1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21
22 /*
23 * Copyright (c) 1989, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012 by Delphix. All rights reserved.
25 * Copyright 2014 Igor Kozhukhov <ikozhukhov@gmail.com>.
26 * Copyright 2017 RackTop Systems.
27 * Copyright 2019 Joyent, Inc.
28 */
29
30 #ifndef _SYS_CPUVAR_H
31 #define _SYS_CPUVAR_H
32
33 #include <sys/thread.h>
34 #include <sys/sysinfo.h> /* has cpu_stat_t definition */
35 #include <sys/disp.h>
36 #include <sys/processor.h>
37 #include <sys/kcpc.h> /* has kcpc_ctx_t definition */
38
39 #include <sys/loadavg.h>
40 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
41 #include <sys/machcpuvar.h>
42 #endif
43
44 #include <sys/types.h>
45 #include <sys/file.h>
46 #include <sys/bitmap.h>
47 #include <sys/rwlock.h>
48 #include <sys/msacct.h>
49 #if defined(__GNUC__) && defined(_ASM_INLINES) && defined(_KERNEL) && \
50 (defined(__i386) || defined(__amd64))
51 #include <asm/cpuvar.h>
52 #endif
53
54 #ifdef __cplusplus
55 extern "C" {
56 #endif
57
58 struct squeue_set_s;
59
60 #define CPU_CACHE_COHERENCE_SIZE 64
61
62 /*
63 * For fast event tracing.
64 */
65 struct ftrace_record;
66 typedef struct ftrace_data {
67 int ftd_state; /* ftrace flags */
68 kmutex_t ftd_unused; /* ftrace buffer lock, unused */
69 struct ftrace_record *ftd_cur; /* current record */
70 struct ftrace_record *ftd_first; /* first record */
71 struct ftrace_record *ftd_last; /* last record */
72 } ftrace_data_t;
73
74 struct cyc_cpu;
75 struct nvlist;
76
77 /*
78 * Per-CPU data.
79 *
80 * Be careful adding new members: if they are not the same in all modules (e.g.
81 * change size depending on a #define), CTF uniquification can fail to work
82 * properly. Furthermore, this is transitive in that it applies recursively to
83 * all types pointed to by cpu_t.
84 */
85 typedef struct cpu {
86 processorid_t cpu_id; /* CPU number */
87 processorid_t cpu_seqid; /* sequential CPU id (0..ncpus-1) */
88 volatile cpu_flag_t cpu_flags; /* flags indicating CPU state */
89 struct cpu *cpu_self; /* pointer to itself */
90 kthread_t *cpu_thread; /* current thread */
91 kthread_t *cpu_idle_thread; /* idle thread for this CPU */
92 kthread_t *cpu_pause_thread; /* pause thread for this CPU */
93 klwp_id_t cpu_lwp; /* current lwp (if any) */
94 klwp_id_t cpu_fpowner; /* currently loaded fpu owner */
95 struct cpupart *cpu_part; /* partition with this CPU */
96 struct lgrp_ld *cpu_lpl; /* pointer to this cpu's load */
97 int cpu_cache_offset; /* see kmem.c for details */
98
99 /*
100 * Links to other CPUs. It is safe to walk these lists if
101 * one of the following is true:
102 * - cpu_lock held
103 * - preemption disabled via kpreempt_disable
104 * - PIL >= DISP_LEVEL
105 * - acting thread is an interrupt thread
106 * - all other CPUs are paused
107 */
108 struct cpu *cpu_next; /* next existing CPU */
109 struct cpu *cpu_prev; /* prev existing CPU */
110 struct cpu *cpu_next_onln; /* next online (enabled) CPU */
111 struct cpu *cpu_prev_onln; /* prev online (enabled) CPU */
112 struct cpu *cpu_next_part; /* next CPU in partition */
113 struct cpu *cpu_prev_part; /* prev CPU in partition */
114 struct cpu *cpu_next_lgrp; /* next CPU in latency group */
115 struct cpu *cpu_prev_lgrp; /* prev CPU in latency group */
116 struct cpu *cpu_next_lpl; /* next CPU in lgrp partition */
117 struct cpu *cpu_prev_lpl;
118
119 struct cpu_pg *cpu_pg; /* cpu's processor groups */
120
121 void *cpu_reserved[4]; /* reserved for future use */
122
123 /*
124 * Scheduling variables.
125 */
126 disp_t *cpu_disp; /* dispatch queue data */
127 /*
128 * Note that cpu_disp is set before the CPU is added to the system
129 * and is never modified. Hence, no additional locking is needed
130 * beyond what's necessary to access the cpu_t structure.
