Merge branch 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-block.git] / kernel / smp.c
CommitLineData
3d442233
JA
1/*
2 * Generic helpers for smp ipi calls
3 *
4 * (C) Jens Axboe <jens.axboe@oracle.com> 2008
3d442233 5 */
ca7dfdbb
ME
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
47885016 9#include <linux/irq_work.h>
3d442233 10#include <linux/rcupdate.h>
59190f42 11#include <linux/rculist.h>
641cd4cf 12#include <linux/kernel.h>
9984de1a 13#include <linux/export.h>
0b13fda1
IM
14#include <linux/percpu.h>
15#include <linux/init.h>
5a0e3ad6 16#include <linux/gfp.h>
3d442233 17#include <linux/smp.h>
8969a5ed 18#include <linux/cpu.h>
c6f4459f 19#include <linux/sched.h>
4c822698 20#include <linux/sched/idle.h>
47ae4b05 21#include <linux/hypervisor.h>
3d442233 22
3bb5d2ee
SS
23#include "smpboot.h"
24
3d442233 25enum {
6e275637 26 CSD_FLAG_LOCK = 0x01,
8053871d 27 CSD_FLAG_SYNCHRONOUS = 0x02,
3d442233
JA
28};
29
30struct call_function_data {
966a9671 31 call_single_data_t __percpu *csd;
0b13fda1 32 cpumask_var_t cpumask;
3fc5b3b6 33 cpumask_var_t cpumask_ipi;
3d442233
JA
34};
35
e03bcb68
MM
36static DEFINE_PER_CPU_SHARED_ALIGNED(struct call_function_data, cfd_data);
37
6897fc22 38static DEFINE_PER_CPU_SHARED_ALIGNED(struct llist_head, call_single_queue);
8969a5ed 39
8d056c48
SB
40static void flush_smp_call_function_queue(bool warn_cpu_offline);
41
31487f83 42int smpcfd_prepare_cpu(unsigned int cpu)
8969a5ed 43{
8969a5ed
PZ
44 struct call_function_data *cfd = &per_cpu(cfd_data, cpu);
45
31487f83
RW
46 if (!zalloc_cpumask_var_node(&cfd->cpumask, GFP_KERNEL,
47 cpu_to_node(cpu)))
48 return -ENOMEM;
3fc5b3b6
AL
49 if (!zalloc_cpumask_var_node(&cfd->cpumask_ipi, GFP_KERNEL,
50 cpu_to_node(cpu))) {
51 free_cpumask_var(cfd->cpumask);
52 return -ENOMEM;
53 }
966a9671 54 cfd->csd = alloc_percpu(call_single_data_t);
31487f83 55 if (!cfd->csd) {
8969a5ed 56 free_cpumask_var(cfd->cpumask);
3fc5b3b6 57 free_cpumask_var(cfd->cpumask_ipi);
31487f83
RW
58 return -ENOMEM;
59 }
60
61 return 0;
8969a5ed
PZ
62}
63
31487f83
RW
64int smpcfd_dead_cpu(unsigned int cpu)
65{
66 struct call_function_data *cfd = &per_cpu(cfd_data, cpu);
67
68 free_cpumask_var(cfd->cpumask);
3fc5b3b6 69 free_cpumask_var(cfd->cpumask_ipi);
31487f83
RW
70 free_percpu(cfd->csd);
71 return 0;
72}
73
74int smpcfd_dying_cpu(unsigned int cpu)
75{
76 /*
77 * The IPIs for the smp-call-function callbacks queued by other
78 * CPUs might arrive late, either due to hardware latencies or
79 * because this CPU disabled interrupts (inside stop-machine)
80 * before the IPIs were sent. So flush out any pending callbacks
81 * explicitly (without waiting for the IPIs to arrive), to
82 * ensure that the outgoing CPU doesn't go offline with work
83 * still pending.
