cpuvar.h revision 277546
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 2014 Igor Kozhukhov <ikozhukhov@gmail.com>. 25 */ 26 27#ifndef _SYS_CPUVAR_H 28#define _SYS_CPUVAR_H 29 30#include <sys/thread.h> 31#include <sys/sysinfo.h> /* has cpu_stat_t definition */ 32#include <sys/disp.h> 33#include <sys/processor.h> 34 35#include <sys/loadavg.h> 36#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) 37#include <sys/machcpuvar.h> 38#endif 39 40#include <sys/types.h> 41#include <sys/file.h> 42#include <sys/bitmap.h> 43#include <sys/rwlock.h> 44#include <sys/msacct.h> 45#if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) && \ 46 (defined(__i386) || defined(__amd64)) 47#include <asm/cpuvar.h> 48#endif 49 50#ifdef __cplusplus 51extern "C" { 52#endif 53 54struct squeue_set_s; 55 56#define CPU_CACHE_COHERENCE_SIZE 64 57 58/* 59 * For fast event tracing. 60 */ 61struct ftrace_record; 62typedef struct ftrace_data { 63 int ftd_state; /* ftrace flags */ 64 kmutex_t ftd_unused; /* ftrace buffer lock, unused */ 65 struct ftrace_record *ftd_cur; /* current record */ 66 struct ftrace_record *ftd_first; /* first record */ 67 struct ftrace_record *ftd_last; /* last record */ 68} ftrace_data_t; 69 70struct cyc_cpu; 71struct nvlist; 72 73/* 74 * Per-CPU data. 75 * 76 * Be careful adding new members: if they are not the same in all modules (e.g. 77 * change size depending on a #define), CTF uniquification can fail to work 78 * properly. Furthermore, this is transitive in that it applies recursively to 79 * all types pointed to by cpu_t. 80 */ 81typedef struct cpu { 82 processorid_t cpu_id; /* CPU number */ 83 processorid_t cpu_seqid; /* sequential CPU id (0..ncpus-1) */ 84 volatile cpu_flag_t cpu_flags; /* flags indicating CPU state */ 85 struct cpu *cpu_self; /* pointer to itself */ 86 kthread_t *cpu_thread; /* current thread */ 87 kthread_t *cpu_idle_thread; /* idle thread for this CPU */ 88 kthread_t *cpu_pause_thread; /* pause thread for this CPU */ 89 klwp_id_t cpu_lwp; /* current lwp (if any) */ 90 klwp_id_t cpu_fpowner; /* currently loaded fpu owner */ 91 struct cpupart *cpu_part; /* partition with this CPU */ 92 struct lgrp_ld *cpu_lpl; /* pointer to this cpu's load */ 93 int cpu_cache_offset; /* see kmem.c for details */ 94 95 /* 96 * Links to other CPUs. It is safe to walk these lists if 97 * one of the following is true: 98 * - cpu_lock held 99 * - preemption disabled via kpreempt_disable 100 * - PIL >= DISP_LEVEL 101 * - acting thread is an interrupt thread 102 * - all other CPUs are paused 103 */ 104 struct cpu *cpu_next; /* next existing CPU */ 105 struct cpu *cpu_prev; /* prev existing CPU */ 106 struct cpu *cpu_next_onln; /* next online (enabled) CPU */ 107 struct cpu *cpu_prev_onln; /* prev online (enabled) CPU */ 108 struct cpu *cpu_next_part; /* next CPU in partition */ 109 struct cpu *cpu_prev_part; /* prev CPU in partition */ 110 struct cpu *cpu_next_lgrp; /* next CPU in latency group */ 111 struct cpu *cpu_prev_lgrp; /* prev CPU in latency group */ 112 struct cpu *cpu_next_lpl; /* next CPU in lgrp partition */ 113 struct cpu *cpu_prev_lpl; 114 115 struct cpu_pg *cpu_pg; /* cpu's processor groups */ 116 117 void *cpu_reserved[4]; /* reserved for future use */ 118 119 /* 120 * Scheduling variables. 121 */ 122 disp_t *cpu_disp; /* dispatch queue data */ 123 /* 124 * Note that cpu_disp is set before the CPU is added to the system 125 * and is never modified. Hence, no additional locking is needed 126 * beyond what's necessary to access the cpu_t structure. 