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 | */ |
b1fc5833 | 10 | #include <linux/compiler.h> |
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> |
ce4f06dc | 23 | #include <linux/nmi.h> |
0b26351b | 24 | #include <linux/sched/wake_q.h> |
1142d810 TH |
25 | |
26 | /* | |
27 | * Structure to determine completion condition and record errors. May | |
28 | * be shared by works on different cpus. | |
29 | */ | |
30 | struct cpu_stop_done { | |
31 | atomic_t nr_todo; /* nr left to execute */ | |
1142d810 TH |
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 | ||
de5b55c1 | 40 | raw_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 | |
e6253970 ON |
50 | /* static data for stop_cpus */ |
51 | static DEFINE_MUTEX(stop_cpus_mutex); | |
52 | static bool stop_cpus_in_progress; | |
7053ea1a | 53 | |
1142d810 TH |
54 | static void cpu_stop_init_done(struct cpu_stop_done *done, unsigned int nr_todo) |
55 | { | |
56 | memset(done, 0, sizeof(*done)); | |
57 | atomic_set(&done->nr_todo, nr_todo); | |
58 | init_completion(&done->completion); | |
59 | } | |
60 | ||
61 | /* signal completion unless @done is NULL */ | |
6fa3b826 | 62 | static void cpu_stop_signal_done(struct cpu_stop_done *done) |
1142d810 | 63 | { |
dd2e3121 ON |
64 | if (atomic_dec_and_test(&done->nr_todo)) |
65 | complete(&done->completion); | |
1142d810 TH |
66 | } |
67 | ||
5caa1c08 | 68 | static void __cpu_stop_queue_work(struct cpu_stopper *stopper, |
0b26351b PZ |
69 | struct cpu_stop_work *work, |
70 | struct wake_q_head *wakeq) | |
5caa1c08 ON |
71 | { |
72 | list_add_tail(&work->list, &stopper->works); | |
0b26351b | 73 | wake_q_add(wakeq, stopper->thread); |
5caa1c08 ON |
74 | } |
75 | ||
1142d810 | 76 | /* queue @work to @stopper. if offline, @work is completed immediately */ |
1b034bd9 | 77 | static bool cpu_stop_queue_work(unsigned int cpu, struct cpu_stop_work *work) |
1142d810 | 78 | { |
860a0ffa | 79 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
0b26351b | 80 | DEFINE_WAKE_Q(wakeq); |
1142d810 | 81 | unsigned long flags; |
1b034bd9 | 82 | bool enabled; |
1142d810 | 83 | |
cfd35514 | 84 | preempt_disable(); |
de5b55c1 | 85 | raw_spin_lock_irqsave(&stopper->lock, flags); |
1b034bd9 ON |
86 | enabled = stopper->enabled; |
87 | if (enabled) | |
0b26351b | 88 | __cpu_stop_queue_work(stopper, work, &wakeq); |
dd2e3121 | 89 | else if (work->done) |
6fa3b826 | 90 | cpu_stop_signal_done(work->done); |
de5b55c1 | 91 | raw_spin_unlock_irqrestore(&stopper->lock, flags); |
1b034bd9 | 92 | |
0b26351b | 93 | wake_up_q(&wakeq); |
cfd35514 | 94 | preempt_enable(); |
0b26351b | 95 | |
1b034bd9 | 96 | return enabled; |
1142d810 TH |
97 | } |
98 | ||
99 | /** | |
100 | * stop_one_cpu - stop a cpu | |
101 | * @cpu: cpu to stop | |
102 | * @fn: function to execute | |
103 | * @arg: argument to @fn | |
104 | * | |
105 | * Execute @fn(@arg) on @cpu. @fn is run in a process context with | |
106 | * the highest priority preempting any task on the cpu and | |
107 | * monopolizing it. This function returns after the execution is | |
108 | * complete. | |
109 | * | |
110 | * This function doesn't guarantee @cpu stays online till @fn | |