84 */
85 flush_smp_call_function_queue(false);
86 return 0;
87}
8969a5ed 88
d8ad7d11 89void __init call_function_init(void)
3d442233
JA
90{
91 int i;
92
6897fc22
CH
93 for_each_possible_cpu(i)
94 init_llist_head(&per_cpu(call_single_queue, i));
8969a5ed 95
31487f83 96 smpcfd_prepare_cpu(smp_processor_id());
3d442233
JA
97}
98
8969a5ed
PZ
99/*
100 * csd_lock/csd_unlock used to serialize access to per-cpu csd resources
101 *
0b13fda1
IM
102 * For non-synchronous ipi calls the csd can still be in use by the
103 * previous function call. For multi-cpu calls its even more interesting
104 * as we'll have to ensure no other cpu is observing our csd.
8969a5ed 105 */
966a9671 106static __always_inline void csd_lock_wait(call_single_data_t *csd)
8969a5ed 107{
1f03e8d2 108 smp_cond_load_acquire(&csd->flags, !(VAL & CSD_FLAG_LOCK));
6e275637
PZ
109}
110
966a9671 111static __always_inline void csd_lock(call_single_data_t *csd)
6e275637 112{
e1d12f32
AM
113 csd_lock_wait(csd);
114 csd->flags |= CSD_FLAG_LOCK;
8969a5ed
PZ
115
116 /*
0b13fda1
IM
117 * prevent CPU from reordering the above assignment
118 * to ->flags with any subsequent assignments to other
966a9671 119 * fields of the specified call_single_data_t structure:
8969a5ed 120 */
8053871d 121 smp_wmb();
8969a5ed
PZ
122}
123
966a9671 124static __always_inline void csd_unlock(call_single_data_t *csd)
8969a5ed 125{
8053871d 126 WARN_ON(!(csd->flags & CSD_FLAG_LOCK));
0b13fda1 127
8969a5ed 128 /*
0b13fda1 129 * ensure we're all done before releasing data:
8969a5ed 130 */
8053871d 131 smp_store_release(&csd->flags, 0);
3d442233
JA
132}
133
966a9671 134static DEFINE_PER_CPU_SHARED_ALIGNED(call_single_data_t, csd_data);
8b28499a 135
3d442233 136/*
966a9671 137 * Insert a previously allocated call_single_data_t element
0b13fda1
IM
138 * for execution on the given CPU. data must already have
139 * ->func, ->info, and ->flags set.
3d442233 140 */
966a9671 141static int generic_exec_single(int cpu, call_single_data_t *csd,
8053871d 142 smp_call_func_t func, void *info)
3d442233 143{
8b28499a 144 if (cpu == smp_processor_id()) {
8053871d
LT
145 unsigned long flags;
146
147 /*
148 * We can unlock early even for the synchronous on-stack case,
149 * since we're doing this from the same CPU..
150 */
151 csd_unlock(csd);
8b28499a
FW
152 local_irq_save(flags);
153 func(info);
154 local_irq_restore(flags);
155 return 0;
156 }
157
158
5224b961
LT
159 if ((unsigned)cpu >= nr_cpu_ids || !cpu_online(cpu)) {
160 csd_unlock(csd);
8b28499a 161 return -ENXIO;
5224b961 162 }
8b28499a 163
8b28499a
FW
164 csd->func = func;
165 csd->info = info;
166
561920a0 167 /*
15d0d3b3
NP
168 * The list addition should be visible before sending the IPI
169 * handler locks the list to pull the entry off it because of
170 * normal cache coherency rules implied by spinlocks.
171 *
172 * If IPIs can go out of order to the cache coherency protocol
173 * in an architecture, sufficient synchronisation should be added
174 * to arch code to make it appear to obey cache coherency WRT
0b13fda1
IM
175 * locking and barrier primitives. Generic code isn't really
176 * equipped to do the right thing...
561920a0 177 */
6897fc22 178 if (llist_add(&csd->llist, &per_cpu(call_single_queue, cpu)))
3d442233
JA
179 arch_send_call_function_single_ipi(cpu);
180
8b28499a 181 return 0;
3d442233
JA
182}
183
8d056c48
SB
184/**
185 * generic_smp_call_function_single_interrupt - Execute SMP IPI callbacks
186 *
187 * Invoked by arch to handle an IPI for call function single.
