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 */ |
1b034bd9 | 84 | static bool 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 | 87 | unsigned long flags; |
1b034bd9 | 88 | bool enabled; |
1142d810 TH |
89 | |
90 | spin_lock_irqsave(&stopper->lock, flags); | |
1b034bd9 ON |
91 | enabled = stopper->enabled; |
92 | if (enabled) | |
5caa1c08 ON |
93 | __cpu_stop_queue_work(stopper, work); |
94 | else | |
1142d810 | 95 | cpu_stop_signal_done(work->done, false); |
1142d810 | 96 | spin_unlock_irqrestore(&stopper->lock, flags); |
1b034bd9 ON |
97 | |
98 | return enabled; | |
1142d810 TH |
99 | } |
100 | ||
101 | /** | |
102 | * stop_one_cpu - stop a cpu | |
103 | * @cpu: cpu to stop | |
104 | * @fn: function to execute | |
105 | * @arg: argument to @fn | |
106 | * | |
107 | * Execute @fn(@arg) on @cpu. @fn is run in a process context with | |
108 | * the highest priority preempting any task on the cpu and | |
109 | * monopolizing it. This function returns after the execution is | |
110 | * complete. | |
111 | * | |
112 | * This function doesn't guarantee @cpu stays online till @fn | |
113 | * completes. If @cpu goes down in the middle, execution may happen | |
114 | * partially or fully on different cpus. @fn should either be ready | |
115 | * for that or the caller should ensure that @cpu stays online until | |
116 | * this function completes. | |
117 | * | |
118 | * CONTEXT: | |
119 | * Might sleep. | |
120 | * | |
121 | * RETURNS: | |
122 | * -ENOENT if @fn(@arg) was not executed because @cpu was offline; | |
123 | * otherwise, the return value of @fn. | |
124 | */ | |
125 | int stop_one_cpu(unsigned int cpu, cpu_stop_fn_t fn, void *arg) | |
126 | { | |
127 | struct cpu_stop_done done; | |
128 | struct cpu_stop_work work = { .fn = fn, .arg = arg, .done = &done }; | |
129 | ||
130 | cpu_stop_init_done(&done, 1); | |
958c5f84 ON |
131 | if (!cpu_stop_queue_work(cpu, &work)) |
132 | return -ENOENT; | |
1142d810 | 133 | wait_for_completion(&done.completion); |
958c5f84 ON |
134 | WARN_ON(!done.executed); |
135 | return done.ret; | |
1142d810 TH |
136 | } |
137 | ||
1be0bd77 PZ |
138 | /* This controls the threads on each CPU. */ |
139 | enum multi_stop_state { | |
140 | /* Dummy starting state for thread. */ | |
141 | MULTI_STOP_NONE, | |
142 | /* Awaiting everyone to be scheduled. */ | |
143 | MULTI_STOP_PREPARE, | |
144 | /* Disable interrupts. */ | |
145 | MULTI_STOP_DISABLE_IRQ, | |
146 | /* Run the function */ | |
147 | MULTI_STOP_RUN, | |
148 | /* Exit */ | |
149 | MULTI_STOP_EXIT, | |
150 | }; | |
151 | ||
152 | struct multi_stop_data { | |
9a301f22 | 153 | cpu_stop_fn_t fn; |
1be0bd77 PZ |
154 | void *data; |
155 | /* Like num_online_cpus(), but hotplug cpu uses us, so we need this. */ | |
156 | unsigned int num_threads; | |
157 | const struct cpumask *active_cpus; | |
158 | ||
159 | enum multi_stop_state state; | |
160 | atomic_t thread_ack; | |
161 | }; | |
162 | ||
163 | static void set_state(struct multi_stop_data *msdata, | |
164 | enum multi_stop_state newstate) | |
165 | { | |
166 | /* Reset ack counter. */ | |
167 | atomic_set(&msdata->thread_ack, msdata->num_threads); | |
168 | smp_wmb(); | |
169 | msdata->state = newstate; | |
170 | } | |
171 | ||
172 | /* Last one to ack a state moves to the next state. */ | |
173 | static void ack_state(struct multi_stop_data *msdata) | |
174 | { | |
175 | if (atomic_dec_and_test(&msdata->thread_ack)) | |
176 | set_state(msdata, msdata->state + 1); | |