111 | * completes. If @cpu goes down in the middle, execution may happen | |
112 | * partially or fully on different cpus. @fn should either be ready | |
113 | * for that or the caller should ensure that @cpu stays online until | |
114 | * this function completes. | |
115 | * | |
116 | * CONTEXT: | |
117 | * Might sleep. | |
118 | * | |
119 | * RETURNS: | |
120 | * -ENOENT if @fn(@arg) was not executed because @cpu was offline; | |
121 | * otherwise, the return value of @fn. | |
122 | */ | |
123 | int stop_one_cpu(unsigned int cpu, cpu_stop_fn_t fn, void *arg) | |
124 | { | |
125 | struct cpu_stop_done done; | |
126 | struct cpu_stop_work work = { .fn = fn, .arg = arg, .done = &done }; | |
127 | ||
128 | cpu_stop_init_done(&done, 1); | |
958c5f84 ON |
129 | if (!cpu_stop_queue_work(cpu, &work)) |
130 | return -ENOENT; | |
bf89a304 CC |
131 | /* |
132 | * In case @cpu == smp_proccessor_id() we can avoid a sleep+wakeup | |
133 | * cycle by doing a preemption: | |
134 | */ | |
135 | cond_resched(); | |
1142d810 | 136 | wait_for_completion(&done.completion); |
958c5f84 | 137 | return done.ret; |
1142d810 TH |
138 | } |
139 | ||
1be0bd77 PZ |
140 | /* This controls the threads on each CPU. */ |
141 | enum multi_stop_state { | |
142 | /* Dummy starting state for thread. */ | |
143 | MULTI_STOP_NONE, | |
144 | /* Awaiting everyone to be scheduled. */ | |
145 | MULTI_STOP_PREPARE, | |
146 | /* Disable interrupts. */ | |
147 | MULTI_STOP_DISABLE_IRQ, | |
148 | /* Run the function */ | |
149 | MULTI_STOP_RUN, | |
150 | /* Exit */ | |
151 | MULTI_STOP_EXIT, | |
152 | }; | |
153 | ||
154 | struct multi_stop_data { | |
9a301f22 | 155 | cpu_stop_fn_t fn; |
1be0bd77 PZ |
156 | void *data; |
157 | /* Like num_online_cpus(), but hotplug cpu uses us, so we need this. */ | |
158 | unsigned int num_threads; | |
159 | const struct cpumask *active_cpus; | |
160 | ||
161 | enum multi_stop_state state; | |
162 | atomic_t thread_ack; | |
163 | }; | |
164 | ||
165 | static void set_state(struct multi_stop_data *msdata, | |
166 | enum multi_stop_state newstate) | |
167 | { | |
168 | /* Reset ack counter. */ | |
169 | atomic_set(&msdata->thread_ack, msdata->num_threads); | |
170 | smp_wmb(); | |
b1fc5833 | 171 | WRITE_ONCE(msdata->state, newstate); |
1be0bd77 PZ |
172 | } |
173 | ||
174 | /* Last one to ack a state moves to the next state. */ | |
175 | static void ack_state(struct multi_stop_data *msdata) | |
176 | { | |
177 | if (atomic_dec_and_test(&msdata->thread_ack)) | |
178 | set_state(msdata, msdata->state + 1); | |
179 | } | |
180 | ||
4ecf0a43 HC |
181 | void __weak stop_machine_yield(const struct cpumask *cpumask) |
182 | { | |
183 | cpu_relax(); | |
184 | } | |
185 | ||
1be0bd77 PZ |
186 | /* This is the cpu_stop function which stops the CPU. */ |
187 | static int multi_cpu_stop(void *data) | |
188 | { | |
189 | struct multi_stop_data *msdata = data; | |
b1fc5833 | 190 | enum multi_stop_state newstate, curstate = MULTI_STOP_NONE; |
1be0bd77 | 191 | int cpu = smp_processor_id(), err = 0; |
38f2c691 | 192 | const struct cpumask *cpumask; |
1be0bd77 PZ |
193 | unsigned long flags; |
194 | bool is_active; | |
195 | ||
196 | /* | |