188 * Must be called with interrupts disabled.
3d442233
JA
189 */
190void generic_smp_call_function_single_interrupt(void)
191{
8d056c48
SB
192 flush_smp_call_function_queue(true);
193}
194
195/**
196 * flush_smp_call_function_queue - Flush pending smp-call-function callbacks
197 *
198 * @warn_cpu_offline: If set to 'true', warn if callbacks were queued on an
199 * offline CPU. Skip this check if set to 'false'.
200 *
201 * Flush any pending smp-call-function callbacks queued on this CPU. This is
202 * invoked by the generic IPI handler, as well as by a CPU about to go offline,
203 * to ensure that all pending IPI callbacks are run before it goes completely
204 * offline.
205 *
206 * Loop through the call_single_queue and run all the queued callbacks.
207 * Must be called with interrupts disabled.
208 */
209static void flush_smp_call_function_queue(bool warn_cpu_offline)
210{
211 struct llist_head *head;
5fd77595 212 struct llist_node *entry;
966a9671 213 call_single_data_t *csd, *csd_next;
a219ccf4
SB
214 static bool warned;
215
83efcbd0 216 lockdep_assert_irqs_disabled();
8d056c48 217
bb964a92 218 head = this_cpu_ptr(&call_single_queue);
8d056c48 219 entry = llist_del_all(head);
a219ccf4 220 entry = llist_reverse_order(entry);
3d442233 221
8d056c48
SB
222 /* There shouldn't be any pending callbacks on an offline CPU. */
223 if (unlikely(warn_cpu_offline && !cpu_online(smp_processor_id()) &&
224 !warned && !llist_empty(head))) {
a219ccf4
SB
225 warned = true;
226 WARN(1, "IPI on offline CPU %d\n", smp_processor_id());
227
228 /*
229 * We don't have to use the _safe() variant here
230 * because we are not invoking the IPI handlers yet.
231 */
232 llist_for_each_entry(csd, entry, llist)
233 pr_warn("IPI callback %pS sent to offline CPU\n",
234 csd->func);
235 }
3d442233 236
5fd77595 237 llist_for_each_entry_safe(csd, csd_next, entry, llist) {
8053871d
LT
238 smp_call_func_t func = csd->func;
239 void *info = csd->info;
240
241 /* Do we wait until *after* callback? */
242 if (csd->flags & CSD_FLAG_SYNCHRONOUS) {
243 func(info);
244 csd_unlock(csd);
245 } else {
246 csd_unlock(csd);
247 func(info);
248 }
3d442233 249 }
47885016
FW
250
251 /*
252 * Handle irq works queued remotely by irq_work_queue_on().
253 * Smp functions above are typically synchronous so they
254 * better run first since some other CPUs may be busy waiting
255 * for them.
256 */
257 irq_work_run();
3d442233
JA
258}
259
260/*
261 * smp_call_function_single - Run a function on a specific CPU
262 * @func: The function to run. This must be fast and non-blocking.
263 * @info: An arbitrary pointer to pass to the function.
3d442233
JA
264 * @wait: If true, wait until function has completed on other CPUs.
265 *
72f279b2 266 * Returns 0 on success, else a negative status code.
3d442233 267 */
3a5f65df 268int smp_call_function_single(int cpu, smp_call_func_t func, void *info,
8691e5a8 269 int wait)
3d442233 270{
966a9671
YH
271 call_single_data_t *csd;
272 call_single_data_t csd_stack = {
273 .flags = CSD_FLAG_LOCK | CSD_FLAG_SYNCHRONOUS,
274 };
0b13fda1 275 int this_cpu;
8b28499a 276 int err;
3d442233 277
0b13fda1
IM
278 /*
279 * prevent preemption and reschedule on another processor,
280 * as well as CPU removal
281 */
282 this_cpu = get_cpu();
283
269c861b
SS
284 /*
285 * Can deadlock when called with interrupts disabled.
286 * We allow cpu's that are not yet online though, as no one else can
287 * send smp call function interrupt to this cpu and as such deadlocks
288 * can't happen.
