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 2009 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 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#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) 36#include <sys/machcpuvar.h> 37#endif 38 39#include <sys/types.h> 40#include <sys/file.h> 41#include <sys/bitmap.h> 42#include <sys/rwlock.h> 43#include <sys/msacct.h> 44#if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) && \ 45 (defined(__i386) || defined(__amd64)) 46#include <asm/cpuvar.h> 47#endif 48 49#ifdef __cplusplus 50extern "C" { 51#endif 52 53struct squeue_set_s; 54 55#define CPU_CACHE_COHERENCE_SIZE 64 56#define S_LOADAVG_SZ 11 57#define S_MOVAVG_SZ 10 58 59struct loadavg_s { 60 int lg_cur; /* current loadavg entry */ 61 unsigned int lg_len; /* number entries recorded */ 62 hrtime_t lg_total; /* used to temporarily hold load totals */ 63 hrtime_t lg_loads[S_LOADAVG_SZ]; /* table of recorded entries */ 64}; 65 66/* 67 * For fast event tracing. 68 */ 69struct ftrace_record; 70typedef struct ftrace_data { 71 int ftd_state; /* ftrace flags */ 72 kmutex_t ftd_unused; /* ftrace buffer lock, unused */ 73 struct ftrace_record *ftd_cur; /* current record */ 74 struct ftrace_record *ftd_first; /* first record */ 75 struct ftrace_record *ftd_last; /* last record */ 76} ftrace_data_t; 77 78struct cyc_cpu; 79struct nvlist; 80 81/* 82 * Per-CPU data. 83 * 84 * Be careful adding new members: if they are not the same in all modules (e.g. 85 * change size depending on a #define), CTF uniquification can fail to work 86 * properly. Furthermore, this is transitive in that it applies recursively to 87 * all types pointed to by cpu_t. 88 */ 89typedef struct cpu { 90 processorid_t cpu_id; /* CPU number */ 91 processorid_t cpu_seqid; /* sequential CPU id (0..ncpus-1) */ 92 volatile cpu_flag_t cpu_flags; /* flags indicating CPU state */ 93 struct cpu *cpu_self; /* pointer to itself */ 94 kthread_t *cpu_thread; /* current thread */ 95 kthread_t *cpu_idle_thread; /* idle thread for this CPU */ 96 kthread_t *cpu_pause_thread; /* pause thread for this CPU */ 97 klwp_id_t cpu_lwp; /* current lwp (if any) */ 98 klwp_id_t cpu_fpowner; /* currently loaded fpu owner */ 99 struct cpupart *cpu_part; /* partition with this CPU */ 100 struct lgrp_ld *cpu_lpl; /* pointer to this cpu's load */ 101 int cpu_cache_offset; /* see kmem.c for details */ 102 103 /* 104 * Links to other CPUs. It is safe to walk these lists if 105 * one of the following is true: 106 * - cpu_lock held 107 * - preemption disabled via kpreempt_disable 108 * - PIL >= DISP_LEVEL 109 * - acting thread is an interrupt thread 110 * - all other CPUs are paused 111 */ 112 struct cpu *cpu_next; /* next existing CPU */ 113 struct cpu *cpu_prev; /* prev existing CPU */ 114 struct cpu *cpu_next_onln; /* next online (enabled) CPU */ 115 struct cpu *cpu_prev_onln; /* prev online (enabled) CPU */ 116 struct cpu *cpu_next_part; /* next CPU in partition */ 117 struct cpu *cpu_prev_part; /* prev CPU in partition */ 118 struct cpu *cpu_next_lgrp; /* next CPU in latency group */ 119 struct cpu *cpu_prev_lgrp; /* prev CPU in latency group */ 120 struct cpu *cpu_next_lpl; /* next CPU in lgrp partition */ 121 struct cpu *cpu_prev_lpl; 122 123 struct cpu_pg *cpu_pg; /* cpu's processor groups */ 124 125 void *cpu_reserved[4]; /* reserved for future use */ 126 127 /* 128 * Scheduling variables. 129 */ 130 disp_t *cpu_disp; /* dispatch queue data */ 131 /* 132 * Note that cpu_disp is set before the CPU is added to the system 133 * and is never modified. Hence, no additional locking is needed 134 * beyond what's necessary to access the cpu_t structure. 