197 | * When called from stop_machine_from_inactive_cpu(), irq might | |
198 | * already be disabled. Save the state and restore it on exit. | |
199 | */ | |
200 | local_save_flags(flags); | |
201 | ||
38f2c691 MS |
202 | if (!msdata->active_cpus) { |
203 | cpumask = cpu_online_mask; | |
204 | is_active = cpu == cpumask_first(cpumask); | |
205 | } else { | |
206 | cpumask = msdata->active_cpus; | |
207 | is_active = cpumask_test_cpu(cpu, cpumask); | |
208 | } | |
1be0bd77 PZ |
209 | |
210 | /* Simple state machine */ | |
211 | do { | |
212 | /* Chill out and ensure we re-read multi_stop_state. */ | |
4ecf0a43 | 213 | stop_machine_yield(cpumask); |
b1fc5833 MR |
214 | newstate = READ_ONCE(msdata->state); |
215 | if (newstate != curstate) { | |
216 | curstate = newstate; | |
1be0bd77 PZ |
217 | switch (curstate) { |
218 | case MULTI_STOP_DISABLE_IRQ: | |
219 | local_irq_disable(); | |
220 | hard_irq_disable(); | |
221 | break; | |
222 | case MULTI_STOP_RUN: | |
223 | if (is_active) | |
224 | err = msdata->fn(msdata->data); | |
225 | break; | |
226 | default: | |
227 | break; | |
228 | } | |
229 | ack_state(msdata); | |
ce4f06dc ON |
230 | } else if (curstate > MULTI_STOP_PREPARE) { |
231 | /* | |
232 | * At this stage all other CPUs we depend on must spin | |
233 | * in the same loop. Any reason for hard-lockup should | |
234 | * be detected and reported on their side. | |
235 | */ | |
236 | touch_nmi_watchdog(); | |
1be0bd77 | 237 | } |
366237e7 | 238 | rcu_momentary_dyntick_idle(); |
1be0bd77 PZ |
239 | } while (curstate != MULTI_STOP_EXIT); |
240 | ||
241 | local_irq_restore(flags); | |
242 | return err; | |
243 | } | |
244 | ||
5caa1c08 ON |
245 | static int cpu_stop_queue_two_works(int cpu1, struct cpu_stop_work *work1, |
246 | int cpu2, struct cpu_stop_work *work2) | |
247 | { | |
d8bc8535 ON |
248 | struct cpu_stopper *stopper1 = per_cpu_ptr(&cpu_stopper, cpu1); |
249 | struct cpu_stopper *stopper2 = per_cpu_ptr(&cpu_stopper, cpu2); | |
0b26351b | 250 | DEFINE_WAKE_Q(wakeq); |
d8bc8535 | 251 | int err; |
b80a2bfc | 252 | |
e6253970 | 253 | retry: |
b80a2bfc PZ |
254 | /* |
255 | * The waking up of stopper threads has to happen in the same | |
256 | * scheduling context as the queueing. Otherwise, there is a | |
257 | * possibility of one of the above stoppers being woken up by another | |
258 | * CPU, and preempting us. This will cause us to not wake up the other | |
259 | * stopper forever. | |
260 | */ | |
261 | preempt_disable(); | |
de5b55c1 TG |
262 | raw_spin_lock_irq(&stopper1->lock); |
263 | raw_spin_lock_nested(&stopper2->lock, SINGLE_DEPTH_NESTING); | |
d8bc8535 | 264 | |
b80a2bfc PZ |
265 | if (!stopper1->enabled || !stopper2->enabled) { |
266 | err = -ENOENT; | |
d8bc8535 | 267 | goto unlock; |
b80a2bfc PZ |
268 | } |
269 | ||
e6253970 ON |
270 | /* |
271 | * Ensure that if we race with __stop_cpus() the stoppers won't get | |
272 | * queued up in reverse order leading to system deadlock. | |
273 | * | |
274 | * We can't miss stop_cpus_in_progress if queue_stop_cpus_work() has | |
275 | * queued a work on cpu1 but not on cpu2, we hold both locks. | |
276 | * | |
277 | * It can be falsely true but it is safe to spin until it is cleared, | |
