Commit | Line | Data |
---|---|---|
6ff3f917 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1142d810 TH |
2 | /* |
3 | * kernel/stop_machine.c | |
4 | * | |
5 | * Copyright (C) 2008, 2005 IBM Corporation. | |
6 | * Copyright (C) 2008, 2005 Rusty Russell rusty@rustcorp.com.au | |
7 | * Copyright (C) 2010 SUSE Linux Products GmbH | |
8 | * Copyright (C) 2010 Tejun Heo <tj@kernel.org> | |
e5582ca2 | 9 | */ |
1142d810 | 10 | #include <linux/completion.h> |
1da177e4 | 11 | #include <linux/cpu.h> |
1142d810 | 12 | #include <linux/init.h> |
ee527cd3 | 13 | #include <linux/kthread.h> |
9984de1a | 14 | #include <linux/export.h> |
1142d810 | 15 | #include <linux/percpu.h> |
ee527cd3 PB |
16 | #include <linux/sched.h> |
17 | #include <linux/stop_machine.h> | |
a12bb444 | 18 | #include <linux/interrupt.h> |
1142d810 | 19 | #include <linux/kallsyms.h> |
14e568e7 | 20 | #include <linux/smpboot.h> |
60063497 | 21 | #include <linux/atomic.h> |
ce4f06dc | 22 | #include <linux/nmi.h> |
0b26351b | 23 | #include <linux/sched/wake_q.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 */ | |
1142d810 TH |
31 | int ret; /* collected return value */ |
32 | struct completion completion; /* fired if nr_todo reaches 0 */ | |
33 | }; | |
34 | ||
35 | /* the actual stopper, one per every possible cpu, enabled on online cpus */ | |
36 | struct cpu_stopper { | |
02cb7aa9 ON |
37 | struct task_struct *thread; |
38 | ||
de5b55c1 | 39 | raw_spinlock_t lock; |
878ae127 | 40 | bool enabled; /* is this stopper enabled? */ |
1142d810 | 41 | struct list_head works; /* list of pending works */ |
02cb7aa9 ON |
42 | |
43 | struct cpu_stop_work stop_work; /* for stop_cpus */ | |
1142d810 TH |
44 | }; |
45 | ||
46 | static DEFINE_PER_CPU(struct cpu_stopper, cpu_stopper); | |
f445027e | 47 | static bool stop_machine_initialized = false; |
1142d810 | 48 | |
e6253970 ON |
49 | /* static data for stop_cpus */ |
50 | static DEFINE_MUTEX(stop_cpus_mutex); | |
51 | static bool stop_cpus_in_progress; | |
7053ea1a | 52 | |
1142d810 TH |
53 | static void cpu_stop_init_done(struct cpu_stop_done *done, unsigned int nr_todo) |
54 | { | |
55 | memset(done, 0, sizeof(*done)); | |
56 | atomic_set(&done->nr_todo, nr_todo); | |
57 | init_completion(&done->completion); | |
58 | } | |
59 | ||
60 | /* signal completion unless @done is NULL */ | |
6fa3b826 | 61 | static void cpu_stop_signal_done(struct cpu_stop_done *done) |
1142d810 | 62 | { |
dd2e3121 ON |
63 | if (atomic_dec_and_test(&done->nr_todo)) |
64 | complete(&done->completion); | |
1142d810 TH |
65 | } |
66 | ||
5caa1c08 | 67 | static void __cpu_stop_queue_work(struct cpu_stopper *stopper, |
0b26351b PZ |
68 | struct cpu_stop_work *work, |
69 | struct wake_q_head *wakeq) | |
5caa1c08 ON |
70 | { |
71 | list_add_tail(&work->list, &stopper->works); | |
0b26351b | 72 | wake_q_add(wakeq, stopper->thread); |
5caa1c08 ON |
73 | } |
74 | ||
1142d810 | 75 | /* queue @work to @stopper. if offline, @work is completed immediately */ |
1b034bd9 | 76 | static bool cpu_stop_queue_work(unsigned int cpu, struct cpu_stop_work *work) |
1142d810 | 77 | { |
860a0ffa | 78 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
0b26351b | 79 | DEFINE_WAKE_Q(wakeq); |
1142d810 | 80 | unsigned long flags; |
1b034bd9 | 81 | bool enabled; |
1142d810 | 82 | |