135 */ 136 char cpu_runrun; /* scheduling flag - set to preempt */ 137 char cpu_kprunrun; /* force kernel preemption */ 138 pri_t cpu_chosen_level; /* priority at which cpu */ 139 /* was chosen for scheduling */ 140 kthread_t *cpu_dispthread; /* thread selected for dispatch */ 141 disp_lock_t cpu_thread_lock; /* dispatcher lock on current thread */ 142 uint8_t cpu_disp_flags; /* flags used by dispatcher */ 143 /* 144 * The following field is updated when ever the cpu_dispthread 145 * changes. Also in places, where the current thread(cpu_dispthread) 146 * priority changes. This is used in disp_lowpri_cpu() 147 */ 148 pri_t cpu_dispatch_pri; /* priority of cpu_dispthread */ 149 clock_t cpu_last_swtch; /* last time switched to new thread */ 150 151 /* 152 * Interrupt data. 153 */ 154 caddr_t cpu_intr_stack; /* interrupt stack */ 155 kthread_t *cpu_intr_thread; /* interrupt thread list */ 156 uint_t cpu_intr_actv; /* interrupt levels active (bitmask) */ 157 int cpu_base_spl; /* priority for highest rupt active */ 158 159 /* 160 * Statistics. 161 */ 162 cpu_stats_t cpu_stats; /* per-CPU statistics */ 163 struct kstat *cpu_info_kstat; /* kstat for cpu info */ 164 165 uintptr_t cpu_profile_pc; /* kernel PC in profile interrupt */ 166 uintptr_t cpu_profile_upc; /* user PC in profile interrupt */ 167 uintptr_t cpu_profile_pil; /* PIL when profile interrupted */ 168 169 ftrace_data_t cpu_ftrace; /* per cpu ftrace data */ 170 171 clock_t cpu_deadman_counter; /* used by deadman() */ 172 uint_t cpu_deadman_countdown; /* used by deadman() */ 173 174 kmutex_t cpu_cpc_ctxlock; /* protects context for idle thread */ 175 kcpc_ctx_t *cpu_cpc_ctx; /* performance counter context */ 176 177 /* 178 * Configuration information for the processor_info system call. 179 */ 180 processor_info_t cpu_type_info; /* config info */ 181 time_t cpu_state_begin; /* when CPU entered current state */ 182 char cpu_cpr_flags; /* CPR related info */ 183 struct cyc_cpu *cpu_cyclic; /* per cpu cyclic subsystem data */ 184 struct squeue_set_s *cpu_squeue_set; /* per cpu squeue set */ 185 struct nvlist *cpu_props; /* pool-related properties */ 186 187 krwlock_t cpu_ft_lock; /* DTrace: fasttrap lock */ 188 uintptr_t cpu_dtrace_caller; /* DTrace: caller, if any */ 189 hrtime_t cpu_dtrace_chillmark; /* DTrace: chill mark time */ 190 hrtime_t cpu_dtrace_chilled; /* DTrace: total chill time */ 191 volatile uint16_t cpu_mstate; /* cpu microstate */ 192 volatile uint16_t cpu_mstate_gen; /* generation counter */ 193 volatile hrtime_t cpu_mstate_start; /* cpu microstate start time */ 194 volatile hrtime_t cpu_acct[NCMSTATES]; /* cpu microstate data */ 195 hrtime_t cpu_intracct[NCMSTATES]; /* interrupt mstate data */ 196 hrtime_t cpu_waitrq; /* cpu run-queue wait time */ 197 struct loadavg_s cpu_loadavg; /* loadavg info for this cpu */ 198 199 char *cpu_idstr; /* for printing and debugging */ 200 char *cpu_brandstr; /* for printing */ 201 202 /* 203 * Sum of all device interrupt weights that are currently directed at 204 * this cpu. Cleared at start of interrupt redistribution. 205 */ 206 int32_t cpu_intr_weight; 207 void *cpu_vm_data; 208 209 struct cpu_physid *cpu_physid; /* physical associations */ 210 211 uint64_t cpu_curr_clock; /* current clock freq in Hz */ 212 char *cpu_supp_freqs; /* supported freqs in Hz */ 213 214 uintptr_t cpu_cpcprofile_pc; /* kernel PC in cpc interrupt */ 215 uintptr_t cpu_cpcprofile_upc; /* user PC in cpc interrupt */ 216 217 /* 218 * Interrupt load factor used by dispatcher & softcall 219 */ 220 hrtime_t cpu_intrlast; /* total interrupt time (nsec) */ 221 int cpu_intrload; /* interrupt load factor (0-99%) */ 222 223 uint_t cpu_rotor; /* for cheap pseudo-random numbers */ 224 225 struct cu_cpu_info *cpu_cu_info; /* capacity & util. info */ 226 227 /* 228 * cpu_generation is updated whenever CPU goes on-line or off-line. 229 * Updates to cpu_generation are protected by cpu_lock. 230 * 231 * See CPU_NEW_GENERATION() macro below. 