1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 2012 by Delphix. All rights reserved.
  25  * Copyright 2014 Igor Kozhukhov <ikozhukhov@gmail.com>.
  26  * Copyright 2018 Joyent, Inc.
  27  * Copyright 2017 RackTop Systems.
  28  */
  29 
  30 #ifndef _SYS_CPUVAR_H
  31 #define _SYS_CPUVAR_H
  32 
  33 #include <sys/thread.h>
  34 #include <sys/sysinfo.h>  /* has cpu_stat_t definition */
  35 #include <sys/disp.h>
  36 #include <sys/processor.h>
  37 #include <sys/kcpc.h>             /* has kcpc_ctx_t definition */
  38 
  39 #include <sys/loadavg.h>
  40 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
  41 #include <sys/machcpuvar.h>
  42 #endif
  43 
  44 #include <sys/types.h>
  45 #include <sys/file.h>
  46 #include <sys/bitmap.h>
  47 #include <sys/rwlock.h>
  48 #include <sys/msacct.h>
  49 #if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) && \
  50         (defined(__i386) || defined(__amd64))
  51 #include <asm/cpuvar.h>
  52 #endif
  53 
  54 #ifdef  __cplusplus
  55 extern "C" {
  56 #endif
  57 
  58 struct squeue_set_s;
  59 
  60 #define CPU_CACHE_COHERENCE_SIZE        64
  61 
  62 /*
  63  * For fast event tracing.
  64  */
  65 struct ftrace_record;
  66 typedef struct ftrace_data {
  67         int                     ftd_state;      /* ftrace flags */
  68         kmutex_t                ftd_unused;     /* ftrace buffer lock, unused */
  69         struct ftrace_record    *ftd_cur;       /* current record */
  70         struct ftrace_record    *ftd_first;     /* first record */
  71         struct ftrace_record    *ftd_last;      /* last record */
  72 } ftrace_data_t;
  73 
  74 struct cyc_cpu;
  75 struct nvlist;
  76 
  77 /*
  78  * Per-CPU data.
  79  *
  80  * Be careful adding new members: if they are not the same in all modules (e.g.
  81  * change size depending on a #define), CTF uniquification can fail to work
  82  * properly.  Furthermore, this is transitive in that it applies recursively to
  83  * all types pointed to by cpu_t.
  84  */
  85 typedef struct cpu {
  86         processorid_t   cpu_id;                 /* CPU number */
  87         processorid_t   cpu_seqid;      /* sequential CPU id (0..ncpus-1) */
  88         volatile cpu_flag_t cpu_flags;          /* flags indicating CPU state */
  89         struct cpu      *cpu_self;              /* pointer to itself */
  90         kthread_t       *cpu_thread;            /* current thread */
  91         kthread_t       *cpu_idle_thread;       /* idle thread for this CPU */
  92         kthread_t       *cpu_pause_thread;      /* pause thread for this CPU */
  93         klwp_id_t       cpu_lwp;                /* current lwp (if any) */
  94         klwp_id_t       cpu_fpowner;            /* currently loaded fpu owner */
  95         struct cpupart  *cpu_part;              /* partition with this CPU */
  96         struct lgrp_ld  *cpu_lpl;               /* pointer to this cpu's load */
  97         int             cpu_cache_offset;       /* see kmem.c for details */
  98 
  99         /*
 100          * Links to other CPUs.  It is safe to walk these lists if
 101          * one of the following is true:
 102          *      - cpu_lock held
 103          *      - preemption disabled via kpreempt_disable
 104          *      - PIL >= DISP_LEVEL
 105          *      - acting thread is an interrupt thread
 106          *      - all other CPUs are paused
 107          */
 108         struct cpu      *cpu_next;              /* next existing CPU */
 109         struct cpu      *cpu_prev;              /* prev existing CPU */
 110         struct cpu      *cpu_next_onln;         /* next online (enabled) CPU */
 111         struct cpu      *cpu_prev_onln;         /* prev online (enabled) CPU */
 112         struct cpu      *cpu_next_part;         /* next CPU in partition */
 113         struct cpu      *cpu_prev_part;         /* prev CPU in partition */
 114         struct cpu      *cpu_next_lgrp;         /* next CPU in latency group */
 115         struct cpu      *cpu_prev_lgrp;         /* prev CPU in latency group */
 116         struct cpu      *cpu_next_lpl;          /* next CPU in lgrp partition */
 117         struct cpu      *cpu_prev_lpl;
 118 
 119         struct cpu_pg   *cpu_pg;                /* cpu's processor groups */
 120 
 121         void            *cpu_reserved[4];       /* reserved for future use */
 122 
 123         /*
 124          * Scheduling variables.
 125          */
 126         disp_t          *cpu_disp;              /* dispatch queue data */
 127         /*
 128          * Note that cpu_disp is set before the CPU is added to the system
 129          * and is never modified.  Hence, no additional locking is needed
 130          * beyond what's necessary to access the cpu_t structure.
