blob: 91fbb109de6c47245f9780a8545c4e97c8ac17f2 [file] [log] [blame]
Tejun Heo1142d812010-05-06 18:49:20 +02001/*
2 * kernel/stop_machine.c
3 *
4 * Copyright (C) 2008, 2005 IBM Corporation.
5 * Copyright (C) 2008, 2005 Rusty Russell rusty@rustcorp.com.au
6 * Copyright (C) 2010 SUSE Linux Products GmbH
7 * Copyright (C) 2010 Tejun Heo <tj@kernel.org>
8 *
9 * This file is released under the GPLv2 and any later version.
Rusty Russelle5582ca2006-09-29 02:01:35 -070010 */
Tejun Heo1142d812010-05-06 18:49:20 +020011#include <linux/completion.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/cpu.h>
Tejun Heo1142d812010-05-06 18:49:20 +020013#include <linux/init.h>
Prarit Bhargavaee527cd2007-05-08 00:25:08 -070014#include <linux/kthread.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040015#include <linux/export.h>
Tejun Heo1142d812010-05-06 18:49:20 +020016#include <linux/percpu.h>
Prarit Bhargavaee527cd2007-05-08 00:25:08 -070017#include <linux/sched.h>
18#include <linux/stop_machine.h>
Benjamin Herrenschmidta12bb442007-05-10 22:22:47 -070019#include <linux/interrupt.h>
Tejun Heo1142d812010-05-06 18:49:20 +020020#include <linux/kallsyms.h>
Thomas Gleixner14e568e2013-01-31 12:11:14 +000021#include <linux/smpboot.h>
Arun Sharma600634972011-07-26 16:09:06 -070022#include <linux/atomic.h>
Rik van Riel7053ea12013-11-01 10:41:46 -040023#include <linux/lglock.h>
Tejun Heo1142d812010-05-06 18:49:20 +020024
25/*
26 * Structure to determine completion condition and record errors. May
27 * be shared by works on different cpus.
28 */
29struct cpu_stop_done {
30 atomic_t nr_todo; /* nr left to execute */
31 bool executed; /* actually executed? */
32 int ret; /* collected return value */
33 struct completion completion; /* fired if nr_todo reaches 0 */
34};
35
36/* the actual stopper, one per every possible cpu, enabled on online cpus */
37struct cpu_stopper {
Oleg Nesterov02cb7aa2015-06-30 03:29:44 +020038 struct task_struct *thread;
39
Tejun Heo1142d812010-05-06 18:49:20 +020040 spinlock_t lock;
Richard Kennedy878ae122010-08-09 17:20:34 -070041 bool enabled; /* is this stopper enabled? */
Tejun Heo1142d812010-05-06 18:49:20 +020042 struct list_head works; /* list of pending works */
Oleg Nesterov02cb7aa2015-06-30 03:29:44 +020043
44 struct cpu_stop_work stop_work; /* for stop_cpus */
Tejun Heo1142d812010-05-06 18:49:20 +020045};
46
47static DEFINE_PER_CPU(struct cpu_stopper, cpu_stopper);
Jeremy Fitzhardingef4450272011-10-31 17:11:15 -070048static bool stop_machine_initialized = false;
Tejun Heo1142d812010-05-06 18:49:20 +020049
Rik van Riel7053ea12013-11-01 10:41:46 -040050/*
51 * Avoids a race between stop_two_cpus and global stop_cpus, where
52 * the stoppers could get queued up in reverse order, leading to
53 * system deadlock. Using an lglock means stop_two_cpus remains
54 * relatively cheap.