232 */ 233 volatile uint_t cpu_generation; /* tracking on/off-line */ 234 235 /* 236 * New members must be added /before/ this member, as the CTF tools 237 * rely on this being the last field before cpu_m, so they can 238 * correctly calculate the offset when synthetically adding the cpu_m 239 * member in objects that do not have it. This fixup is required for 240 * uniquification to work correctly. 241 */ 242 uintptr_t cpu_m_pad; 243 244#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) 245 struct machcpu cpu_m; /* per architecture info */ 246#endif 247} cpu_t; 248 249/* 250 * The cpu_core structure consists of per-CPU state available in any context. 251 * On some architectures, this may mean that the page(s) containing the 252 * NCPU-sized array of cpu_core structures must be locked in the TLB -- it 253 * is up to the platform to assure that this is performed properly. Note that 254 * the structure is sized to avoid false sharing. 255 */ 256#define CPUC_SIZE (sizeof (uint16_t) + sizeof (uint8_t) + \ 257 sizeof (uintptr_t) + sizeof (kmutex_t)) 258#define CPUC_PADSIZE CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE 259 260typedef struct cpu_core { 261 uint16_t cpuc_dtrace_flags; /* DTrace flags */ 262 uint8_t cpuc_dcpc_intr_state; /* DCPC provider intr state */ 263 uint8_t cpuc_pad[CPUC_PADSIZE]; /* padding */ 264 uintptr_t cpuc_dtrace_illval; /* DTrace illegal value */ 265 kmutex_t cpuc_pid_lock; /* DTrace pid provider lock */ 266} cpu_core_t; 267 268#ifdef _KERNEL 269extern cpu_core_t cpu_core[]; 270#endif /* _KERNEL */ 271 272/* 273 * CPU_ON_INTR() macro. Returns non-zero if currently on interrupt stack. 274 * Note that this isn't a test for a high PIL. For example, cpu_intr_actv 275 * does not get updated when we go through sys_trap from TL>0 at high PIL. 276 * getpil() should be used instead to check for PIL levels. 277 */ 278#define CPU_ON_INTR(cpup) ((cpup)->cpu_intr_actv >> (LOCK_LEVEL + 1)) 279 280/* 281 * Check to see if an interrupt thread might be active at a given ipl. 282 * If so return true. 283 * We must be conservative--it is ok to give a false yes, but a false no 284 * will cause disaster. (But if the situation changes after we check it is 285 * ok--the caller is trying to ensure that an interrupt routine has been 286 * exited). 287 * This is used when trying to remove an interrupt handler from an autovector 288 * list in avintr.c. 289 */ 290#define INTR_ACTIVE(cpup, level) \ 291 ((level) <= LOCK_LEVEL ? \ 292 ((cpup)->cpu_intr_actv & (1 << (level))) : (CPU_ON_INTR(cpup))) 293 294/* 295 * CPU_PSEUDO_RANDOM() returns a per CPU value that changes each time one 296 * looks at it. It's meant as a cheap mechanism to be incorporated in routines 297 * wanting to avoid biasing, but where true randomness isn't needed (just 298 * something that changes). 299 */ 300#define CPU_PSEUDO_RANDOM() (CPU->cpu_rotor++) 301 302#if defined(_KERNEL) || defined(_KMEMUSER) 303 304#define INTR_STACK_SIZE MAX(DEFAULTSTKSZ, PAGESIZE) 305 306/* MEMBERS PROTECTED BY "atomicity": cpu_flags */ 307 308/* 309 * Flags in the CPU structure. 310 * 311 * These are protected by cpu_lock (except during creation). 312 * 313 * Offlined-CPUs have three stages of being offline: 314 * 315 * CPU_ENABLE indicates that the CPU is participating in I/O interrupts 316 * that can be directed at a number of different CPUs. If CPU_ENABLE 317 * is off, the CPU will not be given interrupts that can be sent elsewhere, 318 * but will still get interrupts from devices associated with that CPU only, 319 * and from other CPUs. 320 * 321 * CPU_OFFLINE indicates that the dispatcher should not allow any threads 322 * other than interrupt threads to run on that CPU. A CPU will not have 323 * CPU_OFFLINE set if there are any bound threads (besides interrupts). 