 131          */
 132         char            cpu_runrun;     /* scheduling flag - set to preempt */
 133         char            cpu_kprunrun;           /* force kernel preemption */
 134         pri_t           cpu_chosen_level;       /* priority at which cpu */
 135                                                 /* was chosen for scheduling */
 136         kthread_t       *cpu_dispthread; /* thread selected for dispatch */
 137         disp_lock_t     cpu_thread_lock; /* dispatcher lock on current thread */
 138         uint8_t         cpu_disp_flags; /* flags used by dispatcher */
 139         /*
 140          * The following field is updated when ever the cpu_dispthread
 141          * changes. Also in places, where the current thread(cpu_dispthread)
 142          * priority changes. This is used in disp_lowpri_cpu()
 143          */
 144         pri_t           cpu_dispatch_pri; /* priority of cpu_dispthread */
 145         clock_t         cpu_last_swtch; /* last time switched to new thread */
 146 
 147         /*
 148          * Interrupt data.
 149          */
 150         caddr_t         cpu_intr_stack; /* interrupt stack */
 151         kthread_t       *cpu_intr_thread; /* interrupt thread list */
 152         uint_t          cpu_intr_actv;  /* interrupt levels active (bitmask) */
 153         int             cpu_base_spl;   /* priority for highest rupt active */
 154 
 155         /*
 156          * Statistics.
 157          */
 158         cpu_stats_t     cpu_stats;              /* per-CPU statistics */
 159         struct kstat    *cpu_info_kstat;        /* kstat for cpu info */
 160 
 161         uintptr_t       cpu_profile_pc; /* kernel PC in profile interrupt */
 162         uintptr_t       cpu_profile_upc; /* user PC in profile interrupt */
 163         uintptr_t       cpu_profile_pil; /* PIL when profile interrupted */
 164 
 165         ftrace_data_t   cpu_ftrace;             /* per cpu ftrace data */
 166 
 167         clock_t         cpu_deadman_counter;    /* used by deadman() */
 168         uint_t          cpu_deadman_countdown;  /* used by deadman() */
 169 
 170         kmutex_t        cpu_cpc_ctxlock; /* protects context for idle thread */
 171         kcpc_ctx_t      *cpu_cpc_ctx;   /* performance counter context */
 172 
 173         /*
 174          * Configuration information for the processor_info system call.
 175          */
 176         processor_info_t cpu_type_info; /* config info */
 177         time_t          cpu_state_begin; /* when CPU entered current state */
 178         char            cpu_cpr_flags;  /* CPR related info */
 179         struct cyc_cpu  *cpu_cyclic;    /* per cpu cyclic subsystem data */
 180         struct squeue_set_s *cpu_squeue_set;    /* per cpu squeue set */
 181         struct nvlist   *cpu_props;     /* pool-related properties */
 182 
 183         krwlock_t       cpu_ft_lock;            /* DTrace: fasttrap lock */
 184         uintptr_t       cpu_dtrace_caller;      /* DTrace: caller, if any */
 185         hrtime_t        cpu_dtrace_chillmark;   /* DTrace: chill mark time */
 186         hrtime_t        cpu_dtrace_chilled;     /* DTrace: total chill time */
 187         uint64_t        cpu_dtrace_probes;      /* DTrace: total probes fired */
 188         hrtime_t        cpu_dtrace_nsec;        /* DTrace: ns in dtrace_probe */
 189 
 190         volatile uint16_t cpu_mstate;           /* cpu microstate */
 191         volatile uint16_t cpu_mstate_gen;       /* generation counter */
 192         volatile hrtime_t cpu_mstate_start;     /* cpu microstate start time */
 193         volatile hrtime_t cpu_acct[NCMSTATES];  /* cpu microstate data */
 194         hrtime_t        cpu_intracct[NCMSTATES]; /* interrupt mstate data */
 195         hrtime_t        cpu_waitrq;             /* cpu run-queue wait time */
 196         struct loadavg_s cpu_loadavg;           /* loadavg info for this cpu */
 197 
 198         char            *cpu_idstr;     /* for printing and debugging */
 199         char            *cpu_brandstr;  /* for printing */
 200 
 201         /*
 202          * Sum of all device interrupt weights that are currently directed at
 203          * this cpu. Cleared at start of interrupt redistribution.
 204          */
 205         int32_t         cpu_intr_weight;
 206         void            *cpu_vm_data;
 207 
 208         struct cpu_physid *cpu_physid;  /* physical associations */
 209 
 210         uint64_t        cpu_curr_clock;         /* current clock freq in Hz */
 211         char            *cpu_supp_freqs;        /* supported freqs in Hz */
 212 
 213         uintptr_t       cpu_cpcprofile_pc;      /* kernel PC in cpc interrupt */
 214         uintptr_t       cpu_cpcprofile_upc;     /* user PC in cpc interrupt */
 215 
 216         /*
 217          * Interrupt load factor used by dispatcher & softcall
 218          */
 219         hrtime_t        cpu_intrlast;   /* total interrupt time (nsec) */
 220         int             cpu_intrload;   /* interrupt load factor (0-99%) */
 221 
 222         uint_t          cpu_rotor;      /* for cheap pseudo-random numbers */
 223 
 224         struct cu_cpu_info      *cpu_cu_info;   /* capacity & util. info */
 225 
 226         /*
 227          * cpu_generation is updated whenever CPU goes on-line or off-line.
 228          * Updates to cpu_generation are protected by cpu_lock.
 229          *
 230          * See CPU_NEW_GENERATION() macro below.
