840 }
5841 
5842 void
5843 e_ddi_undegrade_finalize(dev_info_t *dip)
5844 {
5845         RIO_DEBUG((CE_NOTE, "e_ddi_undegrade_finalize(): entry: "
5846             "result always = DDI_SUCCESS, dip=%p", (void *)dip));
5847 
5848         contract_device_undegrade(dip, DDI_DEV_T_ANY, 0);
5849         contract_device_negend(dip, DDI_DEV_T_ANY, 0, CT_EV_SUCCESS);
5850 
5851         RIO_VERBOSE((CE_NOTE, "e_ddi_undegrade_finalize(): exit: dip=%p",
5852             (void *)dip));
5853 }
5854 
5855 /*
5856  * detach a node with parent already held busy
5857  */
5858 static int
5859 devi_detach_node(dev_info_t *dip, uint_t flags)
5860 {
5861         dev_info_t *pdip = ddi_get_parent(dip);
5862         int ret = NDI_SUCCESS;
5863         ddi_eventcookie_t cookie;
5864         char *path = NULL;
5865         char *class = NULL;
5866         char *driver = NULL;
5867         int instance = -1;
5868         int post_event = 0;
5869 
5870         ASSERT(pdip && DEVI_BUSY_OWNED(pdip));
5871 
5872         /*
5873          * Invoke notify if offlining
5874          */
5875         if (flags & NDI_DEVI_OFFLINE) {
5876                 RIO_DEBUG((CE_NOTE, "devi_detach_node: offlining dip=%p",
5877                     (void *)dip));
5878                 if (e_ddi_offline_notify(dip) != DDI_SUCCESS) {
5879                         RIO_DEBUG((CE_NOTE, "devi_detach_node: offline NACKed"
5880                             "dip=%p", (void *)dip));
5881                         return (NDI_FAILURE);
5882                 }
5883         }
5884 
5885         if (flags & NDI_POST_EVENT) {
5886                 if (i_ddi_devi_attached(pdip)) {
5887                         if (ddi_get_eventcookie(dip, DDI_DEVI_REMOVE_EVENT,
5888                             &cookie) == NDI_SUCCESS)
5889                                 (void) ndi_post_event(dip, dip, cookie, NULL);
5890                 }
5891         }
5892 
5893         /*
5894          * dv_mknod places a hold on the dev_info_t for each devfs node
5895          * created.  If we're to succeed in detaching this device, we must
5896          * first release all outstanding references held by devfs.
5897          */
5898         (void) devfs_clean(pdip, NULL, DV_CLEAN_FORCE);
5899 
5900         if (i_ddi_detachchild(dip, flags) != DDI_SUCCESS) {
5901                 if (flags & NDI_DEVI_OFFLINE) {
5902                         RIO_DEBUG((CE_NOTE, "devi_detach_node: offline failed."
5903                             " Calling e_ddi_offline_finalize with result=%d. "
5904                             "dip=%p", DDI_FAILURE, (void *)dip));
5905                         e_ddi_offline_finalize(dip, DDI_FAILURE);
5906                 }
5907                 return (NDI_FAILURE);
5908         }
5909 
5910         if (flags & NDI_DEVI_OFFLINE) {
5911                 RIO_DEBUG((CE_NOTE, "devi_detach_node: offline succeeded."
5912                     " Calling e_ddi_offline_finalize with result=%d, "
5913                     "dip=%p", DDI_SUCCESS, (void *)dip));
5914                 e_ddi_offline_finalize(dip, DDI_SUCCESS);
5915         }
5916 
5917         if (flags & NDI_AUTODETACH)
5918                 return (NDI_SUCCESS);
5919 
5920         /*
5921          * For DR, even bound nodes may need to have offline
5922          * flag set.
5923          */
5924         if (flags & NDI_DEVI_OFFLINE) {
5925                 mutex_enter(&(DEVI(dip)->devi_lock));
5926                 DEVI_SET_DEVICE_OFFLINE(dip);
5927                 mutex_exit(&(DEVI(dip)->devi_lock));
5928         }
5929 
5930         if (i_ddi_node_state(dip) == DS_INITIALIZED) {
5931                 struct dev_info *devi = DEVI(dip);
5932 
5933                 if (devi->devi_ev_path == NULL) {
5934                         devi->devi_ev_path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
5935                         (void) ddi_pathname(dip, devi->devi_ev_path);
5936                 }
5937                 if (flags & NDI_DEVI_OFFLINE)
5938                         i_ndi_devi_report_status_change(dip,
5939                             devi->devi_ev_path);
5940 
5941                 if (need_remove_event(dip, flags)) {
5942                         /*
5943                          * instance and path data are lost in call to
5944                          * ddi_uninitchild
5945                          */
5946                         devi->devi_ev_instance = ddi_get_instance(dip);
5947 
5948                         mutex_enter(&(DEVI(dip)->devi_lock));
5949                         DEVI_SET_EVREMOVE(dip);
5950                         mutex_exit(&(DEVI(dip)->devi_lock));
5951                 }
5952         }
5953 
5954         if (flags & (NDI_UNCONFIG | NDI_DEVI_REMOVE)) {
5955                 ret = ddi_uninitchild(dip);
5956                 if (ret == NDI_SUCCESS) {
5957                         /*
5958                          * Remove uninitialized pseudo nodes because
5959                          * system props are lost and the node cannot be
5960                          * reattached.
5961                          */
5962                         if (!ndi_dev_is_persistent_node(dip))
5963                                 flags |= NDI_DEVI_REMOVE;
5964 
5965                         if (flags & NDI_DEVI_REMOVE) {
5966                                 /*
5967                                  * NOTE: If there is a consumer of LDI events,
5968                                  * ddi_uninitchild above would have failed
5969                                  * because of active devi_ref from ldi_open().
5970                                  */
5971 
5972                                 if (DEVI_EVREMOVE(dip)) {
5973                                         path = i_ddi_strdup(
5974                                             DEVI(dip)->devi_ev_path,
5975                                             KM_SLEEP);
5976                                         class =
5977                                             i_ddi_strdup(i_ddi_devi_class(dip),
5978                                             KM_SLEEP);
5979                                         driver =
5980                                             i_ddi_strdup(
5981                                             (char *)ddi_driver_name(dip),
5982                                             KM_SLEEP);
5983                                         instance = DEVI(dip)->devi_ev_instance;
5984                                         post_event = 1;
5985                                 }
5986 
5987                                 ret = ddi_remove_child(dip, 0);
5988                                 if (post_event && ret == NDI_SUCCESS) {
5989                                         /* Generate EC_DEVFS_DEVI_REMOVE */
5990                                         (void) i_log_devfs_remove_devinfo(path,
5991                                             class, driver, instance, flags);
5992                                 }
5993                         }
5994 
5995                 }
5996         }
5997 
5998         if (path)
5999                 strfree(path);
6000         if (class)
6001                 strfree(class);
6002         if (driver)
6003                 strfree(driver);
6004 
6005         return (ret);
6006 }
6007 
6008 /*
6009  * unconfigure immediate children of bus nexus device
6010  */
6011 static int
6012 unconfig_immediate_children(
6013         dev_info_t *dip,
6014         dev_info_t **dipp,
6015         int flags,
6016         major_t major)
6017 {
6018         int rv = NDI_SUCCESS;
6019         int circ, vcirc;
6020         dev_info_t *child;
6021         dev_info_t *vdip = NULL;
6022         dev_info_t *next;
6023 
6024         ASSERT(dipp == NULL || *dipp == NULL);
6025 
6026         /*
6027          * Scan forward to see if we will be processing a pHCI child. If we
6028          * have a child that is a pHCI and vHCI and pHCI are not siblings then
6029          * enter vHCI before parent(pHCI) to prevent deadlock with mpxio
6030          * Client power management operations.
6031          */
6032         ndi_devi_enter(dip, &circ);
6033         for (child = ddi_get_child(dip); child;
6034             child = ddi_get_next_sibling(child)) {
6035                 /* skip same nodes we skip below */
6036                 if (((major != DDI_MAJOR_T_NONE) &&
6037                     (major != ddi_driver_major(child))) ||
6038                     ((flags & NDI_AUTODETACH) && !is_leaf_node(child)))
6039                         continue;
6040 
6041                 if (MDI_PHCI(child)) {
6042                         vdip = mdi_devi_get_vdip(child);
6043                         /*
6044                          * If vHCI and vHCI is not a sibling of pHCI
6045                          * then enter in (vHCI, parent(pHCI)) order.
6046                          */
6047                         if (vdip && (ddi_get_parent(vdip) != dip)) {
6048                                 ndi_devi_exit(dip, circ);
6049 
6050                                 /* use mdi_devi_enter ordering */
6051                                 ndi_devi_enter(vdip, &vcirc);
6052                                 ndi_devi_enter(dip, &circ);
6053                                 break;
6054                         } else
6055                                 vdip = NULL;
6056                 }
6057         }
6058 
6059         child = ddi_get_child(dip);
6060         while (child) {
6061                 next = ddi_get_next_sibling(child);
6062 
6063                 if ((major != DDI_MAJOR_T_NONE) &&
6064                     (major != ddi_driver_major(child))) {
6065                         child = next;
6066                         continue;
6067                 }
6068 
6069                 /* skip nexus nodes during autodetach */
6070                 if ((flags & NDI_AUTODETACH) && !is_leaf_node(child)) {
6071                         child = next;
6072                         continue;
6073                 }
6074 
6075                 if (devi_detach_node(child, flags) != NDI_SUCCESS) {
6076                         if (dipp && *dipp == NULL) {
6077                                 ndi_hold_devi(child);
6078                                 *dipp = child;
6079                         }
6080                         rv = NDI_FAILURE;
6081                 }
6082 
6083                 /*
6084                  * Continue upon failure--best effort algorithm
6085                  */
6086                 child = next;
6087         }
6088 
6089         ndi_devi_exit(dip, circ);
6090         if (vdip)
6091                 ndi_devi_exit(vdip, vcirc);
6092 
6093         return (rv);
6094 }
6095 
6096 /*
6097  * unconfigure grand children of bus nexus device
6098  */
6099 static int
6100 unconfig_grand_children(
6101         dev_info_t *dip,
6102         dev_info_t **dipp,
6103         int flags,
6104         major_t major,
6105         struct brevq_node **brevqp)
6106 {
6107         struct mt_config_handle *hdl;
6108 
6109         if (brevqp)
6110                 *brevqp = NULL;
6111 
6112         /* multi-threaded configuration of child nexus */
6113         hdl = mt_config_init(dip, dipp, flags, major, MT_UNCONFIG_OP, brevqp);
6114         mt_config_children(hdl);
6115 
6116         return (mt_config_fini(hdl));   /* wait for threads to exit */
6117 }
6118 
6119 /*
6120  * Unconfigure children/descendants of the dip.
6121  *
6122  * If brevqp is not NULL, on return *brevqp is set to a queue of dip's
6123  * child devinames for which branch remove events need to be generated.
6124  */
6125 static int
6126 devi_unconfig_common(
6127         dev_info_t *dip,
6128         dev_info_t **dipp,
6129         int flags,
6130         major_t major,
6131         struct brevq_node **brevqp)
6132 {
6133         int rv;
6134         int pm_cookie;
6135         int (*f)();
6136         ddi_bus_config_op_t bus_op;
6137 
6138         if (dipp)
6139                 *dipp = NULL;
6140         if (brevqp)
6141                 *brevqp = NULL;
6142 
6143         /*
6144          * Power up the dip if it is powered off.  If the flag bit
6145          * NDI_AUTODETACH is set and the dip is not at its full power,
6146          * skip the rest of the branch.
6147          */
6148         if (pm_pre_unconfig(dip, flags, &pm_cookie, NULL) != DDI_SUCCESS)
6149                 return ((flags & NDI_AUTODETACH) ? NDI_SUCCESS :
6150                     NDI_FAILURE);
6151 
6152         /*
6153          * Some callers, notably SCSI, need to clear out the devfs
6154          * cache together with the unconfig to prevent stale entries.
6155          */
6156         if (flags & NDI_DEVFS_CLEAN)
6157                 (void) devfs_clean(dip, NULL, 0);
6158 
6159         rv = unconfig_grand_children(dip, dipp, flags, major, brevqp);
6160 
6161         if ((rv != NDI_SUCCESS) && ((flags & NDI_AUTODETACH) == 0)) {
6162                 if (brevqp && *brevqp) {
6163                         log_and_free_br_events_on_grand_children(dip, *brevqp);
6164                         free_brevq(*brevqp);
6165                         *brevqp = NULL;
6166                 }
6167                 pm_post_unconfig(dip, pm_cookie, NULL);
6168                 return (rv);
6169         }
6170 
6171         if (dipp && *dipp) {
6172                 ndi_rele_devi(*dipp);
6173                 *dipp = NULL;
6174         }
6175 
6176         /*
6177          * It is possible to have a detached nexus with children
6178          * and grandchildren (for example: a branch consisting
6179          * entirely of bound nodes.) Since the nexus is detached
6180          * the bus_unconfig entry point cannot be used to remove
6181          * or unconfigure the descendants.
6182          */
6183         if (!i_ddi_devi_attached(dip) ||
6184             (DEVI(dip)->devi_ops->devo_bus_ops == NULL) ||
6185             (DEVI(dip)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
6186             (f = DEVI(dip)->devi_ops->devo_bus_ops->bus_unconfig) == NULL) {
6187                 rv = unconfig_immediate_children(dip, dipp, flags, major);
6188         } else {
6189                 /*
6190                  * call bus_unconfig entry point
6191                  * It should reset nexus flags if unconfigure succeeds.
6192                  */
6193                 bus_op = (major == DDI_MAJOR_T_NONE) ?
6194                     BUS_UNCONFIG_ALL : BUS_UNCONFIG_DRIVER;
6195                 rv = (*f)(dip, flags, bus_op, (void *)(uintptr_t)major);
6196         }
6197 
6198         pm_post_unconfig(dip, pm_cookie, NULL);
6199 
6200         if (brevqp && *brevqp)
6201                 cleanup_br_events_on_grand_children(dip, brevqp);
6202 
6203         return (rv);
6204 }
6205 
6206 /*
6207  * called by devfs/framework to unconfigure children bound to major
6208  * If NDI_AUTODETACH is specified, this is invoked by either the
6209  * moduninstall daemon or the modunload -i 0 command.
6210  */
6211 int
6212 ndi_devi_unconfig_driver(dev_info_t *dip, int flags, major_t major)
6213 {
6214         NDI_CONFIG_DEBUG((CE_CONT,
6215             "ndi_devi_unconfig_driver: par = %s%d (%p), flags = 0x%x\n",
6216             ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
6217 
6218         return (devi_unconfig_common(dip, NULL, flags, major, NULL));
6219 }
6220 
6221 int
6222 ndi_devi_unconfig(dev_info_t *dip, int flags)
6223 {
6224         NDI_CONFIG_DEBUG((CE_CONT,
6225             "ndi_devi_unconfig: par = %s%d (%p), flags = 0x%x\n",
6226             ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
6227 
6228         return (devi_unconfig_common(dip, NULL, flags, DDI_MAJOR_T_NONE, NULL));
6229 }
6230 
6231 int
6232 e_ddi_devi_unconfig(dev_info_t *dip, dev_info_t **dipp, int flags)
6233 {
6234         NDI_CONFIG_DEBUG((CE_CONT,
6235             "e_ddi_devi_unconfig: par = %s%d (%p), flags = 0x%x\n",
6236             ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip, flags));
6237 
6238         return (devi_unconfig_common(dip, dipp, flags, DDI_MAJOR_T_NONE, NULL));
6239 }
6240 
6241 /*
6242  * Unconfigure child by name
6243  */
6244 static int
6245 devi_unconfig_one(dev_info_t *pdip, char *devnm, int flags)
6246 {
6247         int             rv, circ;
6248         dev_info_t      *child;
6249         dev_info_t      *vdip = NULL;
6250         int             v_circ;
6251 
6252         ndi_devi_enter(pdip, &circ);
6253         child = ndi_devi_findchild(pdip, devnm);
6254 
6255         /*
6256          * If child is pHCI and vHCI and pHCI are not siblings then enter vHCI
6257          * before parent(pHCI) to avoid deadlock with mpxio Client power
6258          * management operations.