177 | } | |
178 | ||
179 | /* This is the cpu_stop function which stops the CPU. */ | |
180 | static int multi_cpu_stop(void *data) | |
181 | { | |
182 | struct multi_stop_data *msdata = data; | |
183 | enum multi_stop_state curstate = MULTI_STOP_NONE; | |
184 | int cpu = smp_processor_id(), err = 0; | |
185 | unsigned long flags; | |
186 | bool is_active; | |
187 | ||
188 | /* | |
189 | * When called from stop_machine_from_inactive_cpu(), irq might | |
190 | * already be disabled. Save the state and restore it on exit. | |
191 | */ | |
192 | local_save_flags(flags); | |
193 | ||
194 | if (!msdata->active_cpus) | |
195 | is_active = cpu == cpumask_first(cpu_online_mask); | |
196 | else | |
197 | is_active = cpumask_test_cpu(cpu, msdata->active_cpus); | |
198 | ||
199 | /* Simple state machine */ | |
200 | do { | |
201 | /* Chill out and ensure we re-read multi_stop_state. */ | |
202 | cpu_relax(); | |
203 | if (msdata->state != curstate) { | |
204 | curstate = msdata->state; | |
205 | switch (curstate) { | |
206 | case MULTI_STOP_DISABLE_IRQ: | |
207 | local_irq_disable(); | |
208 | hard_irq_disable(); | |
209 | break; | |
210 | case MULTI_STOP_RUN: | |
211 | if (is_active) | |
212 | err = msdata->fn(msdata->data); | |
213 | break; | |
214 | default: | |
215 | break; | |
216 | } | |
217 | ack_state(msdata); | |
218 | } | |
219 | } while (curstate != MULTI_STOP_EXIT); | |
220 | ||
221 | local_irq_restore(flags); | |
222 | return err; | |
223 | } | |
224 | ||
5caa1c08 ON |
225 | static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1, |
226 | int cpu2, struct cpu_stop_work *work2) | |
227 | { | |
d8bc8535 ON |
228 | struct cpu_stopper *stopper1 = per_cpu_ptr(&cpu_stopper, cpu1); |
229 | struct cpu_stopper *stopper2 = per_cpu_ptr(&cpu_stopper, cpu2); | |
230 | int err; | |
231 | ||
5caa1c08 | 232 | lg_double_lock(&stop_cpus_lock, cpu1, cpu2); |
d8bc8535 ON |
233 | spin_lock_irq(&stopper1->lock); |
234 | spin_lock_nested(&stopper2->lock, SINGLE_DEPTH_NESTING); | |
235 | ||
236 | err = -ENOENT; | |
237 | if (!stopper1->enabled || !stopper2->enabled) | |
238 | goto unlock; | |
239 | ||
240 | err = 0; | |
241 | __cpu_stop_queue_work(stopper1, work1); | |
242 | __cpu_stop_queue_work(stopper2, work2); | |
243 | unlock: | |
244 | spin_unlock(&stopper2->lock); | |
245 | spin_unlock_irq(&stopper1->lock); | |
5caa1c08 ON |
246 | lg_double_unlock(&stop_cpus_lock, cpu1, cpu2); |
247 | ||
d8bc8535 | 248 | return err; |
5caa1c08 | 249 | } |
1be0bd77 PZ |
250 | /** |
251 | * stop_two_cpus - stops two cpus | |
252 | * @cpu1: the cpu to stop | |
253 | * @cpu2: the other cpu to stop | |
254 | * @fn: function to execute | |
255 | * @arg: argument to @fn | |
256 | * | |
257 | * Stops both the current and specified CPU and runs @fn on one of them. | |
258 | * | |
259 | * returns when both are completed. | |
260 | */ | |
261 | int stop_two_cpus(unsigned int cpu1, unsigned int cpu2, cpu_stop_fn_t fn, void *arg) | |
262 | { | |
1be0bd77 PZ |
263 | struct cpu_stop_done done; |
264 | struct cpu_stop_work work1, work2; | |
6acce3ef PZ |
265 | struct multi_stop_data msdata; |
266 | ||
6acce3ef | 267 | msdata = (struct multi_stop_data){ |
1be0bd77 PZ |
268 | .fn = fn, |
269 | .data = arg, | |
270 | .num_threads = 2, | |
271 | .active_cpus = cpumask_of(cpu1), | |
272 | }; | |
273 | ||
274 | work1 = work2 = (struct cpu_stop_work){ | |
275 | .fn = multi_cpu_stop, | |
276 | .arg = &msdata, | |
277 | .done = &done | |