127 */ 128 char cpu_runrun; /* scheduling flag - set to preempt */ 129 char cpu_kprunrun; /* force kernel preemption */ 130 pri_t cpu_chosen_level; /* priority at which cpu */ 131 /* was chosen for scheduling */ 132 kthread_t *cpu_dispthread; /* thread selected for dispatch */ 133 disp_lock_t cpu_thread_lock; /* dispatcher lock on current thread */ 134 uint8_t cpu_disp_flags; /* flags used by dispatcher */ 135 /* 136 * The following field is updated when ever the cpu_dispthread 137 * changes. Also in places, where the current thread(cpu_dispthread) 138 * priority changes. This is used in disp_lowpri_cpu() 139 */ 140 pri_t cpu_dispatch_pri; /* priority of cpu_dispthread */ 141 clock_t cpu_last_swtch; /* last time switched to new thread */ 142 143 /* 144 * Interrupt data. 145 */ 146 caddr_t cpu_intr_stack; /* interrupt stack */ 147 kthread_t *cpu_intr_thread; /* interrupt thread list */ 148 uint_t cpu_intr_actv; /* interrupt levels active (bitmask) */ 149 int cpu_base_spl; /* priority for highest rupt active */ 150 151 /* 152 * Statistics. 153 */ 154 cpu_stats_t cpu_stats; /* per-CPU statistics */ 155 struct kstat *cpu_info_kstat; /* kstat for cpu info */ 156 157 uintptr_t cpu_profile_pc; /* kernel PC in profile interrupt */ 158 uintptr_t cpu_profile_upc; /* user PC in profile interrupt */ 159 uintptr_t cpu_profile_pil; /* PIL when profile interrupted */ 160 161 ftrace_data_t cpu_ftrace; /* per cpu ftrace data */ 162 163 clock_t cpu_deadman_counter; /* used by deadman() */ 164 uint_t cpu_deadman_countdown; /* used by deadman() */ 165 166 kmutex_t cpu_cpc_ctxlock; /* protects context for idle thread */ 167 kcpc_ctx_t *cpu_cpc_ctx; /* performance counter context */ 168 169 /* 170 * Configuration information for the processor_info system call. 171 */ 172 processor_info_t cpu_type_info; /* config info */ 173 time_t cpu_state_begin; /* when CPU entered current state */ 174 char cpu_cpr_flags; /* CPR related info */ 175 struct cyc_cpu *cpu_cyclic; /* per cpu cyclic subsystem data */ 176 struct squeue_set_s *cpu_squeue_set; /* per cpu squeue set */ 177 struct nvlist *cpu_props; /* pool-related properties */ 178 179 krwlock_t cpu_ft_lock; /* DTrace: fasttrap lock */ 180 uintptr_t cpu_dtrace_caller; /* DTrace: caller, if any */ 181 hrtime_t cpu_dtrace_chillmark; /* DTrace: chill mark time */ 182 hrtime_t cpu_dtrace_chilled; /* DTrace: total chill time */ 183 volatile uint16_t cpu_mstate; /* cpu microstate */ 184 volatile uint16_t cpu_mstate_gen; /* generation counter */ 185 volatile hrtime_t cpu_mstate_start; /* cpu microstate start time */ 186 volatile hrtime_t cpu_acct[NCMSTATES]; /* cpu microstate data */ 187 hrtime_t cpu_intracct[NCMSTATES]; /* interrupt mstate data */ 188 hrtime_t cpu_waitrq; /* cpu run-queue wait time */ 189 struct loadavg_s cpu_loadavg; /* loadavg info for this cpu */ 190 191 char *cpu_idstr; /* for printing and debugging */ 192 char *cpu_brandstr; /* for printing */ 193 194 /* 195 * Sum of all device interrupt weights that are currently directed at 196 * this cpu. Cleared at start of interrupt redistribution. 197 */ 198 int32_t cpu_intr_weight; 199 void *cpu_vm_data; 200 201 struct cpu_physid *cpu_physid; /* physical associations */ 202 203 uint64_t cpu_curr_clock; /* current clock freq in Hz */ 204 char *cpu_supp_freqs; /* supported freqs in Hz */ 205 206 uintptr_t cpu_cpcprofile_pc; /* kernel PC in cpc interrupt */ 207 uintptr_t cpu_cpcprofile_upc; /* user PC in cpc interrupt */ 208 209 /* 210 * Interrupt load factor used by dispatcher & softcall 211 */ 212 hrtime_t cpu_intrlast; /* total interrupt time (nsec) */ 213 int cpu_intrload; /* interrupt load factor (0-99%) */ 214 215 uint_t cpu_rotor; /* for cheap pseudo-random numbers */ 216 217 struct cu_cpu_info *cpu_cu_info; /* capacity & util. info */ 218 219 /* 220 * cpu_generation is updated whenever CPU goes on-line or off-line. 221 * Updates to cpu_generation are protected by cpu_lock. 222 * 223 * See CPU_NEW_GENERATION() macro below. 224 */ 225 volatile uint_t cpu_generation; /* tracking on/off-line */ 226 227 /* 228 * New members must be added /before/ this member, as the CTF tools 229 * rely on this being the last field before cpu_m, so they can 230 * correctly calculate the offset when synthetically adding the cpu_m 231 * member in objects that do not have it. This fixup is required for 232 * uniquification to work correctly. 