 231          */
 232         volatile uint_t         cpu_generation; /* tracking on/off-line */
 233 
 234         /*
 235          * New members must be added /before/ this member, as the CTF tools
 236          * rely on this being the last field before cpu_m, so they can
 237          * correctly calculate the offset when synthetically adding the cpu_m
 238          * member in objects that do not have it.  This fixup is required for
 239          * uniquification to work correctly.
 240          */
 241         uintptr_t       cpu_m_pad;
 242 
 243 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
 244         struct machcpu  cpu_m;          /* per architecture info */
 245 #endif
 246 } cpu_t;
 247 
 248 /*
 249  * The cpu_core structure consists of per-CPU state available in any context.
 250  * On some architectures, this may mean that the page(s) containing the
 251  * NCPU-sized array of cpu_core structures must be locked in the TLB -- it
 252  * is up to the platform to assure that this is performed properly.  Note that
 253  * the structure is sized to avoid false sharing.
 254  */
 255 #define CPUC_SIZE               (sizeof (uint16_t) + sizeof (uint8_t) + \
 256                                 sizeof (uintptr_t) + sizeof (kmutex_t))
 257 #define CPUC_PADSIZE            CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE
 258 
 259 typedef struct cpu_core {
 260         uint16_t        cpuc_dtrace_flags;      /* DTrace flags */
 261         uint8_t         cpuc_dcpc_intr_state;   /* DCPC provider intr state */
 262         uint8_t         cpuc_pad[CPUC_PADSIZE]; /* padding */
 263         uintptr_t       cpuc_dtrace_illval;     /* DTrace illegal value */
 264         kmutex_t        cpuc_pid_lock;          /* DTrace pid provider lock */
 265 } cpu_core_t;
 266 
 267 #ifdef _KERNEL
 268 extern cpu_core_t cpu_core[];
 269 #endif /* _KERNEL */
 270 
 271 /*
 272  * CPU_ON_INTR() macro. Returns non-zero if currently on interrupt stack.
 273  * Note that this isn't a test for a high PIL.  For example, cpu_intr_actv
 274  * does not get updated when we go through sys_trap from TL>0 at high PIL.
 275  * getpil() should be used instead to check for PIL levels.
 276  */
 277 #define CPU_ON_INTR(cpup) ((cpup)->cpu_intr_actv >> (LOCK_LEVEL + 1))
 278 
 279 /*
 280  * Check to see if an interrupt thread might be active at a given ipl.
 281  * If so return true.
 282  * We must be conservative--it is ok to give a false yes, but a false no
 283  * will cause disaster.  (But if the situation changes after we check it is
 284  * ok--the caller is trying to ensure that an interrupt routine has been
 285  * exited).
 286  * This is used when trying to remove an interrupt handler from an autovector
 287  * list in avintr.c.
 288  */
 289 #define INTR_ACTIVE(cpup, level)        \
 290         ((level) <= LOCK_LEVEL ?     \
 291         ((cpup)->cpu_intr_actv & (1 << (level))) : (CPU_ON_INTR(cpup)))
 292 
 293 /*
 294  * CPU_PSEUDO_RANDOM() returns a per CPU value that changes each time one
 295  * looks at it. It's meant as a cheap mechanism to be incorporated in routines
 296  * wanting to avoid biasing, but where true randomness isn't needed (just
 297  * something that changes).
 298  */
 299 #define CPU_PSEUDO_RANDOM() (CPU->cpu_rotor++)
 300 
 301 #if defined(_KERNEL) || defined(_KMEMUSER) || defined(_BOOT)
 302 
 303 #define INTR_STACK_SIZE MAX(DEFAULTSTKSZ, PAGESIZE)
 304 
 305 /* MEMBERS PROTECTED BY "atomicity": cpu_flags */
 306 
 307 /*
 308  * Flags in the CPU structure.
 309  *
 310  * These are protected by cpu_lock (except during creation).
 311  *
 312  * Offlined-CPUs have three stages of being offline:
 313  *
 314  * CPU_ENABLE indicates that the CPU is participating in I/O interrupts
 315  * that can be directed at a number of different CPUs.  If CPU_ENABLE
 316  * is off, the CPU will not be given interrupts that can be sent elsewhere,
 317  * but will still get interrupts from devices associated with that CPU only,
 318  * and from other CPUs.
 319  *
 320  * CPU_OFFLINE indicates that the dispatcher should not allow any threads
 321  * other than interrupt threads to run on that CPU.  A CPU will not have
 322  * CPU_OFFLINE set if there are any bound threads (besides interrupts).
 323  *
 324  * CPU_QUIESCED is set if p_offline was able to completely turn idle the
 325  * CPU and it will not have to run interrupt threads.  In this case it'll
 326  * stay in the idle loop until CPU_QUIESCED is turned off.
 327  *
 328  * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully
 329  * suspended (in the suspend path), or have yet to be resumed (in the resume
 330  * case).
 331  *
 332  * On some platforms CPUs can be individually powered off.
 333  * The following flags are set for powered off CPUs: CPU_QUIESCED,
 334  * CPU_OFFLINE, and CPU_POWEROFF.  The following flags are cleared:
 335  * CPU_RUNNING, CPU_READY, CPU_EXISTS, and CPU_ENABLE.