278 | }; | |
279 | ||
1be0bd77 PZ |
280 | cpu_stop_init_done(&done, 2); |
281 | set_state(&msdata, MULTI_STOP_PREPARE); | |
282 | ||
5caa1c08 ON |
283 | if (cpu1 > cpu2) |
284 | swap(cpu1, cpu2); | |
6a190051 | 285 | if (cpu_stop_queue_two_works(cpu1, &work1, cpu2, &work2)) |
5caa1c08 | 286 | return -ENOENT; |
1be0bd77 PZ |
287 | |
288 | wait_for_completion(&done.completion); | |
6a190051 ON |
289 | WARN_ON(!done.executed); |
290 | return done.ret; | |
1be0bd77 PZ |
291 | } |
292 | ||
1142d810 TH |
293 | /** |
294 | * stop_one_cpu_nowait - stop a cpu but don't wait for completion | |
295 | * @cpu: cpu to stop | |
296 | * @fn: function to execute | |
297 | * @arg: argument to @fn | |
cf250040 | 298 | * @work_buf: pointer to cpu_stop_work structure |
1142d810 TH |
299 | * |
300 | * Similar to stop_one_cpu() but doesn't wait for completion. The | |
301 | * caller is responsible for ensuring @work_buf is currently unused | |
302 | * and will remain untouched until stopper starts executing @fn. | |
303 | * | |
304 | * CONTEXT: | |
305 | * Don't care. | |
1b034bd9 ON |
306 | * |
307 | * RETURNS: | |
308 | * true if cpu_stop_work was queued successfully and @fn will be called, | |
309 | * false otherwise. | |
1142d810 | 310 | */ |
1b034bd9 | 311 | bool stop_one_cpu_nowait(unsigned int cpu, cpu_stop_fn_t fn, void *arg, |
1142d810 TH |
312 | struct cpu_stop_work *work_buf) |
313 | { | |
314 | *work_buf = (struct cpu_stop_work){ .fn = fn, .arg = arg, }; | |
1b034bd9 | 315 | return cpu_stop_queue_work(cpu, work_buf); |
1142d810 TH |
316 | } |
317 | ||
318 | /* static data for stop_cpus */ | |
192d8857 | 319 | static DEFINE_MUTEX(stop_cpus_mutex); |
1142d810 | 320 | |
fd7355ba TH |
321 | static void queue_stop_cpus_work(const struct cpumask *cpumask, |
322 | cpu_stop_fn_t fn, void *arg, | |
323 | struct cpu_stop_done *done) | |
1142d810 TH |
324 | { |
325 | struct cpu_stop_work *work; | |
1142d810 TH |
326 | unsigned int cpu; |
327 | ||
1142d810 TH |
328 | /* |
329 | * Disable preemption while queueing to avoid getting | |
330 | * preempted by a stopper which might wait for other stoppers | |
331 | * to enter @fn which can lead to deadlock. | |
332 | */ | |
7053ea1a | 333 | lg_global_lock(&stop_cpus_lock); |
b377c2a0 ON |
334 | for_each_cpu(cpu, cpumask) { |
335 | work = &per_cpu(cpu_stopper.stop_work, cpu); | |
336 | work->fn = fn; | |
337 | work->arg = arg; | |
338 | work->done = done; | |
339 | cpu_stop_queue_work(cpu, work); | |
340 | } | |
7053ea1a | 341 | lg_global_unlock(&stop_cpus_lock); |
fd7355ba | 342 | } |
1142d810 | 343 | |
fd7355ba TH |
344 | static int __stop_cpus(const struct cpumask *cpumask, |
345 | cpu_stop_fn_t fn, void *arg) | |
346 | { | |
347 | struct cpu_stop_done done; | |
348 | ||
349 | cpu_stop_init_done(&done, cpumask_weight(cpumask)); | |
350 | queue_stop_cpus_work(cpumask, fn, arg, &done); | |
1142d810 TH |
351 | wait_for_completion(&done.completion); |
352 | return done.executed ? done.ret : -ENOENT; | |
353 | } | |
354 | ||
355 | /** | |
356 | * stop_cpus - stop multiple cpus | |
357 | * @cpumask: cpus to stop | |
358 | * @fn: function to execute | |
359 | * @arg: argument to @fn | |
360 | * | |
361 | * Execute @fn(@arg) on online cpus in @cpumask. On each target cpu, | |
362 | * @fn is run in a process context with the highest priority | |
363 | * preempting any task on the cpu and monopolizing it. This function | |
364 | * returns after all executions are complete. | |
365 | * | |