cfd35514 | 83 | preempt_disable(); |
de5b55c1 | 84 | raw_spin_lock_irqsave(&stopper->lock, flags); |
1b034bd9 ON |
85 | enabled = stopper->enabled; |
86 | if (enabled) | |
0b26351b | 87 | __cpu_stop_queue_work(stopper, work, &wakeq); |
dd2e3121 | 88 | else if (work->done) |
6fa3b826 | 89 | cpu_stop_signal_done(work->done); |
de5b55c1 | 90 | raw_spin_unlock_irqrestore(&stopper->lock, flags); |
1b034bd9 | 91 | |
0b26351b | 92 | wake_up_q(&wakeq); |
cfd35514 | 93 | preempt_enable(); |
0b26351b | 94 | |
1b034bd9 | 95 | return enabled; |
1142d810 TH |
96 | } |
97 | ||
98 | /** | |
99 | * stop_one_cpu - stop a cpu | |
100 | * @cpu: cpu to stop | |
101 | * @fn: function to execute | |
102 | * @arg: argument to @fn | |
103 | * | |
104 | * Execute @fn(@arg) on @cpu. @fn is run in a process context with | |
105 | * the highest priority preempting any task on the cpu and | |
106 | * monopolizing it. This function returns after the execution is | |
107 | * complete. | |
108 | * | |
109 | * This function doesn't guarantee @cpu stays online till @fn | |
110 | * completes. If @cpu goes down in the middle, execution may happen | |
111 | * partially or fully on different cpus. @fn should either be ready | |
112 | * for that or the caller should ensure that @cpu stays online until | |
113 | * this function completes. | |
114 | * | |
115 | * CONTEXT: | |
116 | * Might sleep. | |
117 | * | |
118 | * RETURNS: | |
119 | * -ENOENT if @fn(@arg) was not executed because @cpu was offline; | |
120 | * otherwise, the return value of @fn. | |
121 | */ | |
122 | int stop_one_cpu(unsigned int cpu, cpu_stop_fn_t fn, void *arg) | |
123 | { | |
124 | struct cpu_stop_done done; | |
125 | struct cpu_stop_work work = { .fn = fn, .arg = arg, .done = &done }; | |
126 | ||
127 | cpu_stop_init_done(&done, 1); | |
958c5f84 ON |
128 | if (!cpu_stop_queue_work(cpu, &work)) |
129 | return -ENOENT; | |
bf89a304 CC |
130 | /* |
131 | * In case @cpu == smp_proccessor_id() we can avoid a sleep+wakeup | |
132 | * cycle by doing a preemption: | |
133 | */ | |
134 | cond_resched(); | |
1142d810 | 135 | wait_for_completion(&done.completion); |
958c5f84 | 136 | return done.ret; |
1142d810 TH |
137 | } |
138 | ||
1be0bd77 PZ |
139 | /* This controls the threads on each CPU. */ |
140 | enum multi_stop_state { | |
141 | /* Dummy starting state for thread. */ | |
142 | MULTI_STOP_NONE, | |
143 | /* Awaiting everyone to be scheduled. */ | |
144 | MULTI_STOP_PREPARE, | |
145 | /* Disable interrupts. */ | |
146 | MULTI_STOP_DISABLE_IRQ, | |
147 | /* Run the function */ | |
148 | MULTI_STOP_RUN, | |
149 | /* Exit */ | |
150 | MULTI_STOP_EXIT, | |
151 | }; | |
152 | ||
153 | struct multi_stop_data { | |
9a301f22 | 154 | cpu_stop_fn_t fn; |
1be0bd77 PZ |
155 | void *data; |
156 | /* Like num_online_cpus(), but hotplug cpu uses us, so we need this. */ | |
157 | unsigned int num_threads; | |
158 | const struct cpumask *active_cpus; | |
159 | ||
160 | enum multi_stop_state state; | |
161 | atomic_t thread_ack; | |
162 | }; | |
163 | ||
164 | static void set_state(struct multi_stop_data *msdata, | |
165 | enum multi_stop_state newstate) | |
166 | { | |
167 | /* Reset ack counter. */ | |
168 | atomic_set(&msdata->thread_ack, msdata->num_threads); | |
169 | smp_wmb(); | |
170 | msdata->state = newstate; | |