6259          */
6260         if (child && MDI_PHCI(child)) {
6261                 vdip = mdi_devi_get_vdip(child);
6262                 if (vdip && (ddi_get_parent(vdip) != pdip)) {
6263                         ndi_devi_exit(pdip, circ);
6264 
6265                         /* use mdi_devi_enter ordering */
6266                         ndi_devi_enter(vdip, &v_circ);
6267                         ndi_devi_enter(pdip, &circ);
6268                         child = ndi_devi_findchild(pdip, devnm);
6269                 } else
6270                         vdip = NULL;
6271         }
6272 
6273         if (child) {
6274                 rv = devi_detach_node(child, flags);
6275         } else {
6276                 NDI_CONFIG_DEBUG((CE_CONT,
6277                     "devi_unconfig_one: %s not found\n", devnm));
6278                 rv = NDI_SUCCESS;
6279         }
6280 
6281         ndi_devi_exit(pdip, circ);
6282         if (vdip)
6283                 ndi_devi_exit(vdip, v_circ);
6284 
6285         return (rv);
6286 }
6287 
6288 int
6289 ndi_devi_unconfig_one(
6290         dev_info_t *pdip,
6291         char *devnm,
6292         dev_info_t **dipp,
6293         int flags)
6294 {
6295         int             (*f)();
6296         int             circ, rv;
6297         int             pm_cookie;
6298         dev_info_t      *child;
6299         dev_info_t      *vdip = NULL;
6300         int             v_circ;
6301         struct brevq_node *brevq = NULL;
6302 
6303         ASSERT(i_ddi_devi_attached(pdip));
6304 
6305         NDI_CONFIG_DEBUG((CE_CONT,
6306             "ndi_devi_unconfig_one: par = %s%d (%p), child = %s\n",
6307             ddi_driver_name(pdip), ddi_get_instance(pdip),
6308             (void *)pdip, devnm));
6309 
6310         if (pm_pre_unconfig(pdip, flags, &pm_cookie, devnm) != DDI_SUCCESS)
6311                 return (NDI_FAILURE);
6312 
6313         if (dipp)
6314                 *dipp = NULL;
6315 
6316         ndi_devi_enter(pdip, &circ);
6317         child = ndi_devi_findchild(pdip, devnm);
6318 
6319         /*
6320          * If child is pHCI and vHCI and pHCI are not siblings then enter vHCI
6321          * before parent(pHCI) to avoid deadlock with mpxio Client power
6322          * management operations.
6323          */
6324         if (child && MDI_PHCI(child)) {
6325                 vdip = mdi_devi_get_vdip(child);
6326                 if (vdip && (ddi_get_parent(vdip) != pdip)) {
6327                         ndi_devi_exit(pdip, circ);
6328 
6329                         /* use mdi_devi_enter ordering */
6330                         ndi_devi_enter(vdip, &v_circ);
6331                         ndi_devi_enter(pdip, &circ);
6332                         child = ndi_devi_findchild(pdip, devnm);
6333                 } else
6334                         vdip = NULL;
6335         }
6336 
6337         if (child == NULL) {
6338                 NDI_CONFIG_DEBUG((CE_CONT, "ndi_devi_unconfig_one: %s"
6339                     " not found\n", devnm));
6340                 rv = NDI_SUCCESS;
6341                 goto out;
6342         }
6343 
6344         /*
6345          * Unconfigure children/descendants of named child
6346          */
6347         rv = devi_unconfig_branch(child, dipp, flags | NDI_UNCONFIG, &brevq);
6348         if (rv != NDI_SUCCESS)
6349                 goto out;
6350 
6351         init_bound_node_ev(pdip, child, flags);
6352 
6353         if ((DEVI(pdip)->devi_ops->devo_bus_ops == NULL) ||
6354             (DEVI(pdip)->devi_ops->devo_bus_ops->busops_rev < BUSO_REV_5) ||
6355             (f = DEVI(pdip)->devi_ops->devo_bus_ops->bus_unconfig) == NULL) {
6356                 rv = devi_detach_node(child, flags);
6357         } else {
6358                 /* call bus_config entry point */
6359                 rv = (*f)(pdip, flags, BUS_UNCONFIG_ONE, (void *)devnm);
6360         }
6361 
6362         if (brevq) {
6363                 if (rv != NDI_SUCCESS)
6364                         log_and_free_brevq_dip(child, brevq);
6365                 else
6366                         free_brevq(brevq);
6367         }
6368 
6369         if (dipp && rv != NDI_SUCCESS) {
6370                 ndi_hold_devi(child);
6371                 ASSERT(*dipp == NULL);
6372                 *dipp = child;
6373         }
6374 
6375 out:
6376         ndi_devi_exit(pdip, circ);
6377         if (vdip)
6378                 ndi_devi_exit(vdip, v_circ);
6379 
6380         pm_post_unconfig(pdip, pm_cookie, devnm);
6381 
6382         return (rv);
6383 }
6384 
6385 struct async_arg {
6386         dev_info_t *dip;
6387         uint_t flags;
6388 };
6389 
6390 /*
6391  * Common async handler for:
6392  *      ndi_devi_bind_driver_async
6393  *      ndi_devi_online_async
6394  */
6395 static int
6396 i_ndi_devi_async_common(dev_info_t *dip, uint_t flags, void (*func)())
6397 {
6398         int tqflag;
6399         int kmflag;
6400         struct async_arg *arg;
6401         dev_info_t *pdip = ddi_get_parent(dip);
6402 
6403         ASSERT(pdip);
6404         ASSERT(DEVI(pdip)->devi_taskq);
6405         ASSERT(ndi_dev_is_persistent_node(dip));
6406 
6407         if (flags & NDI_NOSLEEP) {
6408                 kmflag = KM_NOSLEEP;
6409                 tqflag = TQ_NOSLEEP;
6410         } else {
6411                 kmflag = KM_SLEEP;
6412                 tqflag = TQ_SLEEP;
6413         }
6414 
6415         arg = kmem_alloc(sizeof (*arg), kmflag);
6416         if (arg == NULL)
6417                 goto fail;
6418 
6419         arg->flags = flags;
6420         arg->dip = dip;
6421         if (ddi_taskq_dispatch(DEVI(pdip)->devi_taskq, func, arg, tqflag) ==
6422             DDI_SUCCESS) {
6423                 return (NDI_SUCCESS);
6424         }
6425 
6426 fail:
6427         NDI_CONFIG_DEBUG((CE_CONT, "%s%d: ddi_taskq_dispatch failed",
6428             ddi_driver_name(pdip), ddi_get_instance(pdip)));
6429 
6430         if (arg)
6431                 kmem_free(arg, sizeof (*arg));
6432         return (NDI_FAILURE);
6433 }
6434 
6435 static void
6436 i_ndi_devi_bind_driver_cb(struct async_arg *arg)
6437 {
6438         (void) ndi_devi_bind_driver(arg->dip, arg->flags);
6439         kmem_free(arg, sizeof (*arg));
6440 }
6441 
6442 int
6443 ndi_devi_bind_driver_async(dev_info_t *dip, uint_t flags)
6444 {
6445         return (i_ndi_devi_async_common(dip, flags,
6446             (void (*)())i_ndi_devi_bind_driver_cb));
6447 }
6448 
6449 /*
6450  * place the devinfo in the ONLINE state.
6451  */
6452 int
6453 ndi_devi_online(dev_info_t *dip, uint_t flags)
6454 {
6455         int circ, rv;
6456         dev_info_t *pdip = ddi_get_parent(dip);
6457         int branch_event = 0;
6458 
6459         ASSERT(pdip);
6460 
6461         NDI_CONFIG_DEBUG((CE_CONT, "ndi_devi_online: %s%d (%p)\n",
6462             ddi_driver_name(dip), ddi_get_instance(dip), (void *)dip));
6463 
6464         ndi_devi_enter(pdip, &circ);
6465         /* bind child before merging .conf nodes */
6466         rv = i_ndi_config_node(dip, DS_BOUND, flags);
6467         if (rv != NDI_SUCCESS) {
6468                 ndi_devi_exit(pdip, circ);
6469                 return (rv);
6470         }
6471 
6472         /* merge .conf properties */
6473         (void) i_ndi_make_spec_children(pdip, flags);
6474 
6475         flags |= (NDI_DEVI_ONLINE | NDI_CONFIG);
6476 
6477         if (flags & NDI_NO_EVENT) {
6478                 /*
6479                  * Caller is specifically asking for not to generate an event.
6480                  * Set the following flag so that devi_attach_node() don't
6481                  * change the event state.
6482                  */
6483                 flags |= NDI_NO_EVENT_STATE_CHNG;
6484         }
6485 
6486         if ((flags & (NDI_NO_EVENT | NDI_BRANCH_EVENT_OP)) == 0 &&
6487             ((flags & NDI_CONFIG) || DEVI_NEED_NDI_CONFIG(dip))) {
6488                 flags |= NDI_BRANCH_EVENT_OP;
6489                 branch_event = 1;
6490         }
6491 
6492         /*
6493          * devi_attach_node() may remove dip on failure
6494          */
6495         if ((rv = devi_attach_node(dip, flags)) == NDI_SUCCESS) {
6496                 if ((flags & NDI_CONFIG) || DEVI_NEED_NDI_CONFIG(dip)) {
6497                         /*
6498                          * Hold the attached dip, and exit the parent while
6499                          * we drive configuration of children below the
6500                          * attached dip.
6501                          */
6502                         ndi_hold_devi(dip);
6503                         ndi_devi_exit(pdip, circ);
6504 
6505                         (void) ndi_devi_config(dip, flags);
6506 
6507                         ndi_devi_enter(pdip, &circ);
6508                         ndi_rele_devi(dip);
6509                 }
6510 
6511                 if (branch_event)
6512                         (void) i_log_devfs_branch_add(dip);
6513         }
6514 
6515         ndi_devi_exit(pdip, circ);
6516 
6517         /*
6518          * Notify devfs that we have a new node. Devfs needs to invalidate
6519          * cached directory contents.
6520          *
6521          * For PCMCIA devices, it is possible the pdip is not fully
6522          * attached. In this case, calling back into devfs will
6523          * result in a loop or assertion error. Hence, the check
6524          * on node state.
6525          *
6526          * If we own parent lock, this is part of a branch operation.
6527          * We skip the devfs_clean() step because the cache invalidation
6528          * is done higher up in the device tree.
6529          */
6530         if (rv == NDI_SUCCESS && i_ddi_devi_attached(pdip) &&
6531             !DEVI_BUSY_OWNED(pdip))
6532                 (void) devfs_clean(pdip, NULL, 0);
6533         return (rv);
6534 }
6535 
6536 static void
6537 i_ndi_devi_online_cb(struct async_arg *arg)
6538 {
6539         (void) ndi_devi_online(arg->dip, arg->flags);
6540         kmem_free(arg, sizeof (*arg));
6541 }
6542 
6543 int
6544 ndi_devi_online_async(dev_info_t *dip, uint_t flags)
6545 {
6546         /* mark child as need config if requested. */
6547         if (flags & NDI_CONFIG) {
6548                 mutex_enter(&(DEVI(dip)->devi_lock));
6549                 DEVI_SET_NDI_CONFIG(dip);
6550                 mutex_exit(&(DEVI(dip)->devi_lock));
6551         }
6552 
6553         return (i_ndi_devi_async_common(dip, flags,
6554             (void (*)())i_ndi_devi_online_cb));
6555 }
6556 
6557 /*
6558  * Take a device node Offline
6559  * To take a device Offline means to detach the device instance from
6560  * the driver and prevent devfs requests from re-attaching the device
6561  * instance.
6562  *
6563  * The flag NDI_DEVI_REMOVE causes removes the device node from
6564  * the driver list and the device tree. In this case, the device
6565  * is assumed to be removed from the system.
6566  */
6567 int
6568 ndi_devi_offline(dev_info_t *dip, uint_t flags)
6569 {
6570         int             circ, rval = 0;
6571         dev_info_t      *pdip = ddi_get_parent(dip);
6572         dev_info_t      *vdip = NULL;
6573         int             v_circ;
6574         struct brevq_node *brevq = NULL;
6575 
6576         ASSERT(pdip);
6577 
6578         flags |= NDI_DEVI_OFFLINE;
6579 
6580         /*
6581          * If child is pHCI and vHCI and pHCI are not siblings then enter vHCI
6582          * before parent(pHCI) to avoid deadlock with mpxio Client power
6583          * management operations.
6584          */
6585         if (MDI_PHCI(dip)) {
6586                 vdip = mdi_devi_get_vdip(dip);
6587                 if (vdip && (ddi_get_parent(vdip) != pdip))
6588                         ndi_devi_enter(vdip, &v_circ);
6589                 else
6590                         vdip = NULL;
6591         }
6592         ndi_devi_enter(pdip, &circ);
6593 
6594         if (i_ddi_devi_attached(dip)) {
6595                 /*
6596                  * If dip is in DS_READY state, there may be cached dv_nodes
6597                  * referencing this dip, so we invoke devfs code path.
6598                  * Note that we must release busy changing on pdip to
6599                  * avoid deadlock against devfs.
6600                  */
6601                 char *devname = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
6602                 (void) ddi_deviname(dip, devname);
6603 
6604                 ndi_devi_exit(pdip, circ);
6605                 if (vdip)
6606                         ndi_devi_exit(vdip, v_circ);
6607 
6608                 /*
6609                  * If we are explictly told to clean, then clean. If we own the
6610                  * parent lock then this is part of a branch operation, and we
6611                  * skip the devfs_clean() step.
6612                  *
6613                  * NOTE: A thread performing a devfs file system lookup/
6614                  * bus_config can't call devfs_clean to unconfig without
6615                  * causing rwlock problems in devfs. For ndi_devi_offline, this
6616                  * means that the NDI_DEVFS_CLEAN flag is safe from ioctl code
6617                  * or from an async hotplug thread, but is not safe from a
6618                  * nexus driver's bus_config implementation.
6619                  */
6620                 if ((flags & NDI_DEVFS_CLEAN) ||
6621                     (!DEVI_BUSY_OWNED(pdip)))
6622                         (void) devfs_clean(pdip, devname + 1, DV_CLEAN_FORCE);
6623 
6624                 kmem_free(devname, MAXNAMELEN + 1);
6625 
6626                 rval = devi_unconfig_branch(dip, NULL, flags|NDI_UNCONFIG,
6627                     &brevq);
6628 
6629                 if (rval)
6630                         return (NDI_FAILURE);
6631 
6632                 if (vdip)
6633                         ndi_devi_enter(vdip, &v_circ);
6634                 ndi_devi_enter(pdip, &circ);
6635         }
6636 
6637         init_bound_node_ev(pdip, dip, flags);
6638 
6639         rval = devi_detach_node(dip, flags);
6640         if (brevq) {
6641                 if (rval != NDI_SUCCESS)
6642                         log_and_free_brevq_dip(dip, brevq);
6643                 else
6644                         free_brevq(brevq);
6645         }
6646 
6647         ndi_devi_exit(pdip, circ);
6648         if (vdip)
6649                 ndi_devi_exit(vdip, v_circ);
6650 
6651         return (rval);
6652 }
6653 
6654 /*
6655  * Find the child dev_info node of parent nexus 'p' whose unit address
6656  * matches "cname@caddr".  Recommend use of ndi_devi_findchild() instead.
6657  */
6658 dev_info_t *
6659 ndi_devi_find(dev_info_t *pdip, char *cname, char *caddr)
6660 {
6661         dev_info_t *child;
6662         int circ;
6663 
6664         if (pdip == NULL || cname == NULL || caddr == NULL)
6665                 return ((dev_info_t *)NULL);
6666 
6667         ndi_devi_enter(pdip, &circ);
6668         child = find_sibling(ddi_get_child(pdip), cname, caddr,
6669             FIND_NODE_BY_NODENAME, NULL);
6670         ndi_devi_exit(pdip, circ);
6671         return (child);
6672 }
6673 
6674 /*
6675  * Find the child dev_info node of parent nexus 'p' whose unit address
6676  * matches devname "name@addr".  Permits caller to hold the parent.