324 * 325 * CPU_QUIESCED is set if p_offline was able to completely turn idle the 326 * CPU and it will not have to run interrupt threads. In this case it'll 327 * stay in the idle loop until CPU_QUIESCED is turned off. 328 * 329 * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully 330 * suspended (in the suspend path), or have yet to be resumed (in the resume 331 * case). 332 * 333 * On some platforms CPUs can be individually powered off. 334 * The following flags are set for powered off CPUs: CPU_QUIESCED, 335 * CPU_OFFLINE, and CPU_POWEROFF. The following flags are cleared: 336 * CPU_RUNNING, CPU_READY, CPU_EXISTS, and CPU_ENABLE. 337 */ 338#define CPU_RUNNING 0x001 /* CPU running */ 339#define CPU_READY 0x002 /* CPU ready for cross-calls */ 340#define CPU_QUIESCED 0x004 /* CPU will stay in idle */ 341#define CPU_EXISTS 0x008 /* CPU is configured */ 342#define CPU_ENABLE 0x010 /* CPU enabled for interrupts */ 343#define CPU_OFFLINE 0x020 /* CPU offline via p_online */ 344#define CPU_POWEROFF 0x040 /* CPU is powered off */ 345#define CPU_FROZEN 0x080 /* CPU is frozen via CPR suspend */ 346#define CPU_SPARE 0x100 /* CPU offline available for use */ 347#define CPU_FAULTED 0x200 /* CPU offline diagnosed faulty */ 348 349#define FMT_CPU_FLAGS \ 350 "\20\12fault\11spare\10frozen" \ 351 "\7poweroff\6offline\5enable\4exist\3quiesced\2ready\1run" 352 353#define CPU_ACTIVE(cpu) (((cpu)->cpu_flags & CPU_OFFLINE) == 0) 354 355/* 356 * Flags for cpu_offline(), cpu_faulted(), and cpu_spare(). 357 */ 358#define CPU_FORCED 0x0001 /* Force CPU offline */ 359 360/* 361 * DTrace flags. 362 */ 363#define CPU_DTRACE_NOFAULT 0x0001 /* Don't fault */ 364#define CPU_DTRACE_DROP 0x0002 /* Drop this ECB */ 365#define CPU_DTRACE_BADADDR 0x0004 /* DTrace fault: bad address */ 366#define CPU_DTRACE_BADALIGN 0x0008 /* DTrace fault: bad alignment */ 367#define CPU_DTRACE_DIVZERO 0x0010 /* DTrace fault: divide by zero */ 368#define CPU_DTRACE_ILLOP 0x0020 /* DTrace fault: illegal operation */ 369#define CPU_DTRACE_NOSCRATCH 0x0040 /* DTrace fault: out of scratch */ 370#define CPU_DTRACE_KPRIV 0x0080 /* DTrace fault: bad kernel access */ 371#define CPU_DTRACE_UPRIV 0x0100 /* DTrace fault: bad user access */ 372#define CPU_DTRACE_TUPOFLOW 0x0200 /* DTrace fault: tuple stack overflow */ 373#if defined(__sparc) 374#define CPU_DTRACE_FAKERESTORE 0x0400 /* pid provider hint to getreg */ 375#endif 376#define CPU_DTRACE_ENTRY 0x0800 /* pid provider hint to ustack() */ 377#define CPU_DTRACE_BADSTACK 0x1000 /* DTrace fault: bad stack */ 378 379#define CPU_DTRACE_FAULT (CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \ 380 CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \ 381 CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \ 382 CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \ 383 CPU_DTRACE_BADSTACK) 384#define CPU_DTRACE_ERROR (CPU_DTRACE_FAULT | CPU_DTRACE_DROP) 385 386/* 387 * Dispatcher flags 388 * These flags must be changed only by the current CPU. 389 */ 390#define CPU_DISP_DONTSTEAL 0x01 /* CPU undergoing context swtch */ 391#define CPU_DISP_HALTED 0x02 /* CPU halted waiting for interrupt */ 392 393#endif /* _KERNEL || _KMEMUSER */ 394 395#if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP) 396 397/* 398 * Macros for manipulating sets of CPUs as a bitmap. Note that this 399 * bitmap may vary in size depending on the maximum CPU id a specific 400 * platform supports. This may be different than the number of CPUs 401 * the platform supports, since CPU ids can be sparse. We define two 402 * sets of macros; one for platforms where the maximum CPU id is less 403 * than the number of bits in a single word (32 in a 32-bit kernel, 404 * 64 in a 64-bit kernel), and one for platforms that require bitmaps 405 * of more than one word. 406 */ 407 408#define CPUSET_WORDS BT_BITOUL(NCPU) 409#define CPUSET_NOTINSET ((uint_t)-1) 410 411#if CPUSET_WORDS > 1 412 413typedef struct cpuset { 414 ulong_t cpub[CPUSET_WORDS]; 415} cpuset_t; 416 417/* 418 * Private functions for manipulating cpusets that do not fit in a 419 * single word. These should not be used directly; instead the 420 * CPUSET_* macros should be used so the code will be portable 421 * across different definitions of NCPU. 422 */ 423extern void cpuset_all(cpuset_t *); 424extern void cpuset_all_but(cpuset_t *, uint_t); 425extern int cpuset_isnull(cpuset_t *); 426extern int cpuset_cmp(cpuset_t *, cpuset_t *); 427extern void cpuset_only(cpuset_t *, uint_t); 428extern uint_t cpuset_find(cpuset_t *); 429extern void cpuset_bounds(cpuset_t *, uint_t *, uint_t *); 430 431#define CPUSET_ALL(set) cpuset_all(&(set)) 432#define CPUSET_ALL_BUT(set, cpu) cpuset_all_but(&(set), cpu) 433#define CPUSET_ONLY(set, cpu) cpuset_only(&(set), cpu) 434#define CPU_IN_SET(set, cpu) BT_TEST((set).cpub, cpu) 435#define CPUSET_ADD(set, cpu) BT_SET((set).cpub, cpu) 436#define CPUSET_DEL(set, cpu) BT_CLEAR((set).cpub, cpu) 437#define CPUSET_ISNULL(set) cpuset_isnull(&(set)) 438#define CPUSET_ISEQUAL(set1, set2) cpuset_cmp(&(set1), &(set2)) 439 440/* 441 * Find one CPU in the cpuset. 442 * Sets "cpu" to the id of the found CPU, or CPUSET_NOTINSET if no cpu 443 * could be found. (i.e. empty set) 444 */ 445#define CPUSET_FIND(set, cpu) { \ 446 cpu = cpuset_find(&(set)); \ 447} 448 449/* 450 * Determine the smallest and largest CPU id in the set. Returns 451 * CPUSET_NOTINSET in smallest and largest when set is empty. 452 */ 453#define CPUSET_BOUNDS(set, smallest, largest) { \ 454 cpuset_bounds(&(set), &(smallest), &(largest)); \ 455} 456 457/* 458 * Atomic cpuset operations 459 * These are safe to use for concurrent cpuset manipulations. 460 * "xdel" and "xadd" are exclusive operations, that set "result" to "0" 461 * if the add or del was successful, or "-1" if not successful. 462 * (e.g. attempting to add a cpu to a cpuset that's already there, or 463 * deleting a cpu that's not in the cpuset) 464 */ 465 466#define CPUSET_ATOMIC_DEL(set, cpu) BT_ATOMIC_CLEAR((set).cpub, (cpu)) 467#define CPUSET_ATOMIC_ADD(set, cpu) BT_ATOMIC_SET((set).cpub, (cpu)) 468 469#define CPUSET_ATOMIC_XADD(set, cpu, result) \ 470 BT_ATOMIC_SET_EXCL((set).cpub, cpu, result) 471 472#define CPUSET_ATOMIC_XDEL(set, cpu, result) \ 473 BT_ATOMIC_CLEAR_EXCL((set).cpub, cpu, result) 474 475 476#define CPUSET_OR(set1, set2) { \ 477 int _i; \ 478 for (_i = 0; _i < CPUSET_WORDS; _i++) \ 479 (set1).cpub[_i] |= (set2).cpub[_i]; \ 480} 481 482#define CPUSET_XOR(set1, set2) { \ 483 int _i; \ 484 for (_i = 0; _i < CPUSET_WORDS; _i++) \ 485 (set1).cpub[_i] ^= (set2).cpub[_i]; \ 486} 487 488#define CPUSET_AND(set1, set2) { \ 489 int _i; \ 490 for (_i = 0; _i < CPUSET_WORDS; _i++) \ 491 (set1).cpub[_i] &= (set2).cpub[_i]; \ 492} 493 494#define CPUSET_ZERO(set) { \ 495 int _i; \ 496 for (_i = 0; _i < CPUSET_WORDS; _i++) \ 497 (set).cpub[_i] = 0; \ 498} 499 500#elif CPUSET_WORDS == 1 501 502typedef ulong_t cpuset_t; /* a set of CPUs */ 503 504#define CPUSET(cpu) (1UL << (cpu)) 505 506#define CPUSET_ALL(set) ((void)((set) = ~0UL)) 507#define CPUSET_ALL_BUT(set, cpu) ((void)((set) = ~CPUSET(cpu))) 508#define CPUSET_ONLY(set, cpu) ((void)((set) = CPUSET(cpu))) 509#define CPU_IN_SET(set, cpu) ((set) & CPUSET(cpu)) 510#define CPUSET_ADD(set, cpu) ((void)((set) |= CPUSET(cpu))) 511#define CPUSET_DEL(set, cpu) ((void)((set) &= ~CPUSET(cpu))) 512#define CPUSET_ISNULL(set) ((set) == 0) 513#define CPUSET_ISEQUAL(set1, set2) ((set1) == (set2)) 514#define CPUSET_OR(set1, set2) ((void)((set1) |= (set2))) 515#define CPUSET_XOR(set1, set2) ((void)((set1) ^= (set2))) 516#define CPUSET_AND(set1, set2) ((void)((set1) &= (set2))) 517#define CPUSET_ZERO(set) ((void)((set) = 0)) 518 519#define CPUSET_FIND(set, cpu) { \ 520 cpu = (uint_t)(lowbit(set) - 1); \ 521} 522 523#define CPUSET_BOUNDS(set, smallest, largest) { \ 524 smallest = (uint_t)(lowbit(set) - 1); \ 525 largest = (uint_t)(highbit(set) - 1); \ 526} 527 528#define CPUSET_ATOMIC_DEL(set, cpu) atomic_and_long(&(set), ~CPUSET(cpu)) 529#define CPUSET_ATOMIC_ADD(set, cpu) atomic_or_long(&(set), CPUSET(cpu)) 530 531#define CPUSET_ATOMIC_XADD(set, cpu, result) \ 532 { result = atomic_set_long_excl(&(set), (cpu)); } 533 534#define CPUSET_ATOMIC_XDEL(set, cpu, result) \ 535 { result = atomic_clear_long_excl(&(set), (cpu)); } 536 537#else /* CPUSET_WORDS <= 0 */ 538 539#error NCPU is undefined or invalid 540 541#endif /* CPUSET_WORDS */ 542 543extern cpuset_t cpu_seqid_inuse; 544 545#endif /* (_KERNEL || _KMEMUSER) && _MACHDEP */ 546 547#define CPU_CPR_OFFLINE 0x0 548#define CPU_CPR_ONLINE 0x1 549#define CPU_CPR_IS_OFFLINE(cpu) (((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) == 0) 550#define CPU_CPR_IS_ONLINE(cpu) ((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) 551#define CPU_SET_CPR_FLAGS(cpu, flag) ((cpu)->cpu_cpr_flags |= flag) 552 553#if defined(_KERNEL) || defined(_KMEMUSER) 554 555extern struct cpu *cpu[]; /* indexed by CPU number */ 556extern struct cpu **cpu_seq; /* indexed by sequential CPU id */ 557extern cpu_t *cpu_list; /* list of CPUs */ 558extern cpu_t *cpu_active; /* list of active CPUs */ 559extern int ncpus; /* number of CPUs present */ 560extern int ncpus_online; /* number of CPUs not quiesced */ 561extern int max_ncpus; /* max present before ncpus is known */ 562extern int boot_max_ncpus; /* like max_ncpus but for real */ 563extern int boot_ncpus; /* # cpus present @ boot */ 564extern processorid_t max_cpuid; /* maximum CPU number */ 565extern struct cpu *cpu_inmotion; /* offline or partition move target */ 566extern cpu_t *clock_cpu_list; 567 568#if defined(__i386) || defined(__amd64) 569extern 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 CPU->cpu_id 577 * as the target and to grab cpu_lock instead of requiring the caller 578 * to grab it. 579 */ 580#define CPU_CURRENT -3 581 582/* 583 * Per-CPU statistics 584 * 585 * cpu_stats_t contains numerous system and VM-related statistics, in the form 586 * of gauges or monotonically-increasing event occurrence counts. 587 */ 588 589#define CPU_STATS_ENTER_K() kpreempt_disable() 590#define CPU_STATS_EXIT_K() kpreempt_enable() 591 592#define CPU_STATS_ADD_K(class, stat, amount) \ 593 { kpreempt_disable(); /* keep from switching CPUs */\ 594 CPU_STATS_ADDQ(CPU, class, stat, amount); \ 595 kpreempt_enable(); \ 596 } 597 598#define CPU_STATS_ADDQ(cp, class, stat, amount) { \ 599 extern void __dtrace_probe___cpu_##class##info_##stat(uint_t, \ 600 uint64_t *, cpu_t *); \ 601 uint64_t *stataddr = &((cp)->cpu_stats.class.stat); \ 602 __dtrace_probe___cpu_##class##info_##stat((amount), \ 603 stataddr, cp); \ 604 *(stataddr) += (amount); \ 605} 606 607#define CPU_STATS(cp, stat) \ 608 ((cp)->cpu_stats.stat) 609 610/* 611 * Increment CPU generation value. 612 * This macro should be called whenever CPU goes on-line or off-line. 613 * Updates to cpu_generation should be protected by cpu_lock. 614 */ 615#define CPU_NEW_GENERATION(cp) ((cp)->cpu_generation++) 616 617#endif /* _KERNEL || _KMEMUSER */ 618 619/* 620 * CPU support routines. 621 */ 622#if defined(_KERNEL) && defined(__STDC__) /* not for genassym.c */ 623 624struct zone; 625 626void cpu_list_init(cpu_t *); 627void cpu_add_unit(cpu_t *); 628void cpu_del_unit(int cpuid); 629void cpu_add_active(cpu_t *); 630void cpu_kstat_init(cpu_t *); 631void cpu_visibility_add(cpu_t *, struct zone *); 632void cpu_visibility_remove(cpu_t *, struct zone *); 633void cpu_visibility_configure(cpu_t *, struct zone *); 634void cpu_visibility_unconfigure(cpu_t *, struct zone *); 635void cpu_visibility_online(cpu_t *, struct zone *); 636void cpu_visibility_offline(cpu_t *, struct zone *); 637void cpu_create_intrstat(cpu_t *); 638void cpu_delete_intrstat(cpu_t *); 639int cpu_kstat_intrstat_update(kstat_t *, int); 640void cpu_intr_swtch_enter(kthread_t *); 641void cpu_intr_swtch_exit(kthread_t *); 642 643void mbox_lock_init(void); /* initialize cross-call locks */ 644void mbox_init(int cpun); /* initialize cross-calls */ 645void poke_cpu(int cpun); /* interrupt another CPU (to preempt) */ 646 647/* 648 * values for safe_list. Pause state that CPUs are in. 649 */ 650#define PAUSE_IDLE 0 /* normal state */ 651#define PAUSE_READY 1 /* paused thread ready to spl */ 652#define PAUSE_WAIT 2 /* paused thread is spl-ed high */ 653#define PAUSE_DIE 3 /* tell pause thread to leave */ 654#define PAUSE_DEAD 4 /* pause thread has left */ 655 656void mach_cpu_pause(volatile char *); 657 658void pause_cpus(cpu_t *off_cp); 659void start_cpus(void); 660int cpus_paused(void); 661 662void cpu_pause_init(void); 663cpu_t *cpu_get(processorid_t cpun); /* get the CPU struct associated */ 664 665int cpu_online(cpu_t *cp); /* take cpu online */ 666int cpu_offline(cpu_t *cp, int flags); /* take cpu offline */ 667int cpu_spare(cpu_t *cp, int flags); /* take cpu to spare */ 668int cpu_faulted(cpu_t *cp, int flags); /* take cpu to faulted */ 669int cpu_poweron(cpu_t *cp); /* take powered-off cpu to offline */ 670int cpu_poweroff(cpu_t *cp); /* take offline cpu to powered-off */ 671 672cpu_t *cpu_intr_next(cpu_t *cp); /* get next online CPU taking intrs */ 673int cpu_intr_count(cpu_t *cp); /* count # of CPUs handling intrs */ 674int cpu_intr_on(cpu_t *cp); /* CPU taking I/O interrupts? */ 675void cpu_intr_enable(cpu_t *cp); /* enable I/O interrupts */ 676int cpu_intr_disable(cpu_t *cp); /* disable I/O interrupts */ 677void cpu_intr_alloc(cpu_t *cp, int n); /* allocate interrupt threads */ 678 679/* 680 * Routines for checking CPU states. 681 */ 682int cpu_is_online(cpu_t *); /* check if CPU is online */ 683int cpu_is_nointr(cpu_t *); /* check if CPU can service intrs */ 684int cpu_is_active(cpu_t *); /* check if CPU can run threads */ 685int cpu_is_offline(cpu_t *); /* check if CPU is offline */ 686int cpu_is_poweredoff(cpu_t *); /* check if CPU is powered off */ 687 688int cpu_flagged_online(cpu_flag_t); /* flags show CPU is online */ 689int cpu_flagged_nointr(cpu_flag_t); /* flags show CPU not handling intrs */ 690int cpu_flagged_active(cpu_flag_t); /* flags show CPU scheduling threads */ 691int cpu_flagged_offline(cpu_flag_t); /* flags show CPU is offline */ 692int cpu_flagged_poweredoff(cpu_flag_t); /* flags show CPU is powered off */ 693 694/* 695 * The processor_info(2) state of a CPU is a simplified representation suitable 696 * for use by an application program. Kernel subsystems should utilize the 697 * internal per-CPU state as given by the cpu_flags member of the cpu structure, 698 * as this information may include platform- or architecture-specific state 699 * critical to a subsystem's disposition of a particular CPU. 700 */ 701void cpu_set_state(cpu_t *); /* record/timestamp current state */ 702int cpu_get_state(cpu_t *); /* get current cpu state */ 703const char *cpu_get_state_str(cpu_t *); /* get current cpu state as string */ 704 705 706void cpu_set_curr_clock(uint64_t); /* indicate the current CPU's freq */ 707void cpu_set_supp_freqs(cpu_t *, const char *); /* set the CPU supported */ 708 /* frequencies */ 709 710int cpu_configure(int); 711int cpu_unconfigure(int); 712void cpu_destroy_bound_threads(cpu_t *cp); 713 714extern int cpu_bind_thread(kthread_t *tp, processorid_t bind, 715 processorid_t *obind, int *error); 716extern int cpu_unbind(processorid_t cpu_id, boolean_t force); 717extern void thread_affinity_set(kthread_t *t, int cpu_id); 718extern void thread_affinity_clear(kthread_t *t); 719extern void affinity_set(int cpu_id); 720extern void affinity_clear(void); 721extern void init_cpu_mstate(struct cpu *, int); 722extern void term_cpu_mstate(struct cpu *); 723extern void new_cpu_mstate(int, hrtime_t); 724extern void get_cpu_mstate(struct cpu *, hrtime_t *); 725extern void thread_nomigrate(void); 726extern void thread_allowmigrate(void); 727extern void weakbinding_stop(void); 728extern void weakbinding_start(void); 729 730/* 731 * The following routines affect the CPUs participation in interrupt processing, 732 * if that is applicable on the architecture. This only affects interrupts 733 * which aren't directed at the processor (not cross calls). 