55 */
56DEFINE_STATIC_LGLOCK(stop_cpus_lock);
57
Tejun Heo1142d812010-05-06 18:49:20 +020058static void cpu_stop_init_done(struct cpu_stop_done *done, unsigned int nr_todo)
59{
60 memset(done, 0, sizeof(*done));
61 atomic_set(&done->nr_todo, nr_todo);
62 init_completion(&done->completion);
63}
64
65/* signal completion unless @done is NULL */
66static void cpu_stop_signal_done(struct cpu_stop_done *done, bool executed)
67{
68 if (done) {
69 if (executed)
70 done->executed = true;
71 if (atomic_dec_and_test(&done->nr_todo))
72 complete(&done->completion);
73 }
74}
75
Oleg Nesterov5caa1c02015-10-08 16:51:34 +020076static void __cpu_stop_queue_work(struct cpu_stopper *stopper,
77 struct cpu_stop_work *work)
78{
79 list_add_tail(&work->list, &stopper->works);
80 wake_up_process(stopper->thread);
81}
82
Tejun Heo1142d812010-05-06 18:49:20 +020083/* queue @work to @stopper. if offline, @work is completed immediately */
Thomas Gleixner860a0ff2013-01-31 12:11:13 +000084static void cpu_stop_queue_work(unsigned int cpu, struct cpu_stop_work *work)
Tejun Heo1142d812010-05-06 18:49:20 +020085{
Thomas Gleixner860a0ff2013-01-31 12:11:13 +000086 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
Tejun Heo1142d812010-05-06 18:49:20 +020087 unsigned long flags;
88
89 spin_lock_irqsave(&stopper->lock, flags);
Oleg Nesterov5caa1c02015-10-08 16:51:34 +020090 if (stopper->enabled)
91 __cpu_stop_queue_work(stopper, work);
92 else
Tejun Heo1142d812010-05-06 18:49:20 +020093 cpu_stop_signal_done(work->done, false);
Tejun Heo1142d812010-05-06 18:49:20 +020094 spin_unlock_irqrestore(&stopper->lock, flags);
95}
96
97/**
98 * stop_one_cpu - stop a cpu
99 * @cpu: cpu to stop
100 * @fn: function to execute
101 * @arg: argument to @fn
102 *
103 * Execute @fn(@arg) on @cpu. @fn is run in a process context with
104 * the highest priority preempting any task on the cpu and
105 * monopolizing it. This function returns after the execution is
106 * complete.
107 *
108 * This function doesn't guarantee @cpu stays online till @fn
109 * completes. If @cpu goes down in the middle, execution may happen
110 * partially or fully on different cpus. @fn should either be ready
111 * for that or the caller should ensure that @cpu stays online until
112 * this function completes.
113 *
114 * CONTEXT:
115 * Might sleep.
116 *
117 * RETURNS:
118 * -ENOENT if @fn(@arg) was not executed because @cpu was offline;
119 * otherwise, the return value of @fn.
120 */
121int stop_one_cpu(unsigned int cpu, cpu_stop_fn_t fn, void *arg)
122{
123 struct cpu_stop_done done;
124 struct cpu_stop_work work = { .fn = fn, .arg = arg, .done = &done };
125
126 cpu_stop_init_done(&done, 1);
Thomas Gleixner860a0ff2013-01-31 12:11:13 +0000127 cpu_stop_queue_work(cpu, &work);
Tejun Heo1142d812010-05-06 18:49:20 +0200128 wait_for_completion(&done.completion);
129 return done.executed ? done.ret : -ENOENT;
130}
131
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100132/* This controls the threads on each CPU. */
133enum multi_stop_state {
134 /* Dummy starting state for thread. */
135 MULTI_STOP_NONE,
136 /* Awaiting everyone to be scheduled. */
137 MULTI_STOP_PREPARE,
138 /* Disable interrupts. */
139 MULTI_STOP_DISABLE_IRQ,
140 /* Run the function */
141 MULTI_STOP_RUN,
142 /* Exit */
143 MULTI_STOP_EXIT,
144};
145
146struct multi_stop_data {
Oleg Nesterov9a301f22015-06-30 03:29:55 +0200147 cpu_stop_fn_t fn;
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100148 void *data;
149 /* Like num_online_cpus(), but hotplug cpu uses us, so we need this. */
150 unsigned int num_threads;
151 const struct cpumask *active_cpus;
152
153 enum multi_stop_state state;
154 atomic_t thread_ack;
155};
156
157static void set_state(struct multi_stop_data *msdata,
158 enum multi_stop_state newstate)
159{
160 /* Reset ack counter. */
161 atomic_set(&msdata->thread_ack, msdata->num_threads);
162 smp_wmb();
163 msdata->state = newstate;
164}
165
166/* Last one to ack a state moves to the next state. */
167static void ack_state(struct multi_stop_data *msdata)
168{
169 if (atomic_dec_and_test(&msdata->thread_ack))
170 set_state(msdata, msdata->state + 1);
171}
172
173/* This is the cpu_stop function which stops the CPU. */
174static int multi_cpu_stop(void *data)
175{
176 struct multi_stop_data *msdata = data;
177 enum multi_stop_state curstate = MULTI_STOP_NONE;
178 int cpu = smp_processor_id(), err = 0;
179 unsigned long flags;
180 bool is_active;
181
182 /*
183 * When called from stop_machine_from_inactive_cpu(), irq might
184 * already be disabled. Save the state and restore it on exit.