289 */
290 WARN_ON_ONCE(cpu_online(this_cpu) && irqs_disabled()
291 && !oops_in_progress);
3d442233 292
8053871d
LT
293 csd = &csd_stack;
294 if (!wait) {
295 csd = this_cpu_ptr(&csd_data);
296 csd_lock(csd);
297 }
298
299 err = generic_exec_single(cpu, csd, func, info);
300
301 if (wait)
302 csd_lock_wait(csd);
3d442233
JA
303
304 put_cpu();
0b13fda1 305
f73be6de 306 return err;
3d442233
JA
307}
308EXPORT_SYMBOL(smp_call_function_single);
309
d7877c03 310/**
c46fff2a
FW
311 * smp_call_function_single_async(): Run an asynchronous function on a
312 * specific CPU.
d7877c03
FW
313 * @cpu: The CPU to run on.
314 * @csd: Pre-allocated and setup data structure
d7877c03 315 *
c46fff2a
FW
316 * Like smp_call_function_single(), but the call is asynchonous and
317 * can thus be done from contexts with disabled interrupts.
318 *
319 * The caller passes his own pre-allocated data structure
320 * (ie: embedded in an object) and is responsible for synchronizing it
321 * such that the IPIs performed on the @csd are strictly serialized.
322 *
323 * NOTE: Be careful, there is unfortunately no current debugging facility to
324 * validate the correctness of this serialization.
d7877c03 325 */
966a9671 326int smp_call_function_single_async(int cpu, call_single_data_t *csd)
d7877c03
FW
327{
328 int err = 0;
d7877c03 329
fce8ad15 330 preempt_disable();
8053871d
LT
331
332 /* We could deadlock if we have to wait here with interrupts disabled! */
333 if (WARN_ON_ONCE(csd->flags & CSD_FLAG_LOCK))
334 csd_lock_wait(csd);
335
336 csd->flags = CSD_FLAG_LOCK;
337 smp_wmb();
338
339 err = generic_exec_single(cpu, csd, csd->func, csd->info);
fce8ad15 340 preempt_enable();
d7877c03
FW
341
342 return err;
343}
c46fff2a 344EXPORT_SYMBOL_GPL(smp_call_function_single_async);
d7877c03 345
2ea6dec4
RR
346/*
347 * smp_call_function_any - Run a function on any of the given cpus
348 * @mask: The mask of cpus it can run on.
349 * @func: The function to run. This must be fast and non-blocking.
350 * @info: An arbitrary pointer to pass to the function.
351 * @wait: If true, wait until function has completed.
352 *
353 * Returns 0 on success, else a negative status code (if no cpus were online).
2ea6dec4
RR
354 *
355 * Selection preference:
356 * 1) current cpu if in @mask
357 * 2) any cpu of current node if in @mask
358 * 3) any other online cpu in @mask
359 */
360int smp_call_function_any(const struct cpumask *mask,
3a5f65df 361 smp_call_func_t func, void *info, int wait)
2ea6dec4
RR
362{
363 unsigned int cpu;
364 const struct cpumask *nodemask;
365 int ret;
366
367 /* Try for same CPU (cheapest) */
368 cpu = get_cpu();
369 if (cpumask_test_cpu(cpu, mask))
370 goto call;
371
372 /* Try for same node. */
af2422c4 373 nodemask = cpumask_of_node(cpu_to_node(cpu));
2ea6dec4
RR
374 for (cpu = cpumask_first_and(nodemask, mask); cpu < nr_cpu_ids;
375 cpu = cpumask_next_and(cpu, nodemask, mask)) {
376 if (cpu_online(cpu))
377 goto call;
378 }
379
380 /* Any online will do: smp_call_function_single handles nr_cpu_ids. */
381 cpu = cpumask_any_and(mask, cpu_online_mask);
382call:
383 ret = smp_call_function_single(cpu, func, info, wait);
384 put_cpu();
385 return ret;
386}
387EXPORT_SYMBOL_GPL(smp_call_function_any);
388
3d442233 389/**
54b11e6d
RR
390 * smp_call_function_many(): Run a function on a set of other CPUs.