 336  */
 337 #define CPU_RUNNING     0x001           /* CPU running */
 338 #define CPU_READY       0x002           /* CPU ready for cross-calls */
 339 #define CPU_QUIESCED    0x004           /* CPU will stay in idle */
 340 #define CPU_EXISTS      0x008           /* CPU is configured */
 341 #define CPU_ENABLE      0x010           /* CPU enabled for interrupts */
 342 #define CPU_OFFLINE     0x020           /* CPU offline via p_online */
 343 #define CPU_POWEROFF    0x040           /* CPU is powered off */
 344 #define CPU_FROZEN      0x080           /* CPU is frozen via CPR suspend */
 345 #define CPU_SPARE       0x100           /* CPU offline available for use */
 346 #define CPU_FAULTED     0x200           /* CPU offline diagnosed faulty */
 347 
 348 #define FMT_CPU_FLAGS                                                   \
 349         "\20\12fault\11spare\10frozen"                                  \
 350         "\7poweroff\6offline\5enable\4exist\3quiesced\2ready\1run"
 351 
 352 #define CPU_ACTIVE(cpu) (((cpu)->cpu_flags & CPU_OFFLINE) == 0)
 353 
 354 /*
 355  * Flags for cpu_offline(), cpu_faulted(), and cpu_spare().
 356  */
 357 #define CPU_FORCED      0x0001          /* Force CPU offline */
 358 
 359 /*
 360  * DTrace flags.
 361  */
 362 #define CPU_DTRACE_NOFAULT      0x0001  /* Don't fault */
 363 #define CPU_DTRACE_DROP         0x0002  /* Drop this ECB */
 364 #define CPU_DTRACE_BADADDR      0x0004  /* DTrace fault: bad address */
 365 #define CPU_DTRACE_BADALIGN     0x0008  /* DTrace fault: bad alignment */
 366 #define CPU_DTRACE_DIVZERO      0x0010  /* DTrace fault: divide by zero */
 367 #define CPU_DTRACE_ILLOP        0x0020  /* DTrace fault: illegal operation */
 368 #define CPU_DTRACE_NOSCRATCH    0x0040  /* DTrace fault: out of scratch */
 369 #define CPU_DTRACE_KPRIV        0x0080  /* DTrace fault: bad kernel access */
 370 #define CPU_DTRACE_UPRIV        0x0100  /* DTrace fault: bad user access */
 371 #define CPU_DTRACE_TUPOFLOW     0x0200  /* DTrace fault: tuple stack overflow */
 372 #if defined(__sparc)
 373 #define CPU_DTRACE_FAKERESTORE  0x0400  /* pid provider hint to getreg */
 374 #endif
 375 #define CPU_DTRACE_ENTRY        0x0800  /* pid provider hint to ustack() */
 376 #define CPU_DTRACE_BADSTACK     0x1000  /* DTrace fault: bad stack */
 377 
 378 #define CPU_DTRACE_FAULT        (CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \
 379                                 CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \
 380                                 CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \
 381                                 CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \
 382                                 CPU_DTRACE_BADSTACK)
 383 #define CPU_DTRACE_ERROR        (CPU_DTRACE_FAULT | CPU_DTRACE_DROP)
 384 
 385 /*
 386  * Dispatcher flags
 387  * These flags must be changed only by the current CPU.
 388  */
 389 #define CPU_DISP_DONTSTEAL      0x01    /* CPU undergoing context swtch */
 390 #define CPU_DISP_HALTED         0x02    /* CPU halted waiting for interrupt */
 391 
 392 /* Note: inside ifdef: _KERNEL || _KMEMUSER || _BOOT */
 393 #if defined(_MACHDEP)
 394 
 395 /*
 396  * Macros for manipulating sets of CPUs as a bitmap.  Note that this
 397  * bitmap may vary in size depending on the maximum CPU id a specific
 398  * platform supports.  This may be different than the number of CPUs
 399  * the platform supports, since CPU ids can be sparse.  We define two
 400  * sets of macros; one for platforms where the maximum CPU id is less
 401  * than the number of bits in a single word (32 in a 32-bit kernel,
 402  * 64 in a 64-bit kernel), and one for platforms that require bitmaps
 403  * of more than one word.
 404  */
 405 
 406 #define CPUSET_WORDS    BT_BITOUL(NCPU)
 407 #define CPUSET_NOTINSET ((uint_t)-1)
 408 
 409 #if     CPUSET_WORDS > 1
 410 
 411 typedef struct cpuset {
 412         ulong_t cpub[CPUSET_WORDS];
 413 } cpuset_t;
 414 
 415 /*
 416  * Private functions for manipulating cpusets that do not fit in a
 417  * single word.  These should not be used directly; instead the
 418  * CPUSET_* macros should be used so the code will be portable
 419  * across different definitions of NCPU.