278 | * queue_stop_cpus_work() does everything under preempt_disable(). | |
279 | */ | |
b80a2bfc PZ |
280 | if (unlikely(stop_cpus_in_progress)) { |
281 | err = -EDEADLK; | |
282 | goto unlock; | |
283 | } | |
d8bc8535 ON |
284 | |
285 | err = 0; | |
0b26351b PZ |
286 | __cpu_stop_queue_work(stopper1, work1, &wakeq); |
287 | __cpu_stop_queue_work(stopper2, work2, &wakeq); | |
b80a2bfc | 288 | |
d8bc8535 | 289 | unlock: |
de5b55c1 TG |
290 | raw_spin_unlock(&stopper2->lock); |
291 | raw_spin_unlock_irq(&stopper1->lock); | |
5caa1c08 | 292 | |
e6253970 | 293 | if (unlikely(err == -EDEADLK)) { |
b80a2bfc PZ |
294 | preempt_enable(); |
295 | ||
e6253970 ON |
296 | while (stop_cpus_in_progress) |
297 | cpu_relax(); | |
b80a2bfc | 298 | |
e6253970 ON |
299 | goto retry; |
300 | } | |
0b26351b | 301 | |
b80a2bfc PZ |
302 | wake_up_q(&wakeq); |
303 | preempt_enable(); | |
0b26351b | 304 | |
d8bc8535 | 305 | return err; |
5caa1c08 | 306 | } |
1be0bd77 PZ |
307 | /** |
308 | * stop_two_cpus - stops two cpus | |
309 | * @cpu1: the cpu to stop | |
310 | * @cpu2: the other cpu to stop | |
311 | * @fn: function to execute | |
312 | * @arg: argument to @fn | |
313 | * | |
314 | * Stops both the current and specified CPU and runs @fn on one of them. | |
315 | * | |
316 | * returns when both are completed. | |
317 | */ | |
318 | int stop_two_cpus(unsigned int cpu1, unsigned int cpu2, cpu_stop_fn_t fn, void *arg) | |
319 | { | |
1be0bd77 PZ |
320 | struct cpu_stop_done done; |
321 | struct cpu_stop_work work1, work2; | |
6acce3ef PZ |
322 | struct multi_stop_data msdata; |
323 | ||
6acce3ef | 324 | msdata = (struct multi_stop_data){ |
1be0bd77 PZ |
325 | .fn = fn, |
326 | .data = arg, | |
327 | .num_threads = 2, | |
328 | .active_cpus = cpumask_of(cpu1), | |
329 | }; | |
330 | ||
331 | work1 = work2 = (struct cpu_stop_work){ | |
332 | .fn = multi_cpu_stop, | |
333 | .arg = &msdata, | |
334 | .done = &done | |
335 | }; | |
336 | ||
1be0bd77 PZ |
337 | cpu_stop_init_done(&done, 2); |
338 | set_state(&msdata, MULTI_STOP_PREPARE); | |
339 | ||
5caa1c08 ON |
340 | if (cpu1 > cpu2) |
341 | swap(cpu1, cpu2); | |
6a190051 | 342 | if (cpu_stop_queue_two_works(cpu1, &work1, cpu2, &work2)) |
5caa1c08 | 343 | return -ENOENT; |
1be0bd77 PZ |
344 | |
345 | wait_for_completion(&done.completion); | |
6a190051 | 346 | return done.ret; |
1be0bd77 PZ |
347 | } |
348 | ||
1142d810 TH |
349 | /** |
350 | * stop_one_cpu_nowait - stop a cpu but don't wait for completion | |
351 | * @cpu: cpu to stop | |
352 | * @fn: function to execute | |
353 | * @arg: argument to @fn | |
cf250040 | 354 | * @work_buf: pointer to cpu_stop_work structure |
1142d810 TH |
355 | * |
356 | * Similar to stop_one_cpu() but doesn't wait for completion. The | |
357 | * caller is responsible for ensuring @work_buf is currently unused | |
358 | * and will remain untouched until stopper starts executing @fn. | |
359 | * | |
360 | * CONTEXT: | |
361 | * Don't care. | |
1b034bd9 ON |
362 | * |
363 | * RETURNS: | |
364 | * true if cpu_stop_work was queued successfully and @fn will be called, | |
365 | * false otherwise. | |
1142d810 | 366 | */ |