233 */ 234 uintptr_t cpu_m_pad; 235 236#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) 237 struct machcpu cpu_m; /* per architecture info */ 238#endif 239} cpu_t; 240 241/* 242 * The cpu_core structure consists of per-CPU state available in any context. 243 * On some architectures, this may mean that the page(s) containing the 244 * NCPU-sized array of cpu_core structures must be locked in the TLB -- it 245 * is up to the platform to assure that this is performed properly. Note that 246 * the structure is sized to avoid false sharing. 247 */ 248#define CPUC_SIZE (sizeof (uint16_t) + sizeof (uint8_t) + \ 249 sizeof (uintptr_t) + sizeof (kmutex_t)) 250#define CPUC_PADSIZE CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE 251 252typedef struct cpu_core { 253 uint16_t cpuc_dtrace_flags; /* DTrace flags */ 254 uint8_t cpuc_dcpc_intr_state; /* DCPC provider intr state */ 255 uint8_t cpuc_pad[CPUC_PADSIZE]; /* padding */ 256 uintptr_t cpuc_dtrace_illval; /* DTrace illegal value */ 257 kmutex_t cpuc_pid_lock; /* DTrace pid provider lock */ 258} cpu_core_t; 259 260#ifdef _KERNEL 261extern cpu_core_t cpu_core[]; 262#endif /* _KERNEL */ 263 264/* 265 * CPU_ON_INTR() macro. Returns non-zero if currently on interrupt stack. 266 * Note that this isn't a test for a high PIL. For example, cpu_intr_actv 267 * does not get updated when we go through sys_trap from TL>0 at high PIL. 268 * getpil() should be used instead to check for PIL levels. 269 */ 270#define CPU_ON_INTR(cpup) ((cpup)->cpu_intr_actv >> (LOCK_LEVEL + 1)) 271 272/* 273 * Check to see if an interrupt thread might be active at a given ipl. 274 * If so return true. 275 * We must be conservative--it is ok to give a false yes, but a false no 276 * will cause disaster. (But if the situation changes after we check it is 277 * ok--the caller is trying to ensure that an interrupt routine has been 278 * exited). 279 * This is used when trying to remove an interrupt handler from an autovector 280 * list in avintr.c. 281 */ 282#define INTR_ACTIVE(cpup, level) \ 283 ((level) <= LOCK_LEVEL ? \ 284 ((cpup)->cpu_intr_actv & (1 << (level))) : (CPU_ON_INTR(cpup))) 285 286/* 287 * CPU_PSEUDO_RANDOM() returns a per CPU value that changes each time one 288 * looks at it. It's meant as a cheap mechanism to be incorporated in routines 289 * wanting to avoid biasing, but where true randomness isn't needed (just 290 * something that changes). 291 */ 292#define CPU_PSEUDO_RANDOM() (CPU->cpu_rotor++) 293 294#if defined(_KERNEL) || defined(_KMEMUSER) 295 296#define INTR_STACK_SIZE MAX(DEFAULTSTKSZ, PAGESIZE) 297 298/* MEMBERS PROTECTED BY "atomicity": cpu_flags */ 299 300/* 301 * Flags in the CPU structure. 302 * 303 * These are protected by cpu_lock (except during creation). 304 * 305 * Offlined-CPUs have three stages of being offline: 306 * 307 * CPU_ENABLE indicates that the CPU is participating in I/O interrupts 308 * that can be directed at a number of different CPUs. If CPU_ENABLE 309 * is off, the CPU will not be given interrupts that can be sent elsewhere, 310 * but will still get interrupts from devices associated with that CPU only, 311 * and from other CPUs. 312 * 313 * CPU_OFFLINE indicates that the dispatcher should not allow any threads 314 * other than interrupt threads to run on that CPU. A CPU will not have 315 * CPU_OFFLINE set if there are any bound threads (besides interrupts). 316 * 317 * CPU_QUIESCED is set if p_offline was able to completely turn idle the 318 * CPU and it will not have to run interrupt threads. In this case it'll 319 * stay in the idle loop until CPU_QUIESCED is turned off. 320 * 321 * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully 322 * suspended (in the suspend path), or have yet to be resumed (in the resume 323 * case). 