131 */
132 char cpu_runrun; /* scheduling flag - set to preempt */
133 char cpu_kprunrun; /* force kernel preemption */
134 pri_t cpu_chosen_level; /* priority at which cpu */
135 /* was chosen for scheduling */
136 kthread_t *cpu_dispthread; /* thread selected for dispatch */
137 disp_lock_t cpu_thread_lock; /* dispatcher lock on current thread */
138 uint8_t cpu_disp_flags; /* flags used by dispatcher */
139 /*
140 * The following field is updated when ever the cpu_dispthread
141 * changes. Also in places, where the current thread(cpu_dispthread)
142 * priority changes. This is used in disp_lowpri_cpu()
143 */
144 pri_t cpu_dispatch_pri; /* priority of cpu_dispthread */
145 clock_t cpu_last_swtch; /* last time switched to new thread */
146
147 /*
148 * Interrupt data.
149 */
150 caddr_t cpu_intr_stack; /* interrupt stack */
151 kthread_t *cpu_intr_thread; /* interrupt thread list */
152 uint_t cpu_intr_actv; /* interrupt levels active (bitmask) */
153 int cpu_base_spl; /* priority for highest rupt active */
154
155 /*
156 * Statistics.
157 */
158 cpu_stats_t cpu_stats; /* per-CPU statistics */
159 struct kstat *cpu_info_kstat; /* kstat for cpu info */
160
161 uintptr_t cpu_profile_pc; /* kernel PC in profile interrupt */
162 uintptr_t cpu_profile_upc; /* user PC in profile interrupt */
163 uintptr_t cpu_profile_pil; /* PIL when profile interrupted */
164
165 ftrace_data_t cpu_ftrace; /* per cpu ftrace data */
166
167 clock_t cpu_deadman_counter; /* used by deadman() */
168 uint_t cpu_deadman_countdown; /* used by deadman() */
169
170 kmutex_t cpu_cpc_ctxlock; /* protects context for idle thread */
171 kcpc_ctx_t *cpu_cpc_ctx; /* performance counter context */
172
173 /*
174 * Configuration information for the processor_info system call.
175 */
176 processor_info_t cpu_type_info; /* config info */
177 time_t cpu_state_begin; /* when CPU entered current state */
178 char cpu_cpr_flags; /* CPR related info */
179 struct cyc_cpu *cpu_cyclic; /* per cpu cyclic subsystem data */
180 struct squeue_set_s *cpu_squeue_set; /* per cpu squeue set */
181 struct nvlist *cpu_props; /* pool-related properties */
182
183 krwlock_t cpu_ft_lock; /* DTrace: fasttrap lock */
184 uintptr_t cpu_dtrace_caller; /* DTrace: caller, if any */
185 hrtime_t cpu_dtrace_chillmark; /* DTrace: chill mark time */
186 hrtime_t cpu_dtrace_chilled; /* DTrace: total chill time */
187 uint64_t cpu_dtrace_probes; /* DTrace: total probes fired */
188 hrtime_t cpu_dtrace_nsec; /* DTrace: ns in dtrace_probe */
189
190 volatile uint16_t cpu_mstate; /* cpu microstate */
191 volatile uint16_t cpu_mstate_gen; /* generation counter */
192 volatile hrtime_t cpu_mstate_start; /* cpu microstate start time */
193 volatile hrtime_t cpu_acct[NCMSTATES]; /* cpu microstate data */
194 hrtime_t cpu_intracct[NCMSTATES]; /* interrupt mstate data */
195 hrtime_t cpu_waitrq; /* cpu run-queue wait time */
196 struct loadavg_s cpu_loadavg; /* loadavg info for this cpu */
197
198 char *cpu_idstr; /* for printing and debugging */
199 char *cpu_brandstr; /* for printing */
200
201 /*
202 * Sum of all device interrupt weights that are currently directed at
203 * this cpu. Cleared at start of interrupt redistribution.
204 */
205 int32_t cpu_intr_weight;
206 void *cpu_vm_data;
207
208 struct cpu_physid *cpu_physid; /* physical associations */
209
210 uint64_t cpu_curr_clock; /* current clock freq in Hz */
211 char *cpu_supp_freqs; /* supported freqs in Hz */
212
213 uintptr_t cpu_cpcprofile_pc; /* kernel PC in cpc interrupt */
214 uintptr_t cpu_cpcprofile_upc; /* user PC in cpc interrupt */
215
216 /*
217 * Interrupt load factor used by dispatcher & softcall
218 */
219 hrtime_t cpu_intrlast; /* total interrupt time (nsec) */
220 int cpu_intrload; /* interrupt load factor (0-99%) */
221
222 uint_t cpu_rotor; /* for cheap pseudo-random numbers */
223
224 struct cu_cpu_info *cpu_cu_info; /* capacity & util. info */
225
226 /*
227 * cpu_generation is updated whenever CPU goes on-line or off-line.