1b034bd9 | 367 | bool stop_one_cpu_nowait(unsigned int cpu, cpu_stop_fn_t fn, void *arg, |
1142d810 TH |
368 | struct cpu_stop_work *work_buf) |
369 | { | |
370 | *work_buf = (struct cpu_stop_work){ .fn = fn, .arg = arg, }; | |
1b034bd9 | 371 | return cpu_stop_queue_work(cpu, work_buf); |
1142d810 TH |
372 | } |
373 | ||
4aff1ca6 | 374 | static bool queue_stop_cpus_work(const struct cpumask *cpumask, |
fd7355ba TH |
375 | cpu_stop_fn_t fn, void *arg, |
376 | struct cpu_stop_done *done) | |
1142d810 TH |
377 | { |
378 | struct cpu_stop_work *work; | |
1142d810 | 379 | unsigned int cpu; |
4aff1ca6 | 380 | bool queued = false; |
1142d810 | 381 | |
1142d810 TH |
382 | /* |
383 | * Disable preemption while queueing to avoid getting | |
384 | * preempted by a stopper which might wait for other stoppers | |
385 | * to enter @fn which can lead to deadlock. | |
386 | */ | |
e6253970 ON |
387 | preempt_disable(); |
388 | stop_cpus_in_progress = true; | |
99d84bf8 | 389 | barrier(); |
b377c2a0 ON |
390 | for_each_cpu(cpu, cpumask) { |
391 | work = &per_cpu(cpu_stopper.stop_work, cpu); | |
392 | work->fn = fn; | |
393 | work->arg = arg; | |
394 | work->done = done; | |
4aff1ca6 ON |
395 | if (cpu_stop_queue_work(cpu, work)) |
396 | queued = true; | |
b377c2a0 | 397 | } |
99d84bf8 | 398 | barrier(); |
e6253970 ON |
399 | stop_cpus_in_progress = false; |
400 | preempt_enable(); | |
4aff1ca6 ON |
401 | |
402 | return queued; | |
fd7355ba | 403 | } |
1142d810 | 404 | |
fd7355ba TH |
405 | static int __stop_cpus(const struct cpumask *cpumask, |
406 | cpu_stop_fn_t fn, void *arg) | |
407 | { | |
408 | struct cpu_stop_done done; | |
409 | ||
410 | cpu_stop_init_done(&done, cpumask_weight(cpumask)); | |
4aff1ca6 ON |
411 | if (!queue_stop_cpus_work(cpumask, fn, arg, &done)) |
412 | return -ENOENT; | |
1142d810 | 413 | wait_for_completion(&done.completion); |
4aff1ca6 | 414 | return done.ret; |
1142d810 TH |
415 | } |
416 | ||
417 | /** | |
418 | * stop_cpus - stop multiple cpus | |
419 | * @cpumask: cpus to stop | |
420 | * @fn: function to execute | |
421 | * @arg: argument to @fn | |
422 | * | |
423 | * Execute @fn(@arg) on online cpus in @cpumask. On each target cpu, | |
424 | * @fn is run in a process context with the highest priority | |
425 | * preempting any task on the cpu and monopolizing it. This function | |
426 | * returns after all executions are complete. | |
427 | * | |
428 | * This function doesn't guarantee the cpus in @cpumask stay online | |
429 | * till @fn completes. If some cpus go down in the middle, execution | |
430 | * on the cpu may happen partially or fully on different cpus. @fn | |
431 | * should either be ready for that or the caller should ensure that | |
432 | * the cpus stay online until this function completes. | |
433 | * | |
434 | * All stop_cpus() calls are serialized making it safe for @fn to wait | |
435 | * for all cpus to start executing it. | |
436 | * | |
437 | * CONTEXT: | |
438 | * Might sleep. | |
439 | * | |
440 | * RETURNS: | |
441 | * -ENOENT if @fn(@arg) was not executed at all because all cpus in | |
442 | * @cpumask were offline; otherwise, 0 if all executions of @fn | |
443 | * returned 0, any non zero return value if any returned non zero. | |