324 * 325 * On some platforms CPUs can be individually powered off. 326 * The following flags are set for powered off CPUs: CPU_QUIESCED, 327 * CPU_OFFLINE, and CPU_POWEROFF. The following flags are cleared: 328 * CPU_RUNNING, CPU_READY, CPU_EXISTS, and CPU_ENABLE. 329 */ 330#define CPU_RUNNING 0x001 /* CPU running */ 331#define CPU_READY 0x002 /* CPU ready for cross-calls */ 332#define CPU_QUIESCED 0x004 /* CPU will stay in idle */ 333#define CPU_EXISTS 0x008 /* CPU is configured */ 334#define CPU_ENABLE 0x010 /* CPU enabled for interrupts */ 335#define CPU_OFFLINE 0x020 /* CPU offline via p_online */ 336#define CPU_POWEROFF 0x040 /* CPU is powered off */ 337#define CPU_FROZEN 0x080 /* CPU is frozen via CPR suspend */ 338#define CPU_SPARE 0x100 /* CPU offline available for use */ 339#define CPU_FAULTED 0x200 /* CPU offline diagnosed faulty */ 340 341#define FMT_CPU_FLAGS \ 342 "\20\12fault\11spare\10frozen" \ 343 "\7poweroff\6offline\5enable\4exist\3quiesced\2ready\1run" 344 345#define CPU_ACTIVE(cpu) (((cpu)->cpu_flags & CPU_OFFLINE) == 0) 346 347/* 348 * Flags for cpu_offline(), cpu_faulted(), and cpu_spare(). 349 */ 350#define CPU_FORCED 0x0001 /* Force CPU offline */ 351 352/* 353 * DTrace flags. 354 */ 355#define CPU_DTRACE_NOFAULT 0x0001 /* Don't fault */ 356#define CPU_DTRACE_DROP 0x0002 /* Drop this ECB */ 357#define CPU_DTRACE_BADADDR 0x0004 /* DTrace fault: bad address */ 358#define CPU_DTRACE_BADALIGN 0x0008 /* DTrace fault: bad alignment */ 359#define CPU_DTRACE_DIVZERO 0x0010 /* DTrace fault: divide by zero */ 360#define CPU_DTRACE_ILLOP 0x0020 /* DTrace fault: illegal operation */ 361#define CPU_DTRACE_NOSCRATCH 0x0040 /* DTrace fault: out of scratch */ 362#define CPU_DTRACE_KPRIV 0x0080 /* DTrace fault: bad kernel access */ 363#define CPU_DTRACE_UPRIV 0x0100 /* DTrace fault: bad user access */ 364#define CPU_DTRACE_TUPOFLOW 0x0200 /* DTrace fault: tuple stack overflow */ 365#if defined(__sparc) 366#define CPU_DTRACE_FAKERESTORE 0x0400 /* pid provider hint to getreg */ 367#endif 368#define CPU_DTRACE_ENTRY 0x0800 /* pid provider hint to ustack() */ 369#define CPU_DTRACE_BADSTACK 0x1000 /* DTrace fault: bad stack */ 370 371#define CPU_DTRACE_FAULT (CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \ 372 CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \ 373 CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \ 374 CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \ 375 CPU_DTRACE_BADSTACK) 376#define CPU_DTRACE_ERROR (CPU_DTRACE_FAULT | CPU_DTRACE_DROP) 377 378/* 379 * Dispatcher flags 380 * These flags must be changed only by the current CPU. 381 */ 382#define CPU_DISP_DONTSTEAL 0x01 /* CPU undergoing context swtch */ 383#define CPU_DISP_HALTED 0x02 /* CPU halted waiting for interrupt */ 384 385#endif /* _KERNEL || _KMEMUSER */ 386 387#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) 388 389/* 390 * Macros for manipulating sets of CPUs as a bitmap. Note that this 391 * bitmap may vary in size depending on the maximum CPU id a specific 392 * platform supports. This may be different than the number of CPUs 393 * the platform supports, since CPU ids can be sparse. We define two 394 * sets of macros; one for platforms where the maximum CPU id is less 395 * than the number of bits in a single word (32 in a 32-bit kernel, 396 * 64 in a 64-bit kernel), and one for platforms that require bitmaps 397 * of more than one word. 398 */ 399 400#define CPUSET_WORDS BT_BITOUL(NCPU) 401#define CPUSET_NOTINSET ((uint_t)-1) 402 403#if CPUSET_WORDS > 1 404 405typedef struct cpuset { 406 ulong_t cpub[CPUSET_WORDS]; 407} cpuset_t; 408 409/* 410 * Private functions for manipulating cpusets that do not fit in a 411 * single word. These should not be used