366 | * This function doesn't guarantee the cpus in @cpumask stay online | |
367 | * till @fn completes. If some cpus go down in the middle, execution | |
368 | * on the cpu may happen partially or fully on different cpus. @fn | |
369 | * should either be ready for that or the caller should ensure that | |
370 | * the cpus stay online until this function completes. | |
371 | * | |
372 | * All stop_cpus() calls are serialized making it safe for @fn to wait | |
373 | * for all cpus to start executing it. | |
374 | * | |
375 | * CONTEXT: | |
376 | * Might sleep. | |
377 | * | |
378 | * RETURNS: | |
379 | * -ENOENT if @fn(@arg) was not executed at all because all cpus in | |
380 | * @cpumask were offline; otherwise, 0 if all executions of @fn | |
381 | * returned 0, any non zero return value if any returned non zero. | |
382 | */ | |
383 | int stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) | |
384 | { | |
385 | int ret; | |
386 | ||
387 | /* static works are used, process one request at a time */ | |
388 | mutex_lock(&stop_cpus_mutex); | |
389 | ret = __stop_cpus(cpumask, fn, arg); | |
390 | mutex_unlock(&stop_cpus_mutex); | |
391 | return ret; | |
392 | } | |
393 | ||
394 | /** | |
395 | * try_stop_cpus - try to stop multiple cpus | |
396 | * @cpumask: cpus to stop | |
397 | * @fn: function to execute | |
398 | * @arg: argument to @fn | |
399 | * | |
400 | * Identical to stop_cpus() except that it fails with -EAGAIN if | |
401 | * someone else is already using the facility. | |
402 | * | |
403 | * CONTEXT: | |
404 | * Might sleep. | |
405 | * | |
406 | * RETURNS: | |
407 | * -EAGAIN if someone else is already stopping cpus, -ENOENT if | |
408 | * @fn(@arg) was not executed at all because all cpus in @cpumask were | |
409 | * offline; otherwise, 0 if all executions of @fn returned 0, any non | |
410 | * zero return value if any returned non zero. | |
411 | */ | |
412 | int try_stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) | |
413 | { | |
414 | int ret; | |
415 | ||
416 | /* static works are used, process one request at a time */ | |
417 | if (!mutex_trylock(&stop_cpus_mutex)) | |
418 | return -EAGAIN; | |
419 | ret = __stop_cpus(cpumask, fn, arg); | |
420 | mutex_unlock(&stop_cpus_mutex); | |
421 | return ret; | |
422 | } | |
423 | ||
14e568e7 TG |
424 | static int cpu_stop_should_run(unsigned int cpu) |
425 | { | |
426 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
427 | unsigned long flags; | |
428 | int run; | |
429 | ||
430 | spin_lock_irqsave(&stopper->lock, flags); | |
431 | run = !list_empty(&stopper->works); | |
432 | spin_unlock_irqrestore(&stopper->lock, flags); | |
433 | return run; | |
434 | } | |
435 | ||
436 | static void cpu_stopper_thread(unsigned int cpu) | |
1142d810 | 437 | { |
14e568e7 | 438 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 TH |
439 | struct cpu_stop_work *work; |
440 | int ret; | |
441 | ||
442 | repeat: | |
1142d810 TH |
443 | work = NULL; |
444 | spin_lock_irq(&stopper->lock); | |
445 | if (!list_empty(&stopper->works)) { | |
446 | work = list_first_entry(&stopper->works, | |
447 | struct cpu_stop_work, list); | |
448 | list_del_init(&work->list); | |
449 | } | |
450 | spin_unlock_irq(&stopper->lock); | |
451 | ||
452 | if (work) { | |
453 | cpu_stop_fn_t fn = work->fn; | |
454 | void *arg = work->arg; | |
455 | struct cpu_stop_done *done = work->done; | |
ca51c5a7 | 456 | char ksym_buf[KSYM_NAME_LEN] __maybe_unused; |
1142d810 | 457 | |