6677  */
6678 dev_info_t *
6679 ndi_devi_findchild(dev_info_t *pdip, char *devname)
6680 {
6681         dev_info_t *child;
6682         char    *cname, *caddr;
6683         char    *devstr;
6684 
6685         ASSERT(DEVI_BUSY_OWNED(pdip));
6686 
6687         devstr = i_ddi_strdup(devname, KM_SLEEP);
6688         i_ddi_parse_name(devstr, &cname, &caddr, NULL);
6689 
6690         if (cname == NULL || caddr == NULL) {
6691                 kmem_free(devstr, strlen(devname)+1);
6692                 return ((dev_info_t *)NULL);
6693         }
6694 
6695         child = find_sibling(ddi_get_child(pdip), cname, caddr,
6696             FIND_NODE_BY_NODENAME, NULL);
6697         kmem_free(devstr, strlen(devname)+1);
6698         return (child);
6699 }
6700 
6701 /*
6702  * Misc. routines called by framework only
6703  */
6704 
6705 /*
6706  * Clear the DEVI_MADE_CHILDREN/DEVI_ATTACHED_CHILDREN flags
6707  * if new child spec has been added.
6708  */
6709 static int
6710 reset_nexus_flags(dev_info_t *dip, void *arg)
6711 {
6712         struct hwc_spec *list;
6713         int             circ;
6714 
6715         if (((DEVI(dip)->devi_flags & DEVI_MADE_CHILDREN) == 0) ||
6716             ((list = hwc_get_child_spec(dip, (major_t)(uintptr_t)arg)) == NULL))
6717                 return (DDI_WALK_CONTINUE);
6718 
6719         hwc_free_spec_list(list);
6720 
6721         /* coordinate child state update */
6722         ndi_devi_enter(dip, &circ);
6723         mutex_enter(&DEVI(dip)->devi_lock);
6724         DEVI(dip)->devi_flags &= ~(DEVI_MADE_CHILDREN | DEVI_ATTACHED_CHILDREN);
6725         mutex_exit(&DEVI(dip)->devi_lock);
6726         ndi_devi_exit(dip, circ);
6727 
6728         return (DDI_WALK_CONTINUE);
6729 }
6730 
6731 /*
6732  * Helper functions, returns NULL if no memory.
6733  */
6734 
6735 /*
6736  * path_to_major:
6737  *
6738  * Return an alternate driver name binding for the leaf device
6739  * of the given pathname, if there is one. The purpose of this
6740  * function is to deal with generic pathnames. The default action
6741  * for platforms that can't do this (ie: x86 or any platform that
6742  * does not have prom_finddevice functionality, which matches
6743  * nodenames and unit-addresses without the drivers participation)
6744  * is to return DDI_MAJOR_T_NONE.
6745  *
6746  * Used in loadrootmodules() in the swapgeneric module to
6747  * associate a given pathname with a given leaf driver.
6748  *
6749  */
6750 major_t
6751 path_to_major(char *path)
6752 {
6753         dev_info_t *dip;
6754         char *p, *q;
6755         pnode_t nodeid;
6756         major_t major;
6757 
6758         /* check for path-oriented alias */
6759         major = ddi_name_to_major(path);
6760         if (driver_active(major)) {
6761                 NDI_CONFIG_DEBUG((CE_NOTE, "path_to_major: %s path bound %s\n",
6762                     path, ddi_major_to_name(major)));
6763                 return (major);
6764         }
6765 
6766         /*
6767          * Get the nodeid of the given pathname, if such a mapping exists.
6768          */
6769         dip = NULL;
6770         nodeid = prom_finddevice(path);
6771         if (nodeid != OBP_BADNODE) {
6772                 /*
6773                  * Find the nodeid in our copy of the device tree and return
6774                  * whatever name we used to bind this node to a driver.
6775                  */
6776                 dip = e_ddi_nodeid_to_dip(nodeid);
6777         }
6778 
6779         if (dip == NULL) {
6780                 NDI_CONFIG_DEBUG((CE_WARN,
6781                     "path_to_major: can't bind <%s>\n", path));
6782                 return (DDI_MAJOR_T_NONE);
6783         }
6784 
6785         /*
6786          * If we're bound to something other than the nodename,
6787          * note that in the message buffer and system log.
6788          */
6789         p = ddi_binding_name(dip);
6790         q = ddi_node_name(dip);
6791         if (p && q && (strcmp(p, q) != 0))
6792                 NDI_CONFIG_DEBUG((CE_NOTE, "path_to_major: %s bound to %s\n",
6793                     path, p));
6794 
6795         major = ddi_name_to_major(p);
6796 
6797         ndi_rele_devi(dip);             /* release e_ddi_nodeid_to_dip hold */
6798 
6799         return (major);
6800 }
6801 
6802 /*
6803  * Return the held dip for the specified major and instance, attempting to do
6804  * an attach if specified. Return NULL if the devi can't be found or put in
6805  * the proper state. The caller must release the hold via ddi_release_devi if
6806  * a non-NULL value is returned.
6807  *
6808  * Some callers expect to be able to perform a hold_devi() while in a context
6809  * where using ndi_devi_enter() to ensure the hold might cause deadlock (see
6810  * open-from-attach code in consconfig_dacf.c). Such special-case callers
6811  * must ensure that an ndi_devi_enter(parent)/ndi_hold_devi() from a safe
6812  * context is already active. The hold_devi() implementation must accommodate
6813  * these callers.
6814  */
6815 static dev_info_t *
6816 hold_devi(major_t major, int instance, int flags)
6817 {
6818         struct devnames *dnp;
6819         dev_info_t      *dip;
6820         char            *path;
6821         char            *vpath;
6822 
6823         if ((major >= devcnt) || (instance == -1))
6824                 return (NULL);
6825 
6826         /* try to find the instance in the per driver list */
6827         dnp = &(devnamesp[major]);
6828         LOCK_DEV_OPS(&(dnp->dn_lock));
6829         for (dip = dnp->dn_head; dip;
6830             dip = (dev_info_t *)DEVI(dip)->devi_next) {
6831                 /* skip node if instance field is not valid */
6832                 if (i_ddi_node_state(dip) < DS_INITIALIZED)
6833                         continue;
6834 
6835                 /* look for instance match */
6836                 if (DEVI(dip)->devi_instance == instance) {
6837                         /*
6838                          * To accommodate callers that can't block in
6839                          * ndi_devi_enter() we do an ndi_hold_devi(), and
6840                          * afterwards check that the node is in a state where
6841                          * the hold prevents detach(). If we did not manage to
6842                          * prevent detach then we ndi_rele_devi() and perform
6843                          * the slow path below (which can result in a blocking
6844                          * ndi_devi_enter() while driving attach top-down).
6845                          * This code depends on the ordering of
6846                          * DEVI_SET_DETACHING and the devi_ref check in the
6847                          * detach_node() code path.
6848                          */
6849                         ndi_hold_devi(dip);
6850                         if (i_ddi_devi_attached(dip) &&
6851                             !DEVI_IS_DETACHING(dip)) {
6852                                 UNLOCK_DEV_OPS(&(dnp->dn_lock));
6853                                 return (dip);   /* fast-path with devi held */
6854                         }
6855                         ndi_rele_devi(dip);
6856 
6857                         /* try slow-path */
6858                         dip = NULL;
6859                         break;
6860                 }
6861         }
6862         ASSERT(dip == NULL);
6863         UNLOCK_DEV_OPS(&(dnp->dn_lock));
6864 
6865         if (flags & E_DDI_HOLD_DEVI_NOATTACH)
6866                 return (NULL);          /* told not to drive attach */
6867 
6868         /* slow-path may block, so it should not occur from interrupt */
6869         ASSERT(!servicing_interrupt());
6870         if (servicing_interrupt())
6871                 return (NULL);
6872 
6873         /* reconstruct the path and drive attach by path through devfs. */
6874         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6875         if (e_ddi_majorinstance_to_path(major, instance, path) == 0) {
6876                 dip = e_ddi_hold_devi_by_path(path, flags);
6877 
6878                 /*
6879                  * Verify that we got the correct device - a path_to_inst file
6880                  * with a bogus/corrupt path (or a nexus that changes its
6881                  * unit-address format) could result in an incorrect answer
6882                  *
6883                  * Verify major, instance, and path.
6884                  */
6885                 vpath = kmem_alloc(MAXPATHLEN, KM_SLEEP);
6886                 if (dip &&
6887                     ((DEVI(dip)->devi_major != major) ||
6888                     ((DEVI(dip)->devi_instance != instance)) ||
6889                     (strcmp(path, ddi_pathname(dip, vpath)) != 0))) {
6890                         ndi_rele_devi(dip);
6891                         dip = NULL;     /* no answer better than wrong answer */
6892                 }
6893                 kmem_free(vpath, MAXPATHLEN);
6894         }
6895         kmem_free(path, MAXPATHLEN);
6896         return (dip);                   /* with devi held */
6897 }
6898 
6899 /*
6900  * The {e_}ddi_hold_devi{_by_{instance|dev|path}} hold the devinfo node
6901  * associated with the specified arguments.  This hold should be released
6902  * by calling ddi_release_devi.
6903  *
6904  * The E_DDI_HOLD_DEVI_NOATTACH flag argument allows the caller to to specify
6905  * a failure return if the node is not already attached.
6906  *
6907  * NOTE: by the time we make e_ddi_hold_devi public, we should be able to reuse
6908  * ddi_hold_devi again.
6909  */
6910 dev_info_t *
6911 ddi_hold_devi_by_instance(major_t major, int instance, int flags)
6912 {
6913         return (hold_devi(major, instance, flags));
6914 }
6915 
6916 dev_info_t *
6917 e_ddi_hold_devi_by_dev(dev_t dev, int flags)
6918 {
6919         major_t major = getmajor(dev);
6920         dev_info_t      *dip;
6921         struct dev_ops  *ops;
6922         dev_info_t      *ddip = NULL;
6923 
6924         dip = hold_devi(major, dev_to_instance(dev), flags);
6925 
6926         /*
6927          * The rest of this routine is legacy support for drivers that
6928          * have broken DDI_INFO_DEVT2INSTANCE implementations but may have
6929          * functional DDI_INFO_DEVT2DEVINFO implementations.  This code will
6930          * diagnose inconsistency and, for maximum compatibility with legacy
6931          * drivers, give preference to the drivers DDI_INFO_DEVT2DEVINFO
6932          * implementation over the above derived dip based the driver's
6933          * DDI_INFO_DEVT2INSTANCE implementation. This legacy support should
6934          * be removed when DDI_INFO_DEVT2DEVINFO is deprecated.
6935          *
6936          * NOTE: The following code has a race condition. DEVT2DEVINFO
6937          *      returns a dip which is not held. By the time we ref ddip,
6938          *      it could have been freed. The saving grace is that for
6939          *      most drivers, the dip returned from hold_devi() is the
6940          *      same one as the one returned by DEVT2DEVINFO, so we are
6941          *      safe for drivers with the correct getinfo(9e) impl.
6942          */
6943         if (((ops = ddi_hold_driver(major)) != NULL) &&
6944             CB_DRV_INSTALLED(ops) && ops->devo_getinfo)  {
6945                 if ((*ops->devo_getinfo)(NULL, DDI_INFO_DEVT2DEVINFO,
6946                     (void *)dev, (void **)&ddip) != DDI_SUCCESS)
6947                         ddip = NULL;
6948         }
6949 
6950         /* give preference to the driver returned DEVT2DEVINFO dip */
6951         if (ddip && (dip != ddip)) {
6952 #ifdef  DEBUG
6953                 cmn_err(CE_WARN, "%s: inconsistent getinfo(9E) implementation",
6954                     ddi_driver_name(ddip));
6955 #endif  /* DEBUG */
6956                 ndi_hold_devi(ddip);
6957                 if (dip)
6958                         ndi_rele_devi(dip);
6959                 dip = ddip;
6960         }
6961 
6962         if (ops)
6963                 ddi_rele_driver(major);
6964 
6965         return (dip);
6966 }
6967 
6968 /*
6969  * For compatibility only. Do not call this function!
6970  */
6971 dev_info_t *
6972 e_ddi_get_dev_info(dev_t dev, vtype_t type)
6973 {
6974         dev_info_t *dip = NULL;
6975         if (getmajor(dev) >= devcnt)
6976                 return (NULL);
6977 
6978         switch (type) {
6979         case VCHR:
6980         case VBLK:
6981                 dip = e_ddi_hold_devi_by_dev(dev, 0);
6982         default:
6983                 break;
6984         }
6985 
6986         /*
6987          * For compatibility reasons, we can only return the dip with
6988          * the driver ref count held. This is not a safe thing to do.
6989          * For certain broken third-party software, we are willing
6990          * to venture into unknown territory.
6991          */
6992         if (dip) {
6993                 (void) ndi_hold_driver(dip);
6994                 ndi_rele_devi(dip);
6995         }
6996         return (dip);
6997 }
6998 
6999 dev_info_t *
7000 e_ddi_hold_devi_by_path(char *path, int flags)
7001 {
7002         dev_info_t      *dip;
7003 
7004         /* can't specify NOATTACH by path */
7005         ASSERT(!(flags & E_DDI_HOLD_DEVI_NOATTACH));
7006 
7007         return (resolve_pathname(path, &dip, NULL, NULL) ? NULL : dip);
7008 }
7009 
7010 void
7011 e_ddi_hold_devi(dev_info_t *dip)
7012 {
7013         ndi_hold_devi(dip);
7014 }
7015 
7016 void
7017 ddi_release_devi(dev_info_t *dip)
7018 {
7019         ndi_rele_devi(dip);
7020 }
7021 
7022 /*
7023  * Associate a streams queue with a devinfo node
7024  * NOTE: This function is called by STREAM driver's put procedure.
7025  *      It cannot block.
7026  */
7027 void
7028 ddi_assoc_queue_with_devi(queue_t *q, dev_info_t *dip)
7029 {
7030         queue_t *rq = _RD(q);
7031         struct stdata *stp;
7032         vnode_t *vp;
7033 
7034         /* set flag indicating that ddi_assoc_queue_with_devi was called */
7035         mutex_enter(QLOCK(rq));
7036         rq->q_flag |= _QASSOCIATED;
7037         mutex_exit(QLOCK(rq));
7038 
7039         /* get the vnode associated with the queue */
7040         stp = STREAM(rq);
7041         vp = stp->sd_vnode;
7042         ASSERT(vp);
7043 
7044         /* change the hardware association of the vnode */
7045         spec_assoc_vp_with_devi(vp, dip);
7046 }
7047 
7048 /*
7049  * ddi_install_driver(name)
7050  *
7051  * Driver installation is currently a byproduct of driver loading.  This
7052  * may change.
7053  */
7054 int
7055 ddi_install_driver(char *name)
7056 {
7057         major_t major = ddi_name_to_major(name);
7058 
7059         if ((major == DDI_MAJOR_T_NONE) ||
7060             (ddi_hold_installed_driver(major) == NULL)) {
7061                 return (DDI_FAILURE);
7062         }
7063         ddi_rele_driver(major);
7064         return (DDI_SUCCESS);
7065 }
7066 
7067 struct dev_ops *
7068 ddi_hold_driver(major_t major)
7069 {
7070         return (mod_hold_dev_by_major(major));
7071 }
7072 
7073 
7074 void
7075 ddi_rele_driver(major_t major)
7076 {
7077         mod_rele_dev_by_major(major);
7078 }
7079 
7080 
7081 /*
7082  * This is called during boot to force attachment order of special dips
7083  * dip must be referenced via ndi_hold_devi()
7084  */
7085 int
7086 i_ddi_attach_node_hierarchy(dev_info_t *dip)
7087 {
7088         dev_info_t      *parent;
7089         int             ret, circ;
7090 
7091         /*
7092          * Recurse up until attached parent is found.
7093          */
7094         if (i_ddi_devi_attached(dip))
7095                 return (DDI_SUCCESS);
7096         parent = ddi_get_parent(dip);
7097         if (i_ddi_attach_node_hierarchy(parent) != DDI_SUCCESS)
7098                 return (DDI_FAILURE);
7099 
7100         /*
7101          * Come top-down, expanding .conf nodes under this parent
7102          * and driving attach.