185 */
186 local_save_flags(flags);
187
188 if (!msdata->active_cpus)
189 is_active = cpu == cpumask_first(cpu_online_mask);
190 else
191 is_active = cpumask_test_cpu(cpu, msdata->active_cpus);
192
193 /* Simple state machine */
194 do {
195 /* Chill out and ensure we re-read multi_stop_state. */
196 cpu_relax();
197 if (msdata->state != curstate) {
198 curstate = msdata->state;
199 switch (curstate) {
200 case MULTI_STOP_DISABLE_IRQ:
201 local_irq_disable();
202 hard_irq_disable();
203 break;
204 case MULTI_STOP_RUN:
205 if (is_active)
206 err = msdata->fn(msdata->data);
207 break;
208 default:
209 break;
210 }
211 ack_state(msdata);
212 }
213 } while (curstate != MULTI_STOP_EXIT);
214
215 local_irq_restore(flags);
216 return err;
217}
218
Oleg Nesterov5caa1c02015-10-08 16:51:34 +0200219static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1,
220 int cpu2, struct cpu_stop_work *work2)
221{
Oleg Nesterovd8bc8532015-10-08 19:01:41 +0200222 struct cpu_stopper *stopper1 = per_cpu_ptr(&cpu_stopper, cpu1);
223 struct cpu_stopper *stopper2 = per_cpu_ptr(&cpu_stopper, cpu2);
224 int err;
225
Oleg Nesterov5caa1c02015-10-08 16:51:34 +0200226 lg_double_lock(&stop_cpus_lock, cpu1, cpu2);
Oleg Nesterovd8bc8532015-10-08 19:01:41 +0200227 spin_lock_irq(&stopper1->lock);
228 spin_lock_nested(&stopper2->lock, SINGLE_DEPTH_NESTING);
229
230 err = -ENOENT;
231 if (!stopper1->enabled || !stopper2->enabled)
232 goto unlock;
233
234 err = 0;
235 __cpu_stop_queue_work(stopper1, work1);
236 __cpu_stop_queue_work(stopper2, work2);
237unlock:
238 spin_unlock(&stopper2->lock);
239 spin_unlock_irq(&stopper1->lock);
Oleg Nesterov5caa1c02015-10-08 16:51:34 +0200240 lg_double_unlock(&stop_cpus_lock, cpu1, cpu2);
241
Oleg Nesterovd8bc8532015-10-08 19:01:41 +0200242 return err;
Oleg Nesterov5caa1c02015-10-08 16:51:34 +0200243}
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100244/**
245 * stop_two_cpus - stops two cpus
246 * @cpu1: the cpu to stop
247 * @cpu2: the other cpu to stop
248 * @fn: function to execute
249 * @arg: argument to @fn
250 *
251 * Stops both the current and specified CPU and runs @fn on one of them.
252 *
253 * returns when both are completed.
254 */
255int stop_two_cpus(unsigned int cpu1, unsigned int cpu2, cpu_stop_fn_t fn, void *arg)
256{
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100257 struct cpu_stop_done done;
258 struct cpu_stop_work work1, work2;
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200259 struct multi_stop_data msdata;
260
261 preempt_disable();
262 msdata = (struct multi_stop_data){
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100263 .fn = fn,
264 .data = arg,
265 .num_threads = 2,
266 .active_cpus = cpumask_of(cpu1),
267 };
268
269 work1 = work2 = (struct cpu_stop_work){
270 .fn = multi_cpu_stop,
271 .arg = &msdata,
272 .done = &done
273 };
274
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100275 cpu_stop_init_done(&done, 2);
276 set_state(&msdata, MULTI_STOP_PREPARE);
277
278 /*
Oleg Nesterovd8bc8532015-10-08 19:01:41 +0200279 * We do not want to migrate to inactive CPU. FIXME: move this
280 * into migrate_swap_stop() callback.
