]>
Commit | Line | Data |
---|---|---|
1142d810 TH |
1 | /* |
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. | |
e5582ca2 | 10 | */ |
1142d810 | 11 | #include <linux/completion.h> |
1da177e4 | 12 | #include <linux/cpu.h> |
1142d810 | 13 | #include <linux/init.h> |
ee527cd3 | 14 | #include <linux/kthread.h> |
9984de1a | 15 | #include <linux/export.h> |
1142d810 | 16 | #include <linux/percpu.h> |
ee527cd3 PB |
17 | #include <linux/sched.h> |
18 | #include <linux/stop_machine.h> | |
a12bb444 | 19 | #include <linux/interrupt.h> |
1142d810 | 20 | #include <linux/kallsyms.h> |
14e568e7 | 21 | #include <linux/smpboot.h> |
60063497 | 22 | #include <linux/atomic.h> |
7053ea1a | 23 | #include <linux/lglock.h> |
1142d810 TH |
24 | |
25 | /* | |
26 | * Structure to determine completion condition and record errors. May | |
27 | * be shared by works on different cpus. | |
28 | */ | |
29 | struct 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 */ | |
37 | struct cpu_stopper { | |
02cb7aa9 ON |
38 | struct task_struct *thread; |
39 | ||
1142d810 | 40 | spinlock_t lock; |
878ae127 | 41 | bool enabled; /* is this stopper enabled? */ |
1142d810 | 42 | struct list_head works; /* list of pending works */ |
02cb7aa9 ON |
43 | |
44 | struct cpu_stop_work stop_work; /* for stop_cpus */ | |
1142d810 TH |
45 | }; |
46 | ||
47 | static DEFINE_PER_CPU(struct cpu_stopper, cpu_stopper); | |
f445027e | 48 | static bool stop_machine_initialized = false; |
1142d810 | 49 | |
7053ea1a RR |
50 | /* |
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 | */ | |
56 | DEFINE_STATIC_LGLOCK(stop_cpus_lock); | |
57 | ||
1142d810 TH |
58 | static 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 */ | |
66 | static 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 | ||
5caa1c08 ON |
76 | static 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 | ||
1142d810 | 83 | /* queue @work to @stopper. if offline, @work is completed immediately */ |
860a0ffa | 84 | static void cpu_stop_queue_work(unsigned int cpu, struct cpu_stop_work *work) |
1142d810 | 85 | { |
860a0ffa | 86 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 TH |
87 | unsigned long flags; |
88 | ||
89 | spin_lock_irqsave(&stopper->lock, flags); | |
5caa1c08 ON |
90 | if (stopper->enabled) |
91 | __cpu_stop_queue_work(stopper, work); | |
92 | else | |
1142d810 | 93 | cpu_stop_signal_done(work->done, false); |
1142d810 TH |
94 | 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 | */ | |
121 | int 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); | |
860a0ffa | 127 | cpu_stop_queue_work(cpu, &work); |
1142d810 TH |
128 | wait_for_completion(&done.completion); |
129 | return done.executed ? done.ret : -ENOENT; | |
130 | } | |
131 | ||
1be0bd77 PZ |
132 | /* This controls the threads on each CPU. */ |
133 | enum 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 | ||
146 | struct multi_stop_data { | |
9a301f22 | 147 | cpu_stop_fn_t fn; |
1be0bd77 PZ |
148 | 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 | ||
157 | static 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. */ | |
167 | static 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. */ | |
174 | static 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 | ||
5caa1c08 ON |
219 | static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1, |
220 | int cpu2, struct cpu_stop_work *work2) | |
221 | { | |
d8bc8535 ON |
222 | 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 | ||
5caa1c08 | 226 | lg_double_lock(&stop_cpus_lock, cpu1, cpu2); |
d8bc8535 ON |
227 | 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); | |
237 | unlock: | |
238 | spin_unlock(&stopper2->lock); | |
239 | spin_unlock_irq(&stopper1->lock); | |