7103          */
7104         ndi_devi_enter(parent, &circ);
7105         (void) i_ndi_make_spec_children(parent, 0);
7106         ret = i_ddi_attachchild(dip);
7107         ndi_devi_exit(parent, circ);
7108 
7109         return (ret);
7110 }
7111 
7112 /* keep this function static */
7113 static int
7114 attach_driver_nodes(major_t major)
7115 {
7116         struct devnames *dnp;
7117         dev_info_t *dip;
7118         int error = DDI_FAILURE;
7119 
7120         dnp = &devnamesp[major];
7121         LOCK_DEV_OPS(&dnp->dn_lock);
7122         dip = dnp->dn_head;
7123         while (dip) {
7124                 ndi_hold_devi(dip);
7125                 UNLOCK_DEV_OPS(&dnp->dn_lock);
7126                 if (i_ddi_attach_node_hierarchy(dip) == DDI_SUCCESS)
7127                         error = DDI_SUCCESS;
7128                 /*
7129                  * Set the 'ddi-config-driver-node' property on a nexus
7130                  * node to cause attach_driver_nodes() to configure all
7131                  * immediate children of the nexus. This property should
7132                  * be set on nodes with immediate children that bind to
7133                  * the same driver as parent.
7134                  */
7135                 if ((error == DDI_SUCCESS) && (ddi_prop_exists(DDI_DEV_T_ANY,
7136                     dip, DDI_PROP_DONTPASS, "ddi-config-driver-node"))) {
7137                         (void) ndi_devi_config(dip, NDI_NO_EVENT);
7138                 }
7139                 LOCK_DEV_OPS(&dnp->dn_lock);
7140                 ndi_rele_devi(dip);
7141                 dip = ddi_get_next(dip);
7142         }
7143         if (error == DDI_SUCCESS)
7144                 dnp->dn_flags |= DN_NO_AUTODETACH;
7145         UNLOCK_DEV_OPS(&dnp->dn_lock);
7146 
7147 
7148         return (error);
7149 }
7150 
7151 /*
7152  * i_ddi_attach_hw_nodes configures and attaches all hw nodes
7153  * bound to a specific driver. This function replaces calls to
7154  * ddi_hold_installed_driver() for drivers with no .conf
7155  * enumerated nodes.
7156  *
7157  * This facility is typically called at boot time to attach
7158  * platform-specific hardware nodes, such as ppm nodes on xcal
7159  * and grover and keyswitch nodes on cherrystone. It does not
7160  * deal with .conf enumerated node. Calling it beyond the boot
7161  * process is strongly discouraged.
7162  */
7163 int
7164 i_ddi_attach_hw_nodes(char *driver)
7165 {
7166         major_t major;
7167 
7168         major = ddi_name_to_major(driver);
7169         if (major == DDI_MAJOR_T_NONE)
7170                 return (DDI_FAILURE);
7171 
7172         return (attach_driver_nodes(major));
7173 }
7174 
7175 /*
7176  * i_ddi_attach_pseudo_node configures pseudo drivers which
7177  * has a single node. The .conf nodes must be enumerated
7178  * before calling this interface. The dip is held attached
7179  * upon returning.
7180  *
7181  * This facility should only be called only at boot time
7182  * by the I/O framework.
7183  */
7184 dev_info_t *
7185 i_ddi_attach_pseudo_node(char *driver)
7186 {
7187         major_t major;
7188         dev_info_t *dip;
7189 
7190         major = ddi_name_to_major(driver);
7191         if (major == DDI_MAJOR_T_NONE)
7192                 return (NULL);
7193 
7194         if (attach_driver_nodes(major) != DDI_SUCCESS)
7195                 return (NULL);
7196 
7197         dip = devnamesp[major].dn_head;
7198         ASSERT(dip && ddi_get_next(dip) == NULL);
7199         ndi_hold_devi(dip);
7200         return (dip);
7201 }
7202 
7203 static void
7204 diplist_to_parent_major(dev_info_t *head, char parents[])
7205 {
7206         major_t major;
7207         dev_info_t *dip, *pdip;
7208 
7209         for (dip = head; dip != NULL; dip = ddi_get_next(dip)) {
7210                 pdip = ddi_get_parent(dip);
7211                 ASSERT(pdip);   /* disallow rootnex.conf nodes */
7212                 major = ddi_driver_major(pdip);
7213                 if ((major != DDI_MAJOR_T_NONE) && parents[major] == 0)
7214                         parents[major] = 1;
7215         }
7216 }
7217 
7218 /*
7219  * Call ddi_hold_installed_driver() on each parent major
7220  * and invoke mt_config_driver() to attach child major.
7221  * This is part of the implementation of ddi_hold_installed_driver.
7222  */
7223 static int
7224 attach_driver_by_parent(major_t child_major, char parents[])
7225 {
7226         major_t par_major;
7227         struct mt_config_handle *hdl;
7228         int flags = NDI_DEVI_PERSIST | NDI_NO_EVENT;
7229 
7230         hdl = mt_config_init(NULL, NULL, flags, child_major, MT_CONFIG_OP,
7231             NULL);
7232         for (par_major = 0; par_major < devcnt; par_major++) {
7233                 /* disallow recursion on the same driver */
7234                 if (parents[par_major] == 0 || par_major == child_major)
7235                         continue;
7236                 if (ddi_hold_installed_driver(par_major) == NULL)
7237                         continue;
7238                 hdl->mtc_parmajor = par_major;
7239                 mt_config_driver(hdl);
7240                 ddi_rele_driver(par_major);
7241         }
7242         (void) mt_config_fini(hdl);
7243 
7244         return (i_ddi_devs_attached(child_major));
7245 }
7246 
7247 int
7248 i_ddi_devs_attached(major_t major)
7249 {
7250         dev_info_t *dip;
7251         struct devnames *dnp;
7252         int error = DDI_FAILURE;
7253 
7254         /* check for attached instances */
7255         dnp = &devnamesp[major];
7256         LOCK_DEV_OPS(&dnp->dn_lock);
7257         for (dip = dnp->dn_head; dip != NULL; dip = ddi_get_next(dip)) {
7258                 if (i_ddi_devi_attached(dip)) {
7259                         error = DDI_SUCCESS;
7260                         break;
7261                 }
7262         }
7263         UNLOCK_DEV_OPS(&dnp->dn_lock);
7264 
7265         return (error);
7266 }
7267 
7268 int
7269 i_ddi_minor_node_count(dev_info_t *ddip, const char *node_type)
7270 {
7271         int                     circ;
7272         struct ddi_minor_data   *dp;
7273         int                     count = 0;
7274 
7275         ndi_devi_enter(ddip, &circ);
7276         for (dp = DEVI(ddip)->devi_minor; dp != NULL; dp = dp->next) {
7277                 if (strcmp(dp->ddm_node_type, node_type) == 0)
7278                         count++;
7279         }
7280         ndi_devi_exit(ddip, circ);
7281         return (count);
7282 }
7283 
7284 /*
7285  * ddi_hold_installed_driver configures and attaches all
7286  * instances of the specified driver. To accomplish this
7287  * it configures and attaches all possible parents of
7288  * the driver, enumerated both in h/w nodes and in the
7289  * driver's .conf file.
7290  *
7291  * NOTE: This facility is for compatibility purposes only and will
7292  *      eventually go away. Its usage is strongly discouraged.
7293  */
7294 static void
7295 enter_driver(struct devnames *dnp)
7296 {
7297         mutex_enter(&dnp->dn_lock);
7298         ASSERT(dnp->dn_busy_thread != curthread);
7299         while (dnp->dn_flags & DN_DRIVER_BUSY)
7300                 cv_wait(&dnp->dn_wait, &dnp->dn_lock);
7301         dnp->dn_flags |= DN_DRIVER_BUSY;
7302         dnp->dn_busy_thread = curthread;
7303         mutex_exit(&dnp->dn_lock);
7304 }
7305 
7306 static void
7307 exit_driver(struct devnames *dnp)
7308 {
7309         mutex_enter(&dnp->dn_lock);
7310         ASSERT(dnp->dn_busy_thread == curthread);
7311         dnp->dn_flags &= ~DN_DRIVER_BUSY;
7312         dnp->dn_busy_thread = NULL;
7313         cv_broadcast(&dnp->dn_wait);
7314         mutex_exit(&dnp->dn_lock);
7315 }
7316 
7317 struct dev_ops *
7318 ddi_hold_installed_driver(major_t major)
7319 {
7320         struct dev_ops *ops;
7321         struct devnames *dnp;
7322         char *parents;
7323         int error;
7324 
7325         ops = ddi_hold_driver(major);
7326         if (ops == NULL)
7327                 return (NULL);
7328 
7329         /*
7330          * Return immediately if all the attach operations associated
7331          * with a ddi_hold_installed_driver() call have already been done.
7332          */
7333         dnp = &devnamesp[major];
7334         enter_driver(dnp);
7335         ASSERT(driver_active(major));
7336 
7337         if (dnp->dn_flags & DN_DRIVER_HELD) {
7338                 exit_driver(dnp);
7339                 if (i_ddi_devs_attached(major) == DDI_SUCCESS)
7340                         return (ops);
7341                 ddi_rele_driver(major);
7342                 return (NULL);
7343         }
7344 
7345         LOCK_DEV_OPS(&dnp->dn_lock);
7346         dnp->dn_flags |= (DN_DRIVER_HELD | DN_NO_AUTODETACH);
7347         UNLOCK_DEV_OPS(&dnp->dn_lock);
7348 
7349         DCOMPATPRINTF((CE_CONT,
7350             "ddi_hold_installed_driver: %s\n", dnp->dn_name));
7351 
7352         /*
7353          * When the driver has no .conf children, it is sufficient
7354          * to attach existing nodes in the device tree. Nodes not
7355          * enumerated by the OBP are not attached.
7356          */
7357         if (dnp->dn_pl == NULL) {
7358                 if (attach_driver_nodes(major) == DDI_SUCCESS) {
7359                         exit_driver(dnp);
7360                         return (ops);
7361                 }
7362                 exit_driver(dnp);
7363                 ddi_rele_driver(major);
7364                 return (NULL);
7365         }
7366 
7367         /*
7368          * Driver has .conf nodes. We find all possible parents
7369          * and recursively all ddi_hold_installed_driver on the
7370          * parent driver; then we invoke ndi_config_driver()
7371          * on all possible parent node in parallel to speed up
7372          * performance.
7373          */
7374         parents = kmem_zalloc(devcnt * sizeof (char), KM_SLEEP);
7375 
7376         LOCK_DEV_OPS(&dnp->dn_lock);
7377         /* find .conf parents */
7378         (void) impl_parlist_to_major(dnp->dn_pl, parents);
7379         /* find hw node parents */
7380         diplist_to_parent_major(dnp->dn_head, parents);
7381         UNLOCK_DEV_OPS(&dnp->dn_lock);
7382 
7383         error = attach_driver_by_parent(major, parents);
7384         kmem_free(parents, devcnt * sizeof (char));
7385         if (error == DDI_SUCCESS) {
7386                 exit_driver(dnp);
7387                 return (ops);
7388         }
7389 
7390         exit_driver(dnp);
7391         ddi_rele_driver(major);
7392         return (NULL);
7393 }
7394 
7395 /*
7396  * Default bus_config entry point for nexus drivers
7397  */
7398 int
7399 ndi_busop_bus_config(dev_info_t *pdip, uint_t flags, ddi_bus_config_op_t op,
7400     void *arg, dev_info_t **child, clock_t timeout)
7401 {
7402         major_t major;
7403 
7404         /*
7405          * A timeout of 30 minutes or more is probably a mistake
7406          * This is intended to catch uses where timeout is in
7407          * the wrong units.  timeout must be in units of ticks.
7408          */
7409         ASSERT(timeout < SEC_TO_TICK(1800));
7410 
7411         major = DDI_MAJOR_T_NONE;
7412         switch (op) {
7413         case BUS_CONFIG_ONE:
7414                 NDI_DEBUG(flags, (CE_CONT, "%s%d: bus config %s timeout=%ld\n",
7415                     ddi_driver_name(pdip), ddi_get_instance(pdip),
7416                     (char *)arg, timeout));
7417                 return (devi_config_one(pdip, (char *)arg, child, flags,
7418                     timeout));
7419 
7420         case BUS_CONFIG_DRIVER:
7421                 major = (major_t)(uintptr_t)arg;
7422                 /*FALLTHROUGH*/
7423         case BUS_CONFIG_ALL:
7424                 NDI_DEBUG(flags, (CE_CONT, "%s%d: bus config timeout=%ld\n",
7425                     ddi_driver_name(pdip), ddi_get_instance(pdip),
7426                     timeout));
7427                 if (timeout > 0) {
7428                         NDI_DEBUG(flags, (CE_CONT,
7429                             "%s%d: bus config all timeout=%ld\n",
7430                             ddi_driver_name(pdip), ddi_get_instance(pdip),
7431                             timeout));
7432                         delay(timeout);
7433                 }
7434                 return (config_immediate_children(pdip, flags, major));
7435 
7436         default:
7437                 return (NDI_FAILURE);
7438         }
7439         /*NOTREACHED*/
7440 }
7441 
7442 /*
7443  * Default busop bus_unconfig handler for nexus drivers
7444  */
7445 int
7446 ndi_busop_bus_unconfig(dev_info_t *pdip, uint_t flags, ddi_bus_config_op_t op,
7447     void *arg)
7448 {
7449         major_t major;
7450 
7451         major = DDI_MAJOR_T_NONE;
7452         switch (op) {
7453         case BUS_UNCONFIG_ONE:
7454                 NDI_DEBUG(flags, (CE_CONT, "%s%d: bus unconfig %s\n",
7455                     ddi_driver_name(pdip), ddi_get_instance(pdip),
7456                     (char *)arg));
7457                 return (devi_unconfig_one(pdip, (char *)arg, flags));
7458 
7459         case BUS_UNCONFIG_DRIVER:
7460                 major = (major_t)(uintptr_t)arg;
7461                 /*FALLTHROUGH*/
7462         case BUS_UNCONFIG_ALL:
7463                 NDI_DEBUG(flags, (CE_CONT, "%s%d: bus unconfig all\n",
7464                     ddi_driver_name(pdip), ddi_get_instance(pdip)));
7465                 return (unconfig_immediate_children(pdip, NULL, flags, major));
7466 
7467         default:
7468                 return (NDI_FAILURE);
7469         }
7470         /*NOTREACHED*/
7471 }
7472 
7473 /*
7474  * dummy functions to be removed
7475  */
7476 void
7477 impl_rem_dev_props(dev_info_t *dip)
7478 {
7479         _NOTE(ARGUNUSED(dip))
7480         /* do nothing */
7481 }
7482 
7483 /*
7484  * Determine if a node is a leaf node. If not sure, return false (0).