228 * Updates to cpu_generation are protected by cpu_lock.
229 *
230 * See CPU_NEW_GENERATION() macro below.
231 */
232 volatile uint_t cpu_generation; /* tracking on/off-line */
233
234 /*
235 * New members must be added /before/ this member, as the CTF tools
236 * rely on this being the last field before cpu_m, so they can
237 * correctly calculate the offset when synthetically adding the cpu_m
238 * member in objects that do not have it. This fixup is required for
239 * uniquification to work correctly.
240 */
241 uintptr_t cpu_m_pad;
242
243 #if (defined(_KERNEL) || defined(_KMEMUSER)) && defined(_MACHDEP)
244 struct machcpu cpu_m; /* per architecture info */
245 #endif
246 } cpu_t;
247
248 /*
249 * The cpu_core structure consists of per-CPU state available in any context.
250 * On some architectures, this may mean that the page(s) containing the
251 * NCPU-sized array of cpu_core structures must be locked in the TLB -- it
252 * is up to the platform to assure that this is performed properly. Note that
253 * the structure is sized to avoid false sharing.
254 */
255 #define CPUC_SIZE (sizeof (uint16_t) + sizeof (uint8_t) + \
256 sizeof (uintptr_t) + sizeof (kmutex_t))
257 #define CPUC_PADSIZE CPU_CACHE_COHERENCE_SIZE - CPUC_SIZE
258
259 typedef struct cpu_core {
260 uint16_t cpuc_dtrace_flags; /* DTrace flags */
261 uint8_t cpuc_dcpc_intr_state; /* DCPC provider intr state */
262 uint8_t cpuc_pad[CPUC_PADSIZE]; /* padding */
263 uintptr_t cpuc_dtrace_illval; /* DTrace illegal value */
264 kmutex_t cpuc_pid_lock; /* DTrace pid provider lock */
265 } cpu_core_t;
266
267 #ifdef _KERNEL
268 extern cpu_core_t cpu_core[];
269 #endif /* _KERNEL */
270
271 /*
272 * CPU_ON_INTR() macro. Returns non-zero if currently on interrupt stack.
273 * Note that this isn't a test for a high PIL. For example, cpu_intr_actv
274 * does not get updated when we go through sys_trap from TL>0 at high PIL.
275 * getpil() should be used instead to check for PIL levels.
276 */
277 #define CPU_ON_INTR(cpup) ((cpup)->cpu_intr_actv >> (LOCK_LEVEL + 1))
278
279 /*
280 * Check to see if an interrupt thread might be active at a given ipl.
281 * If so return true.
282 * We must be conservative--it is ok to give a false yes, but a false no
283 * will cause disaster. (But if the situation changes after we check it is
284 * ok--the caller is trying to ensure that an interrupt routine has been
285 * exited).
286 * This is used when trying to remove an interrupt handler from an autovector
287 * list in avintr.c.
288 */
289 #define INTR_ACTIVE(cpup, level) \
290 ((level) <= LOCK_LEVEL ? \
291 ((cpup)->cpu_intr_actv & (1 << (level))) : (CPU_ON_INTR(cpup)))
292
293 /*
294 * CPU_PSEUDO_RANDOM() returns a per CPU value that changes each time one
295 * looks at it. It's meant as a cheap mechanism to be incorporated in routines
296 * wanting to avoid biasing, but where true randomness isn't needed (just
297 * something that changes).
298 */
299 #define CPU_PSEUDO_RANDOM() (CPU->cpu_rotor++)
300
301 #if defined(_KERNEL) || defined(_KMEMUSER) || defined(_BOOT)
302
303 #define INTR_STACK_SIZE MAX(DEFAULTSTKSZ, PAGESIZE)
304
305 /* MEMBERS PROTECTED BY "atomicity": cpu_flags */
306
307 /*
308 * Flags in the CPU structure.
309 *
310 * These are protected by cpu_lock (except during creation).
311 *
312 * Offlined-CPUs have three stages of being offline:
313 *
314 * CPU_ENABLE indicates that the CPU is participating in I/O interrupts
315 * that can be directed at a number of different CPUs. If CPU_ENABLE
316 * is off, the CPU will not be given interrupts that can be sent elsewhere,
317 * but will still get interrupts from devices associated with that CPU only,
318 * and from other CPUs.
319 *
320 * CPU_OFFLINE indicates that the dispatcher should not allow any threads
321 * other than interrupt threads to run on that CPU. A CPU will not have
322 * CPU_OFFLINE set if there are any bound threads (besides interrupts).