5caa1c08 ON |
240 | lg_double_unlock(&stop_cpus_lock, cpu1, cpu2); |
241 | ||
d8bc8535 | 242 | return err; |
5caa1c08 | 243 | } |
1be0bd77 PZ |
244 | /** |
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 | */ | |
255 | int stop_two_cpus(unsigned int cpu1, unsigned int cpu2, cpu_stop_fn_t fn, void *arg) | |
256 | { | |
1be0bd77 PZ |
257 | struct cpu_stop_done done; |
258 | struct cpu_stop_work work1, work2; | |
6acce3ef PZ |
259 | struct multi_stop_data msdata; |
260 | ||
261 | preempt_disable(); | |
262 | msdata = (struct multi_stop_data){ | |
1be0bd77 PZ |
263 | .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 | ||
1be0bd77 PZ |
275 | cpu_stop_init_done(&done, 2); |
276 | set_state(&msdata, MULTI_STOP_PREPARE); | |
277 | ||
6acce3ef | 278 | /* |
d8bc8535 ON |
279 | * We do not want to migrate to inactive CPU. FIXME: move this |
280 | * into migrate_swap_stop() callback. | |
6acce3ef PZ |
281 | */ |
282 | if (!cpu_active(cpu1) || !cpu_active(cpu2)) { | |
283 | preempt_enable(); | |
284 | return -ENOENT; | |
285 | } | |
286 | ||
5caa1c08 ON |
287 | 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 | } | |
b17718d0 | 293 | |
6acce3ef | 294 | preempt_enable(); |
1be0bd77 PZ |
295 | |
296 | wait_for_completion(&done.completion); | |
6acce3ef | 297 | |
1be0bd77 PZ |
298 | return done.executed ? done.ret : -ENOENT; |
299 | } | |
300 | ||
1142d810 TH |
301 | /** |
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 | |
cf250040 | 306 | * @work_buf: pointer to cpu_stop_work structure |
1142d810 TH |
307 | * |
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 | */ | |
315 | void 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, }; | |
860a0ffa | 319 | cpu_stop_queue_work(cpu, work_buf); |
1142d810 TH |
320 | } |
321 | ||
322 | /* static data for stop_cpus */ | |
192d8857 | 323 | static DEFINE_MUTEX(stop_cpus_mutex); |
1142d810 | 324 | |
fd7355ba TH |
325 | static void queue_stop_cpus_work(const struct cpumask *cpumask, |
326 | cpu_stop_fn_t fn, void *arg, | |
327 | struct cpu_stop_done *done) | |
1142d810 TH |
328 | { |
329 | struct cpu_stop_work *work; | |
1142d810 TH |
330 | unsigned int cpu; |
331 | ||
1142d810 TH |
332 | /* |
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 | */ | |
7053ea1a | 337 | lg_global_lock(&stop_cpus_lock); |
b377c2a0 ON |
338 | 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 | } | |
7053ea1a | 345 | lg_global_unlock(&stop_cpus_lock); |
fd7355ba | 346 | } |
1142d810 | 347 | |
fd7355ba TH |
348 | static 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); | |
1142d810 TH |
355 | 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 | */ | |
387 | int 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 | */ | |
416 | int 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 | ||
14e568e7 TG |
428 | static int cpu_stop_should_run(unsigned int cpu) |
429 | { | |
430 | 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 | ||
440 | static void cpu_stopper_thread(unsigned int cpu) | |
1142d810 | 441 | { |
14e568e7 | 442 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 TH |
443 | struct cpu_stop_work *work; |
444 | int ret; | |
445 | ||
446 | repeat: | |
1142d810 TH |
447 | 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; | |
ca51c5a7 | 460 | char ksym_buf[KSYM_NAME_LEN] __maybe_unused; |
1142d810 | 461 | |
1142d810 TH |
462 | /* 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); | |
14e568e7 TG |
477 | goto repeat; |
478 | } | |
1142d810 TH |
479 | } |
480 | ||
233e7f26 ON |
481 | void 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 | ||
34f971f6 PZ |
493 | extern void sched_set_stop_task(int cpu, struct task_struct *stop); |
494 | ||