391 * @mask: The set of cpus to run on (only runs on online subset).
3d442233
JA
392 * @func: The function to run. This must be fast and non-blocking.
393 * @info: An arbitrary pointer to pass to the function.
0b13fda1
IM
394 * @wait: If true, wait (atomically) until function has completed
395 * on other CPUs.
3d442233 396 *
72f279b2 397 * If @wait is true, then returns once @func has returned.
3d442233
JA
398 *
399 * You must not call this function with disabled interrupts or from a
400 * hardware interrupt handler or from a bottom half handler. Preemption
401 * must be disabled when calling this function.
402 */
54b11e6d 403void smp_call_function_many(const struct cpumask *mask,
3a5f65df 404 smp_call_func_t func, void *info, bool wait)
3d442233 405{
e1d12f32 406 struct call_function_data *cfd;
9a46ad6d 407 int cpu, next_cpu, this_cpu = smp_processor_id();
3d442233 408
269c861b
SS
409 /*
410 * Can deadlock when called with interrupts disabled.
411 * We allow cpu's that are not yet online though, as no one else can
412 * send smp call function interrupt to this cpu and as such deadlocks
413 * can't happen.
414 */
415 WARN_ON_ONCE(cpu_online(this_cpu) && irqs_disabled()
bd924e8c 416 && !oops_in_progress && !early_boot_irqs_disabled);
3d442233 417
723aae25 418 /* Try to fastpath. So, what's a CPU they want? Ignoring this one. */
54b11e6d 419 cpu = cpumask_first_and(mask, cpu_online_mask);
0b13fda1 420 if (cpu == this_cpu)
54b11e6d 421 cpu = cpumask_next_and(cpu, mask, cpu_online_mask);
0b13fda1 422
54b11e6d
RR
423 /* No online cpus? We're done. */
424 if (cpu >= nr_cpu_ids)
425 return;
426
427 /* Do we have another CPU which isn't us? */
428 next_cpu = cpumask_next_and(cpu, mask, cpu_online_mask);
0b13fda1 429 if (next_cpu == this_cpu)
54b11e6d
RR
430 next_cpu = cpumask_next_and(next_cpu, mask, cpu_online_mask);
431
432 /* Fastpath: do that cpu by itself. */
433 if (next_cpu >= nr_cpu_ids) {
434 smp_call_function_single(cpu, func, info, wait);
435 return;
3d442233
JA
436 }
437
bb964a92 438 cfd = this_cpu_ptr(&cfd_data);
45a57919 439
e1d12f32 440 cpumask_and(cfd->cpumask, mask, cpu_online_mask);
6c8557bd 441 __cpumask_clear_cpu(this_cpu, cfd->cpumask);
723aae25
MM
442
443 /* Some callers race with other cpus changing the passed mask */
e1d12f32 444 if (unlikely(!cpumask_weight(cfd->cpumask)))
723aae25 445 return;
3d442233 446
3fc5b3b6 447 cpumask_clear(cfd->cpumask_ipi);
e1d12f32 448 for_each_cpu(cpu, cfd->cpumask) {
966a9671 449 call_single_data_t *csd = per_cpu_ptr(cfd->csd, cpu);
9a46ad6d
SL
450
451 csd_lock(csd);
8053871d
LT
452 if (wait)
453 csd->flags |= CSD_FLAG_SYNCHRONOUS;
9a46ad6d
SL
454 csd->func = func;
455 csd->info = info;
3fc5b3b6 456 if (llist_add(&csd->llist, &per_cpu(call_single_queue, cpu)))
6c8557bd 457 __cpumask_set_cpu(cpu, cfd->cpumask_ipi);
9a46ad6d 458 }
561920a0 459
3d442233 460 /* Send a message to all CPUs in the map */
3fc5b3b6 461 arch_send_call_function_ipi_mask(cfd->cpumask_ipi);
3d442233 462
9a46ad6d 463 if (wait) {
e1d12f32 464 for_each_cpu(cpu, cfd->cpumask) {
966a9671 465 call_single_data_t *csd;
e1d12f32
AM
466
467 csd = per_cpu_ptr(cfd->csd, cpu);
9a46ad6d
SL
468 csd_lock_wait(csd);
469 }
470 }
3d442233 471}
54b11e6d 472EXPORT_SYMBOL(smp_call_function_many);
3d442233
JA
473
474/**
475 * smp_call_function(): Run a function on all other CPUs.