444 | */ | |
445 | int stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) | |
446 | { | |
447 | int ret; | |
448 | ||
449 | /* static works are used, process one request at a time */ | |
450 | mutex_lock(&stop_cpus_mutex); | |
451 | ret = __stop_cpus(cpumask, fn, arg); | |
452 | mutex_unlock(&stop_cpus_mutex); | |
453 | return ret; | |
454 | } | |
455 | ||
456 | /** | |
457 | * try_stop_cpus - try to stop multiple cpus | |
458 | * @cpumask: cpus to stop | |
459 | * @fn: function to execute | |
460 | * @arg: argument to @fn | |
461 | * | |
462 | * Identical to stop_cpus() except that it fails with -EAGAIN if | |
463 | * someone else is already using the facility. | |
464 | * | |
465 | * CONTEXT: | |
466 | * Might sleep. | |
467 | * | |
468 | * RETURNS: | |
469 | * -EAGAIN if someone else is already stopping cpus, -ENOENT if | |
470 | * @fn(@arg) was not executed at all because all cpus in @cpumask were | |
471 | * offline; otherwise, 0 if all executions of @fn returned 0, any non | |
472 | * zero return value if any returned non zero. | |
473 | */ | |
474 | int try_stop_cpus(const struct cpumask *cpumask, cpu_stop_fn_t fn, void *arg) | |
475 | { | |
476 | int ret; | |
477 | ||
478 | /* static works are used, process one request at a time */ | |
479 | if (!mutex_trylock(&stop_cpus_mutex)) | |
480 | return -EAGAIN; | |
481 | ret = __stop_cpus(cpumask, fn, arg); | |
482 | mutex_unlock(&stop_cpus_mutex); | |
483 | return ret; | |
484 | } | |
485 | ||
14e568e7 TG |
486 | static int cpu_stop_should_run(unsigned int cpu) |
487 | { | |
488 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
489 | unsigned long flags; | |
490 | int run; | |
491 | ||
de5b55c1 | 492 | raw_spin_lock_irqsave(&stopper->lock, flags); |
14e568e7 | 493 | run = !list_empty(&stopper->works); |
de5b55c1 | 494 | raw_spin_unlock_irqrestore(&stopper->lock, flags); |
14e568e7 TG |
495 | return run; |
496 | } | |
497 | ||
498 | static void cpu_stopper_thread(unsigned int cpu) | |
1142d810 | 499 | { |
14e568e7 | 500 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 | 501 | struct cpu_stop_work *work; |
1142d810 TH |
502 | |
503 | repeat: | |
1142d810 | 504 | work = NULL; |
de5b55c1 | 505 | raw_spin_lock_irq(&stopper->lock); |
1142d810 TH |
506 | if (!list_empty(&stopper->works)) { |
507 | work = list_first_entry(&stopper->works, | |
508 | struct cpu_stop_work, list); | |
509 | list_del_init(&work->list); | |
510 | } | |
de5b55c1 | 511 | raw_spin_unlock_irq(&stopper->lock); |
1142d810 TH |
512 | |
513 | if (work) { | |
514 | cpu_stop_fn_t fn = work->fn; | |
515 | void *arg = work->arg; | |
516 | struct cpu_stop_done *done = work->done; | |
accaf6ea | 517 | int ret; |
1142d810 | 518 | |
accaf6ea ON |
519 | /* cpu stop callbacks must not sleep, make in_atomic() == T */ |
520 | preempt_count_inc(); | |
1142d810 | 521 | ret = fn(arg); |
dd2e3121 ON |
522 | if (done) { |
523 | if (ret) | |
524 | done->ret = ret; | |
525 | cpu_stop_signal_done(done); | |
526 | } | |
accaf6ea | 527 | preempt_count_dec(); |
1142d810 | 528 | WARN_ONCE(preempt_count(), |
d75f773c | 529 | "cpu_stop: %ps(%p) leaked preempt count\n", fn, arg); |
14e568e7 TG |