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200281 */
282 if (!cpu_active(cpu1) || !cpu_active(cpu2)) {
283 preempt_enable();
284 return -ENOENT;
285 }
286
Oleg Nesterov5caa1c02015-10-08 16:51:34 +0200287 if (cpu1 > cpu2)
288 swap(cpu1, cpu2);
289 if (cpu_stop_queue_two_works(cpu1, &work1, cpu2, &work2)) {
290 preempt_enable();
291 return -ENOENT;
292 }
Peter Zijlstrab17718d2015-06-05 17:30:23 +0200293
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200294 preempt_enable();
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100295
296 wait_for_completion(&done.completion);
Peter Zijlstra6acce3e2013-10-11 14:38:20 +0200297
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100298 return done.executed ? done.ret : -ENOENT;
299}
300
Tejun Heo1142d812010-05-06 18:49:20 +0200301/**
302 * stop_one_cpu_nowait - stop a cpu but don't wait for completion
303 * @cpu: cpu to stop
304 * @fn: function to execute
305 * @arg: argument to @fn
Fabian Frederickcf250042014-06-04 16:11:22 -0700306 * @work_buf: pointer to cpu_stop_work structure
Tejun Heo1142d812010-05-06 18:49:20 +0200307 *
308 * Similar to stop_one_cpu() but doesn't wait for completion. The
309 * caller is responsible for ensuring @work_buf is currently unused
310 * and will remain untouched until stopper starts executing @fn.
311 *
312 * CONTEXT:
313 * Don't care.
314 */
315void stop_one_cpu_nowait(unsigned int cpu, cpu_stop_fn_t fn, void *arg,
316 struct cpu_stop_work *work_buf)
317{
318 *work_buf = (struct cpu_stop_work){ .fn = fn, .arg = arg, };
Thomas Gleixner860a0ff2013-01-31 12:11:13 +0000319 cpu_stop_queue_work(cpu, work_buf);
Tejun Heo1142d812010-05-06 18:49:20 +0200320}
321
322/* static data for stop_cpus */
Suresh Siddha192d8852011-06-23 11:19:29 -0700323static DEFINE_MUTEX(stop_cpus_mutex);
Tejun Heo1142d812010-05-06 18:49:20 +0200324
Tejun Heofd7355b2011-06-23 11:19:27 -0700325static void queue_stop_cpus_work(const struct cpumask *cpumask,
326 cpu_stop_fn_t fn, void *arg,
327 struct cpu_stop_done *done)
Tejun Heo1142d812010-05-06 18:49:20 +0200328{
329 struct cpu_stop_work *work;
Tejun Heo1142d812010-05-06 18:49:20 +0200330 unsigned int cpu;
331
Tejun Heo1142d812010-05-06 18:49:20 +0200332 /*
333 * Disable preemption while queueing to avoid getting
334 * preempted by a stopper which might wait for other stoppers
335 * to enter @fn which can lead to deadlock.
336 */
Rik van Riel7053ea12013-11-01 10:41:46 -0400337 lg_global_lock(&stop_cpus_lock);
Oleg Nesterovb377c2a2015-06-30 03:29:48 +0200338 for_each_cpu(cpu, cpumask) {
339 work = &per_cpu(cpu_stopper.stop_work, cpu);
340 work->fn = fn;
341 work->arg = arg;
342 work->done = done;
343 cpu_stop_queue_work(cpu, work);
344 }
Rik van Riel7053ea12013-11-01 10:41:46 -0400345 lg_global_unlock(&stop_cpus_lock);
Tejun Heofd7355b2011-06-23 11:19:27 -0700346}
Tejun Heo1142d812010-05-06 18:49:20 +0200347
Tejun Heofd7355b2011-06-23 11:19:27 -0700348static int __stop_cpus(const struct cpumask *cpumask,
349 cpu_stop_fn_t fn, void *arg)
350{
351 struct cpu_stop_done done;
352
353 cpu_stop_init_done(&done, cpumask_weight(cpumask));
354 queue_stop_cpus_work(cpumask, fn, arg, &done);
Tejun Heo1142d812010-05-06 18:49:20 +0200355 wait_for_completion(&done.completion);
356 return done.executed ? done.ret : -ENOENT;
357}
358
359/**
360 * stop_cpus - stop multiple cpus
361 * @cpumask: cpus to stop
362 * @fn: function to execute
363 * @arg: argument to @fn
364 *
365 * Execute @fn(@arg) on online cpus in @cpumask. On each target cpu,
366 * @fn is run in a process context with the highest priority
367 * preempting any task on the cpu and monopolizing it. This function
368 * returns after all executions are complete.