476 * @func: The function to run. This must be fast and non-blocking.
477 * @info: An arbitrary pointer to pass to the function.
0b13fda1
IM
478 * @wait: If true, wait (atomically) until function has completed
479 * on other CPUs.
3d442233 480 *
54b11e6d 481 * Returns 0.
3d442233
JA
482 *
483 * If @wait is true, then returns once @func has returned; otherwise
72f279b2 484 * it returns just before the target cpu calls @func.
3d442233
JA
485 *
486 * You must not call this function with disabled interrupts or from a
487 * hardware interrupt handler or from a bottom half handler.
488 */
3a5f65df 489int smp_call_function(smp_call_func_t func, void *info, int wait)
3d442233 490{
3d442233 491 preempt_disable();
54b11e6d 492 smp_call_function_many(cpu_online_mask, func, info, wait);
3d442233 493 preempt_enable();
0b13fda1 494
54b11e6d 495 return 0;
3d442233
JA
496}
497EXPORT_SYMBOL(smp_call_function);
351f8f8e 498
34db18a0
AW
499/* Setup configured maximum number of CPUs to activate */
500unsigned int setup_max_cpus = NR_CPUS;
501EXPORT_SYMBOL(setup_max_cpus);
502
503
504/*
505 * Setup routine for controlling SMP activation
506 *
507 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
508 * activation entirely (the MPS table probe still happens, though).
509 *
510 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
511 * greater than 0, limits the maximum number of CPUs activated in
512 * SMP mode to <NUM>.
513 */
514
515void __weak arch_disable_smp_support(void) { }
516
517static int __init nosmp(char *str)
518{
519 setup_max_cpus = 0;
520 arch_disable_smp_support();
521
522 return 0;
523}
524
525early_param("nosmp", nosmp);
526
527/* this is hard limit */
528static int __init nrcpus(char *str)
529{
530 int nr_cpus;
531
532 get_option(&str, &nr_cpus);
533 if (nr_cpus > 0 && nr_cpus < nr_cpu_ids)
534 nr_cpu_ids = nr_cpus;
535
536 return 0;
537}
538
539early_param("nr_cpus", nrcpus);
540
541static int __init maxcpus(char *str)
542{
543 get_option(&str, &setup_max_cpus);
544 if (setup_max_cpus == 0)
545 arch_disable_smp_support();
546
547 return 0;
548}
549
550early_param("maxcpus", maxcpus);
551
552/* Setup number of possible processor ids */
9b130ad5 553unsigned int nr_cpu_ids __read_mostly = NR_CPUS;
34db18a0
AW
554EXPORT_SYMBOL(nr_cpu_ids);
555
556/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
557void __init setup_nr_cpu_ids(void)
558{
559 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
560}
561
562/* Called by boot processor to activate the rest. */
563void __init smp_init(void)
564{
92b23278 565 int num_nodes, num_cpus;
34db18a0
AW
566 unsigned int cpu;
567
3bb5d2ee 568 idle_threads_init();
4cb28ced 569 cpuhp_threads_init();
3bb5d2ee 570
51111dce
ME
571 pr_info("Bringing up secondary CPUs ...\n");
572
34db18a0
AW
573 /* FIXME: This should be done in userspace --RR */
574 for_each_present_cpu(cpu) {
575 if (num_online_cpus() >= setup_max_cpus)
576 break;
577 if (!cpu_online(cpu))
578 cpu_up(cpu);
579 }
580
92b23278
ME
581 num_nodes = num_online_nodes();
582 num_cpus = num_online_cpus();
583 pr_info("Brought up %d node%s, %d CPU%s\n",
584 num_nodes, (num_nodes > 1 ? "s" : ""),
585 num_cpus, (num_cpus > 1 ? "s" : ""));
586
34db18a0 587 /* Any cleanup work */
34db18a0
AW
588 smp_cpus_done(setup_max_cpus);
589}
590
351f8f8e 591/*
bd924e8c
TH
592 * Call a function on all processors. May be used during early boot while
593 * early_boot_irqs_disabled is set. Use local_irq_save/restore() instead
594 * of local_irq_disable/enable().