directly; instead the 412 * CPUSET_* macros should be used so the code will be portable 413 * across different definitions of NCPU. 414 */ 415extern void cpuset_all(cpuset_t *); 416extern void cpuset_all_but(cpuset_t *, uint_t); 417extern int cpuset_isnull(cpuset_t *); 418extern int cpuset_cmp(cpuset_t *, cpuset_t *); 419extern void cpuset_only(cpuset_t *, uint_t); 420extern uint_t cpuset_find(cpuset_t *); 421extern void cpuset_bounds(cpuset_t *, uint_t *, uint_t *); 422 423#define CPUSET_ALL(set) cpuset_all(&(set)) 424#define CPUSET_ALL_BUT(set, cpu) cpuset_all_but(&(set), cpu) 425#define CPUSET_ONLY(set, cpu) cpuset_only(&(set), cpu) 426#define CPU_IN_SET(set, cpu) BT_TEST((set).cpub, cpu) 427#define CPUSET_ADD(set, cpu) BT_SET((set).cpub, cpu) 428#define CPUSET_DEL(set, cpu) BT_CLEAR((set).cpub, cpu) 429#define CPUSET_ISNULL(set) cpuset_isnull(&(set)) 430#define CPUSET_ISEQUAL(set1, set2) cpuset_cmp(&(set1), &(set2)) 431 432/* 433 * Find one CPU in the cpuset. 434 * Sets "cpu" to the id of the found CPU, or CPUSET_NOTINSET if no cpu 435 * could be found. (i.e. empty set) 436 */ 437#define CPUSET_FIND(set, cpu) { \ 438 cpu = cpuset_find(&(set)); \ 439} 440 441/* 442 * Determine the smallest and largest CPU id in the set. Returns 443 * CPUSET_NOTINSET in smallest and largest when set is empty. 444 */ 445#define CPUSET_BOUNDS(set, smallest, largest) { \ 446 cpuset_bounds(&(set), &(smallest), &(largest)); \ 447} 448 449/* 450 * Atomic cpuset operations 451 * These are safe to use for concurrent cpuset manipulations. 452 * "xdel" and "xadd" are exclusive operations, that set "result" to "0" 453 * if the add or del was successful, or "-1" if not successful. 454 * (e.g. attempting to add a cpu to a cpuset that's already there, or 455 * deleting a cpu that's not in the cpuset) 456 */ 457 458#define CPUSET_ATOMIC_DEL(set, cpu) BT_ATOMIC_CLEAR((set).cpub, (cpu)) 459#define CPUSET_ATOMIC_ADD(set, cpu) BT_ATOMIC_SET((set).cpub, (cpu)) 460 461#define CPUSET_ATOMIC_XADD(set, cpu, result) \ 462 BT_ATOMIC_SET_EXCL((set).cpub, cpu, result) 463 464#define CPUSET_ATOMIC_XDEL(set, cpu, result) \ 465 BT_ATOMIC_CLEAR_EXCL((set).cpub, cpu, result) 466 467 468#define CPUSET_OR(set1, set2) { \ 469 int _i; \ 470 for (_i = 0; _i < CPUSET_WORDS; _i++) \ 471 (set1).cpub[_i] |= (set2).cpub[_i]; \ 472} 473 474#define CPUSET_XOR(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_AND(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_ZERO(set) { \ 487 int _i; \ 488 for (_i = 0; _i < CPUSET_WORDS; _i++) \ 489 (set).cpub[_i] = 0; \ 490} 491 492#elif CPUSET_WORDS == 1 493 494typedef ulong_t cpuset_t; /* a set of CPUs */ 495 496#define CPUSET(cpu) (1UL << (cpu)) 497 498#define CPUSET_ALL(set) ((void)((set) = ~0UL)) 499#define CPUSET_ALL_BUT(set, cpu) ((void)((set) = ~CPUSET(cpu))) 500#define CPUSET_ONLY(set, cpu) ((void)((set) = CPUSET(cpu))) 501#define CPU_IN_SET(set, cpu) ((set) & CPUSET(cpu)) 502#define CPUSET_ADD(set, cpu) ((void)((set) |= CPUSET(cpu))) 503#define CPUSET_DEL(set, cpu) ((void)((set) &= ~CPUSET(cpu))) 504#define CPUSET_ISNULL(set) ((set) == 0) 505#define CPUSET_ISEQUAL(set1, set2) ((set1) == (set2)) 506#define CPUSET_OR(set1, set2) ((void)((set1) |= (set2))) 507#define CPUSET_XOR(set1, set2) ((void)((set1) ^= (set2))) 508#define CPUSET_AND(set1, set2) ((void)((set1) &= (set2))) 509#define CPUSET_ZERO(set) ((void)((set) = 0)) 510 511#define CPUSET_FIND(set, cpu) { \ 512 cpu = (uint_t)(lowbit(set) - 1); \ 513} 514 515#define CPUSET_BOUNDS(set, smallest, largest) { \ 516 smallest = (uint_t)(lowbit(set) - 1); \ 517 largest = (uint_t)(highbit(set) - 1); \ 518} 519 520#define CPUSET_ATOMIC_DEL(set, cpu) atomic_and_ulong(&(set), ~CPUSET(cpu)) 521#define CPUSET_ATOMIC_ADD(set, cpu) atomic_or_ulong(&(set), CPUSET(cpu)) 522 523#define CPUSET_ATOMIC_XADD(set, cpu, result) \ 524 { result = atomic_set_long_excl(&(set), (cpu)); } 