171 | } | |
172 | ||
173 | /* Last one to ack a state moves to the next state. */ | |
174 | static void ack_state(struct multi_stop_data *msdata) | |
175 | { | |
176 | if (atomic_dec_and_test(&msdata->thread_ack)) | |
177 | set_state(msdata, msdata->state + 1); | |
178 | } | |
179 | ||
4ecf0a43 HC |
180 | void __weak stop_machine_yield(const struct cpumask *cpumask) |
181 | { | |
182 | cpu_relax(); | |
183 | } | |
184 | ||
1be0bd77 PZ |
185 | /* This is the cpu_stop function which stops the CPU. */ |
186 | static int multi_cpu_stop(void *data) | |
187 | { | |
188 | struct multi_stop_data *msdata = data; | |
189 | enum multi_stop_state curstate = MULTI_STOP_NONE; | |
190 | int cpu = smp_processor_id(), err = 0; | |
38f2c691 | 191 | const struct cpumask *cpumask; |
1be0bd77 PZ |
192 | unsigned long flags; |
193 | bool is_active; | |
194 | ||
195 | /* | |
196 | * When called from stop_machine_from_inactive_cpu(), irq might | |
197 | * already be disabled. Save the state and restore it on exit. | |
198 | */ | |
199 | local_save_flags(flags); | |
200 | ||
38f2c691 MS |
201 | if (!msdata->active_cpus) { |
202 | cpumask = cpu_online_mask; | |
203 | is_active = cpu == cpumask_first(cpumask); | |
204 | } else { | |
205 | cpumask = msdata->active_cpus; | |
206 | is_active = cpumask_test_cpu(cpu, cpumask); | |
207 | } | |
1be0bd77 PZ |
208 | |
209 | /* Simple state machine */ | |
210 | do { | |
211 | /* Chill out and ensure we re-read multi_stop_state. */ | |
4ecf0a43 | 212 | stop_machine_yield(cpumask); |
1be0bd77 PZ |
213 | if (msdata->state != curstate) { |
214 | curstate = msdata->state; | |
215 | switch (curstate) { | |
216 | case MULTI_STOP_DISABLE_IRQ: | |
217 | local_irq_disable(); | |
218 | hard_irq_disable(); | |
219 | break; | |
220 | case MULTI_STOP_RUN: | |
221 | if (is_active) | |
222 | err = msdata->fn(msdata->data); | |
223 | break; | |
224 | default: | |
225 | break; | |
226 | } | |
227 | ack_state(msdata); | |
ce4f06dc ON |
228 | } else if (curstate > MULTI_STOP_PREPARE) { |
229 | /* | |
230 | * At this stage all other CPUs we depend on must spin | |
231 | * in the same loop. Any reason for hard-lockup should | |
232 | * be detected and reported on their side. | |
233 | */ | |
234 | touch_nmi_watchdog(); | |
1be0bd77 PZ |
235 | } |
236 | } while (curstate != MULTI_STOP_EXIT); | |
237 | ||
238 | local_irq_restore(flags); | |
239 | return err; | |
240 | } | |
241 | ||
5caa1c08 ON |
242 | static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1, |
243 | int cpu2, struct cpu_stop_work *work2) | |
244 | { | |
d8bc8535 ON |
245 | struct cpu_stopper *stopper1 = per_cpu_ptr(&cpu_stopper, cpu1); |
246 | struct cpu_stopper *stopper2 = per_cpu_ptr(&cpu_stopper, cpu2); | |
0b26351b | 247 | DEFINE_WAKE_Q(wakeq); |
d8bc8535 | 248 | int err; |
b80a2bfc | 249 | |
e6253970 | 250 | retry: |
b80a2bfc PZ |
251 | /* |
252 | * The waking up of stopper threads has to happen in the same | |
253 | * scheduling context as the queueing. Otherwise, there is a | |
254 | * possibility of one of the above stoppers being woken up by another | |
255 | * CPU, and preempting us. This will cause us to not wake up the other | |
256 | * stopper forever. | |
257 | */ | |
258 | preempt_disable(); | |
de5b55c1 TG |
259 | raw_spin_lock_irq(&stopper1->lock); |