351f8f8e
AW
595 */
596int on_each_cpu(void (*func) (void *info), void *info, int wait)
597{
bd924e8c 598 unsigned long flags;
351f8f8e
AW
599 int ret = 0;
600
601 preempt_disable();
602 ret = smp_call_function(func, info, wait);
bd924e8c 603 local_irq_save(flags);
351f8f8e 604 func(info);
bd924e8c 605 local_irq_restore(flags);
351f8f8e
AW
606 preempt_enable();
607 return ret;
608}
609EXPORT_SYMBOL(on_each_cpu);
3fc498f1
GBY
610
611/**
612 * on_each_cpu_mask(): Run a function on processors specified by
613 * cpumask, which may include the local processor.
614 * @mask: The set of cpus to run on (only runs on online subset).
615 * @func: The function to run. This must be fast and non-blocking.
616 * @info: An arbitrary pointer to pass to the function.
617 * @wait: If true, wait (atomically) until function has completed
618 * on other CPUs.
619 *
620 * If @wait is true, then returns once @func has returned.
621 *
202da400
DD
622 * You must not call this function with disabled interrupts or from a
623 * hardware interrupt handler or from a bottom half handler. The
624 * exception is that it may be used during early boot while
625 * early_boot_irqs_disabled is set.
3fc498f1
GBY
626 */
627void on_each_cpu_mask(const struct cpumask *mask, smp_call_func_t func,
628 void *info, bool wait)
629{
630 int cpu = get_cpu();
631
632 smp_call_function_many(mask, func, info, wait);
633 if (cpumask_test_cpu(cpu, mask)) {
202da400
DD
634 unsigned long flags;
635 local_irq_save(flags);
3fc498f1 636 func(info);
202da400 637 local_irq_restore(flags);
3fc498f1
GBY
638 }
639 put_cpu();
640}
641EXPORT_SYMBOL(on_each_cpu_mask);
b3a7e98e
GBY
642
643/*
644 * on_each_cpu_cond(): Call a function on each processor for which
645 * the supplied function cond_func returns true, optionally waiting
646 * for all the required CPUs to finish. This may include the local
647 * processor.
648 * @cond_func: A callback function that is passed a cpu id and
649 * the the info parameter. The function is called
650 * with preemption disabled. The function should
651 * return a blooean value indicating whether to IPI
652 * the specified CPU.
653 * @func: The function to run on all applicable CPUs.
654 * This must be fast and non-blocking.
655 * @info: An arbitrary pointer to pass to both functions.
656 * @wait: If true, wait (atomically) until function has
657 * completed on other CPUs.
658 * @gfp_flags: GFP flags to use when allocating the cpumask
659 * used internally by the function.
660 *
661 * The function might sleep if the GFP flags indicates a non
662 * atomic allocation is allowed.
663 *
664 * Preemption is disabled to protect against CPUs going offline but not online.
665 * CPUs going online during the call will not be seen or sent an IPI.
666 *
667 * You must not call this function with disabled interrupts or
668 * from a hardware interrupt handler or from a bottom half handler.
669 */
670void on_each_cpu_cond(bool (*cond_func)(int cpu, void *info),
671 smp_call_func_t func, void *info, bool wait,
672 gfp_t gfp_flags)
673{
674 cpumask_var_t cpus;
675 int cpu, ret;
676
d0164adc 677 might_sleep_if(gfpflags_allow_blocking(gfp_flags));
b3a7e98e
GBY
678
679 if (likely(zalloc_cpumask_var(&cpus, (gfp_flags|__GFP_NOWARN)))) {
680 preempt_disable();
681 for_each_online_cpu(cpu)
682 if (cond_func(cpu, info))
683 cpumask_set_cpu(cpu, cpus);
684 on_each_cpu_mask(cpus, func, info, wait);
685 preempt_enable();
686 free_cpumask_var(cpus);
687 } else {
688 /*
689 * No free cpumask, bother. No matter, we'll
690 * just have to IPI them one by one.