 420  */
 421 extern  void    cpuset_all(cpuset_t *);
 422 extern  void    cpuset_all_but(cpuset_t *, uint_t);
 423 extern  int     cpuset_isnull(cpuset_t *);
 424 extern  int     cpuset_cmp(cpuset_t *, cpuset_t *);
 425 extern  void    cpuset_only(cpuset_t *, uint_t);
 426 extern  uint_t  cpuset_find(cpuset_t *);
 427 extern  void    cpuset_bounds(cpuset_t *, uint_t *, uint_t *);
 428 
 429 #define CPUSET_ALL(set)                 cpuset_all(&(set))
 430 #define CPUSET_ALL_BUT(set, cpu)        cpuset_all_but(&(set), cpu)
 431 #define CPUSET_ONLY(set, cpu)           cpuset_only(&(set), cpu)
 432 #define CPU_IN_SET(set, cpu)            BT_TEST((set).cpub, cpu)
 433 #define CPUSET_ADD(set, cpu)            BT_SET((set).cpub, cpu)
 434 #define CPUSET_DEL(set, cpu)            BT_CLEAR((set).cpub, cpu)
 435 #define CPUSET_ISNULL(set)              cpuset_isnull(&(set))
 436 #define CPUSET_ISEQUAL(set1, set2)      cpuset_cmp(&(set1), &(set2))
 437 
 438 /*
 439  * Find one CPU in the cpuset.
 440  * Sets "cpu" to the id of the found CPU, or CPUSET_NOTINSET if no cpu
 441  * could be found. (i.e. empty set)
 442  */
 443 #define CPUSET_FIND(set, cpu)           {               \
 444         cpu = cpuset_find(&(set));                  \
 445 }
 446 
 447 /*
 448  * Determine the smallest and largest CPU id in the set. Returns
 449  * CPUSET_NOTINSET in smallest and largest when set is empty.
 450  */
 451 #define CPUSET_BOUNDS(set, smallest, largest)   {               \
 452         cpuset_bounds(&(set), &(smallest), &(largest));             \
 453 }
 454 
 455 /*
 456  * Atomic cpuset operations
 457  * These are safe to use for concurrent cpuset manipulations.
 458  * "xdel" and "xadd" are exclusive operations, that set "result" to "0"
 459  * if the add or del was successful, or "-1" if not successful.
 460  * (e.g. attempting to add a cpu to a cpuset that's already there, or
 461  * deleting a cpu that's not in the cpuset)
 462  */
 463 
 464 #define CPUSET_ATOMIC_DEL(set, cpu)     BT_ATOMIC_CLEAR((set).cpub, (cpu))
 465 #define CPUSET_ATOMIC_ADD(set, cpu)     BT_ATOMIC_SET((set).cpub, (cpu))
 466 
 467 #define CPUSET_ATOMIC_XADD(set, cpu, result) \
 468         BT_ATOMIC_SET_EXCL((set).cpub, cpu, result)
 469 
 470 #define CPUSET_ATOMIC_XDEL(set, cpu, result) \
 471         BT_ATOMIC_CLEAR_EXCL((set).cpub, cpu, result)
 472 
 473 
 474 #define CPUSET_OR(set1, set2)           {               \
 475         int _i;                                         \
 476         for (_i = 0; _i < CPUSET_WORDS; _i++)                \
 477                 (set1).cpub[_i] |= (set2).cpub[_i];     \
 478 }
 479 
 480 #define CPUSET_XOR(set1, set2)          {               \
 481         int _i;                                         \
 482         for (_i = 0; _i < CPUSET_WORDS; _i++)                \
 483                 (set1).cpub[_i] ^= (set2).cpub[_i];     \
 484 }
 485 
 486 #define CPUSET_AND(set1, set2)          {               \
 487         int _i;                                         \
 488         for (_i = 0; _i < CPUSET_WORDS; _i++)                \
 489                 (set1).cpub[_i] &= (set2).cpub[_i]; \
 490 }
 491 
 492 #define CPUSET_ZERO(set)                {               \
 493         int _i;                                         \
 494         for (_i = 0; _i < CPUSET_WORDS; _i++)                \
 495                 (set).cpub[_i] = 0;                     \
 496 }
 497 
 498 #elif   CPUSET_WORDS == 1
 499 
 500 typedef ulong_t cpuset_t;       /* a set of CPUs */
 501 
 502 #define CPUSET(cpu)                     (1UL << (cpu))
 503 
 504 #define CPUSET_ALL(set)                 ((void)((set) = ~0UL))
 505 #define CPUSET_ALL_BUT(set, cpu)        ((void)((set) = ~CPUSET(cpu)))
 506 #define CPUSET_ONLY(set, cpu)           ((void)((set) = CPUSET(cpu)))
 507 #define CPU_IN_SET(set, cpu)            ((set) & CPUSET(cpu))
 508 #define CPUSET_ADD(set, cpu)            ((void)((set) |= CPUSET(cpu)))
 509 #define CPUSET_DEL(set, cpu)            ((void)((set) &= ~CPUSET(cpu)))
 510 #define CPUSET_ISNULL(set)              ((set) == 0)
 511 #define CPUSET_ISEQUAL(set1, set2)      ((set1) == (set2))
 512 #define CPUSET_OR(set1, set2)           ((void)((set1) |= (set2)))
 513 #define CPUSET_XOR(set1, set2)          ((void)((set1) ^= (set2)))
 514 #define CPUSET_AND(set1, set2)          ((void)((set1) &= (set2)))
 515 #define CPUSET_ZERO(set)                ((void)((set) = 0))
 516 
 517 #define CPUSET_FIND(set, cpu)           {               \