369 *
370 * This function doesn't guarantee the cpus in @cpumask stay online
371 * till @fn completes. If some cpus go down in the middle, execution
372 * on the cpu may happen partially or fully on different cpus. @fn
373 * should either be ready for that or the caller should ensure that
374 * the cpus stay online until this function completes.
375 *
376 * All stop_cpus() calls are serialized making it safe for @fn to wait
377 * for all cpus to start executing it.
378 *
379 * CONTEXT:
380 * Might sleep.
381 *
382 * RETURNS:
383 * -ENOENT if @fn(@arg) was not executed at all because all cpus in
384 * @cpumask were offline; otherwise, 0 if all executions of @fn
385 * returned 0, any non zero return value if any returned non zero.
386 */
387int stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg)
388{
389 int ret;
390
391 /* static works are used, process one request at a time */
392 mutex_lock(&stop_cpus_mutex);
393 ret = __stop_cpus(cpumask, fn, arg);
394 mutex_unlock(&stop_cpus_mutex);
395 return ret;
396}
397
398/**
399 * try_stop_cpus - try to stop multiple cpus
400 * @cpumask: cpus to stop
401 * @fn: function to execute
402 * @arg: argument to @fn
403 *
404 * Identical to stop_cpus() except that it fails with -EAGAIN if
405 * someone else is already using the facility.
406 *
407 * CONTEXT:
408 * Might sleep.
409 *
410 * RETURNS:
411 * -EAGAIN if someone else is already stopping cpus, -ENOENT if
412 * @fn(@arg) was not executed at all because all cpus in @cpumask were
413 * offline; otherwise, 0 if all executions of @fn returned 0, any non
414 * zero return value if any returned non zero.
415 */
416int try_stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg)
417{
418 int ret;
419
420 /* static works are used, process one request at a time */
421 if (!mutex_trylock(&stop_cpus_mutex))
422 return -EAGAIN;
423 ret = __stop_cpus(cpumask, fn, arg);
424 mutex_unlock(&stop_cpus_mutex);
425 return ret;
426}
427
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000428static int cpu_stop_should_run(unsigned int cpu)
Tejun Heo1142d812010-05-06 18:49:20 +0200429{
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000430 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
431 unsigned long flags;
432 int run;
433
434 spin_lock_irqsave(&stopper->lock, flags);
435 run = !list_empty(&stopper->works);
436 spin_unlock_irqrestore(&stopper->lock, flags);
437 return run;
438}
439
440static void cpu_stopper_thread(unsigned int cpu)
441{
442 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
Tejun Heo1142d812010-05-06 18:49:20 +0200443 struct cpu_stop_work *work;
444 int ret;
445
446repeat:
Tejun Heo1142d812010-05-06 18:49:20 +0200447 work = NULL;
448 spin_lock_irq(&stopper->lock);
449 if (!list_empty(&stopper->works)) {
450 work = list_first_entry(&stopper->works,
451 struct cpu_stop_work, list);
452 list_del_init(&work->list);
453 }
454 spin_unlock_irq(&stopper->lock);
455
456 if (work) {
457 cpu_stop_fn_t fn = work->fn;
458 void *arg = work->arg;
459 struct cpu_stop_done *done = work->done;
Rakib Mullickca51c5a2010-10-26 14:22:44 -0700460 char ksym_buf[KSYM_NAME_LEN] __maybe_unused;
Tejun Heo1142d812010-05-06 18:49:20 +0200461
Tejun Heo1142d812010-05-06 18:49:20 +0200462 /* cpu stop callbacks are not allowed to sleep */
463 preempt_disable();
464
465 ret = fn(arg);
466 if (ret)
467 done->ret = ret;
468
469 /* restore preemption and check it's still balanced */
470 preempt_enable();
471 WARN_ONCE(preempt_count(),
472 "cpu_stop: %s(%p) leaked preempt count\n",
473 kallsyms_lookup((unsigned long)fn, NULL, NULL, NULL,
474 ksym_buf), arg);
475
476 cpu_stop_signal_done(done, true);
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000477 goto repeat;
478 }
Tejun Heo1142d812010-05-06 18:49:20 +0200479}
480
Oleg Nesterov233e7f22015-10-08 16:51:31 +0200481void stop_machine_park(int cpu)
482{
483 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
484 /*
485 * Lockless. cpu_stopper_thread() will take stopper->lock and flush
486 * the pending works before it parks, until then it is fine to queue
487 * the new works.