260 | raw_spin_lock_nested(&stopper2->lock, SINGLE_DEPTH_NESTING); | |
d8bc8535 | 261 | |
b80a2bfc PZ |
262 | if (!stopper1->enabled || !stopper2->enabled) { |
263 | err = -ENOENT; | |
d8bc8535 | 264 | goto unlock; |
b80a2bfc PZ |
265 | } |
266 | ||
e6253970 ON |
267 | /* |
268 | * Ensure that if we race with __stop_cpus() the stoppers won't get | |
269 | * queued up in reverse order leading to system deadlock. | |
270 | * | |
271 | * We can't miss stop_cpus_in_progress if queue_stop_cpus_work() has | |
272 | * queued a work on cpu1 but not on cpu2, we hold both locks. | |
273 | * | |
274 | * It can be falsely true but it is safe to spin until it is cleared, | |
275 | * queue_stop_cpus_work() does everything under preempt_disable(). | |
276 | */ | |
b80a2bfc PZ |
277 | if (unlikely(stop_cpus_in_progress)) { |
278 | err = -EDEADLK; | |
279 | goto unlock; | |
280 | } | |
d8bc8535 ON |
281 | |
282 | err = 0; | |
0b26351b PZ |
283 | __cpu_stop_queue_work(stopper1, work1, &wakeq); |
284 | __cpu_stop_queue_work(stopper2, work2, &wakeq); | |
b80a2bfc | 285 | |
d8bc8535 | 286 | unlock: |
de5b55c1 TG |
287 | raw_spin_unlock(&stopper2->lock); |
288 | raw_spin_unlock_irq(&stopper1->lock); | |
5caa1c08 | 289 | |
e6253970 | 290 | if (unlikely(err == -EDEADLK)) { |
b80a2bfc PZ |
291 | preempt_enable(); |
292 | ||
e6253970 ON |
293 | while (stop_cpus_in_progress) |
294 | cpu_relax(); | |
b80a2bfc | 295 | |
e6253970 ON |
296 | goto retry; |
297 | } | |
0b26351b | 298 | |
b80a2bfc PZ |
299 | wake_up_q(&wakeq); |
300 | preempt_enable(); | |
0b26351b | 301 | |
d8bc8535 | 302 | return err; |
5caa1c08 | 303 | } |
1be0bd77 PZ |
304 | /** |
305 | * stop_two_cpus - stops two cpus | |
306 | * @cpu1: the cpu to stop | |
307 | * @cpu2: the other cpu to stop | |
308 | * @fn: function to execute | |
309 | * @arg: argument to @fn | |
310 | * | |
311 | * Stops both the current and specified CPU and runs @fn on one of them. | |
312 | * | |
313 | * returns when both are completed. | |
314 | */ | |
315 | int stop_two_cpus(unsigned int cpu1, unsigned int cpu2, cpu_stop_fn_t fn, void *arg) | |
316 | { | |
1be0bd77 PZ |
317 | struct cpu_stop_done done; |
318 | struct cpu_stop_work work1, work2; | |
6acce3ef PZ |
319 | struct multi_stop_data msdata; |
320 | ||
6acce3ef | 321 | msdata = (struct multi_stop_data){ |
1be0bd77 PZ |
322 | .fn = fn, |
323 | .data = arg, | |
324 | .num_threads = 2, | |
325 | .active_cpus = cpumask_of(cpu1), | |
326 | }; | |
327 | ||
328 | work1 = work2 = (struct cpu_stop_work){ | |
329 | .fn = multi_cpu_stop, | |
330 | .arg = &msdata, | |
331 | .done = &done | |
332 | }; | |
333 | ||
1be0bd77 PZ |
334 | cpu_stop_init_done(&done, 2); |
335 | set_state(&msdata, MULTI_STOP_PREPARE); | |
336 | ||
5caa1c08 ON |
337 | if (cpu1 > cpu2) |
338 | swap(cpu1, cpu2); | |
6a190051 | 339 | if (cpu_stop_queue_two_works(cpu1, &work1, cpu2, &work2)) |
5caa1c08 | 340 | return -ENOENT; |
1be0bd77 PZ |
341 | |
342 | wait_for_completion(&done.completion); | |
6a190051 | 343 | return done.ret; |
1be0bd77 PZ |
344 | } |
345 | ||
1142d810 TH |
346 | /** |
347 | * stop_one_cpu_nowait - stop a cpu but don't wait for completion | |
348 | * @cpu: cpu to stop | |
349 | * @fn: function to execute | |
350 | * @arg: argument to @fn | |
cf250040 | 351 | * @work_buf: pointer to cpu_stop_work structure |