14e568e7 TG |
495 | static void cpu_stop_create(unsigned int cpu) |
496 | { | |
02cb7aa9 | 497 | sched_set_stop_task(cpu, per_cpu(cpu_stopper.thread, cpu)); |
14e568e7 TG |
498 | } |
499 | ||
500 | static void cpu_stop_park(unsigned int cpu) | |
1142d810 | 501 | { |
1142d810 | 502 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 | 503 | |
233e7f26 | 504 | WARN_ON(!list_empty(&stopper->works)); |
14e568e7 | 505 | } |
1142d810 | 506 | |
14e568e7 TG |
507 | static 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); | |
1142d810 TH |
514 | } |
515 | ||
14e568e7 | 516 | static struct smp_hotplug_thread cpu_stop_threads = { |
02cb7aa9 | 517 | .store = &cpu_stopper.thread, |
14e568e7 TG |
518 | .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, | |
46c498c2 | 524 | .pre_unpark = cpu_stop_unpark, |
14e568e7 | 525 | .selfparking = true, |
1142d810 TH |
526 | }; |
527 | ||
528 | static int __init cpu_stop_init(void) | |
529 | { | |
1142d810 | 530 | unsigned int cpu; |
1142d810 TH |
531 | |
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 | ||
14e568e7 | 539 | BUG_ON(smpboot_register_percpu_thread(&cpu_stop_threads)); |
f445027e | 540 | stop_machine_initialized = true; |
1142d810 TH |
541 | return 0; |
542 | } | |
543 | early_initcall(cpu_stop_init); | |
1da177e4 | 544 | |
bbf1bb3e TH |
545 | #ifdef CONFIG_STOP_MACHINE |
546 | ||
9a301f22 | 547 | static int __stop_machine(cpu_stop_fn_t fn, void *data, const struct cpumask *cpus) |
1da177e4 | 548 | { |
1be0bd77 PZ |
549 | struct multi_stop_data msdata = { |
550 | .fn = fn, | |
551 | .data = data, | |
552 | .num_threads = num_online_cpus(), | |
553 | .active_cpus = cpus, | |
554 | }; | |
3fc1f1e2 | 555 | |
f445027e JF |
556 | 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 | ||
1be0bd77 | 565 | WARN_ON_ONCE(msdata.num_threads != 1); |
f445027e JF |
566 | |
567 | local_irq_save(flags); | |
568 | hard_irq_disable(); | |
569 | ret = (*fn)(data); | |
570 | local_irq_restore(flags); | |
571 | ||
572 | return ret; | |
573 | } | |
574 | ||
3fc1f1e2 | 575 | /* Set the initial state and stop all online cpus. */ |
1be0bd77 PZ |
576 | set_state(&msdata, MULTI_STOP_PREPARE); |
577 | return stop_cpus(cpu_online_mask, multi_cpu_stop, &msdata); | |
1da177e4 LT |
578 | } |
579 | ||
9a301f22 | 580 | int stop_machine(cpu_stop_fn_t fn, void *data, const struct cpumask *cpus) |
1da177e4 | 581 | { |
1da177e4 LT |
582 | int ret; |
583 | ||
584 | /* No CPUs can come up or down during this. */ | |
86ef5c9a | 585 | get_online_cpus(); |
eeec4fad | 586 | ret = __stop_machine(fn, data, cpus); |
86ef5c9a | 587 | put_online_cpus(); |
1da177e4 LT |
588 | return ret; |
589 | } | |
eeec4fad | 590 | EXPORT_SYMBOL_GPL(stop_machine); |
bbf1bb3e | 591 | |
f740e6cd TH |
592 | /** |
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 | */ | |
9a301f22 | 614 | int stop_machine_from_inactive_cpu(cpu_stop_fn_t fn, void *data, |
f740e6cd TH |
615 | const struct cpumask *cpus) |
616 | { | |
1be0bd77 | 617 | struct multi_stop_data msdata = { .fn = fn, .data = data, |
f740e6cd TH |
618 | .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())); | |
1be0bd77 | 624 | msdata.num_threads = num_active_cpus() + 1; /* +1 for local */ |
f740e6cd TH |
625 | |
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 */ | |
1be0bd77 | 631 | set_state(&msdata, MULTI_STOP_PREPARE); |
f740e6cd | 632 | cpu_stop_init_done(&done, num_active_cpus()); |
1be0bd77 | 633 | queue_stop_cpus_work(cpu_active_mask, multi_cpu_stop, &msdata, |
f740e6cd | 634 | &done); |
1be0bd77 | 635 | ret = multi_cpu_stop(&msdata); |
f740e6cd TH |
636 | |
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 | ||
bbf1bb3e | 645 | #endif /* CONFIG_STOP_MACHINE */ |