7485  */
7486 static int
7487 is_leaf_node(dev_info_t *dip)
7488 {
7489         major_t major = ddi_driver_major(dip);
7490 
7491         if (major == DDI_MAJOR_T_NONE)
7492                 return (0);
7493 
7494         return (devnamesp[major].dn_flags & DN_LEAF_DRIVER);
7495 }
7496 
7497 /*
7498  * Multithreaded [un]configuration
7499  */
7500 static struct mt_config_handle *
7501 mt_config_init(dev_info_t *pdip, dev_info_t **dipp, int flags,
7502     major_t major, int op, struct brevq_node **brevqp)
7503 {
7504         struct mt_config_handle *hdl = kmem_alloc(sizeof (*hdl), KM_SLEEP);
7505 
7506         mutex_init(&hdl->mtc_lock, NULL, MUTEX_DEFAULT, NULL);
7507         cv_init(&hdl->mtc_cv, NULL, CV_DEFAULT, NULL);
7508         hdl->mtc_pdip = pdip;
7509         hdl->mtc_fdip = dipp;
7510         hdl->mtc_parmajor = DDI_MAJOR_T_NONE;
7511         hdl->mtc_flags = flags;
7512         hdl->mtc_major = major;
7513         hdl->mtc_thr_count = 0;
7514         hdl->mtc_op = op;
7515         hdl->mtc_error = 0;
7516         hdl->mtc_brevqp = brevqp;
7517 
7518 #ifdef DEBUG
7519         gethrestime(&hdl->start_time);
7520         hdl->total_time = 0;
7521 #endif /* DEBUG */
7522 
7523         return (hdl);
7524 }
7525 
7526 #ifdef DEBUG
7527 static int
7528 time_diff_in_msec(timestruc_t start, timestruc_t end)
7529 {
7530         int     nsec, sec;
7531 
7532         sec = end.tv_sec - start.tv_sec;
7533         nsec = end.tv_nsec - start.tv_nsec;
7534         if (nsec < 0) {
7535                 nsec += NANOSEC;
7536                 sec -= 1;
7537         }
7538 
7539         return (sec * (NANOSEC >> 20) + (nsec >> 20));
7540 }
7541 
7542 #endif  /* DEBUG */
7543 
7544 static int
7545 mt_config_fini(struct mt_config_handle *hdl)
7546 {
7547         int             rv;
7548 #ifdef DEBUG
7549         int             real_time;
7550         timestruc_t     end_time;
7551 #endif /* DEBUG */
7552 
7553         mutex_enter(&hdl->mtc_lock);
7554         while (hdl->mtc_thr_count > 0)
7555                 cv_wait(&hdl->mtc_cv, &hdl->mtc_lock);
7556         rv = hdl->mtc_error;
7557         mutex_exit(&hdl->mtc_lock);
7558 
7559 #ifdef DEBUG
7560         gethrestime(&end_time);
7561         real_time = time_diff_in_msec(hdl->start_time, end_time);
7562         if ((ddidebug & DDI_MTCONFIG) && hdl->mtc_pdip)
7563                 cmn_err(CE_NOTE,
7564                     "config %s%d: total time %d msec, real time %d msec",
7565                     ddi_driver_name(hdl->mtc_pdip),
7566                     ddi_get_instance(hdl->mtc_pdip),
7567                     hdl->total_time, real_time);
7568 #endif /* DEBUG */
7569 
7570         cv_destroy(&hdl->mtc_cv);
7571         mutex_destroy(&hdl->mtc_lock);
7572         kmem_free(hdl, sizeof (*hdl));
7573 
7574         return (rv);
7575 }
7576 
7577 struct mt_config_data {
7578         struct mt_config_handle *mtc_hdl;
7579         dev_info_t              *mtc_dip;
7580         major_t                 mtc_major;
7581         int                     mtc_flags;
7582         struct brevq_node       *mtc_brn;
7583         struct mt_config_data   *mtc_next;
7584 };
7585 
7586 static void
7587 mt_config_thread(void *arg)
7588 {
7589         struct mt_config_data   *mcd = (struct mt_config_data *)arg;
7590         struct mt_config_handle *hdl = mcd->mtc_hdl;
7591         dev_info_t              *dip = mcd->mtc_dip;
7592         dev_info_t              *rdip, **dipp;
7593         major_t                 major = mcd->mtc_major;
7594         int                     flags = mcd->mtc_flags;
7595         int                     rv = 0;
7596 
7597 #ifdef DEBUG
7598         timestruc_t start_time, end_time;
7599         gethrestime(&start_time);
7600 #endif /* DEBUG */
7601 
7602         rdip = NULL;
7603         dipp = hdl->mtc_fdip ? &rdip : NULL;
7604 
7605         switch (hdl->mtc_op) {
7606         case MT_CONFIG_OP:
7607                 rv = devi_config_common(dip, flags, major);
7608                 break;
7609         case MT_UNCONFIG_OP:
7610                 if (mcd->mtc_brn) {
7611                         struct brevq_node *brevq = NULL;
7612                         rv = devi_unconfig_common(dip, dipp, flags, major,
7613                             &brevq);
7614                         mcd->mtc_brn->brn_child = brevq;
7615                 } else
7616                         rv = devi_unconfig_common(dip, dipp, flags, major,
7617                             NULL);
7618                 break;
7619         }
7620 
7621         mutex_enter(&hdl->mtc_lock);
7622 #ifdef DEBUG
7623         gethrestime(&end_time);
7624         hdl->total_time += time_diff_in_msec(start_time, end_time);
7625 #endif /* DEBUG */
7626 
7627         if ((rv != NDI_SUCCESS) && (hdl->mtc_error == 0)) {
7628                 hdl->mtc_error = rv;
7629 #ifdef  DEBUG
7630                 if ((ddidebug & DDI_DEBUG) && (major != DDI_MAJOR_T_NONE)) {
7631                         char    *path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
7632 
7633                         (void) ddi_pathname(dip, path);
7634                         cmn_err(CE_NOTE, "mt_config_thread: "
7635                             "op %d.%d.%x at %s failed %d",
7636                             hdl->mtc_op, major, flags, path, rv);
7637                         kmem_free(path, MAXPATHLEN);
7638                 }
7639 #endif  /* DEBUG */
7640         }
7641 
7642         if (hdl->mtc_fdip && *hdl->mtc_fdip == NULL) {
7643                 *hdl->mtc_fdip = rdip;
7644                 rdip = NULL;
7645         }
7646 
7647         if (rdip) {
7648                 ASSERT(rv != NDI_SUCCESS);
7649                 ndi_rele_devi(rdip);
7650         }
7651 
7652         ndi_rele_devi(dip);
7653 
7654         if (--hdl->mtc_thr_count == 0)
7655                 cv_broadcast(&hdl->mtc_cv);
7656         mutex_exit(&hdl->mtc_lock);
7657         kmem_free(mcd, sizeof (*mcd));
7658 }
7659 
7660 /*
7661  * Multi-threaded config/unconfig of child nexus
7662  */
7663 static void
7664 mt_config_children(struct mt_config_handle *hdl)
7665 {
7666         dev_info_t              *pdip = hdl->mtc_pdip;
7667         major_t                 major = hdl->mtc_major;
7668         dev_info_t              *dip;
7669         int                     circ;
7670         struct brevq_node       *brn;
7671         struct mt_config_data   *mcd_head = NULL;
7672         struct mt_config_data   *mcd_tail = NULL;
7673         struct mt_config_data   *mcd;
7674 #ifdef DEBUG
7675         timestruc_t             end_time;
7676 
7677         /* Update total_time in handle */
7678         gethrestime(&end_time);
7679         hdl->total_time += time_diff_in_msec(hdl->start_time, end_time);
7680 #endif
7681 
7682         ndi_devi_enter(pdip, &circ);
7683         dip = ddi_get_child(pdip);
7684         while (dip) {
7685                 if (hdl->mtc_op == MT_UNCONFIG_OP && hdl->mtc_brevqp &&
7686                     !(DEVI_EVREMOVE(dip)) &&
7687                     i_ddi_node_state(dip) >= DS_INITIALIZED) {
7688                         /*
7689                          * Enqueue this dip's deviname.
7690                          * No need to hold a lock while enqueuing since this
7691                          * is the only thread doing the enqueue and no one
7692                          * walks the queue while we are in multithreaded
7693                          * unconfiguration.
7694                          */
7695                         brn = brevq_enqueue(hdl->mtc_brevqp, dip, NULL);
7696                 } else
7697                         brn = NULL;
7698 
7699                 /*
7700                  * Hold the child that we are processing so he does not get
7701                  * removed. The corrisponding ndi_rele_devi() for children
7702                  * that are not being skipped is done at the end of
7703                  * mt_config_thread().
7704                  */
7705                 ndi_hold_devi(dip);
7706 
7707                 /*
7708                  * skip leaf nodes and (for configure) nodes not
7709                  * fully attached.
7710                  */
7711                 if (is_leaf_node(dip) ||
7712                     (hdl->mtc_op == MT_CONFIG_OP &&
7713                     i_ddi_node_state(dip) < DS_READY)) {
7714                         ndi_rele_devi(dip);
7715                         dip = ddi_get_next_sibling(dip);
7716                         continue;
7717                 }
7718 
7719                 mcd = kmem_alloc(sizeof (*mcd), KM_SLEEP);
7720                 mcd->mtc_dip = dip;
7721                 mcd->mtc_hdl = hdl;
7722                 mcd->mtc_brn = brn;
7723 
7724                 /*
7725                  * Switch a 'driver' operation to an 'all' operation below a
7726                  * node bound to the driver.
7727                  */
7728                 if ((major == DDI_MAJOR_T_NONE) ||
7729                     (major == ddi_driver_major(dip)))
7730                         mcd->mtc_major = DDI_MAJOR_T_NONE;
7731                 else
7732                         mcd->mtc_major = major;
7733 
7734                 /*
7735                  * The unconfig-driver to unconfig-all conversion above
7736                  * constitutes an autodetach for NDI_DETACH_DRIVER calls,
7737                  * set NDI_AUTODETACH.
7738                  */
7739                 mcd->mtc_flags = hdl->mtc_flags;
7740                 if ((mcd->mtc_flags & NDI_DETACH_DRIVER) &&
7741                     (hdl->mtc_op == MT_UNCONFIG_OP) &&
7742                     (major == ddi_driver_major(pdip)))
7743                         mcd->mtc_flags |= NDI_AUTODETACH;
7744 
7745                 mutex_enter(&hdl->mtc_lock);
7746                 hdl->mtc_thr_count++;
7747                 mutex_exit(&hdl->mtc_lock);
7748 
7749                 /*
7750                  * Add to end of list to process after ndi_devi_exit to avoid
7751                  * locking differences depending on value of mtc_off.
7752                  */
7753                 mcd->mtc_next = NULL;
7754                 if (mcd_head == NULL)
7755                         mcd_head = mcd;
7756                 else
7757                         mcd_tail->mtc_next = mcd;
7758                 mcd_tail = mcd;
7759 
7760                 dip = ddi_get_next_sibling(dip);
7761         }
7762         ndi_devi_exit(pdip, circ);
7763 
7764         /* go through the list of held children */
7765         for (mcd = mcd_head; mcd; mcd = mcd_head) {
7766                 mcd_head = mcd->mtc_next;
7767                 if (mtc_off || (mcd->mtc_flags & NDI_MTC_OFF))
7768                         mt_config_thread(mcd);
7769                 else
7770                         (void) thread_create(NULL, 0, mt_config_thread, mcd,
7771                             0, &p0, TS_RUN, minclsyspri);
7772         }
7773 }
7774 
7775 static void
7776 mt_config_driver(struct mt_config_handle *hdl)
7777 {
7778         major_t                 par_major = hdl->mtc_parmajor;
7779         major_t                 major = hdl->mtc_major;
7780         struct devnames         *dnp = &devnamesp[par_major];
7781         dev_info_t              *dip;
7782         struct mt_config_data   *mcd_head = NULL;
7783         struct mt_config_data   *mcd_tail = NULL;
7784         struct mt_config_data   *mcd;
7785 #ifdef DEBUG
7786         timestruc_t             end_time;
7787 
7788         /* Update total_time in handle */
7789         gethrestime(&end_time);
7790         hdl->total_time += time_diff_in_msec(hdl->start_time, end_time);
7791 #endif
7792         ASSERT(par_major != DDI_MAJOR_T_NONE);
7793         ASSERT(major != DDI_MAJOR_T_NONE);
7794 
7795         LOCK_DEV_OPS(&dnp->dn_lock);
7796         dip = devnamesp[par_major].dn_head;
7797         while (dip) {
7798                 /*
7799                  * Hold the child that we are processing so he does not get
7800                  * removed. The corrisponding ndi_rele_devi() for children
7801                  * that are not being skipped is done at the end of
7802                  * mt_config_thread().
7803                  */
7804                 ndi_hold_devi(dip);
7805 
7806                 /* skip leaf nodes and nodes not fully attached */
7807                 if (!i_ddi_devi_attached(dip) || is_leaf_node(dip)) {
7808                         ndi_rele_devi(dip);
7809                         dip = ddi_get_next(dip);
7810                         continue;
7811                 }
7812 
7813                 mcd = kmem_alloc(sizeof (*mcd), KM_SLEEP);
7814                 mcd->mtc_dip = dip;
7815                 mcd->mtc_hdl = hdl;
7816                 mcd->mtc_major = major;
7817                 mcd->mtc_flags = hdl->mtc_flags;
7818 
7819                 mutex_enter(&hdl->mtc_lock);
7820                 hdl->mtc_thr_count++;
7821                 mutex_exit(&hdl->mtc_lock);
7822 
7823                 /*
7824                  * Add to end of list to process after UNLOCK_DEV_OPS to avoid
7825                  * locking differences depending on value of mtc_off.
7826                  */
7827                 mcd->mtc_next = NULL;
7828                 if (mcd_head == NULL)
7829                         mcd_head = mcd;
7830                 else
7831                         mcd_tail->mtc_next = mcd;
7832                 mcd_tail = mcd;
7833 
7834                 dip = ddi_get_next(dip);
7835         }
7836         UNLOCK_DEV_OPS(&dnp->dn_lock);
7837 
7838         /* go through the list of held children */
7839         for (mcd = mcd_head; mcd; mcd = mcd_head) {
7840                 mcd_head = mcd->mtc_next;
7841                 if (mtc_off || (mcd->mtc_flags & NDI_MTC_OFF))
7842                         mt_config_thread(mcd);
7843                 else
7844                         (void) thread_create(NULL, 0, mt_config_thread, mcd,
7845                             0, &p0, TS_RUN, minclsyspri);
7846         }
7847 }
7848 
7849 /*
7850  * Given the nodeid for a persistent (PROM or SID) node, return
7851  * the corresponding devinfo node
7852  * NOTE: This function will return NULL for .conf nodeids.
7853  */
7854 dev_info_t *
7855 e_ddi_nodeid_to_dip(pnode_t nodeid)
7856 {
7857         dev_info_t              *dip = NULL;
7858         struct devi_nodeid      *prev, *elem;
7859 
7860         mutex_enter(&devimap->dno_lock);
7861 
7862         prev = NULL;
7863         for (elem = devimap->dno_head; elem; elem = elem->next) {
7864                 if (elem->nodeid == nodeid) {
7865                         ndi_hold_devi(elem->dip);
7866                         dip = elem->dip;
7867                         break;
7868                 }
7869                 prev = elem;
7870         }
7871 
7872         /*
7873          * Move to head for faster lookup next time
7874          */
7875         if (elem && prev) {
7876                 prev->next = elem->next;
7877                 elem->next = devimap->dno_head;
7878                 devimap->dno_head = elem;
7879         }
7880 
7881         mutex_exit(&devimap->dno_lock);
7882         return (dip);
7883 }
7884 
7885 static void
7886 free_cache_task(void *arg)
7887 {
7888         ASSERT(arg == NULL);
7889 
7890         mutex_enter(&di_cache.cache_lock);
7891 
7892         /*
7893          * The cache can be invalidated without holding the lock
7894          * but it can be made valid again only while the lock is held.
7895          * So if the cache is invalid when the lock is held, it will
7896          * stay invalid until lock is released.