1142d810 TH |
458 | /* cpu stop callbacks are not allowed to sleep */ |
459 | preempt_disable(); | |
460 | ||
461 | ret = fn(arg); | |
64038f29 | 462 | if (ret && done) |
1142d810 TH |
463 | done->ret = ret; |
464 | ||
465 | /* restore preemption and check it's still balanced */ | |
466 | preempt_enable(); | |
467 | WARN_ONCE(preempt_count(), | |
468 | "cpu_stop: %s(%p) leaked preempt count\n", | |
469 | kallsyms_lookup((unsigned long)fn, NULL, NULL, NULL, | |
470 | ksym_buf), arg); | |
471 | ||
472 | cpu_stop_signal_done(done, true); | |
14e568e7 TG |
473 | goto repeat; |
474 | } | |
1142d810 TH |
475 | } |
476 | ||
233e7f26 ON |
477 | void stop_machine_park(int cpu) |
478 | { | |
479 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
480 | /* | |
481 | * Lockless. cpu_stopper_thread() will take stopper->lock and flush | |
482 | * the pending works before it parks, until then it is fine to queue | |
483 | * the new works. | |
484 | */ | |
485 | stopper->enabled = false; | |
486 | kthread_park(stopper->thread); | |
487 | } | |
488 | ||
34f971f6 PZ |
489 | extern void sched_set_stop_task(int cpu, struct task_struct *stop); |
490 | ||
14e568e7 TG |
491 | static void cpu_stop_create(unsigned int cpu) |
492 | { | |
02cb7aa9 | 493 | sched_set_stop_task(cpu, per_cpu(cpu_stopper.thread, cpu)); |
14e568e7 TG |
494 | } |
495 | ||
496 | static void cpu_stop_park(unsigned int cpu) | |
1142d810 | 497 | { |
1142d810 | 498 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 | 499 | |
233e7f26 | 500 | WARN_ON(!list_empty(&stopper->works)); |
14e568e7 | 501 | } |
1142d810 | 502 | |
c00166d8 ON |
503 | void stop_machine_unpark(int cpu) |
504 | { | |
505 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
506 | ||
f0cf16cb | 507 | stopper->enabled = true; |
c00166d8 ON |
508 | kthread_unpark(stopper->thread); |
509 | } | |
510 | ||
14e568e7 | 511 | static struct smp_hotplug_thread cpu_stop_threads = { |
02cb7aa9 | 512 | .store = &cpu_stopper.thread, |
14e568e7 TG |
513 | .thread_should_run = cpu_stop_should_run, |
514 | .thread_fn = cpu_stopper_thread, | |
515 | .thread_comm = "migration/%u", | |
516 | .create = cpu_stop_create, | |
14e568e7 | 517 | .park = cpu_stop_park, |
14e568e7 | 518 | .selfparking = true, |
1142d810 TH |
519 | }; |
520 | ||
521 | static int __init cpu_stop_init(void) | |
522 | { | |
1142d810 | 523 | unsigned int cpu; |
1142d810 TH |
524 | |
525 | for_each_possible_cpu(cpu) { | |
526 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
527 | ||
528 | spin_lock_init(&stopper->lock); | |
529 | INIT_LIST_HEAD(&stopper->works); | |
530 | } | |
531 | ||
14e568e7 | 532 | BUG_ON(smpboot_register_percpu_thread(&cpu_stop_threads)); |
c00166d8 | 533 | stop_machine_unpark(raw_smp_processor_id()); |
f445027e | 534 | stop_machine_initialized = true; |
1142d810 TH |
535 | return 0; |
536 | } | |
537 | early_initcall(cpu_stop_init); | |
1da177e4 | 538 | |
bbf1bb3e TH |
539 | #ifdef CONFIG_STOP_MACHINE |
540 | ||
9a301f22 | 541 | static int __stop_machine(cpu_stop_fn_t fn, void *data, const struct cpumask *cpus) |
1da177e4 | 542 | { |
1be0bd77 PZ |
543 | struct multi_stop_data msdata = { |
544 | .fn = fn, | |
545 | .data = data, | |
546 | .num_threads = num_online_cpus(), | |
547 | .active_cpus = cpus, | |
548 | }; | |
3fc1f1e2 | 549 | |
f445027e JF |
550 | if (!stop_machine_initialized) { |
551 | /* | |
552 | * Handle the case where stop_machine() is called | |
553 | * early in boot before stop_machine() has been | |