488 */
489 stopper->enabled = false;
490 kthread_park(stopper->thread);
491}
492
Peter Zijlstra34f971f2010-09-22 13:53:15 +0200493extern void sched_set_stop_task(int cpu, struct task_struct *stop);
494
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000495static void cpu_stop_create(unsigned int cpu)
Tejun Heo1142d812010-05-06 18:49:20 +0200496{
Oleg Nesterov02cb7aa2015-06-30 03:29:44 +0200497 sched_set_stop_task(cpu, per_cpu(cpu_stopper.thread, cpu));
Tejun Heo1142d812010-05-06 18:49:20 +0200498}
499
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000500static void cpu_stop_park(unsigned int cpu)
501{
502 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000503
Oleg Nesterov233e7f22015-10-08 16:51:31 +0200504 WARN_ON(!list_empty(&stopper->works));
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000505}
506
507static void cpu_stop_unpark(unsigned int cpu)
508{
509 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
510
511 spin_lock_irq(&stopper->lock);
512 stopper->enabled = true;
513 spin_unlock_irq(&stopper->lock);
514}
515
516static struct smp_hotplug_thread cpu_stop_threads = {
Oleg Nesterov02cb7aa2015-06-30 03:29:44 +0200517 .store = &cpu_stopper.thread,
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000518 .thread_should_run = cpu_stop_should_run,
519 .thread_fn = cpu_stopper_thread,
520 .thread_comm = "migration/%u",
521 .create = cpu_stop_create,
522 .setup = cpu_stop_unpark,
523 .park = cpu_stop_park,
Thomas Gleixner46c498c2013-02-26 18:44:33 +0100524 .pre_unpark = cpu_stop_unpark,
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000525 .selfparking = true,
Tejun Heo1142d812010-05-06 18:49:20 +0200526};
527
528static int __init cpu_stop_init(void)
529{
Tejun Heo1142d812010-05-06 18:49:20 +0200530 unsigned int cpu;
Tejun Heo1142d812010-05-06 18:49:20 +0200531
532 for_each_possible_cpu(cpu) {
533 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
534
535 spin_lock_init(&stopper->lock);
536 INIT_LIST_HEAD(&stopper->works);
537 }
538
Thomas Gleixner14e568e2013-01-31 12:11:14 +0000539 BUG_ON(smpboot_register_percpu_thread(&cpu_stop_threads));
Jeremy Fitzhardingef4450272011-10-31 17:11:15 -0700540 stop_machine_initialized = true;
Tejun Heo1142d812010-05-06 18:49:20 +0200541 return 0;
542}
543early_initcall(cpu_stop_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
Tejun Heobbf1bb32010-05-08 16:20:53 +0200545#ifdef CONFIG_STOP_MACHINE
546
Oleg Nesterov9a301f22015-06-30 03:29:55 +0200547static int __stop_machine(cpu_stop_fn_t fn, void *data, const struct cpumask *cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100549 struct multi_stop_data msdata = {
550 .fn = fn,
551 .data = data,
552 .num_threads = num_online_cpus(),
553 .active_cpus = cpus,
554 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
Jeremy Fitzhardingef4450272011-10-31 17:11:15 -0700556 if (!stop_machine_initialized) {
557 /*
558 * Handle the case where stop_machine() is called
559 * early in boot before stop_machine() has been
560 * initialized.