7897          */
7898         if (!di_cache.cache_valid)
7899                 i_ddi_di_cache_free(&di_cache);
7900 
7901         mutex_exit(&di_cache.cache_lock);
7902 
7903         if (di_cache_debug)
7904                 cmn_err(CE_NOTE, "system_taskq: di_cache freed");
7905 }
7906 
7907 extern int modrootloaded;
7908 
7909 void
7910 i_ddi_di_cache_free(struct di_cache *cache)
7911 {
7912         int     error;
7913         extern int sys_shutdown;
7914 
7915         ASSERT(mutex_owned(&cache->cache_lock));
7916 
7917         if (cache->cache_size) {
7918                 ASSERT(cache->cache_size > 0);
7919                 ASSERT(cache->cache_data);
7920 
7921                 kmem_free(cache->cache_data, cache->cache_size);
7922                 cache->cache_data = NULL;
7923                 cache->cache_size = 0;
7924 
7925                 if (di_cache_debug)
7926                         cmn_err(CE_NOTE, "i_ddi_di_cache_free: freed cachemem");
7927         } else {
7928                 ASSERT(cache->cache_data == NULL);
7929                 if (di_cache_debug)
7930                         cmn_err(CE_NOTE, "i_ddi_di_cache_free: NULL cache");
7931         }
7932 
7933         if (!modrootloaded || rootvp == NULL ||
7934             vn_is_readonly(rootvp) || sys_shutdown) {
7935                 if (di_cache_debug) {
7936                         cmn_err(CE_WARN, "/ not mounted/RDONLY. Skip unlink");
7937                 }
7938                 return;
7939         }
7940 
7941         error = vn_remove(DI_CACHE_FILE, UIO_SYSSPACE, RMFILE);
7942         if (di_cache_debug && error && error != ENOENT) {
7943                 cmn_err(CE_WARN, "%s: unlink failed: %d", DI_CACHE_FILE, error);
7944         } else if (di_cache_debug && !error) {
7945                 cmn_err(CE_NOTE, "i_ddi_di_cache_free: unlinked cache file");
7946         }
7947 }
7948 
7949 void
7950 i_ddi_di_cache_invalidate()
7951 {
7952         int     cache_valid;
7953 
7954         if (!modrootloaded || !i_ddi_io_initialized()) {
7955                 if (di_cache_debug)
7956                         cmn_err(CE_NOTE, "I/O not inited. Skipping invalidate");
7957                 return;
7958         }
7959 
7960         /* Increment devtree generation number. */
7961         atomic_inc_ulong(&devtree_gen);
7962 
7963         /* Invalidate the in-core cache and dispatch free on valid->invalid */
7964         cache_valid = atomic_swap_uint(&di_cache.cache_valid, 0);
7965         if (cache_valid) {
7966                 /*
7967                  * This is an optimization to start cleaning up a cached
7968                  * snapshot early.  For this reason, it is OK for
7969                  * taskq_dispatach to fail (and it is OK to not track calling
7970                  * context relative to sleep, and assume NOSLEEP).
7971                  */
7972                 (void) taskq_dispatch(system_taskq, free_cache_task, NULL,
7973                     TQ_NOSLEEP);
7974         }
7975 
7976         if (di_cache_debug) {
7977                 cmn_err(CE_NOTE, "invalidation");
7978         }
7979 }
7980 
7981 
7982 static void
7983 i_bind_vhci_node(dev_info_t *dip)
7984 {
7985         DEVI(dip)->devi_major = ddi_name_to_major(ddi_node_name(dip));
7986         i_ddi_set_node_state(dip, DS_BOUND);
7987 }
7988 
7989 static char vhci_node_addr[2];
7990 
7991 static int
7992 i_init_vhci_node(dev_info_t *dip)
7993 {
7994         add_global_props(dip);
7995         DEVI(dip)->devi_ops = ndi_hold_driver(dip);
7996         if (DEVI(dip)->devi_ops == NULL)
7997                 return (-1);
7998 
7999         DEVI(dip)->devi_instance = e_ddi_assign_instance(dip);
8000         e_ddi_keep_instance(dip);
8001         vhci_node_addr[0]       = '\0';
8002         ddi_set_name_addr(dip, vhci_node_addr);
8003         i_ddi_set_node_state(dip, DS_INITIALIZED);
8004         return (0);
8005 }
8006 
8007 static void
8008 i_link_vhci_node(dev_info_t *dip)
8009 {
8010         ASSERT(MUTEX_HELD(&global_vhci_lock));
8011 
8012         /*
8013          * scsi_vhci should be kept left most of the device tree.
8014          */
8015         if (scsi_vhci_dip) {
8016                 DEVI(dip)->devi_sibling = DEVI(scsi_vhci_dip)->devi_sibling;
8017                 DEVI(scsi_vhci_dip)->devi_sibling = DEVI(dip);
8018         } else {
8019                 DEVI(dip)->devi_sibling = DEVI(top_devinfo)->devi_child;
8020                 DEVI(top_devinfo)->devi_child = DEVI(dip);
8021         }
8022 }
8023 
8024 
8025 /*
8026  * This a special routine to enumerate vhci node (child of rootnex
8027  * node) without holding the ndi_devi_enter() lock. The device node
8028  * is allocated, initialized and brought into DS_READY state before
8029  * inserting into the device tree. The VHCI node is handcrafted
8030  * here to bring the node to DS_READY, similar to rootnex node.
8031  *
8032  * The global_vhci_lock protects linking the node into the device
8033  * as same lock is held before linking/unlinking any direct child
8034  * of rootnex children.
8035  *
8036  * This routine is a workaround to handle a possible deadlock
8037  * that occurs while trying to enumerate node in a different sub-tree
8038  * during _init/_attach entry points.
8039  */
8040 /*ARGSUSED*/
8041 dev_info_t *
8042 ndi_devi_config_vhci(char *drvname, int flags)
8043 {
8044         struct devnames         *dnp;
8045         dev_info_t              *dip;
8046         major_t                 major = ddi_name_to_major(drvname);
8047 
8048         if (major == -1)
8049                 return (NULL);
8050 
8051         /* Make sure we create the VHCI node only once */
8052         dnp = &devnamesp[major];
8053         LOCK_DEV_OPS(&dnp->dn_lock);
8054         if (dnp->dn_head) {
8055                 dip = dnp->dn_head;
8056                 UNLOCK_DEV_OPS(&dnp->dn_lock);
8057                 return (dip);
8058         }
8059         UNLOCK_DEV_OPS(&dnp->dn_lock);
8060 
8061         /* Allocate the VHCI node */
8062         ndi_devi_alloc_sleep(top_devinfo, drvname, DEVI_SID_NODEID, &dip);
8063         ndi_hold_devi(dip);
8064 
8065         /* Mark the node as VHCI */
8066         DEVI(dip)->devi_node_attributes |= DDI_VHCI_NODE;
8067 
8068         i_ddi_add_devimap(dip);
8069         i_bind_vhci_node(dip);
8070         if (i_init_vhci_node(dip) == -1) {
8071                 ndi_rele_devi(dip);
8072                 (void) ndi_devi_free(dip);
8073                 return (NULL);
8074         }
8075 
8076         mutex_enter(&(DEVI(dip)->devi_lock));
8077         DEVI_SET_ATTACHING(dip);
8078         mutex_exit(&(DEVI(dip)->devi_lock));
8079 
8080         if (devi_attach(dip, DDI_ATTACH) != DDI_SUCCESS) {
8081                 cmn_err(CE_CONT, "Could not attach %s driver", drvname);
8082                 e_ddi_free_instance(dip, vhci_node_addr);
8083                 ndi_rele_devi(dip);
8084                 (void) ndi_devi_free(dip);
8085                 return (NULL);
8086         }
8087         mutex_enter(&(DEVI(dip)->devi_lock));
8088         DEVI_CLR_ATTACHING(dip);
8089         mutex_exit(&(DEVI(dip)->devi_lock));
8090 
8091         mutex_enter(&global_vhci_lock);
8092         i_link_vhci_node(dip);
8093         mutex_exit(&global_vhci_lock);
8094         i_ddi_set_node_state(dip, DS_READY);
8095 
8096         LOCK_DEV_OPS(&dnp->dn_lock);
8097         dnp->dn_flags |= DN_DRIVER_HELD;
8098         dnp->dn_head = dip;
8099         UNLOCK_DEV_OPS(&dnp->dn_lock);
8100 
8101         i_ndi_devi_report_status_change(dip, NULL);
8102 
8103         return (dip);
8104 }
8105 
8106 /*
8107  * Maintain DEVI_DEVICE_REMOVED hotplug devi_state for remove/reinsert hotplug
8108  * of open devices. Currently, because of tight coupling between the devfs file
8109  * system and the Solaris device tree, a driver can't always make the device
8110  * tree state (esp devi_node_state) match device hardware hotplug state. Until
8111  * resolved, to overcome this deficiency we use the following interfaces that
8112  * maintain the DEVI_DEVICE_REMOVED devi_state status bit.  These interface
8113  * report current state, and drive operation (like events and cache
8114  * invalidation) when a driver changes remove/insert state of an open device.
8115  *
8116  * The ndi_devi_device_isremoved() returns 1 if the device is currently removed.
8117  *
8118  * The ndi_devi_device_remove() interface declares the device as removed, and
8119  * returns 1 if there was a state change associated with this declaration.
8120  *
8121  * The ndi_devi_device_insert() declares the device as inserted, and returns 1
8122  * if there was a state change associated with this declaration.
8123  */
8124 int
8125 ndi_devi_device_isremoved(dev_info_t *dip)
8126 {
8127         return (DEVI_IS_DEVICE_REMOVED(dip));
8128 }
8129 
8130 int
8131 ndi_devi_device_remove(dev_info_t *dip)
8132 {
8133         ASSERT(dip && ddi_get_parent(dip) &&
8134             DEVI_BUSY_OWNED(ddi_get_parent(dip)));
8135 
8136         /* Return if already marked removed. */
8137         if (ndi_devi_device_isremoved(dip))
8138                 return (0);
8139 
8140         /* Mark the device as having been physically removed. */
8141         mutex_enter(&(DEVI(dip)->devi_lock));
8142         ndi_devi_set_hidden(dip);       /* invisible: lookup/snapshot */
8143         DEVI_SET_DEVICE_REMOVED(dip);
8144         DEVI_SET_EVREMOVE(dip);         /* this clears EVADD too */
8145         mutex_exit(&(DEVI(dip)->devi_lock));
8146 
8147         /* report remove (as 'removed') */
8148         i_ndi_devi_report_status_change(dip, NULL);
8149 
8150         /*
8151          * Invalidate the cache to ensure accurate
8152          * (di_state() & DI_DEVICE_REMOVED).
8153          */
8154         i_ddi_di_cache_invalidate();
8155 
8156         /*
8157          * Generate sysevent for those interested in removal (either
8158          * directly via private EC_DEVFS or indirectly via devfsadmd
8159          * generated EC_DEV). This will generate LDI DEVICE_REMOVE
8160          * event too.
8161          */
8162         i_ddi_log_devfs_device_remove(dip);
8163 
8164         return (1);             /* DEVICE_REMOVED state changed */
8165 }
8166 
8167 int
8168 ndi_devi_device_insert(dev_info_t *dip)
8169 {
8170         ASSERT(dip && ddi_get_parent(dip) &&
8171             DEVI_BUSY_OWNED(ddi_get_parent(dip)));
8172 
8173         /* Return if not marked removed. */
8174         if (!ndi_devi_device_isremoved(dip))
8175                 return (0);
8176 
8177         /* Mark the device as having been physically reinserted. */
8178         mutex_enter(&(DEVI(dip)->devi_lock));
8179         ndi_devi_clr_hidden(dip);       /* visible: lookup/snapshot */
8180         DEVI_SET_DEVICE_REINSERTED(dip);
8181         DEVI_SET_EVADD(dip);            /* this clears EVREMOVE too */
8182         mutex_exit(&(DEVI(dip)->devi_lock));
8183 
8184         /* report insert (as 'online') */
8185         i_ndi_devi_report_status_change(dip, NULL);
8186 
8187         /*
8188          * Invalidate the cache to ensure accurate
8189          * (di_state() & DI_DEVICE_REMOVED).
8190          */
8191         i_ddi_di_cache_invalidate();
8192 
8193         /*
8194          * Generate sysevent for those interested in removal (either directly
8195          * via EC_DEVFS or indirectly via devfsadmd generated EC_DEV).
8196          */
8197         i_ddi_log_devfs_device_insert(dip);
8198 
8199         return (1);             /* DEVICE_REMOVED state changed */
8200 }
8201 
8202 /*
8203  * ibt_hw_is_present() returns 0 when there is no IB hardware actively
8204  * running.  This is primarily useful for modules like rpcmod which
8205  * needs a quick check to decide whether or not it should try to use
8206  * InfiniBand
8207  */
8208 int ib_hw_status = 0;
8209 int
8210 ibt_hw_is_present()
8211 {
8212         return (ib_hw_status);
8213 }
8214 
8215 /*
8216  * ASSERT that constraint flag is not set and then set the "retire attempt"
8217  * flag.
8218  */
8219 int
8220 e_ddi_mark_retiring(dev_info_t *dip, void *arg)
8221 {
8222         char    **cons_array = (char **)arg;
8223         char    *path;
8224         int     constraint;
8225         int     i;
8226 
8227         constraint = 0;
8228         if (cons_array) {
8229                 path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8230                 (void) ddi_pathname(dip, path);
8231                 for (i = 0; cons_array[i] != NULL; i++) {
8232                         if (strcmp(path, cons_array[i]) == 0) {
8233                                 constraint = 1;
8234                                 break;
8235                         }
8236                 }
8237                 kmem_free(path, MAXPATHLEN);
8238         }
8239 
8240         mutex_enter(&DEVI(dip)->devi_lock);
8241         ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8242         DEVI(dip)->devi_flags |= DEVI_RETIRING;
8243         if (constraint)
8244                 DEVI(dip)->devi_flags |= DEVI_R_CONSTRAINT;
8245         mutex_exit(&DEVI(dip)->devi_lock);
8246 
8247         RIO_VERBOSE((CE_NOTE, "marked dip as undergoing retire process dip=%p",
8248             (void *)dip));
8249 
8250         if (constraint)
8251                 RIO_DEBUG((CE_NOTE, "marked dip as constrained, dip=%p",
8252                     (void *)dip));
8253 
8254         if (MDI_PHCI(dip))
8255                 mdi_phci_mark_retiring(dip, cons_array);
8256 
8257         return (DDI_WALK_CONTINUE);
8258 }
8259 
8260 static void
8261 free_array(char **cons_array)
8262 {
8263         int     i;
8264 
8265         if (cons_array == NULL)
8266                 return;
8267 
8268         for (i = 0; cons_array[i] != NULL; i++) {
8269                 kmem_free(cons_array[i], strlen(cons_array[i]) + 1);
8270         }
8271         kmem_free(cons_array, (i+1) * sizeof (char *));
8272 }
8273 
8274 /*
8275  * Walk *every* node in subtree and check if it blocks, allows or has no
8276  * comment on a proposed retire.