530 | goto repeat; |
531 | } | |
1142d810 TH |
532 | } |
533 | ||
233e7f26 ON |
534 | void stop_machine_park(int cpu) |
535 | { | |
536 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
537 | /* | |
538 | * Lockless. cpu_stopper_thread() will take stopper->lock and flush | |
539 | * the pending works before it parks, until then it is fine to queue | |
540 | * the new works. | |
541 | */ | |
542 | stopper->enabled = false; | |
543 | kthread_park(stopper->thread); | |
544 | } | |
545 | ||
34f971f6 PZ |
546 | extern void sched_set_stop_task(int cpu, struct task_struct *stop); |
547 | ||
14e568e7 TG |
548 | static void cpu_stop_create(unsigned int cpu) |
549 | { | |
02cb7aa9 | 550 | sched_set_stop_task(cpu, per_cpu(cpu_stopper.thread, cpu)); |
14e568e7 TG |
551 | } |
552 | ||
553 | static void cpu_stop_park(unsigned int cpu) | |
1142d810 | 554 | { |
1142d810 | 555 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); |
1142d810 | 556 | |
233e7f26 | 557 | WARN_ON(!list_empty(&stopper->works)); |
14e568e7 | 558 | } |
1142d810 | 559 | |
c00166d8 ON |
560 | void stop_machine_unpark(int cpu) |
561 | { | |
562 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
563 | ||
f0cf16cb | 564 | stopper->enabled = true; |
c00166d8 ON |
565 | kthread_unpark(stopper->thread); |
566 | } | |
567 | ||
14e568e7 | 568 | static struct smp_hotplug_thread cpu_stop_threads = { |
02cb7aa9 | 569 | .store = &cpu_stopper.thread, |
14e568e7 TG |
570 | .thread_should_run = cpu_stop_should_run, |
571 | .thread_fn = cpu_stopper_thread, | |
572 | .thread_comm = "migration/%u", | |
573 | .create = cpu_stop_create, | |
14e568e7 | 574 | .park = cpu_stop_park, |
14e568e7 | 575 | .selfparking = true, |
1142d810 TH |
576 | }; |
577 | ||
578 | static int __init cpu_stop_init(void) | |
579 | { | |
1142d810 | 580 | unsigned int cpu; |
1142d810 TH |
581 | |
582 | for_each_possible_cpu(cpu) { | |
583 | struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); | |
584 | ||
de5b55c1 | 585 | raw_spin_lock_init(&stopper->lock); |
1142d810 TH |
586 | INIT_LIST_HEAD(&stopper->works); |
587 | } | |
588 | ||
14e568e7 | 589 | BUG_ON(smpboot_register_percpu_thread(&cpu_stop_threads)); |
c00166d8 | 590 | stop_machine_unpark(raw_smp_processor_id()); |
f445027e | 591 | stop_machine_initialized = true; |
1142d810 TH |
592 | return 0; |
593 | } | |
594 | early_initcall(cpu_stop_init); | |
1da177e4 | 595 | |
fe5595c0 SAS |
596 | int stop_machine_cpuslocked(cpu_stop_fn_t fn, void *data, |
597 | const struct cpumask *cpus) | |
1da177e4 | 598 | { |
1be0bd77 PZ |
599 | struct multi_stop_data msdata = { |
600 | .fn = fn, | |
601 | .data = data, | |
602 | .num_threads = num_online_cpus(), | |
603 | .active_cpus = cpus, | |
604 | }; | |
3fc1f1e2 | 605 | |
fe5595c0 SAS |
606 | lockdep_assert_cpus_held(); |
607 | ||
f445027e JF |
608 | if (!stop_machine_initialized) { |
609 | /* | |
610 | * Handle the case where stop_machine() is called | |
611 | * early in boot before stop_machine() has been | |
612 | * initialized. | |
613 | */ | |
614 | unsigned long flags; | |
615 | int ret; | |
616 | ||
1be0bd77 | 617 | WARN_ON_ONCE(msdata.num_threads != 1); |
f445027e JF |
618 | |
619 | local_irq_save(flags); | |