561 */
562 unsigned long flags;
563 int ret;
564
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100565 WARN_ON_ONCE(msdata.num_threads != 1);
Jeremy Fitzhardingef4450272011-10-31 17:11:15 -0700566
567 local_irq_save(flags);
568 hard_irq_disable();
569 ret = (*fn)(data);
570 local_irq_restore(flags);
571
572 return ret;
573 }
574
Tejun Heo3fc1f1e2010-05-06 18:49:20 +0200575 /* Set the initial state and stop all online cpus. */
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100576 set_state(&msdata, MULTI_STOP_PREPARE);
577 return stop_cpus(cpu_online_mask, multi_cpu_stop, &msdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
Oleg Nesterov9a301f22015-06-30 03:29:55 +0200580int stop_machine(cpu_stop_fn_t fn, void *data, const struct cpumask *cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 int ret;
583
584 /* No CPUs can come up or down during this. */
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100585 get_online_cpus();
Rusty Russelleeec4fa2008-07-28 12:16:30 -0500586 ret = __stop_machine(fn, data, cpus);
Gautham R Shenoy86ef5c92008-01-25 21:08:02 +0100587 put_online_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 return ret;
589}
Rusty Russelleeec4fa2008-07-28 12:16:30 -0500590EXPORT_SYMBOL_GPL(stop_machine);
Tejun Heobbf1bb32010-05-08 16:20:53 +0200591
Tejun Heof740e6cd2011-06-23 11:19:28 -0700592/**
593 * stop_machine_from_inactive_cpu - stop_machine() from inactive CPU
594 * @fn: the function to run
595 * @data: the data ptr for the @fn()
596 * @cpus: the cpus to run the @fn() on (NULL = any online cpu)
597 *
598 * This is identical to stop_machine() but can be called from a CPU which
599 * is not active. The local CPU is in the process of hotplug (so no other
600 * CPU hotplug can start) and not marked active and doesn't have enough
601 * context to sleep.
602 *
603 * This function provides stop_machine() functionality for such state by
604 * using busy-wait for synchronization and executing @fn directly for local
605 * CPU.
606 *
607 * CONTEXT:
608 * Local CPU is inactive. Temporarily stops all active CPUs.
609 *
610 * RETURNS:
611 * 0 if all executions of @fn returned 0, any non zero return value if any
612 * returned non zero.
613 */
Oleg Nesterov9a301f22015-06-30 03:29:55 +0200614int stop_machine_from_inactive_cpu(cpu_stop_fn_t fn, void *data,
Tejun Heof740e6cd2011-06-23 11:19:28 -0700615 const struct cpumask *cpus)
616{
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100617 struct multi_stop_data msdata = { .fn = fn, .data = data,
Tejun Heof740e6cd2011-06-23 11:19:28 -0700618 .active_cpus = cpus };
619 struct cpu_stop_done done;
620 int ret;
621
622 /* Local CPU must be inactive and CPU hotplug in progress. */
623 BUG_ON(cpu_active(raw_smp_processor_id()));
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100624 msdata.num_threads = num_active_cpus() + 1; /* +1 for local */
Tejun Heof740e6cd2011-06-23 11:19:28 -0700625
626 /* No proper task established and can't sleep - busy wait for lock. */
627 while (!mutex_trylock(&stop_cpus_mutex))
628 cpu_relax();
629
630 /* Schedule work on other CPUs and execute directly for local CPU */
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100631 set_state(&msdata, MULTI_STOP_PREPARE);
Tejun Heof740e6cd2011-06-23 11:19:28 -0700632 cpu_stop_init_done(&done, num_active_cpus());
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100633 queue_stop_cpus_work(cpu_active_mask, multi_cpu_stop, &msdata,
Tejun Heof740e6cd2011-06-23 11:19:28 -0700634 &done);
Peter Zijlstra1be0bd772013-10-07 11:29:15 +0100635 ret = multi_cpu_stop(&msdata);
Tejun Heof740e6cd2011-06-23 11:19:28 -0700636
637 /* Busy wait for completion. */
638 while (!completion_done(&done.completion))
639 cpu_relax();
640
641 mutex_unlock(&stop_cpus_mutex);
642 return ret ?: done.ret;
643}
644
Tejun Heobbf1bb32010-05-08 16:20:53 +0200645#endif /* CONFIG_STOP_MACHINE */