525 526#define CPUSET_ATOMIC_XDEL(set, cpu, result) \ 527 { result = atomic_clear_long_excl(&(set), (cpu)); } 528 529#else /* CPUSET_WORDS <= 0 */ 530 531#error NCPU is undefined or invalid 532 533#endif /* CPUSET_WORDS */ 534 535extern cpuset_t cpu_seqid_inuse; 536 537#endif /* (_KERNEL || _KMEMUSER) && _MACHDEP */ 538 539#define CPU_CPR_OFFLINE 0x0 540#define CPU_CPR_ONLINE 0x1 541#define CPU_CPR_IS_OFFLINE(cpu) (((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) == 0) 542#define CPU_CPR_IS_ONLINE(cpu) ((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) 543#define CPU_SET_CPR_FLAGS(cpu, flag) ((cpu)->cpu_cpr_flags |= flag) 544 545#if defined(_KERNEL) || defined(_KMEMUSER) 546 547extern struct cpu *cpu[]; /* indexed by CPU number */ 548extern struct cpu **cpu_seq; /* indexed by sequential CPU id */ 549extern cpu_t *cpu_list; /* list of CPUs */ 550extern cpu_t *cpu_active; /* list of active CPUs */ 551extern int ncpus; /* number of CPUs present */ 552extern int ncpus_online; /* number of CPUs not quiesced */ 553extern int max_ncpus; /* max present before ncpus is known */ 554extern int boot_max_ncpus; /* like max_ncpus but for real */ 555extern int boot_ncpus; /* # cpus present @ boot */ 556extern processorid_t max_cpuid; /* maximum CPU number */ 557extern struct cpu *cpu_inmotion; /* offline or partition move target */ 558extern cpu_t *clock_cpu_list; 559extern processorid_t max_cpu_seqid_ever; /* maximum seqid ever given */ 560 561#if defined(__i386) || defined(__amd64) 562extern struct cpu *curcpup(void); 563#define CPU (curcpup()) /* Pointer to current CPU */ 564#else 565#define CPU (curthread->t_cpu) /* Pointer to current CPU */ 566#endif 567 568/* 569 * CPU_CURRENT indicates to thread_affinity_set to use CPU->cpu_id 570 * as the target and to grab cpu_lock instead of requiring the caller 571 * to grab it. 572 */ 573#define CPU_CURRENT -3 574 575/* 576 * Per-CPU statistics 577 * 578 * cpu_stats_t contains numerous system and VM-related statistics, in the form 579 * of gauges or monotonically-increasing event occurrence counts. 580 */ 581 582#define CPU_STATS_ENTER_K() kpreempt_disable() 583#define CPU_STATS_EXIT_K() kpreempt_enable() 584 585#define CPU_STATS_ADD_K(class, stat, amount) \ 586 { kpreempt_disable(); /* keep from switching CPUs */\ 587 CPU_STATS_ADDQ(CPU, class, stat, amount); \ 588 kpreempt_enable(); \ 589 } 590 591#define CPU_STATS_ADDQ(cp, class, stat, amount) { \ 592 extern void __dtrace_probe___cpu_##class##info_##stat(uint_t, \ 593 uint64_t *, cpu_t *); \ 594 uint64_t *stataddr = &((cp)->cpu_stats.class.stat); \ 595 __dtrace_probe___cpu_##class##info_##stat((amount), \ 596 stataddr, cp); \ 597 *(stataddr) += (amount); \ 598} 599 600#define CPU_STATS(cp, stat) \ 601 ((cp)->cpu_stats.stat) 602 603/* 604 * Increment CPU generation value. 605 * This macro should be called whenever CPU goes on-line or off-line. 606 * Updates to cpu_generation should be protected by cpu_lock. 607 */ 608#define CPU_NEW_GENERATION(cp) ((cp)->cpu_generation++) 609 610#endif /* _KERNEL || _KMEMUSER */ 611 612/* 613 * CPU support routines. 614 */ 615#if defined(_KERNEL) && defined(__STDC__) /* not for genassym.c */ 616 617struct zone; 618 619void cpu_list_init(cpu_t *); 620void cpu_add_unit(cpu_t *); 621void cpu_del_unit(int cpuid); 622void cpu_add_active(cpu_t *); 623void cpu_kstat_init(cpu_t *); 624void cpu_visibility_add(cpu_t *, struct zone *); 625void cpu_visibility_remove(cpu_t *, struct zone *); 626void cpu_visibility_configure(cpu_t *, struct zone *); 627void cpu_visibility_unconfigure(cpu_t *, struct zone *); 628void cpu_visibility_online(cpu_t *, struct zone *); 629void cpu_visibility_offline(cpu_t *, struct zone *); 630void cpu_create_intrstat(cpu_t *); 631void cpu_delete_intrstat(cpu_t *); 632int cpu_kstat_intrstat_update(kstat_t *, int); 633void