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