1142d810 TH |
352 | * |
353 | * Similar to stop_one_cpu() but doesn't wait for completion. The | |
354 | * caller is responsible for ensuring @work_buf is currently unused | |
355 | * and will remain untouched until stopper starts executing @fn. | |
356 | * | |
357 | * CONTEXT: | |
358 | * Don't care. | |
1b034bd9 ON |
359 | * |
360 | * RETURNS: | |
361 | * true if cpu_stop_work was queued successfully and @fn will be called, | |
362 | * false otherwise. | |
1142d810 | 363 | */ |
1b034bd9 | 364 | bool stop_one_cpu_nowait(unsigned int cpu, cpu_stop_fn_t fn, void *arg, |
1142d810 TH |
365 | struct cpu_stop_work *work_buf) |
366 | { | |
367 | *work_buf = (struct cpu_stop_work){ .fn = fn, .arg = arg, }; | |
1b034bd9 | 368 | return cpu_stop_queue_work(cpu, work_buf); |
1142d810 TH |
369 | } |
370 | ||
4aff1ca6 | 371 | static bool queue_stop_cpus_work(const struct cpumask *cpumask, |
fd7355ba TH |
372 | cpu_stop_fn_t fn, void *arg, |
373 | struct cpu_stop_done *done) | |
1142d810 TH |
374 | { |
375 | struct cpu_stop_work *work; | |
1142d810 | 376 | unsigned int cpu; |
4aff1ca6 | 377 | bool queued = false; |
1142d810 | 378 | |
1142d810 TH |
379 | /* |
380 | * Disable preemption while queueing to avoid getting | |
381 | * preempted by a stopper which might wait for other stoppers | |
382 | * to enter @fn which can lead to deadlock. | |
383 | */ | |
e6253970 ON |
384 | preempt_disable(); |
385 | stop_cpus_in_progress = true; | |
b377c2a0 ON |
386 | for_each_cpu(cpu, cpumask) { |
387 | work = &per_cpu(cpu_stopper.stop_work, cpu); | |
388 | work->fn = fn; | |
389 | work->arg = arg; | |
390 | work->done = done; | |
4aff1ca6 ON |
391 | if (cpu_stop_queue_work(cpu, work)) |
392 | queued = true; | |
b377c2a0 | 393 | } |
e6253970 ON |
394 | stop_cpus_in_progress = false; |
395 | preempt_enable(); | |
4aff1ca6 ON |
396 | |
397 | return queued; | |
fd7355ba | 398 | } |
1142d810 | 399 | |
fd7355ba TH |
400 | static int __stop_cpus(const struct cpumask *cpumask, |
401 | cpu_stop_fn_t fn, void *arg) | |
402 | { | |
403 | struct cpu_stop_done done; | |
404 | ||
405 | cpu_stop_init_done(&done, cpumask_weight(cpumask)); | |
4aff1ca6 ON |
406 | if (!queue_stop_cpus_work(cpumask, fn, arg, &done)) |
407 | return -ENOENT; | |
1142d810 | 408 | wait_for_completion(&done.completion); |
4aff1ca6 | 409 | return done.ret; |
1142d810 TH |
410 | } |
411 | ||
412 | /** | |
413 | * stop_cpus - stop multiple cpus | |
414 | * @cpumask: cpus to stop | |
415 | * @fn: function to execute | |
416 | * @arg: argument to @fn | |
417 | * | |
418 | * Execute @fn(@arg) on online cpus in @cpumask. On each target cpu, | |
419 | * @fn is run in a process context with the highest priority | |
420 | * preempting any task on the cpu and monopolizing it. This function | |
421 | * returns after all executions are complete. | |
422 | * | |
423 | * This function doesn't guarantee the cpus in @cpumask stay online | |
424 | * till @fn completes. If some cpus go down in the middle, execution | |
425 | * on the cpu may happen partially or fully on different cpus. @fn | |
426 | * should either be ready for that or the caller should ensure that | |
427 | * the cpus stay online until this function completes. | |
428 | * | |
429 | * All stop_cpus() calls are serialized making it safe for @fn to wait | |