323 *
324 * CPU_QUIESCED is set if p_offline was able to completely turn idle the
325 * CPU and it will not have to run interrupt threads. In this case it'll
326 * stay in the idle loop until CPU_QUIESCED is turned off.
327 *
328 * CPU_FROZEN is used only by CPR to mark CPUs that have been successfully
329 * suspended (in the suspend path), or have yet to be resumed (in the resume
330 * case).
331 *
332 * On some platforms CPUs can be individually powered off.
333 * The following flags are set for powered off CPUs: CPU_QUIESCED,
334 * CPU_OFFLINE, and CPU_POWEROFF. The following flags are cleared:
335 * CPU_RUNNING, CPU_READY, CPU_EXISTS, and CPU_ENABLE.
336 */
337 #define CPU_RUNNING 0x001 /* CPU running */
338 #define CPU_READY 0x002 /* CPU ready for cross-calls */
339 #define CPU_QUIESCED 0x004 /* CPU will stay in idle */
340 #define CPU_EXISTS 0x008 /* CPU is configured */
341 #define CPU_ENABLE 0x010 /* CPU enabled for interrupts */
342 #define CPU_OFFLINE 0x020 /* CPU offline via p_online */
343 #define CPU_POWEROFF 0x040 /* CPU is powered off */
344 #define CPU_FROZEN 0x080 /* CPU is frozen via CPR suspend */
345 #define CPU_SPARE 0x100 /* CPU offline available for use */
346 #define CPU_FAULTED 0x200 /* CPU offline diagnosed faulty */
347
348 #define FMT_CPU_FLAGS \
349 "\20\12fault\11spare\10frozen" \
350 "\7poweroff\6offline\5enable\4exist\3quiesced\2ready\1run"
351
352 #define CPU_ACTIVE(cpu) (((cpu)->cpu_flags & CPU_OFFLINE) == 0)
353
354 /*
355 * Flags for cpu_offline(), cpu_faulted(), and cpu_spare().
356 */
357 #define CPU_FORCED 0x0001 /* Force CPU offline */
358
359 /*
360 * DTrace flags.
361 */
362 #define CPU_DTRACE_NOFAULT 0x0001 /* Don't fault */
363 #define CPU_DTRACE_DROP 0x0002 /* Drop this ECB */
364 #define CPU_DTRACE_BADADDR 0x0004 /* DTrace fault: bad address */
365 #define CPU_DTRACE_BADALIGN 0x0008 /* DTrace fault: bad alignment */
366 #define CPU_DTRACE_DIVZERO 0x0010 /* DTrace fault: divide by zero */
367 #define CPU_DTRACE_ILLOP 0x0020 /* DTrace fault: illegal operation */
368 #define CPU_DTRACE_NOSCRATCH 0x0040 /* DTrace fault: out of scratch */
369 #define CPU_DTRACE_KPRIV 0x0080 /* DTrace fault: bad kernel access */
370 #define CPU_DTRACE_UPRIV 0x0100 /* DTrace fault: bad user access */
371 #define CPU_DTRACE_TUPOFLOW 0x0200 /* DTrace fault: tuple stack overflow */
372 #if defined(__sparc)
373 #define CPU_DTRACE_FAKERESTORE 0x0400 /* pid provider hint to getreg */
374 #endif
375 #define CPU_DTRACE_ENTRY 0x0800 /* pid provider hint to ustack() */
376 #define CPU_DTRACE_BADSTACK 0x1000 /* DTrace fault: bad stack */
377
378 #define CPU_DTRACE_FAULT (CPU_DTRACE_BADADDR | CPU_DTRACE_BADALIGN | \
379 CPU_DTRACE_DIVZERO | CPU_DTRACE_ILLOP | \
380 CPU_DTRACE_NOSCRATCH | CPU_DTRACE_KPRIV | \
381 CPU_DTRACE_UPRIV | CPU_DTRACE_TUPOFLOW | \
382 CPU_DTRACE_BADSTACK)
383 #define CPU_DTRACE_ERROR (CPU_DTRACE_FAULT | CPU_DTRACE_DROP)
384
385 /*
386 * Dispatcher flags
387 * These flags must be changed only by the current CPU.
388 */
389 #define CPU_DISP_DONTSTEAL 0x01 /* CPU undergoing context swtch */
390 #define CPU_DISP_HALTED 0x02 /* CPU halted waiting for interrupt */
391
392 /*
393 * Macros for manipulating sets of CPUs as a bitmap. Note that this
394 * bitmap may vary in size depending on the maximum CPU id a specific
395 * platform supports. This may be different than the number of CPUs
396 * the platform supports, since CPU ids can be sparse. We define two
397 * sets of macros; one for platforms where the maximum CPU id is less
398 * than the number of bits in a single word (32 in a 32-bit kernel,
399 * 64 in a 64-bit kernel), and one for platforms that require bitmaps
400 * of more than one word.