cpu_intr_swtch_enter(kthread_t *); 634void cpu_intr_swtch_exit(kthread_t *); 635 636void mbox_lock_init(void); /* initialize cross-call locks */ 637void mbox_init(int cpun); /* initialize cross-calls */ 638void poke_cpu(int cpun); /* interrupt another CPU (to preempt) */ 639 640/* 641 * values for safe_list. Pause state that CPUs are in. 642 */ 643#define PAUSE_IDLE 0 /* normal state */ 644#define PAUSE_READY 1 /* paused thread ready to spl */ 645#define PAUSE_WAIT 2 /* paused thread is spl-ed high */ 646#define PAUSE_DIE 3 /* tell pause thread to leave */ 647#define PAUSE_DEAD 4 /* pause thread has left */ 648 649void mach_cpu_pause(volatile char *); 650 651void pause_cpus(cpu_t *off_cp, void *(*func)(void *)); 652void start_cpus(void); 653int cpus_paused(void); 654 655void cpu_pause_init(void); 656cpu_t *cpu_get(processorid_t cpun); /* get the CPU struct associated */ 657 658int cpu_online(cpu_t *cp); /* take cpu online */ 659int cpu_offline(cpu_t *cp, int flags); /* take cpu offline */ 660int cpu_spare(cpu_t *cp, int flags); /* take cpu to spare */ 661int cpu_faulted(cpu_t *cp, int flags); /* take cpu to faulted */ 662int cpu_poweron(cpu_t *cp); /* take powered-off cpu to offline */ 663int cpu_poweroff(cpu_t *cp); /* take offline cpu to powered-off */ 664 665cpu_t *cpu_intr_next(cpu_t *cp); /* get next online CPU taking intrs */ 666int cpu_intr_count(cpu_t *cp); /* count # of CPUs handling intrs */ 667int cpu_intr_on(cpu_t *cp); /* CPU taking I/O interrupts? */ 668void cpu_intr_enable(cpu_t *cp); /* enable I/O interrupts */ 669int cpu_intr_disable(cpu_t *cp); /* disable I/O interrupts */ 670void cpu_intr_alloc(cpu_t *cp, int n); /* allocate interrupt threads */ 671 672/* 673 * Routines for checking CPU states. 674 */ 675int cpu_is_online(cpu_t *); /* check if CPU is online */ 676int cpu_is_nointr(cpu_t *); /* check if CPU can service intrs */ 677int cpu_is_active(cpu_t *); /* check if CPU can run threads */ 678int cpu_is_offline(cpu_t *); /* check if CPU is offline */ 679int cpu_is_poweredoff(cpu_t *); /* check if CPU is powered off */ 680 681int cpu_flagged_online(cpu_flag_t); /* flags show CPU is online */ 682int cpu_flagged_nointr(cpu_flag_t); /* flags show CPU not handling intrs */ 683int cpu_flagged_active(cpu_flag_t); /* flags show CPU scheduling threads */ 684int cpu_flagged_offline(cpu_flag_t); /* flags show CPU is offline */ 685int cpu_flagged_poweredoff(cpu_flag_t); /* flags show CPU is powered off */ 686 687/* 688 * The processor_info(2) state of a CPU is a simplified representation suitable 689 * for use by an application program. Kernel subsystems should utilize the 690 * internal per-CPU state as given by the cpu_flags member of the cpu structure, 691 * as this information may include platform- or architecture-specific state 692 * critical to a subsystem's disposition of a particular CPU. 693 */ 694void cpu_set_state(cpu_t *); /* record/timestamp current state */ 695int cpu_get_state(cpu_t *); /* get current cpu state */ 696const char *cpu_get_state_str(cpu_t *); /* get current cpu state as string */ 697 698 699void cpu_set_curr_clock(uint64_t); /* indicate the current CPU's freq */ 700void cpu_set_supp_freqs(cpu_t *, const char *); /* set the CPU supported */ 701 /* frequencies */ 702 703int cpu_configure(int); 704int cpu_unconfigure(int); 705void cpu_destroy_bound_threads(cpu_t *cp); 706 707extern int cpu_bind_thread(kthread_t *tp, processorid_t bind, 708 processorid_t *obind, int *error); 709extern int cpu_unbind(processorid_t cpu_id, boolean_t force); 710extern void thread_affinity_set(kthread_t *t, int cpu_id); 711extern void thread_affinity_clear(kthread_t *t); 712extern void affinity_set(int cpu_id); 713extern void affinity_clear(void); 714extern void init_cpu_mstate(struct cpu *, int); 715extern void term_cpu_mstate(struct cpu *); 716extern void new_cpu_mstate(int, hrtime_t); 717extern void get_cpu_mstate(struct cpu *, hrtime_t *); 718extern void thread_nomigrate(void); 719extern void thread_allowmigrate(void); 720extern void weakbinding_stop(void); 721extern void weakbinding_start(void); 722 723/* 724 * The following routines affect the CPUs participation in interrupt processing, 725 * if that is applicable on the architecture. This only affects interrupts 726 * which aren't directed at the processor (not cross calls). 