8277  */
8278 int
8279 e_ddi_retire_notify(dev_info_t *dip, void *arg)
8280 {
8281         int     *constraint = (int *)arg;
8282 
8283         RIO_DEBUG((CE_NOTE, "retire notify: dip = %p", (void *)dip));
8284 
8285         (void) e_ddi_offline_notify(dip);
8286 
8287         mutex_enter(&(DEVI(dip)->devi_lock));
8288         if (!(DEVI(dip)->devi_flags & DEVI_RETIRING)) {
8289                 RIO_DEBUG((CE_WARN, "retire notify: dip in retire "
8290                     "subtree is not marked: dip = %p", (void *)dip));
8291                 *constraint = 0;
8292         } else if (DEVI(dip)->devi_flags & DEVI_R_BLOCKED) {
8293                 ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8294                 RIO_DEBUG((CE_NOTE, "retire notify: BLOCKED: dip = %p",
8295                     (void *)dip));
8296                 *constraint = 0;
8297         } else if (!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT)) {
8298                 RIO_DEBUG((CE_NOTE, "retire notify: NO CONSTRAINT: "
8299                     "dip = %p", (void *)dip));
8300                 *constraint = 0;
8301         } else {
8302                 RIO_DEBUG((CE_NOTE, "retire notify: CONSTRAINT set: "
8303                     "dip = %p", (void *)dip));
8304         }
8305         mutex_exit(&DEVI(dip)->devi_lock);
8306 
8307         if (MDI_PHCI(dip))
8308                 mdi_phci_retire_notify(dip, constraint);
8309 
8310         return (DDI_WALK_CONTINUE);
8311 }
8312 
8313 int
8314 e_ddi_retire_finalize(dev_info_t *dip, void *arg)
8315 {
8316         int constraint = *(int *)arg;
8317         int finalize;
8318         int phci_only;
8319 
8320         mutex_enter(&DEVI(dip)->devi_lock);
8321         if (!(DEVI(dip)->devi_flags & DEVI_RETIRING)) {
8322                 RIO_DEBUG((CE_WARN,
8323                     "retire: unmarked dip(%p) in retire subtree",
8324                     (void *)dip));
8325                 ASSERT(!(DEVI(dip)->devi_flags & DEVI_RETIRED));
8326                 ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8327                 ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_BLOCKED));
8328                 mutex_exit(&DEVI(dip)->devi_lock);
8329                 return (DDI_WALK_CONTINUE);
8330         }
8331 
8332         /*
8333          * retire the device if constraints have been applied
8334          * or if the device is not in use
8335          */
8336         finalize = 0;
8337         if (constraint) {
8338                 ASSERT(DEVI_BUSY_OWNED(ddi_get_parent(dip)));
8339 
8340                 ASSERT(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT);
8341                 ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_BLOCKED));
8342                 DEVI(dip)->devi_flags &= ~DEVI_R_CONSTRAINT;
8343                 DEVI(dip)->devi_flags &= ~DEVI_RETIRING;
8344                 DEVI(dip)->devi_flags |= DEVI_RETIRED;
8345                 mutex_exit(&DEVI(dip)->devi_lock);
8346                 (void) spec_fence_snode(dip, NULL);
8347                 RIO_DEBUG((CE_NOTE, "Fenced off: dip = %p", (void *)dip));
8348                 e_ddi_offline_finalize(dip, DDI_SUCCESS);
8349         } else {
8350                 if (DEVI(dip)->devi_flags & DEVI_R_BLOCKED) {
8351                         ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8352                         DEVI(dip)->devi_flags &= ~DEVI_R_BLOCKED;
8353                         DEVI(dip)->devi_flags &= ~DEVI_RETIRING;
8354                         /* we have already finalized during notify */
8355                 } else if (DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT) {
8356                         DEVI(dip)->devi_flags &= ~DEVI_R_CONSTRAINT;
8357                         DEVI(dip)->devi_flags &= ~DEVI_RETIRING;
8358                         finalize = 1;
8359                 } else {
8360                         DEVI(dip)->devi_flags &= ~DEVI_RETIRING;
8361                         /*
8362                          * even if no contracts, need to call finalize
8363                          * to clear the contract barrier on the dip
8364                          */
8365                         finalize = 1;
8366                 }
8367                 mutex_exit(&DEVI(dip)->devi_lock);
8368                 RIO_DEBUG((CE_NOTE, "finalize: NOT retired: dip = %p",
8369                     (void *)dip));
8370                 if (finalize)
8371                         e_ddi_offline_finalize(dip, DDI_FAILURE);
8372         }
8373 
8374         /*
8375          * phci_only variable indicates no client checking, just
8376          * offline the PHCI. We set that to 0 to enable client
8377          * checking
8378          */
8379         phci_only = 0;
8380         if (MDI_PHCI(dip))
8381                 mdi_phci_retire_finalize(dip, phci_only, arg);
8382 
8383         return (DDI_WALK_CONTINUE);
8384 }
8385 
8386 /*
8387  * Returns
8388  *      DDI_SUCCESS if constraints allow retire
8389  *      DDI_FAILURE if constraints don't allow retire.
8390  * cons_array is a NULL terminated array of node paths for
8391  * which constraints have already been applied.
8392  */
8393 int
8394 e_ddi_retire_device(char *path, char **cons_array)
8395 {
8396         dev_info_t      *dip;
8397         dev_info_t      *pdip;
8398         int             circ;
8399         int             circ2;
8400         int             constraint;
8401         char            *devnm;
8402 
8403         /*
8404          * First, lookup the device
8405          */
8406         dip = e_ddi_hold_devi_by_path(path, 0);
8407         if (dip == NULL) {
8408                 /*
8409                  * device does not exist. This device cannot be
8410                  * a critical device since it is not in use. Thus
8411                  * this device is always retireable. Return DDI_SUCCESS
8412                  * to indicate this. If this device is ever
8413                  * instantiated, I/O framework will consult the
8414                  * the persistent retire store, mark it as
8415                  * retired and fence it off.
8416                  */
8417                 RIO_DEBUG((CE_NOTE, "Retire device: device doesn't exist."
8418                     " NOP. Just returning SUCCESS. path=%s", path));
8419                 free_array(cons_array);
8420                 return (DDI_SUCCESS);
8421         }
8422 
8423         RIO_DEBUG((CE_NOTE, "Retire device: found dip = %p.", (void *)dip));
8424 
8425         pdip = ddi_get_parent(dip);
8426         ndi_hold_devi(pdip);
8427 
8428         /*
8429          * Run devfs_clean() in case dip has no constraints and is
8430          * not in use, so is retireable but there are dv_nodes holding
8431          * ref-count on the dip. Note that devfs_clean() always returns
8432          * success.
8433          */
8434         devnm = kmem_alloc(MAXNAMELEN + 1, KM_SLEEP);
8435         (void) ddi_deviname(dip, devnm);
8436         (void) devfs_clean(pdip, devnm + 1, DV_CLEAN_FORCE);
8437         kmem_free(devnm, MAXNAMELEN + 1);
8438 
8439         ndi_devi_enter(pdip, &circ);
8440 
8441         /* release hold from e_ddi_hold_devi_by_path */
8442         ndi_rele_devi(dip);
8443 
8444         /*
8445          * If it cannot make a determination, is_leaf_node() assumes
8446          * dip is a nexus.
8447          */
8448         (void) e_ddi_mark_retiring(dip, cons_array);
8449         if (!is_leaf_node(dip)) {
8450                 ndi_devi_enter(dip, &circ2);
8451                 ddi_walk_devs(ddi_get_child(dip), e_ddi_mark_retiring,
8452                     cons_array);
8453                 ndi_devi_exit(dip, circ2);
8454         }
8455         free_array(cons_array);
8456 
8457         /*
8458          * apply constraints
8459          */
8460         RIO_DEBUG((CE_NOTE, "retire: subtree retire notify: path = %s", path));
8461 
8462         constraint = 1; /* assume constraints allow retire */
8463         (void) e_ddi_retire_notify(dip, &constraint);
8464         if (!is_leaf_node(dip)) {
8465                 ndi_devi_enter(dip, &circ2);
8466                 ddi_walk_devs(ddi_get_child(dip), e_ddi_retire_notify,
8467                     &constraint);
8468                 ndi_devi_exit(dip, circ2);
8469         }
8470 
8471         /*
8472          * Now finalize the retire
8473          */
8474         (void) e_ddi_retire_finalize(dip, &constraint);
8475         if (!is_leaf_node(dip)) {
8476                 ndi_devi_enter(dip, &circ2);
8477                 ddi_walk_devs(ddi_get_child(dip), e_ddi_retire_finalize,
8478                     &constraint);
8479                 ndi_devi_exit(dip, circ2);
8480         }
8481 
8482         if (!constraint) {
8483                 RIO_DEBUG((CE_WARN, "retire failed: path = %s", path));
8484         } else {
8485                 RIO_DEBUG((CE_NOTE, "retire succeeded: path = %s", path));
8486         }
8487 
8488         ndi_devi_exit(pdip, circ);
8489         ndi_rele_devi(pdip);
8490         return (constraint ? DDI_SUCCESS : DDI_FAILURE);
8491 }
8492 
8493 static int
8494 unmark_and_unfence(dev_info_t *dip, void *arg)
8495 {
8496         char    *path = (char *)arg;
8497 
8498         ASSERT(path);
8499 
8500         (void) ddi_pathname(dip, path);
8501 
8502         mutex_enter(&DEVI(dip)->devi_lock);
8503         DEVI(dip)->devi_flags &= ~DEVI_RETIRED;
8504         DEVI_SET_DEVICE_ONLINE(dip);
8505         mutex_exit(&DEVI(dip)->devi_lock);
8506 
8507         RIO_VERBOSE((CE_NOTE, "Cleared RETIRED flag: dip=%p, path=%s",
8508             (void *)dip, path));
8509 
8510         (void) spec_unfence_snode(dip);
8511         RIO_DEBUG((CE_NOTE, "Unfenced device: %s", path));
8512 
8513         if (MDI_PHCI(dip))
8514                 mdi_phci_unretire(dip);
8515 
8516         return (DDI_WALK_CONTINUE);
8517 }
8518 
8519 struct find_dip {
8520         char    *fd_buf;
8521         char    *fd_path;
8522         dev_info_t *fd_dip;
8523 };
8524 
8525 static int
8526 find_dip_fcn(dev_info_t *dip, void *arg)
8527 {
8528         struct find_dip *findp = (struct find_dip *)arg;
8529 
8530         (void) ddi_pathname(dip, findp->fd_buf);
8531 
8532         if (strcmp(findp->fd_path, findp->fd_buf) != 0)
8533                 return (DDI_WALK_CONTINUE);
8534 
8535         ndi_hold_devi(dip);
8536         findp->fd_dip = dip;
8537 
8538         return (DDI_WALK_TERMINATE);
8539 }
8540 
8541 int
8542 e_ddi_unretire_device(char *path)
8543 {
8544         int             circ;
8545         int             circ2;
8546         char            *path2;
8547         dev_info_t      *pdip;
8548         dev_info_t      *dip;
8549         struct find_dip  find_dip;
8550 
8551         ASSERT(path);
8552         ASSERT(*path == '/');
8553 
8554         if (strcmp(path, "/") == 0) {
8555                 cmn_err(CE_WARN, "Root node cannot be retired. Skipping "
8556                     "device unretire: %s", path);
8557                 return (0);
8558         }
8559 
8560         /*
8561          * We can't lookup the dip (corresponding to path) via
8562          * e_ddi_hold_devi_by_path() because the dip may be offline
8563          * and may not attach. Use ddi_walk_devs() instead;
8564          */
8565         find_dip.fd_buf = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8566         find_dip.fd_path = path;
8567         find_dip.fd_dip = NULL;
8568 
8569         pdip = ddi_root_node();
8570 
8571         ndi_devi_enter(pdip, &circ);
8572         ddi_walk_devs(ddi_get_child(pdip), find_dip_fcn, &find_dip);
8573         ndi_devi_exit(pdip, circ);
8574 
8575         kmem_free(find_dip.fd_buf, MAXPATHLEN);
8576 
8577         if (find_dip.fd_dip == NULL) {
8578                 cmn_err(CE_WARN, "Device not found in device tree. Skipping "
8579                     "device unretire: %s", path);
8580                 return (0);
8581         }
8582 
8583         dip = find_dip.fd_dip;
8584 
8585         pdip = ddi_get_parent(dip);
8586 
8587         ndi_hold_devi(pdip);
8588 
8589         ndi_devi_enter(pdip, &circ);
8590 
8591         path2 = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8592 
8593         (void) unmark_and_unfence(dip, path2);
8594         if (!is_leaf_node(dip)) {
8595                 ndi_devi_enter(dip, &circ2);
8596                 ddi_walk_devs(ddi_get_child(dip), unmark_and_unfence, path2);
8597                 ndi_devi_exit(dip, circ2);
8598         }
8599 
8600         kmem_free(path2, MAXPATHLEN);
8601 
8602         /* release hold from find_dip_fcn() */
8603         ndi_rele_devi(dip);
8604 
8605         ndi_devi_exit(pdip, circ);
8606 
8607         ndi_rele_devi(pdip);
8608 
8609         return (0);
8610 }
8611 
8612 /*
8613  * Called before attach on a dip that has been retired.
8614  */
8615 static int
8616 mark_and_fence(dev_info_t *dip, void *arg)
8617 {
8618         char    *fencepath = (char *)arg;
8619 
8620         /*
8621          * We have already decided to retire this device. The various
8622          * constraint checking should not be set.
8623          * NOTE that the retire flag may already be set due to
8624          * fenced -> detach -> fenced transitions.
8625          */
8626         mutex_enter(&DEVI(dip)->devi_lock);
8627         ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_CONSTRAINT));
8628         ASSERT(!(DEVI(dip)->devi_flags & DEVI_R_BLOCKED));
8629         ASSERT(!(DEVI(dip)->devi_flags & DEVI_RETIRING));
8630         DEVI(dip)->devi_flags |= DEVI_RETIRED;
8631         mutex_exit(&DEVI(dip)->devi_lock);
8632         RIO_VERBOSE((CE_NOTE, "marked as RETIRED dip=%p", (void *)dip));
8633 
8634         if (fencepath) {
8635                 (void) spec_fence_snode(dip, NULL);
8636                 RIO_DEBUG((CE_NOTE, "Fenced: %s",
8637                     ddi_pathname(dip, fencepath)));
8638         }
8639 
8640         return (DDI_WALK_CONTINUE);
8641 }
8642 
8643 /*
8644  * Checks the retire database and:
8645  *
8646  * - if device is present in the retire database, marks the device retired
8647  *   and fences it off.
8648  * - if device is not in retire database, allows the device to attach normally
8649  *
8650  * To be called only by framework attach code on first attach attempt.
8651  *
8652  */
8653 static int
8654 i_ddi_check_retire(dev_info_t *dip)
8655 {
8656         char            *path;
8657         dev_info_t      *pdip;
8658         int             circ;
8659         int             phci_only;
8660         int             constraint;
8661 
8662         pdip = ddi_get_parent(dip);
8663 
8664         /*
8665          * Root dip is treated special and doesn't take this code path.
8666          * Also root can never be retired.
8667          */
8668         ASSERT(pdip);
8669         ASSERT(DEVI_BUSY_OWNED(pdip));
8670         ASSERT(i_ddi_node_state(dip) < DS_ATTACHED);
8671 
8672         path = kmem_alloc(MAXPATHLEN, KM_SLEEP);
8673 
8674         (void) ddi_pathname(dip, path);
8675 
8676         RIO_VERBOSE((CE_NOTE, "Checking if dip should attach: dip=%p, path=%s",
8677             (void *)dip, path));
8678 
8679         /*
8680          * Check if this device is in the "retired" store i.e.  should
8681          * be retired. If not, we have nothing to do.