401 */
402
403 #define CPUSET_WORDS BT_BITOUL(NCPU)
404 #define CPUSET_NOTINSET ((uint_t)-1)
405
406 #if defined(_MACHDEP)
407 struct cpuset {
408 ulong_t cpub[CPUSET_WORDS];
409 };
410 #else
411 struct cpuset;
412 #endif
413
414 typedef struct cpuset cpuset_t;
415
416 extern cpuset_t *cpuset_alloc(int);
417 extern void cpuset_free(cpuset_t *);
418
419 /*
420 * Functions for manipulating cpusets. These were previously considered
421 * private when some cpuset_t handling was performed in the CPUSET_* macros.
422 * They are now acceptable to use in non-_MACHDEP code.
423 */
424 extern void cpuset_all(cpuset_t *);
425 extern void cpuset_all_but(cpuset_t *, const uint_t);
426 extern int cpuset_isnull(const cpuset_t *);
427 extern int cpuset_isequal(const cpuset_t *, const cpuset_t *);
428 extern void cpuset_only(cpuset_t *, const uint_t);
429 extern long cpu_in_set(const cpuset_t *, const uint_t);
430 extern void cpuset_add(cpuset_t *, const uint_t);
431 extern void cpuset_del(cpuset_t *, const uint_t);
432 extern uint_t cpuset_find(const cpuset_t *);
433 extern void cpuset_bounds(const cpuset_t *, uint_t *, uint_t *);
434 extern void cpuset_atomic_del(cpuset_t *, const uint_t);
435 extern void cpuset_atomic_add(cpuset_t *, const uint_t);
436 extern long cpuset_atomic_xadd(cpuset_t *, const uint_t);
437 extern long cpuset_atomic_xdel(cpuset_t *, const uint_t);
438 extern void cpuset_or(cpuset_t *, cpuset_t *);
439 extern void cpuset_xor(cpuset_t *, cpuset_t *);
440 extern void cpuset_and(cpuset_t *, cpuset_t *);
441 extern void cpuset_zero(cpuset_t *);
442
443
444 #if defined(_MACHDEP)
445
446 /*
447 * Prior to the cpuset_t restructuring, the CPUSET_* macros contained
448 * significant logic, rather than directly invoking the backend functions.
449 * They are maintained here so that existing _MACHDEP code can use them.
450 */
451
452 #define CPUSET_ALL(set) cpuset_all(&(set))
453 #define CPUSET_ALL_BUT(set, cpu) cpuset_all_but(&(set), cpu)
454 #define CPUSET_ONLY(set, cpu) cpuset_only(&(set), cpu)
455 #define CPU_IN_SET(set, cpu) cpu_in_set(&(set), cpu)
456 #define CPUSET_ADD(set, cpu) cpuset_add(&(set), cpu)
457 #define CPUSET_DEL(set, cpu) cpuset_del(&(set), cpu)
458 #define CPUSET_ISNULL(set) cpuset_isnull(&(set))
459 #define CPUSET_ISEQUAL(set1, set2) cpuset_isequal(&(set1), &(set2))
460
461 /*
462 * Find one CPU in the cpuset.
463 * Sets "cpu" to the id of the found CPU, or CPUSET_NOTINSET if no cpu
464 * could be found. (i.e. empty set)
465 */
466 #define CPUSET_FIND(set, cpu) { \
467 cpu = cpuset_find(&(set)); \
468 }
469
470 /*
471 * Determine the smallest and largest CPU id in the set. Returns
472 * CPUSET_NOTINSET in smallest and largest when set is empty.
473 */
474 #define CPUSET_BOUNDS(set, smallest, largest) { \
475 cpuset_bounds(&(set), &(smallest), &(largest)); \
476 }
477
478 /*
479 * Atomic cpuset operations
480 * These are safe to use for concurrent cpuset manipulations.
481 * "xdel" and "xadd" are exclusive operations, that set "result" to "0"
482 * if the add or del was successful, or "-1" if not successful.