430 | * for all cpus to start executing it. | |
431 | * | |
432 | * CONTEXT: | |
433 | * Might sleep. | |
434 | * | |
435 | * RETURNS: | |
436 | * -ENOENT if @fn(@arg) was not executed at all because all cpus in | |
437 | * @cpumask were offline; otherwise, 0 if all executions of @fn | |
438 | * returned 0, any non zero return value if any returned non zero. | |
439 | */ | |
440 | int stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) | |
441 | { | |
442 | int ret; | |
443 | ||
444 | /* static works are used, process one request at a time */ | |
445 | mutex_lock(&stop_cpus_mutex); | |
446 | ret = __stop_cpus(cpumask, fn, arg); | |
447 | mutex_unlock(&stop_cpus_mutex); | |
448 | return ret; | |
449 | } | |
450 | ||
451 | /** | |
452 | * try_stop_cpus - try to stop multiple cpus | |
453 | * @cpumask: cpus to stop | |
454 | * @fn: function to execute | |
455 | * @arg: argument to @fn | |
456 | * | |
457 | * Identical to stop_cpus() except that it fails with -EAGAIN if | |
458 | * someone else is already using the facility. | |
459 | * | |
460 | * CONTEXT: | |
461 | * Might sleep. | |
462 | * | |
463 | * RETURNS: | |
464 | * -EAGAIN if someone else is already stopping cpus, -ENOENT if | |
465 | * @fn(@arg) was not executed at all because all cpus in @cpumask were | |
466 | * offline; otherwise, 0 if all executions of @fn returned 0, any non | |
467 | * zero return value if any returned non zero. | |
468 | */ | |
469 | int try_stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) | |
470 | { | |
471 | int ret; | |
472 | ||
473 | /* static works are used, process one request at a time */ | |
474 | if (!mutex_trylock(&stop_cpus_mutex)) | |
475 | return -EAGAIN; | |
476 | ret = __stop_cpus(cpumask, fn, arg); | |
477 | mutex_unlock(&stop_cpus_mutex); | |
478 | return ret; | |
479 | } | |
480 | ||
14e568e7 TG |
481 | static int cpu_stop_should_run(unsigned int cpu) |
482 | { | |
483 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
484 | unsigned long flags; | |
485 | int run; | |
486 | ||
de5b55c1 | 487 | raw_spin_lock_irqsave(&stopper->lock, flags); |
14e568e7 | 488 | run = !list_empty(&stopper->works); |
de5b55c1 | 489 | raw_spin_unlock_irqrestore(&stopper->lock, flags); |
14e568e7 TG |
490 | return run; |
491 | } | |
492 | ||
493 | static void cpu_stopper_thread(unsigned int cpu) | |
1142d810 | 494 | { |
14e568e7 | 495 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 | 496 | struct cpu_stop_work *work; |
1142d810 TH |
497 | |
498 | repeat: | |
1142d810 | 499 | work = NULL; |
de5b55c1 | 500 | raw_spin_lock_irq(&stopper->lock); |
1142d810 TH |
501 | if (!list_empty(&stopper->works)) { |
502 | work = list_first_entry(&stopper->works, | |
503 | struct cpu_stop_work, list); | |
504 | list_del_init(&work->list); | |
505 | } | |
de5b55c1 | 506 | raw_spin_unlock_irq(&stopper->lock); |
1142d810 TH |
507 | |
508 | if (work) { | |
509 | cpu_stop_fn_t fn = work->fn; | |
510 | void *arg = work->arg; | |
511 | struct cpu_stop_done *done = work->done; | |
accaf6ea | 512 | int ret; |
1142d810 | 513 | |
accaf6ea ON |
514 | /* cpu stop callbacks must not sleep, make in_atomic() == T */ |
515 | preempt_count_inc(); | |
1142d810 | 516 | ret = fn(arg); |
dd2e3121 ON |
517 | if (done) { |
518 | if (ret) | |
