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