483 * (e.g. attempting to add a cpu to a cpuset that's already there, or
484 * deleting a cpu that's not in the cpuset)
485 */
486
487 #define CPUSET_ATOMIC_DEL(set, cpu) cpuset_atomic_del(&(set), cpu)
488 #define CPUSET_ATOMIC_ADD(set, cpu) cpuset_atomic_add(&(set), cpu)
489
490 #define CPUSET_ATOMIC_XADD(set, cpu, result) \
491 (result) = cpuset_atomic_xadd(&(set), cpu)
492
493 #define CPUSET_ATOMIC_XDEL(set, cpu, result) \
494 (result) = cpuset_atomic_xdel(&(set), cpu)
495
496 #define CPUSET_OR(set1, set2) cpuset_or(&(set1), &(set2))
497
498 #define CPUSET_XOR(set1, set2) cpuset_xor(&(set1), &(set2))
499
500 #define CPUSET_AND(set1, set2) cpuset_and(&(set1), &(set2))
501
502 #define CPUSET_ZERO(set) cpuset_zero(&(set))
503
504 #endif /* defined(_MACHDEP) */
505
506
507 extern cpuset_t cpu_seqid_inuse;
508
509 extern struct cpu *cpu[]; /* indexed by CPU number */
510 extern struct cpu **cpu_seq; /* indexed by sequential CPU id */
511 extern cpu_t *cpu_list; /* list of CPUs */
512 extern cpu_t *cpu_active; /* list of active CPUs */
513 extern cpuset_t cpu_active_set; /* cached set of active CPUs */
514 extern int ncpus; /* number of CPUs present */
515 extern int ncpus_online; /* number of CPUs not quiesced */
516 extern int max_ncpus; /* max present before ncpus is known */
517 extern int boot_max_ncpus; /* like max_ncpus but for real */
518 extern int boot_ncpus; /* # cpus present @ boot */
519 extern processorid_t max_cpuid; /* maximum CPU number */
520 extern struct cpu *cpu_inmotion; /* offline or partition move target */
521 extern cpu_t *clock_cpu_list;
522 extern processorid_t max_cpu_seqid_ever; /* maximum seqid ever given */
523
524 #if defined(__i386) || defined(__amd64)
525 extern struct cpu *curcpup(void);
526 #define CPU (curcpup()) /* Pointer to current CPU */
527 #else
528 #define CPU (curthread->t_cpu) /* Pointer to current CPU */
529 #endif
530
531 /*
532 * CPU_CURRENT indicates to thread_affinity_set to use CPU->cpu_id
533 * as the target and to grab cpu_lock instead of requiring the caller
534 * to grab it.
535 */
536 #define CPU_CURRENT -3
537
538 /*
539 * Per-CPU statistics
540 *
541 * cpu_stats_t contains numerous system and VM-related statistics, in the form
542 * of gauges or monotonically-increasing event occurrence counts.
543 */
544
545 #define CPU_STATS_ENTER_K() kpreempt_disable()
546 #define CPU_STATS_EXIT_K() kpreempt_enable()
547
548 #define CPU_STATS_ADD_K(class, stat, amount) \
549 { kpreempt_disable(); /* keep from switching CPUs */\
550 CPU_STATS_ADDQ(CPU, class, stat, amount); \
551 kpreempt_enable(); \
552 }
553
554 #define CPU_STATS_ADDQ(cp, class, stat, amount) { \
555 extern void __dtrace_probe___cpu_##class##info_##stat(uint_t, \
556 uint64_t *, cpu_t *); \
557 uint64_t *stataddr = &((cp)->cpu_stats.class.stat); \
558 __dtrace_probe___cpu_##class##info_##stat((amount), \
559 stataddr, cp); \
560 *(stataddr) += (amount); \
561 }
562
563 #define CPU_STATS(cp, stat) \
564 ((cp)->cpu_stats.stat)
565
566 /*
567 * Increment CPU generation value.
568 * This macro should be called whenever CPU goes on-line or off-line.
569 * Updates to cpu_generation should be protected by cpu_lock.
570 */
571 #define CPU_NEW_GENERATION(cp) ((cp)->cpu_generation++)
572
573 #endif /* defined(_KERNEL) || defined(_KMEMUSER) */
574
575 #define CPU_CPR_OFFLINE 0x0
576 #define CPU_CPR_ONLINE 0x1
577 #define CPU_CPR_IS_OFFLINE(cpu) (((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE) == 0)
578 #define CPU_CPR_IS_ONLINE(cpu) ((cpu)->cpu_cpr_flags & CPU_CPR_ONLINE)
579 #define CPU_SET_CPR_FLAGS(cpu, flag) ((cpu)->cpu_cpr_flags |= flag)
580
581 /*
582 * CPU support routines (not for genassym.c)
583 */
584 #if (defined(_KERNEL) || defined(_FAKE_KERNEL)) && defined(__STDC__)
585
586 struct zone;
587
588 void cpu_list_init(cpu_t *);
589 void cpu_add_unit(cpu_t *);
590 void cpu_del_unit(int cpuid);
591 void cpu_add_active(cpu_t *);
592 void cpu_kstat_init(cpu_t *);
593 void cpu_visibility_add(cpu_t *, struct zone *);
594 void cpu_visibility_remove(cpu_t *, struct zone *);
595 void cpu_visibility_configure(cpu_t *, struct zone *);
596 void cpu_visibility_unconfigure(cpu_t *, struct zone *);
597 void cpu_visibility_online(cpu_t *, struct zone *);
598 void cpu_visibility_offline(cpu_t *, struct zone *);
599 void cpu_create_intrstat(cpu_t *);
600 void cpu_delete_intrstat(cpu_t *);
601 int cpu_kstat_intrstat_update(kstat_t *, int);
602 void cpu_intr_swtch_enter(kthread_t *);
603 void cpu_intr_swtch_exit(kthread_t *);
604
605 void mbox_lock_init(void); /* initialize cross-call locks */
606 void mbox_init(int cpun); /* initialize cross-calls */
607 void poke_cpu(int cpun); /* interrupt another CPU (to preempt) */
608
609 /*
610 * values for safe_list. Pause state that CPUs are in.