 518         cpu = (uint_t)(lowbit(set) - 1);                                \
 519 }
 520 
 521 #define CPUSET_BOUNDS(set, smallest, largest)   {       \
 522         smallest = (uint_t)(lowbit(set) - 1);           \
 523         largest = (uint_t)(highbit(set) - 1);           \
 524 }
 525 
 526 #define CPUSET_ATOMIC_DEL(set, cpu)     atomic_and_ulong(&(set), ~CPUSET(cpu))
 527 #define CPUSET_ATOMIC_ADD(set, cpu)     atomic_or_ulong(&(set), CPUSET(cpu))
 528 
 529 #define CPUSET_ATOMIC_XADD(set, cpu, result) \
 530         { result = atomic_set_long_excl(&(set), (cpu)); }
 531 
 532 #define CPUSET_ATOMIC_XDEL(set, cpu, result) \
 533         { result = atomic_clear_long_excl(&(set), (cpu)); }
 534 
 535 #else   /* CPUSET_WORDS <= 0 */
 536 
 537 #error NCPU is undefined or invalid
 538 
 539 #endif  /* CPUSET_WORDS */
 540 
 541 extern cpuset_t cpu_seqid_inuse;
 542 
 543 #endif  /* _MACHDEP */
 544 #endif /* _KERNEL || _KMEMUSER || _BOOT */
 545 
 546 #define CPU_CPR_OFFLINE         0x0
 547 #define CPU_CPR_ONLINE          0x1
 548 #define CPU_CPR_IS_OFFLINE(cpu) (((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) == 0)
 549 #define CPU_CPR_IS_ONLINE(cpu)  ((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE)
 550 #define CPU_SET_CPR_FLAGS(cpu, flag)    ((cpu)->cpu_cpr_flags |= flag)
 551 
 552 #if defined(_KERNEL) || defined(_KMEMUSER)
 553 
 554 extern struct cpu       *cpu[];         /* indexed by CPU number */
 555 extern struct cpu       **cpu_seq;      /* indexed by sequential CPU id */
 556 extern cpu_t            *cpu_list;      /* list of CPUs */
 557 extern cpu_t            *cpu_active;    /* list of active CPUs */
 558 extern int              ncpus;          /* number of CPUs present */
 559 extern int              ncpus_online;   /* number of CPUs not quiesced */
 560 extern int              max_ncpus;      /* max present before ncpus is known */
 561 extern int              boot_max_ncpus; /* like max_ncpus but for real */
 562 extern int              boot_ncpus;     /* # cpus present @ boot */
 563 extern processorid_t    max_cpuid;      /* maximum CPU number */
 564 extern struct cpu       *cpu_inmotion;  /* offline or partition move target */
 565 extern cpu_t            *clock_cpu_list;
 566 extern processorid_t    max_cpu_seqid_ever;     /* maximum seqid ever given */
 567 
 568 #if defined(__i386) || defined(__amd64)
 569 extern struct cpu *curcpup(void);
 570 #define CPU             (curcpup())     /* Pointer to current CPU */
 571 #else
 572 #define CPU             (curthread->t_cpu)   /* Pointer to current CPU */
 573 #endif
 574 
 575 /*
 576  * CPU_CURRENT indicates to thread_affinity_set() to use whatever curthread's
 577  * current CPU is; holding cpu_lock is not required.
 578  */
 579 #define CPU_CURRENT     -3
 580 
 581 /*
 582  * CPU_BEST can be used by thread_affinity_set() callers to set affinity to a
 583  * good CPU (in particular, an ht_acquire()-friendly choice); holding cpu_lock
 584  * is not required.
 585  */
 586 #define CPU_BEST        -4
 587 
 588 /*
 589  * Per-CPU statistics
 590  *
 591  * cpu_stats_t contains numerous system and VM-related statistics, in the form
 592  * of gauges or monotonically-increasing event occurrence counts.
 593  */
 594 
 595 #define CPU_STATS_ENTER_K()     kpreempt_disable()
 596 #define CPU_STATS_EXIT_K()      kpreempt_enable()
 597 
 598 #define CPU_STATS_ADD_K(class, stat, amount) \
 599         {       kpreempt_disable(); /* keep from switching CPUs */\
 600                 CPU_STATS_ADDQ(CPU, class, stat, amount); \
 601                 kpreempt_enable(); \
 602         }
 603 
 604 #define CPU_STATS_ADDQ(cp, class, stat, amount) {                       \
 605         extern void __dtrace_probe___cpu_##class##info_##stat(uint_t,   \
 606             uint64_t *, cpu_t *);                                       \
 607         uint64_t *stataddr = &((cp)->cpu_stats.class.stat);              \
 608         __dtrace_probe___cpu_##class##info_##stat((amount),             \
 609             stataddr, cp);                                              \
 610         *(stataddr) += (amount);                                        \
 611 }
 612 
 613 #define CPU_STATS(cp, stat)                                       \
 614         ((cp)->cpu_stats.stat)
 615 
 616 /*
 617  * Increment CPU generation value.
 618  * This macro should be called whenever CPU goes on-line or off-line.
 619  * Updates to cpu_generation should be protected by cpu_lock.