519 | done->ret = ret; | |
520 | cpu_stop_signal_done(done); | |
521 | } | |
accaf6ea | 522 | preempt_count_dec(); |
1142d810 | 523 | WARN_ONCE(preempt_count(), |
d75f773c | 524 | "cpu_stop: %ps(%p) leaked preempt count\n", fn, arg); |
14e568e7 TG |
525 | goto repeat; |
526 | } | |
1142d810 TH |
527 | } |
528 | ||
233e7f26 ON |
529 | void stop_machine_park(int cpu) |
530 | { | |
531 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
532 | /* | |
533 | * Lockless. cpu_stopper_thread() will take stopper->lock and flush | |
534 | * the pending works before it parks, until then it is fine to queue | |
535 | * the new works. | |
536 | */ | |
537 | stopper->enabled = false; | |
538 | kthread_park(stopper->thread); | |
539 | } | |
540 | ||
34f971f6 PZ |
541 | extern void sched_set_stop_task(int cpu, struct task_struct *stop); |
542 | ||
14e568e7 TG |
543 | static void cpu_stop_create(unsigned int cpu) |
544 | { | |
02cb7aa9 | 545 | sched_set_stop_task(cpu, per_cpu(cpu_stopper.thread, cpu)); |
14e568e7 TG |
546 | } |
547 | ||
548 | static void cpu_stop_park(unsigned int cpu) | |
1142d810 | 549 | { |
1142d810 | 550 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 | 551 | |
233e7f26 | 552 | WARN_ON(!list_empty(&stopper->works)); |
14e568e7 | 553 | } |
1142d810 | 554 | |
c00166d8 ON |
555 | void stop_machine_unpark(int cpu) |
556 | { | |
557 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
558 | ||
f0cf16cb | 559 | stopper->enabled = true; |
c00166d8 ON |
560 | kthread_unpark(stopper->thread); |
561 | } | |
562 | ||
14e568e7 | 563 | static struct smp_hotplug_thread cpu_stop_threads = { |
02cb7aa9 | 564 | .store = &cpu_stopper.thread, |
14e568e7 TG |
565 | .thread_should_run = cpu_stop_should_run, |
566 | .thread_fn = cpu_stopper_thread, | |
567 | .thread_comm = "migration/%u", | |
568 | .create = cpu_stop_create, | |
14e568e7 | 569 | .park = cpu_stop_park, |
14e568e7 | 570 | .selfparking = true, |
1142d810 TH |
571 | }; |
572 | ||
573 | static int __init cpu_stop_init(void) | |
574 | { | |
1142d810 | 575 | unsigned int cpu; |
1142d810 TH |
576 | |
577 | for_each_possible_cpu(cpu) { | |
578 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
579 | ||
de5b55c1 | 580 | raw_spin_lock_init(&stopper->lock); |
1142d810 TH |
581 | INIT_LIST_HEAD(&stopper->works); |
582 | } | |
583 | ||
14e568e7 | 584 | BUG_ON(smpboot_register_percpu_thread(&cpu_stop_threads)); |
c00166d8 | 585 | stop_machine_unpark(raw_smp_processor_id()); |
f445027e | 586 | stop_machine_initialized = true; |
1142d810 TH |
587 | return 0; |
588 | } | |
589 | early_initcall(cpu_stop_init); | |
1da177e4 | 590 | |
fe5595c0 SAS |
591 | int stop_machine_cpuslocked(cpu_stop_fn_t fn, void *data, |
592 | const struct cpumask *cpus) | |
1da177e4 | 593 | { |
1be0bd77 PZ |
594 | struct multi_stop_data msdata = { |
595 | .fn = fn, | |
596 | .data = data, | |
597 | .num_threads = num_online_cpus(), | |
598 | .active_cpus = cpus, | |
599 | }; | |
3fc1f1e2 | 600 | |
fe5595c0 SAS |
601 | lockdep_assert_cpus_held(); |
602 | ||
f445027e JF |
603 | if (!stop_machine_initialized) { |
604 | /* | |
605 | * Handle the case where stop_machine() is called | |
606 | * early in boot before stop_machine() has been | |
607 | * initialized. | |
608 | */ | |
609 | unsigned long flags; | |
610 | int ret; | |