611 */
612 #define PAUSE_IDLE 0 /* normal state */
613 #define PAUSE_READY 1 /* paused thread ready to spl */
614 #define PAUSE_WAIT 2 /* paused thread is spl-ed high */
615 #define PAUSE_DIE 3 /* tell pause thread to leave */
616 #define PAUSE_DEAD 4 /* pause thread has left */
617
618 void mach_cpu_pause(volatile char *);
619
620 void pause_cpus(cpu_t *off_cp, void *(*func)(void *));
621 void start_cpus(void);
622 int cpus_paused(void);
623
624 void cpu_pause_init(void);
625 cpu_t *cpu_get(processorid_t cpun); /* get the CPU struct associated */
626
627 int cpu_online(cpu_t *cp); /* take cpu online */
628 int cpu_offline(cpu_t *cp, int flags); /* take cpu offline */
629 int cpu_spare(cpu_t *cp, int flags); /* take cpu to spare */
630 int cpu_faulted(cpu_t *cp, int flags); /* take cpu to faulted */
631 int cpu_poweron(cpu_t *cp); /* take powered-off cpu to offline */
632 int cpu_poweroff(cpu_t *cp); /* take offline cpu to powered-off */
633
634 cpu_t *cpu_intr_next(cpu_t *cp); /* get next online CPU taking intrs */
635 int cpu_intr_count(cpu_t *cp); /* count # of CPUs handling intrs */
636 int cpu_intr_on(cpu_t *cp); /* CPU taking I/O interrupts? */
637 void cpu_intr_enable(cpu_t *cp); /* enable I/O interrupts */
638 int cpu_intr_disable(cpu_t *cp); /* disable I/O interrupts */
639 void cpu_intr_alloc(cpu_t *cp, int n); /* allocate interrupt threads */
640
641 /*
642 * Routines for checking CPU states.
643 */
644 int cpu_is_online(cpu_t *); /* check if CPU is online */
645 int cpu_is_nointr(cpu_t *); /* check if CPU can service intrs */
646 int cpu_is_active(cpu_t *); /* check if CPU can run threads */
647 int cpu_is_offline(cpu_t *); /* check if CPU is offline */
648 int cpu_is_poweredoff(cpu_t *); /* check if CPU is powered off */
649
650 int cpu_flagged_online(cpu_flag_t); /* flags show CPU is online */
651 int cpu_flagged_nointr(cpu_flag_t); /* flags show CPU not handling intrs */
652 int cpu_flagged_active(cpu_flag_t); /* flags show CPU scheduling threads */
653 int cpu_flagged_offline(cpu_flag_t); /* flags show CPU is offline */
654 int cpu_flagged_poweredoff(cpu_flag_t); /* flags show CPU is powered off */
655
656 /*
657 * The processor_info(2) state of a CPU is a simplified representation suitable
658 * for use by an application program. Kernel subsystems should utilize the
659 * internal per-CPU state as given by the cpu_flags member of the cpu structure,
660 * as this information may include platform- or architecture-specific state
661 * critical to a subsystem's disposition of a particular CPU.