727 * 728 * cpu_disable_intr returns non-zero if interrupts were previously enabled. 729 */ 730int cpu_disable_intr(struct cpu *cp); /* stop issuing interrupts to cpu */ 731void cpu_enable_intr(struct cpu *cp); /* start issuing interrupts to cpu */ 732 733/* 734 * The mutex cpu_lock protects cpu_flags for all CPUs, as well as the ncpus 735 * and ncpus_online counts. 736 */ 737extern kmutex_t cpu_lock; /* lock protecting CPU data */ 738 739/* 740 * CPU state change events 741 * 742 * Various subsystems need to know when CPUs change their state. They get this 743 * information by registering CPU state change callbacks using 744 * register_cpu_setup_func(). Whenever any CPU changes its state, the callback 745 * function is called. The callback function is passed three arguments: 746 * 747 * Event, described by cpu_setup_t 748 * CPU ID 749 * Transparent pointer passed when registering the callback 750 * 751 * The callback function is called with cpu_lock held. The return value from the 752 * callback function is usually ignored, except for CPU_CONFIG and CPU_UNCONFIG 753 * events. For these two events, non-zero return value indicates a failure and 754 * prevents successful completion of the operation. 755 * 756 * New events may be added in the future. Callback functions should ignore any 757 * events that they do not understand. 758 * 759 * The following events provide notification callbacks: 760 * 761 * CPU_INIT A new CPU is started and added to the list of active CPUs 762 * This event is only used during boot 763 * 764 * CPU_CONFIG A newly inserted CPU is prepared for starting running code 765 * This event is called by DR code 766 * 767 * CPU_UNCONFIG CPU has been powered off and needs cleanup 768 * This event is called by DR code 769 * 770 * CPU_ON CPU is enabled but does not run anything yet 771 * 772 * CPU_INTR_ON CPU is enabled and has interrupts enabled 773 * 774 * CPU_OFF CPU is going offline but can still run threads 775 * 776 * CPU_CPUPART_OUT CPU is going to move out of its partition 777 * 778 * CPU_CPUPART_IN CPU is going to move to a new partition 779 * 780 * CPU_SETUP CPU is set up during boot and can run threads 781 */ 782typedef enum { 783 CPU_INIT, 784 CPU_CONFIG, 785 CPU_UNCONFIG, 786 CPU_ON, 787 CPU_OFF, 788 CPU_CPUPART_IN, 789 CPU_CPUPART_OUT, 790 CPU_SETUP, 791 CPU_INTR_ON 792} cpu_setup_t; 793 794typedef int cpu_setup_func_t(cpu_setup_t, int, void *); 795 796/* 797 * Routines used to register interest in cpu's being added to or removed 798 * from the system. 799 */ 800extern void register_cpu_setup_func(cpu_setup_func_t *, void *); 801extern void unregister_cpu_setup_func(cpu_setup_func_t *, void *); 802extern void cpu_state_change_notify(int, cpu_setup_t); 803 804/* 805 * Call specified function on the given CPU 806 */ 807typedef void (*cpu_call_func_t)(uintptr_t, uintptr_t); 808extern void cpu_call(cpu_t *, cpu_call_func_t, uintptr_t, uintptr_t); 809 810 811/* 812 * Create various strings that describe the given CPU for the 813 * processor_info system call and configuration-related kstats. 814 */ 815#define CPU_IDSTRLEN 100 816 817extern void init_cpu_info(struct cpu *); 818extern void populate_idstr(struct cpu *); 819extern void cpu_vm_data_init(struct cpu *); 820extern void cpu_vm_data_destroy(struct cpu *); 821 822#endif /* _KERNEL */ 823 824#ifdef __cplusplus 825} 826#endif 827 828#endif /* _SYS_CPUVAR_H */ 829