554 | * initialized. | |
555 | */ | |
556 | unsigned long flags; | |
557 | int ret; | |
558 | ||
1be0bd77 | 559 | WARN_ON_ONCE(msdata.num_threads != 1); |
f445027e JF |
560 | |
561 | local_irq_save(flags); | |
562 | hard_irq_disable(); | |
563 | ret = (*fn)(data); | |
564 | local_irq_restore(flags); | |
565 | ||
566 | return ret; | |
567 | } | |
568 | ||
3fc1f1e2 | 569 | /* Set the initial state and stop all online cpus. */ |
1be0bd77 PZ |
570 | set_state(&msdata, MULTI_STOP_PREPARE); |
571 | return stop_cpus(cpu_online_mask, multi_cpu_stop, &msdata); | |
1da177e4 LT |
572 | } |
573 | ||
9a301f22 | 574 | int stop_machine(cpu_stop_fn_t fn, void *data, const struct cpumask *cpus) |
1da177e4 | 575 | { |
1da177e4 LT |
576 | int ret; |
577 | ||
578 | /* No CPUs can come up or down during this. */ | |
86ef5c9a | 579 | get_online_cpus(); |
eeec4fad | 580 | ret = __stop_machine(fn, data, cpus); |
86ef5c9a | 581 | put_online_cpus(); |
1da177e4 LT |
582 | return ret; |
583 | } | |
eeec4fad | 584 | EXPORT_SYMBOL_GPL(stop_machine); |
bbf1bb3e | 585 | |
f740e6cd TH |
586 | /** |
587 | * stop_machine_from_inactive_cpu - stop_machine() from inactive CPU | |
588 | * @fn: the function to run | |
589 | * @data: the data ptr for the @fn() | |
590 | * @cpus: the cpus to run the @fn() on (NULL = any online cpu) | |
591 | * | |
592 | * This is identical to stop_machine() but can be called from a CPU which | |
593 | * is not active. The local CPU is in the process of hotplug (so no other | |
594 | * CPU hotplug can start) and not marked active and doesn't have enough | |
595 | * context to sleep. | |
596 | * | |
597 | * This function provides stop_machine() functionality for such state by | |
598 | * using busy-wait for synchronization and executing @fn directly for local | |
599 | * CPU. | |
600 | * | |
601 | * CONTEXT: | |
602 | * Local CPU is inactive. Temporarily stops all active CPUs. | |
603 | * | |
604 | * RETURNS: | |
605 | * 0 if all executions of @fn returned 0, any non zero return value if any | |
606 | * returned non zero. | |
607 | */ | |
9a301f22 | 608 | int stop_machine_from_inactive_cpu(cpu_stop_fn_t fn, void *data, |
f740e6cd TH |
609 | const struct cpumask *cpus) |
610 | { | |
1be0bd77 | 611 | struct multi_stop_data msdata = { .fn = fn, .data = data, |
f740e6cd TH |
612 | .active_cpus = cpus }; |
613 | struct cpu_stop_done done; | |
614 | int ret; | |
615 | ||
616 | /* Local CPU must be inactive and CPU hotplug in progress. */ | |
617 | BUG_ON(cpu_active(raw_smp_processor_id())); | |
1be0bd77 | 618 | msdata.num_threads = num_active_cpus() + 1; /* +1 for local */ |
f740e6cd TH |
619 | |
620 | /* No proper task established and can't sleep - busy wait for lock. */ | |
621 | while (!mutex_trylock(&stop_cpus_mutex)) | |
622 | cpu_relax(); | |
623 | ||
624 | /* Schedule work on other CPUs and execute directly for local CPU */ | |
1be0bd77 | 625 | set_state(&msdata, MULTI_STOP_PREPARE); |
f740e6cd | 626 | cpu_stop_init_done(&done, num_active_cpus()); |
1be0bd77 | 627 | queue_stop_cpus_work(cpu_active_mask, multi_cpu_stop, &msdata, |
f740e6cd | 628 | &done); |
1be0bd77 | 629 | ret = multi_cpu_stop(&msdata); |
f740e6cd TH |
630 | |
631 | /* Busy wait for completion. */ | |
632 | while (!completion_done(&done.completion)) | |
633 | cpu_relax(); | |
634 | ||
635 | mutex_unlock(&stop_cpus_mutex); | |
636 | return ret ?: done.ret; | |
637 | } | |
638 | ||
bbf1bb3e | 639 | #endif /* CONFIG_STOP_MACHINE */ |