620 | hard_irq_disable(); | |
621 | ret = (*fn)(data); | |
622 | local_irq_restore(flags); | |
623 | ||
624 | return ret; | |
625 | } | |
626 | ||
3fc1f1e2 | 627 | /* Set the initial state and stop all online cpus. */ |
1be0bd77 PZ |
628 | set_state(&msdata, MULTI_STOP_PREPARE); |
629 | return stop_cpus(cpu_online_mask, multi_cpu_stop, &msdata); | |
1da177e4 LT |
630 | } |
631 | ||
9a301f22 | 632 | int stop_machine(cpu_stop_fn_t fn, void *data, const struct cpumask *cpus) |
1da177e4 | 633 | { |
1da177e4 LT |
634 | int ret; |
635 | ||
636 | /* No CPUs can come up or down during this. */ | |
fe5595c0 SAS |
637 | cpus_read_lock(); |
638 | ret = stop_machine_cpuslocked(fn, data, cpus); | |
639 | cpus_read_unlock(); | |
1da177e4 LT |
640 | return ret; |
641 | } | |
eeec4fad | 642 | EXPORT_SYMBOL_GPL(stop_machine); |
bbf1bb3e | 643 | |
f740e6cd TH |
644 | /** |
645 | * stop_machine_from_inactive_cpu - stop_machine() from inactive CPU | |
646 | * @fn: the function to run | |
647 | * @data: the data ptr for the @fn() | |
648 | * @cpus: the cpus to run the @fn() on (NULL = any online cpu) | |
649 | * | |
650 | * This is identical to stop_machine() but can be called from a CPU which | |
651 | * is not active. The local CPU is in the process of hotplug (so no other | |
652 | * CPU hotplug can start) and not marked active and doesn't have enough | |
653 | * context to sleep. | |
654 | * | |
655 | * This function provides stop_machine() functionality for such state by | |
656 | * using busy-wait for synchronization and executing @fn directly for local | |
657 | * CPU. | |
658 | * | |
659 | * CONTEXT: | |
660 | * Local CPU is inactive. Temporarily stops all active CPUs. | |
661 | * | |
662 | * RETURNS: | |
663 | * 0 if all executions of @fn returned 0, any non zero return value if any | |
664 | * returned non zero. | |
665 | */ | |
9a301f22 | 666 | int stop_machine_from_inactive_cpu(cpu_stop_fn_t fn, void *data, |
f740e6cd TH |
667 | const struct cpumask *cpus) |
668 | { | |
1be0bd77 | 669 | struct multi_stop_data msdata = { .fn = fn, .data = data, |
f740e6cd TH |
670 | .active_cpus = cpus }; |
671 | struct cpu_stop_done done; | |
672 | int ret; | |
673 | ||
674 | /* Local CPU must be inactive and CPU hotplug in progress. */ | |
675 | BUG_ON(cpu_active(raw_smp_processor_id())); | |
1be0bd77 | 676 | msdata.num_threads = num_active_cpus() + 1; /* +1 for local */ |
f740e6cd TH |
677 | |
678 | /* No proper task established and can't sleep - busy wait for lock. */ | |
679 | while (!mutex_trylock(&stop_cpus_mutex)) | |
680 | cpu_relax(); | |
681 | ||
682 | /* Schedule work on other CPUs and execute directly for local CPU */ | |
1be0bd77 | 683 | set_state(&msdata, MULTI_STOP_PREPARE); |
f740e6cd | 684 | cpu_stop_init_done(&done, num_active_cpus()); |
1be0bd77 | 685 | queue_stop_cpus_work(cpu_active_mask, multi_cpu_stop, &msdata, |
f740e6cd | 686 | &done); |
1be0bd77 | 687 | ret = multi_cpu_stop(&msdata); |
f740e6cd TH |
688 | |
689 | /* Busy wait for completion. */ | |
690 | while (!completion_done(&done.completion)) | |
691 | cpu_relax(); | |
692 | ||
693 | mutex_unlock(&stop_cpus_mutex); | |
694 | return ret ?: done.ret; | |
695 | } |