 620  */
 621 #define CPU_NEW_GENERATION(cp)  ((cp)->cpu_generation++)
 622 
 623 #endif /* _KERNEL || _KMEMUSER */
 624 
 625 /*
 626  * CPU support routines (not for genassym.c)
 627  */
 628 #if     (defined(_KERNEL) || defined(_FAKE_KERNEL)) && defined(__STDC__)
 629 
 630 struct zone;
 631 
 632 void    cpu_list_init(cpu_t *);
 633 void    cpu_add_unit(cpu_t *);
 634 void    cpu_del_unit(int cpuid);
 635 void    cpu_add_active(cpu_t *);
 636 void    cpu_kstat_init(cpu_t *);
 637 void    cpu_visibility_add(cpu_t *, struct zone *);
 638 void    cpu_visibility_remove(cpu_t *, struct zone *);
 639 void    cpu_visibility_configure(cpu_t *, struct zone *);
 640 void    cpu_visibility_unconfigure(cpu_t *, struct zone *);
 641 void    cpu_visibility_online(cpu_t *, struct zone *);
 642 void    cpu_visibility_offline(cpu_t *, struct zone *);
 643 void    cpu_create_intrstat(cpu_t *);
 644 void    cpu_delete_intrstat(cpu_t *);
 645 int     cpu_kstat_intrstat_update(kstat_t *, int);
 646 void    cpu_intr_swtch_enter(kthread_t *);
 647 void    cpu_intr_swtch_exit(kthread_t *);
 648 
 649 void    mbox_lock_init(void);    /* initialize cross-call locks */
 650 void    mbox_init(int cpun);     /* initialize cross-calls */
 651 void    poke_cpu(int cpun);      /* interrupt another CPU (to preempt) */
 652 
 653 /*
 654  * values for safe_list.  Pause state that CPUs are in.
 655  */
 656 #define PAUSE_IDLE      0               /* normal state */
 657 #define PAUSE_READY     1               /* paused thread ready to spl */
 658 #define PAUSE_WAIT      2               /* paused thread is spl-ed high */
 659 #define PAUSE_DIE       3               /* tell pause thread to leave */
 660 #define PAUSE_DEAD      4               /* pause thread has left */
 661 
 662 void    mach_cpu_pause(volatile char *);
 663 
 664 void    pause_cpus(cpu_t *off_cp, void *(*func)(void *));
 665 void    start_cpus(void);
 666 int     cpus_paused(void);
 667 
 668 void    cpu_pause_init(void);
 669 cpu_t   *cpu_get(processorid_t cpun);   /* get the CPU struct associated */
 670 
 671 int     cpu_online(cpu_t *cp);                  /* take cpu online */
 672 int     cpu_offline(cpu_t *cp, int flags);      /* take cpu offline */
 673 int     cpu_spare(cpu_t *cp, int flags);        /* take cpu to spare */
 674 int     cpu_faulted(cpu_t *cp, int flags);      /* take cpu to faulted */
 675 int     cpu_poweron(cpu_t *cp);         /* take powered-off cpu to offline */
 676 int     cpu_poweroff(cpu_t *cp);        /* take offline cpu to powered-off */
 677 
 678 cpu_t   *cpu_intr_next(cpu_t *cp);      /* get next online CPU taking intrs */
 679 int     cpu_intr_count(cpu_t *cp);      /* count # of CPUs handling intrs */
 680 int     cpu_intr_on(cpu_t *cp);         /* CPU taking I/O interrupts? */
 681 void    cpu_intr_enable(cpu_t *cp);     /* enable I/O interrupts */
 682 int     cpu_intr_disable(cpu_t *cp);    /* disable I/O interrupts */
 683 void    cpu_intr_alloc(cpu_t *cp, int n); /* allocate interrupt threads */
 684 
 685 /*
 686  * Routines for checking CPU states.
 687  */
 688 int     cpu_is_online(cpu_t *);         /* check if CPU is online */
 689 int     cpu_is_nointr(cpu_t *);         /* check if CPU can service intrs */
 690 int     cpu_is_active(cpu_t *);         /* check if CPU can run threads */
 691 int     cpu_is_offline(cpu_t *);        /* check if CPU is offline */
 692 int     cpu_is_poweredoff(cpu_t *);     /* check if CPU is powered off */
 693 
 694 int     cpu_flagged_online(cpu_flag_t); /* flags show CPU is online */
 695 int     cpu_flagged_nointr(cpu_flag_t); /* flags show CPU not handling intrs */
 696 int     cpu_flagged_active(cpu_flag_t); /* flags show CPU scheduling threads */
 697 int     cpu_flagged_offline(cpu_flag_t); /* flags show CPU is offline */
 698 int     cpu_flagged_poweredoff(cpu_flag_t); /* flags show CPU is powered off */
 699 
 700 /*
 701  * The processor_info(2) state of a CPU is a simplified representation suitable
 702  * for use by an application program.  Kernel subsystems should utilize the
 703  * internal per-CPU state as given by the cpu_flags member of the cpu structure,
 704  * as this information may include platform- or architecture-specific state
 705  * critical to a subsystem's disposition of a particular CPU.