691 */
692 preempt_disable();
693 for_each_online_cpu(cpu)
694 if (cond_func(cpu, info)) {
695 ret = smp_call_function_single(cpu, func,
696 info, wait);
618fde87 697 WARN_ON_ONCE(ret);
b3a7e98e
GBY
698 }
699 preempt_enable();
700 }
701}
702EXPORT_SYMBOL(on_each_cpu_cond);
f37f435f
TG
703
704static void do_nothing(void *unused)
705{
706}
707
708/**
709 * kick_all_cpus_sync - Force all cpus out of idle
710 *
711 * Used to synchronize the update of pm_idle function pointer. It's
712 * called after the pointer is updated and returns after the dummy
713 * callback function has been executed on all cpus. The execution of
714 * the function can only happen on the remote cpus after they have
715 * left the idle function which had been called via pm_idle function
716 * pointer. So it's guaranteed that nothing uses the previous pointer
717 * anymore.
718 */
719void kick_all_cpus_sync(void)
720{
721 /* Make sure the change is visible before we kick the cpus */
722 smp_mb();
723 smp_call_function(do_nothing, NULL, 1);
724}
725EXPORT_SYMBOL_GPL(kick_all_cpus_sync);
c6f4459f
CL
726
727/**
728 * wake_up_all_idle_cpus - break all cpus out of idle
729 * wake_up_all_idle_cpus try to break all cpus which is in idle state even
730 * including idle polling cpus, for non-idle cpus, we will do nothing
731 * for them.
732 */
733void wake_up_all_idle_cpus(void)
734{
735 int cpu;
736
737 preempt_disable();
738 for_each_online_cpu(cpu) {
739 if (cpu == smp_processor_id())
740 continue;
741
742 wake_up_if_idle(cpu);
743 }
744 preempt_enable();
745}
746EXPORT_SYMBOL_GPL(wake_up_all_idle_cpus);
df8ce9d7
JG
747
748/**
749 * smp_call_on_cpu - Call a function on a specific cpu
750 *
751 * Used to call a function on a specific cpu and wait for it to return.
752 * Optionally make sure the call is done on a specified physical cpu via vcpu
753 * pinning in order to support virtualized environments.
754 */
755struct smp_call_on_cpu_struct {
756 struct work_struct work;
757 struct completion done;
758 int (*func)(void *);
759 void *data;
760 int ret;
761 int cpu;
762};
763
764static void smp_call_on_cpu_callback(struct work_struct *work)
765{
766 struct smp_call_on_cpu_struct *sscs;
767
768 sscs = container_of(work, struct smp_call_on_cpu_struct, work);
769 if (sscs->cpu >= 0)
770 hypervisor_pin_vcpu(sscs->cpu);
771 sscs->ret = sscs->func(sscs->data);
772 if (sscs->cpu >= 0)
773 hypervisor_pin_vcpu(-1);
774
775 complete(&sscs->done);
776}
777
778int smp_call_on_cpu(unsigned int cpu, int (*func)(void *), void *par, bool phys)
779{
780 struct smp_call_on_cpu_struct sscs = {
df8ce9d7
JG
781 .done = COMPLETION_INITIALIZER_ONSTACK(sscs.done),
782 .func = func,
783 .data = par,
784 .cpu = phys ? cpu : -1,
785 };
786
8db54949
PZ
787 INIT_WORK_ONSTACK(&sscs.work, smp_call_on_cpu_callback);
788
df8ce9d7
JG
789 if (cpu >= nr_cpu_ids || !cpu_online(cpu))
790 return -ENXIO;
791
792 queue_work_on(cpu, system_wq, &sscs.work);
793 wait_for_completion(&sscs.done);
794
795 return sscs.ret;
796}
797EXPORT_SYMBOL_GPL(smp_call_on_cpu);