8682          */
8683         if (e_ddi_device_retired(path) == 0) {
8684                 RIO_VERBOSE((CE_NOTE, "device is NOT retired: path=%s", path));
8685                 if (DEVI(dip)->devi_flags & DEVI_RETIRED)
8686                         (void) e_ddi_unretire_device(path);
8687                 kmem_free(path, MAXPATHLEN);
8688                 return (0);
8689         }
8690 
8691         RIO_DEBUG((CE_NOTE, "attach: device is retired: path=%s", path));
8692 
8693         /*
8694          * Mark dips and fence off snodes (if any)
8695          */
8696         RIO_DEBUG((CE_NOTE, "attach: Mark and fence subtree: path=%s", path));
8697         (void) mark_and_fence(dip, path);
8698         if (!is_leaf_node(dip)) {
8699                 ndi_devi_enter(dip, &circ);
8700                 ddi_walk_devs(ddi_get_child(dip), mark_and_fence, path);
8701                 ndi_devi_exit(dip, circ);
8702         }
8703 
8704         kmem_free(path, MAXPATHLEN);
8705 
8706         /*
8707          * We don't want to check the client. We just want to
8708          * offline the PHCI
8709          */
8710         phci_only = 1;
8711         constraint = 1;
8712         if (MDI_PHCI(dip))
8713                 mdi_phci_retire_finalize(dip, phci_only, &constraint);
8714         return (1);
8715 }
8716 
8717 
8718 #define VAL_ALIAS(array, x)     (strlen(array[x].pair_alias))
8719 #define VAL_CURR(array, x)      (strlen(array[x].pair_curr))
8720 #define SWAP(array, x, y)                       \
8721 {                                               \
8722         alias_pair_t tmpair = array[x];         \
8723         array[x] = array[y];                    \
8724         array[y] = tmpair;                      \
8725 }
8726 
8727 static int
8728 partition_curr(alias_pair_t *array, int start, int end)
8729 {
8730         int     i = start - 1;
8731         int     j = end + 1;
8732         int     pivot = start;
8733 
8734         for (;;) {
8735                 do {
8736                         j--;
8737                 } while (VAL_CURR(array, j) > VAL_CURR(array, pivot));
8738 
8739                 do {
8740                         i++;
8741                 } while (VAL_CURR(array, i) < VAL_CURR(array, pivot));
8742 
8743                 if (i < j)
8744                         SWAP(array, i, j)
8745                 else
8746                         return (j);
8747         }
8748 }
8749 
8750 static int
8751 partition_aliases(alias_pair_t *array, int start, int end)
8752 {
8753         int     i = start - 1;
8754         int     j = end + 1;
8755         int     pivot = start;
8756 
8757         for (;;) {
8758                 do {
8759                         j--;
8760                 } while (VAL_ALIAS(array, j) > VAL_ALIAS(array, pivot));
8761 
8762                 do {
8763                         i++;
8764                 } while (VAL_ALIAS(array, i) < VAL_ALIAS(array, pivot));
8765 
8766                 if (i < j)
8767                         SWAP(array, i, j)
8768                 else
8769                         return (j);
8770         }
8771 }
8772 static void
8773 sort_alias_pairs(alias_pair_t *array, int start, int end)
8774 {
8775         int mid;
8776 
8777         if (start < end) {
8778                 mid = partition_aliases(array, start, end);
8779                 sort_alias_pairs(array, start, mid);
8780                 sort_alias_pairs(array, mid + 1, end);
8781         }
8782 }
8783 
8784 static void
8785 sort_curr_pairs(alias_pair_t *array, int start, int end)
8786 {
8787         int mid;
8788 
8789         if (start < end) {
8790                 mid = partition_curr(array, start, end);
8791                 sort_curr_pairs(array, start, mid);
8792                 sort_curr_pairs(array, mid + 1, end);
8793         }
8794 }
8795 
8796 static void
8797 create_sorted_pairs(plat_alias_t *pali, int npali)
8798 {
8799         int             i;
8800         int             j;
8801         int             k;
8802         int             count;
8803 
8804         count = 0;
8805         for (i = 0; i < npali; i++) {
8806                 count += pali[i].pali_naliases;
8807         }
8808 
8809         ddi_aliases.dali_alias_pairs = kmem_zalloc(
8810             (sizeof (alias_pair_t)) * count, KM_NOSLEEP);
8811         if (ddi_aliases.dali_alias_pairs == NULL) {
8812                 cmn_err(CE_PANIC, "alias path-pair alloc failed");
8813                 /*NOTREACHED*/
8814         }
8815 
8816         ddi_aliases.dali_curr_pairs = kmem_zalloc(
8817             (sizeof (alias_pair_t)) * count, KM_NOSLEEP);
8818         if (ddi_aliases.dali_curr_pairs == NULL) {
8819                 cmn_err(CE_PANIC, "curr path-pair alloc failed");
8820                 /*NOTREACHED*/
8821         }
8822 
8823         for (i = 0, k = 0; i < npali; i++) {
8824                 for (j = 0; j < pali[i].pali_naliases; j++, k++) {
8825                         ddi_aliases.dali_alias_pairs[k].pair_curr =
8826                             ddi_aliases.dali_curr_pairs[k].pair_curr =
8827                             pali[i].pali_current;
8828                         ddi_aliases.dali_alias_pairs[k].pair_alias =
8829                             ddi_aliases.dali_curr_pairs[k].pair_alias =
8830                             pali[i].pali_aliases[j];
8831                 }
8832         }
8833 
8834         ASSERT(k == count);
8835 
8836         ddi_aliases.dali_num_pairs = count;
8837 
8838         /* Now sort the array based on length of pair_alias */
8839         sort_alias_pairs(ddi_aliases.dali_alias_pairs, 0, count - 1);
8840         sort_curr_pairs(ddi_aliases.dali_curr_pairs, 0, count - 1);
8841 }
8842 
8843 void
8844 ddi_register_aliases(plat_alias_t *pali, uint64_t npali)
8845 {
8846 
8847         ASSERT((pali == NULL) ^ (npali != 0));
8848 
8849         if (npali == 0) {
8850                 ddi_err(DER_PANIC, NULL, "npali == 0");
8851                 /*NOTREACHED*/
8852         }
8853 
8854         if (ddi_aliases_present == B_TRUE) {
8855                 ddi_err(DER_PANIC, NULL, "multiple init");
8856                 /*NOTREACHED*/
8857         }
8858 
8859         ddi_aliases.dali_alias_TLB = mod_hash_create_strhash(
8860             "ddi-alias-tlb", DDI_ALIAS_HASH_SIZE, mod_hash_null_valdtor);
8861         if (ddi_aliases.dali_alias_TLB == NULL) {
8862                 ddi_err(DER_PANIC, NULL, "alias TLB hash alloc failed");
8863                 /*NOTREACHED*/
8864         }
8865 
8866         ddi_aliases.dali_curr_TLB = mod_hash_create_strhash(
8867             "ddi-curr-tlb", DDI_ALIAS_HASH_SIZE, mod_hash_null_valdtor);
8868         if (ddi_aliases.dali_curr_TLB == NULL) {
8869                 ddi_err(DER_PANIC, NULL, "curr TLB hash alloc failed");
8870                 /*NOTREACHED*/
8871         }
8872 
8873         create_sorted_pairs(pali, npali);
8874 
8875         tsd_create(&tsd_ddi_redirect, NULL);
8876 
8877         ddi_aliases_present = B_TRUE;
8878 }
8879 
8880 static dev_info_t *
8881 path_to_dip(char *path)
8882 {
8883         dev_info_t      *currdip;
8884         int             error;
8885         char            *pdup;
8886 
8887         pdup = ddi_strdup(path, KM_NOSLEEP);
8888         if (pdup == NULL) {
8889                 cmn_err(CE_PANIC, "path strdup failed: %s", path);
8890                 /*NOTREACHED*/
8891         }
8892 
8893         error = resolve_pathname(pdup, &currdip, NULL, NULL);
8894 
8895         kmem_free(pdup, strlen(path) + 1);
8896 
8897         return (error ? NULL : currdip);
8898 }
8899 
8900 dev_info_t *
8901 ddi_alias_to_currdip(char *alias, int i)
8902 {
8903         alias_pair_t *pair;
8904         char *curr;
8905         dev_info_t *currdip = NULL;
8906         char *aliasdup;
8907         int rv, len;
8908 
8909         pair = &(ddi_aliases.dali_alias_pairs[i]);
8910         len = strlen(pair->pair_alias);
8911 
8912         curr = NULL;
8913         aliasdup = ddi_strdup(alias, KM_NOSLEEP);
8914         if (aliasdup == NULL) {
8915                 cmn_err(CE_PANIC, "aliasdup alloc failed");
8916                 /*NOTREACHED*/
8917         }
8918 
8919         if (strncmp(alias, pair->pair_alias, len)  != 0)
8920                 goto out;
8921 
8922         if (alias[len] != '/' && alias[len] != '\0')
8923                 goto out;
8924 
8925         curr = kmem_alloc(MAXPATHLEN, KM_NOSLEEP);
8926         if (curr == NULL) {
8927                 cmn_err(CE_PANIC, "curr alloc failed");
8928                 /*NOTREACHED*/
8929         }
8930         (void) strlcpy(curr, pair->pair_curr, MAXPATHLEN);
8931         if (alias[len] == '/') {
8932                 (void) strlcat(curr, "/", MAXPATHLEN);
8933                 (void) strlcat(curr, &alias[len + 1], MAXPATHLEN);
8934         }
8935 
8936         currdip = path_to_dip(curr);
8937 
8938 out:
8939         if (currdip) {
8940                 rv = mod_hash_insert(ddi_aliases.dali_alias_TLB,
8941                     (mod_hash_key_t)aliasdup, (mod_hash_val_t)curr);
8942                 if (rv != 0) {
8943                         kmem_free(curr, MAXPATHLEN);
8944                         strfree(aliasdup);
8945                 }
8946         } else {
8947                 rv = mod_hash_insert(ddi_aliases.dali_alias_TLB,
8948                     (mod_hash_key_t)aliasdup, (mod_hash_val_t)NULL);
8949                 if (rv != 0) {
8950                         strfree(aliasdup);
8951                 }
8952                 if (curr)
8953                         kmem_free(curr, MAXPATHLEN);
8954         }
8955 
8956         return (currdip);
8957 }
8958 
8959 char *
8960 ddi_curr_to_alias(char *curr, int i)
8961 {
8962         alias_pair_t    *pair;
8963         char            *alias;
8964         char            *currdup;
8965         int             len;
8966         int             rv;
8967 
8968         pair = &(ddi_aliases.dali_curr_pairs[i]);
8969 
8970         len = strlen(pair->pair_curr);
8971 
8972         alias = NULL;
8973 
8974         currdup = ddi_strdup(curr, KM_NOSLEEP);
8975         if (currdup == NULL) {
8976                 cmn_err(CE_PANIC, "currdup alloc failed");
8977                 /*NOTREACHED*/
8978         }
8979 
8980         if (strncmp(curr, pair->pair_curr, len) != 0)
8981                 goto out;
8982 
8983         if (curr[len] != '/' && curr[len] != '\0')
8984                 goto out;
8985 
8986         alias = kmem_alloc(MAXPATHLEN, KM_NOSLEEP);
8987         if (alias == NULL) {
8988                 cmn_err(CE_PANIC, "alias alloc failed");
8989                 /*NOTREACHED*/
8990         }
8991 
8992         (void) strlcpy(alias, pair->pair_alias, MAXPATHLEN);
8993         if (curr[len] == '/') {
8994                 (void) strlcat(alias, "/", MAXPATHLEN);
8995                 (void) strlcat(alias, &curr[len + 1], MAXPATHLEN);
8996         }
8997 
8998         if (e_ddi_path_to_instance(alias) == NULL) {
8999                 kmem_free(alias, MAXPATHLEN);
9000                 alias = NULL;
9001         }
9002 
9003 out:
9004         rv = mod_hash_insert(ddi_aliases.dali_curr_TLB,
9005             (mod_hash_key_t)currdup, (mod_hash_val_t)alias);
9006         if (rv != 0) {
9007                 strfree(currdup);
9008         }
9009 
9010         return (alias);
9011 }
9012 
9013 dev_info_t *
9014 ddi_alias_redirect(char *alias)
9015 {
9016         char            *curr;
9017         dev_info_t      *currdip;
9018         int             i;
9019 
9020         if (ddi_aliases_present == B_FALSE)
9021                 return (NULL);
9022 
9023         if (tsd_get(tsd_ddi_redirect))
9024                 return (NULL);
9025 
9026         (void) tsd_set(tsd_ddi_redirect, (void *)1);
9027 
9028         ASSERT(ddi_aliases.dali_alias_TLB);
9029         ASSERT(ddi_aliases.dali_alias_pairs);
9030 
9031         curr = NULL;
9032         if (mod_hash_find(ddi_aliases.dali_alias_TLB,
9033             (mod_hash_key_t)alias, (mod_hash_val_t *)&curr) == 0) {
9034                 currdip = curr ? path_to_dip(curr) : NULL;
9035                 goto out;
9036         }
9037 
9038         /* The TLB has no translation, do it the hard way */
9039         currdip = NULL;
9040         for (i = ddi_aliases.dali_num_pairs - 1; i >= 0; i--) {
9041                 currdip = ddi_alias_to_currdip(alias, i);
9042                 if (currdip)
9043                         break;
9044         }
9045 out:
9046         (void) tsd_set(tsd_ddi_redirect, NULL);
9047 
9048         return (currdip);
9049 }
9050 
9051 char *
9052 ddi_curr_redirect(char *curr)
9053 {
9054         char    *alias;
9055         int i;
9056 
9057         if (ddi_aliases_present == B_FALSE)
9058                 return (NULL);
9059 
9060         if (tsd_get(tsd_ddi_redirect))
9061                 return (NULL);
9062 
9063         (void) tsd_set(tsd_ddi_redirect, (void *)1);
9064 
9065         ASSERT(ddi_aliases.dali_curr_TLB);
9066         ASSERT(ddi_aliases.dali_curr_pairs);
9067 
9068         alias = NULL;
9069         if (mod_hash_find(ddi_aliases.dali_curr_TLB,
9070             (mod_hash_key_t)curr, (mod_hash_val_t *)&alias) == 0) {
9071                 goto out;
9072         }
9073 
9074 
9075         /* The TLB has no translation, do it the slow way */
9076         alias = NULL;
9077         for (i = ddi_aliases.dali_num_pairs - 1; i >= 0; i--) {
9078                 alias = ddi_curr_to_alias(curr, i);
9079                 if (alias)
9080                         break;
9081         }
9082 
9083 out:
9084         (void) tsd_set(tsd_ddi_redirect, NULL);
9085 
9086         return (alias);
9087 }
9088 
9089 void
9090 ddi_err(ddi_err_t ade, dev_info_t *rdip, const char *fmt, ...)
9091 {
9092         va_list ap;
9093         char strbuf[256];
9094         char *buf;
9095         size_t buflen, tlen;
9096         int ce;
9097         int de;
9098         const char *fmtbad = "Invalid arguments to ddi_err()";
9099 
9100         de = DER_CONT;
9101         strbuf[1] = '\0';
9102 
9103         switch (ade) {
9104         case DER_CONS:
9105                 strbuf[0] = '^';
9106                 break;
9107         case DER_LOG:
9108                 strbuf[0] = '!';
9109                 break;
9110         case DER_VERB:
9111                 strbuf[0] = '?';
9112                 break;
9113         default:
9114                 strbuf[0] = '\0';
9115                 de = ade;
9116                 break;
9117         }
9118 
9119         tlen = strlen(strbuf);
9120         buf = strbuf + tlen;
9121         buflen = sizeof (strbuf) - tlen;
9122 
9123         if (rdip && ddi_get_instance(rdip) == -1) {
9124                 (void) snprintf(buf, buflen, "%s: ",
9125                     ddi_driver_name(rdip));
9126         } else if (rdip) {
9127                 (void) snprintf(buf, buflen, "%s%d: ",
9128                     ddi_driver_name(rdip), ddi_get_instance(rdip));
9129         }
9130 
9131         tlen = strlen(strbuf);
9132         buf = strbuf + tlen;
9133         buflen = sizeof (strbuf) - tlen;
9134 
9135         va_start(ap, fmt);
9136         switch (de) {
9137         case DER_CONT:
9138                 (void) vsnprintf(buf, buflen, fmt, ap);
9139                 if (ade != DER_CONT) {
9140                         (void) strlcat(strbuf, "\n", sizeof (strbuf));
9141                 }
9142                 ce = CE_CONT;
9143                 break;
9144         case DER_NOTE:
9145                 (void) vsnprintf(buf, buflen, fmt, ap);
9146                 ce = CE_NOTE;
9147                 break;
9148         case DER_WARN:
9149                 (void) vsnprintf(buf, buflen, fmt, ap);
9150                 ce = CE_WARN;
9151                 break;
9152         case DER_MODE:
9153                 (void) vsnprintf(buf, buflen, fmt, ap);
9154                 if (ddi_err_panic == B_TRUE) {
9155                         ce = CE_PANIC;
9156                 } else {
9157                         ce = CE_WARN;
9158                 }
9159                 break;
9160         case DER_DEBUG:
9161                 (void) snprintf(buf, buflen, "DEBUG: ");
9162                 tlen = strlen("DEBUG: ");
9163                 (void) vsnprintf(buf + tlen, buflen - tlen, fmt, ap);
9164                 ce = CE_CONT;
9165                 break;
9166         case DER_PANIC:
9167                 (void) vsnprintf(buf, buflen, fmt, ap);
9168                 ce = CE_PANIC;
9169                 break;
9170         case DER_INVALID:
9171         default:
9172                 (void) snprintf(buf, buflen, fmtbad);
9173                 tlen = strlen(fmtbad);
9174                 (void) vsnprintf(buf + tlen, buflen - tlen, fmt, ap);
9175                 ce = CE_PANIC;
9176                 break;
9177         }
9178         va_end(ap);
9179 
9180         cmn_err(ce, strbuf);
9181 }
9182 
9183 /*ARGSUSED*/
9184 void
9185 ddi_mem_update(uint64_t addr, uint64_t size)
9186 {
9187 #if defined(__x86) && !defined(__xpv)
9188         extern void immu_physmem_update(uint64_t addr, uint64_t size);
9189         immu_physmem_update(addr, size);
9190 #else
9191         /*LINTED*/
9192         ;
9193 #endif
9194 }