611 | ||
1be0bd77 | 612 | WARN_ON_ONCE(msdata.num_threads != 1); |
f445027e JF |
613 | |
614 | local_irq_save(flags); | |
615 | hard_irq_disable(); | |
616 | ret = (*fn)(data); | |
617 | local_irq_restore(flags); | |
618 | ||
619 | return ret; | |
620 | } | |
621 | ||
3fc1f1e2 | 622 | /* Set the initial state and stop all online cpus. */ |
1be0bd77 PZ |
623 | set_state(&msdata, MULTI_STOP_PREPARE); |
624 | return stop_cpus(cpu_online_mask, multi_cpu_stop, &msdata); | |
1da177e4 LT |
625 | } |
626 | ||
9a301f22 | 627 | int stop_machine(cpu_stop_fn_t fn, void *data, const struct cpumask *cpus) |
1da177e4 | 628 | { |
1da177e4 LT |
629 | int ret; |
630 | ||
631 | /* No CPUs can come up or down during this. */ | |
fe5595c0 SAS |
632 | cpus_read_lock(); |
633 | ret = stop_machine_cpuslocked(fn, data, cpus); | |
634 | cpus_read_unlock(); | |
1da177e4 LT |
635 | return ret; |
636 | } | |
eeec4fad | 637 | EXPORT_SYMBOL_GPL(stop_machine); |
bbf1bb3e | 638 | |
f740e6cd TH |
639 | /** |
640 | * stop_machine_from_inactive_cpu - stop_machine() from inactive CPU | |
641 | * @fn: the function to run | |
642 | * @data: the data ptr for the @fn() | |
643 | * @cpus: the cpus to run the @fn() on (NULL = any online cpu) | |
644 | * | |
645 | * This is identical to stop_machine() but can be called from a CPU which | |
646 | * is not active. The local CPU is in the process of hotplug (so no other | |
647 | * CPU hotplug can start) and not marked active and doesn't have enough | |
648 | * context to sleep. | |
649 | * | |
650 | * This function provides stop_machine() functionality for such state by | |
651 | * using busy-wait for synchronization and executing @fn directly for local | |
652 | * CPU. | |
653 | * | |
654 | * CONTEXT: | |
655 | * Local CPU is inactive. Temporarily stops all active CPUs. | |
656 | * | |
657 | * RETURNS: | |
658 | * 0 if all executions of @fn returned 0, any non zero return value if any | |
659 | * returned non zero. | |
660 | */ | |
9a301f22 | 661 | int stop_machine_from_inactive_cpu(cpu_stop_fn_t fn, void *data, |
f740e6cd TH |
662 | const struct cpumask *cpus) |
663 | { | |
1be0bd77 | 664 | struct multi_stop_data msdata = { .fn = fn, .data = data, |
f740e6cd TH |
665 | .active_cpus = cpus }; |
666 | struct cpu_stop_done done; | |
667 | int ret; | |
668 | ||
669 | /* Local CPU must be inactive and CPU hotplug in progress. */ | |
670 | BUG_ON(cpu_active(raw_smp_processor_id())); | |
1be0bd77 | 671 | msdata.num_threads = num_active_cpus() + 1; /* +1 for local */ |
f740e6cd TH |
672 | |
673 | /* No proper task established and can't sleep - busy wait for lock. */ | |
674 | while (!mutex_trylock(&stop_cpus_mutex)) | |
675 | cpu_relax(); | |
676 | ||
677 | /* Schedule work on other CPUs and execute directly for local CPU */ | |
1be0bd77 | 678 | set_state(&msdata, MULTI_STOP_PREPARE); |
f740e6cd | 679 | cpu_stop_init_done(&done, num_active_cpus()); |
1be0bd77 | 680 | queue_stop_cpus_work(cpu_active_mask, multi_cpu_stop, &msdata, |
f740e6cd | 681 | &done); |
1be0bd77 | 682 | ret = multi_cpu_stop(&msdata); |
f740e6cd TH |
683 | |
684 | /* Busy wait for completion. */ | |
685 | while (!completion_done(&done.completion)) | |
686 | cpu_relax(); | |
687 | ||
688 | mutex_unlock(&stop_cpus_mutex); | |
689 | return ret ?: done.ret; | |
690 | } |