734 * 735 * cpu_disable_intr returns non-zero if interrupts were previously enabled. 736 */ 737int cpu_disable_intr(struct cpu *cp); /* stop issuing interrupts to cpu */ 738void cpu_enable_intr(struct cpu *cp); /* start issuing interrupts to cpu */ 739 740/* 741 * The mutex cpu_lock protects cpu_flags for all CPUs, as well as the ncpus 742 * and ncpus_online counts. 743 */ 744extern kmutex_t cpu_lock; /* lock protecting CPU data */ 745 746/* 747 * CPU state change events 748 * 749 * Various subsystems need to know when CPUs change their state. They get this 750 * information by registering CPU state change callbacks using 751 * register_cpu_setup_func(). Whenever any CPU changes its state, the callback 752 * function is called. The callback function is passed three arguments: 753 * 754 * Event, described by cpu_setup_t 755 * CPU ID 756 * Transparent pointer passed when registering the callback 757 * 758 * The callback function is called with cpu_lock held. The return value from the 759 * callback function is usually ignored, except for CPU_CONFIG and CPU_UNCONFIG 760 * events. For these two events, non-zero return value indicates a failure and 761 * prevents successful completion of the operation. 762 * 763 * New events may be added in the future. Callback functions should ignore any 764 * events that they do not understand. 765 * 766 * The following events provide notification callbacks: 767 * 768 * CPU_INIT A new CPU is started and added to the list of active CPUs 769 * This event is only used during boot 770 * 771 * CPU_CONFIG A newly inserted CPU is prepared for starting running code 772 * This event is called by DR code 773 * 774 * CPU_UNCONFIG CPU has been powered off and needs cleanup 775 * This event is called by DR code 776 * 777 * CPU_ON CPU is enabled but does not run anything yet 778 * 779 * CPU_INTR_ON CPU is enabled and has interrupts enabled 780 * 781 * CPU_OFF CPU is going offline but can still run threads 782 * 783 * CPU_CPUPART_OUT CPU is going to move out of its partition 784 * 785 * CPU_CPUPART_IN CPU is going to move to a new partition 786 * 787 * CPU_SETUP CPU is set up during boot and can run threads 788 */ 789typedef enum { 790 CPU_INIT, 791 CPU_CONFIG, 792 CPU_UNCONFIG, 793 CPU_ON, 794 CPU_OFF, 795 CPU_CPUPART_IN, 796 CPU_CPUPART_OUT, 797 CPU_SETUP, 798 CPU_INTR_ON 799} cpu_setup_t; 800 801typedef int cpu_setup_func_t(cpu_setup_t, int, void *); 802 803/* 804 * Routines used to register interest in cpu's being added to or removed 805 * from the system. 806 */ 807extern void register_cpu_setup_func(cpu_setup_func_t *, void *); 808extern void unregister_cpu_setup_func(cpu_setup_func_t *, void *); 809extern void cpu_state_change_notify(int, cpu_setup_t); 810 811/* 812 * Call specified function on the given CPU 813 */ 814typedef void (*cpu_call_func_t)(uintptr_t, uintptr_t); 815extern void cpu_call(cpu_t *, cpu_call_func_t, uintptr_t, uintptr_t); 816 817 818/* 819 * Create various strings that describe the given CPU for the 820 * processor_info system call and configuration-related kstats. 821 */ 822#define CPU_IDSTRLEN 100 823 824extern void init_cpu_info(struct cpu *); 825extern void populate_idstr(struct cpu *); 826extern void cpu_vm_data_init(struct cpu *); 827extern void cpu_vm_data_destroy(struct cpu *); 828 829#endif /* _KERNEL */ 830 831#ifdef __cplusplus 832} 833#endif 834 835#endif /* _SYS_CPUVAR_H */ 836