662 */
663 void cpu_set_state(cpu_t *); /* record/timestamp current state */
664 int cpu_get_state(cpu_t *); /* get current cpu state */
665 const char *cpu_get_state_str(cpu_t *); /* get current cpu state as string */
666
667
668 void cpu_set_curr_clock(uint64_t); /* indicate the current CPU's freq */
669 void cpu_set_supp_freqs(cpu_t *, const char *); /* set the CPU supported */
670 /* frequencies */
671
672 int cpu_configure(int);
673 int cpu_unconfigure(int);
674 void cpu_destroy_bound_threads(cpu_t *cp);
675
676 extern int cpu_bind_thread(kthread_t *tp, processorid_t bind,
677 processorid_t *obind, int *error);
678 extern int cpu_unbind(processorid_t cpu_id, boolean_t force);
679 extern void thread_affinity_set(kthread_t *t, int cpu_id);
680 extern void thread_affinity_clear(kthread_t *t);
681 extern void affinity_set(int cpu_id);
682 extern void affinity_clear(void);
683 extern void init_cpu_mstate(struct cpu *, int);
684 extern void term_cpu_mstate(struct cpu *);
685 extern void new_cpu_mstate(int, hrtime_t);
686 extern void get_cpu_mstate(struct cpu *, hrtime_t *);
687 extern void thread_nomigrate(void);
688 extern void thread_allowmigrate(void);
689 extern void weakbinding_stop(void);
690 extern void weakbinding_start(void);
691
692 /*
693 * The following routines affect the CPUs participation in interrupt processing,
694 * if that is applicable on the architecture. This only affects interrupts
695 * which aren't directed at the processor (not cross calls).
696 *
697 * cpu_disable_intr returns non-zero if interrupts were previously enabled.
698 */
699 int cpu_disable_intr(struct cpu *cp); /* stop issuing interrupts to cpu */
700 void cpu_enable_intr(struct cpu *cp); /* start issuing interrupts to cpu */
701
702 /*
703 * The mutex cpu_lock protects cpu_flags for all CPUs, as well as the ncpus
704 * and ncpus_online counts.
705 */
706 extern kmutex_t cpu_lock; /* lock protecting CPU data */
707
708 /*
709 * CPU state change events
710 *
711 * Various subsystems need to know when CPUs change their state. They get this
712 * information by registering CPU state change callbacks using
713 * register_cpu_setup_func(). Whenever any CPU changes its state, the callback
714 * function is called. The callback function is passed three arguments:
715 *
716 * Event, described by cpu_setup_t
717 * CPU ID
718 * Transparent pointer passed when registering the callback
719 *
720 * The callback function is called with cpu_lock held. The return value from the
721 * callback function is usually ignored, except for CPU_CONFIG and CPU_UNCONFIG
722 * events. For these two events, non-zero return value indicates a failure and
723 * prevents successful completion of the operation.
724 *
725 * New events may be added in the future. Callback functions should ignore any
726 * events that they do not understand.
727 *
728 * The following events provide notification callbacks:
729 *
730 * CPU_INIT A new CPU is started and added to the list of active CPUs
731 * This event is only used during boot
732 *
733 * CPU_CONFIG A newly inserted CPU is prepared for starting running code
734 * This event is called by DR code
735 *
736 * CPU_UNCONFIG CPU has been powered off and needs cleanup
737 * This event is called by DR code
738 *
739 * CPU_ON CPU is enabled but does not run anything yet
740 *
741 * CPU_INTR_ON CPU is enabled and has interrupts enabled
742 *
743 * CPU_OFF CPU is going offline but can still run threads
744 *
745 * CPU_CPUPART_OUT CPU is going to move out of its partition
746 *
747 * CPU_CPUPART_IN CPU is going to move to a new partition
748 *
749 * CPU_SETUP CPU is set up during boot and can run threads
750 */
751 typedef enum {
752 CPU_INIT,
753 CPU_CONFIG,
754 CPU_UNCONFIG,
755 CPU_ON,
756 CPU_OFF,
757 CPU_CPUPART_IN,
758 CPU_CPUPART_OUT,
759 CPU_SETUP,
760 CPU_INTR_ON
761 } cpu_setup_t;
762
763 typedef int cpu_setup_func_t(cpu_setup_t, int, void *);
764
765 /*
766 * Routines used to register interest in cpu's being added to or removed
767 * from the system.
768 */
769 extern void register_cpu_setup_func(cpu_setup_func_t *, void *);
770 extern void unregister_cpu_setup_func(cpu_setup_func_t *, void *);
771 extern void cpu_state_change_notify(int, cpu_setup_t);
772
773 /*
774 * Call specified function on the given CPU
775 */
776 typedef void (*cpu_call_func_t)(uintptr_t, uintptr_t);
777 extern void cpu_call(cpu_t *, cpu_call_func_t, uintptr_t, uintptr_t);
778
779
780 /*
781 * Create various strings that describe the given CPU for the
782 * processor_info system call and configuration-related kstats.
783 */
784 #define CPU_IDSTRLEN 100
785
786 extern void init_cpu_info(struct cpu *);
787 extern void populate_idstr(struct cpu *);
788 extern void cpu_vm_data_init(struct cpu *);
789 extern void cpu_vm_data_destroy(struct cpu *);
790
791 #endif /* _KERNEL || _FAKE_KERNEL */
792
793 #ifdef __cplusplus
794 }
795 #endif
796
797 #endif /* _SYS_CPUVAR_H */