 706  */
 707 void    cpu_set_state(cpu_t *);         /* record/timestamp current state */
 708 int     cpu_get_state(cpu_t *);         /* get current cpu state */
 709 const char *cpu_get_state_str(cpu_t *); /* get current cpu state as string */
 710 
 711 
 712 void    cpu_set_curr_clock(uint64_t);   /* indicate the current CPU's freq */
 713 void    cpu_set_supp_freqs(cpu_t *, const char *); /* set the CPU supported */
 714                                                 /* frequencies */
 715 
 716 int     cpu_configure(int);
 717 int     cpu_unconfigure(int);
 718 void    cpu_destroy_bound_threads(cpu_t *cp);
 719 
 720 extern int cpu_bind_thread(kthread_t *tp, processorid_t bind,
 721     processorid_t *obind, int *error);
 722 extern int cpu_unbind(processorid_t cpu_id, boolean_t force);
 723 extern void thread_affinity_set(kthread_t *t, int cpu_id);
 724 extern void thread_affinity_clear(kthread_t *t);
 725 extern void affinity_set(int cpu_id);
 726 extern void affinity_clear(void);
 727 extern void init_cpu_mstate(struct cpu *, int);
 728 extern void term_cpu_mstate(struct cpu *);
 729 extern void new_cpu_mstate(int, hrtime_t);
 730 extern void get_cpu_mstate(struct cpu *, hrtime_t *);
 731 extern void thread_nomigrate(void);
 732 extern void thread_allowmigrate(void);
 733 extern void weakbinding_stop(void);
 734 extern void weakbinding_start(void);
 735 
 736 /*
 737  * The following routines affect the CPUs participation in interrupt processing,
 738  * if that is applicable on the architecture.  This only affects interrupts
 739  * which aren't directed at the processor (not cross calls).
 740  *
 741  * cpu_disable_intr returns non-zero if interrupts were previously enabled.
 742  */
 743 int     cpu_disable_intr(struct cpu *cp); /* stop issuing interrupts to cpu */
 744 void    cpu_enable_intr(struct cpu *cp); /* start issuing interrupts to cpu */
 745 
 746 /*
 747  * The mutex cpu_lock protects cpu_flags for all CPUs, as well as the ncpus
 748  * and ncpus_online counts.
 749  */
 750 extern kmutex_t cpu_lock;       /* lock protecting CPU data */
 751 
 752 /*
 753  * CPU state change events
 754  *
 755  * Various subsystems need to know when CPUs change their state. They get this
 756  * information by registering  CPU state change callbacks using
 757  * register_cpu_setup_func(). Whenever any CPU changes its state, the callback
 758  * function is called. The callback function is passed three arguments:
 759  *
 760  *   Event, described by cpu_setup_t
 761  *   CPU ID
 762  *   Transparent pointer passed when registering the callback
 763  *
 764  * The callback function is called with cpu_lock held. The return value from the
 765  * callback function is usually ignored, except for CPU_CONFIG and CPU_UNCONFIG
 766  * events. For these two events, non-zero return value indicates a failure and
 767  * prevents successful completion of the operation.
 768  *
 769  * New events may be added in the future. Callback functions should ignore any
 770  * events that they do not understand.
 771  *
 772  * The following events provide notification callbacks:
 773  *
 774  *  CPU_INIT    A new CPU is started and added to the list of active CPUs
 775  *                This event is only used during boot
 776  *
 777  *  CPU_CONFIG  A newly inserted CPU is prepared for starting running code
 778  *                This event is called by DR code
 779  *
 780  *  CPU_UNCONFIG CPU has been powered off and needs cleanup
 781  *                This event is called by DR code
 782  *
 783  *  CPU_ON      CPU is enabled but does not run anything yet
 784  *
 785  *  CPU_INTR_ON CPU is enabled and has interrupts enabled
 786  *
 787  *  CPU_OFF     CPU is going offline but can still run threads
 788  *
 789  *  CPU_CPUPART_OUT     CPU is going to move out of its partition
 790  *
 791  *  CPU_CPUPART_IN      CPU is going to move to a new partition
 792  *
 793  *  CPU_SETUP   CPU is set up during boot and can run threads
 794  */
 795 typedef enum {
 796         CPU_INIT,
 797         CPU_CONFIG,
 798         CPU_UNCONFIG,
 799         CPU_ON,
 800         CPU_OFF,
 801         CPU_CPUPART_IN,
 802         CPU_CPUPART_OUT,
 803         CPU_SETUP,
 804         CPU_INTR_ON
 805 } cpu_setup_t;
 806 
 807 typedef int cpu_setup_func_t(cpu_setup_t, int, void *);
 808 
 809 /*
 810  * Routines used to register interest in cpu's being added to or removed
 811  * from the system.
 812  */
 813 extern void register_cpu_setup_func(cpu_setup_func_t *, void *);
 814 extern void unregister_cpu_setup_func(cpu_setup_func_t *, void *);
 815 extern void cpu_state_change_notify(int, cpu_setup_t);
 816 
 817 /*
 818  * Call specified function on the given CPU
 819  */
 820 typedef void (*cpu_call_func_t)(uintptr_t, uintptr_t);
 821 extern void cpu_call(cpu_t *, cpu_call_func_t, uintptr_t, uintptr_t);
 822 
 823 
 824 /*
 825  * Create various strings that describe the given CPU for the
 826  * processor_info system call and configuration-related kstats.
 827  */
 828 #define CPU_IDSTRLEN    100
 829 
 830 extern void init_cpu_info(struct cpu *);
 831 extern void populate_idstr(struct cpu *);
 832 extern void cpu_vm_data_init(struct cpu *);
 833 extern void cpu_vm_data_destroy(struct cpu *);
 834 
 835 #endif  /* _KERNEL || _FAKE_KERNEL */
 836 
 837 #ifdef  __cplusplus
 838 }
 839 #endif
 840 
 841 #endif /* _SYS_CPUVAR_H */