blob: fbc230e4196905504cc0a0ccd5d9b74e6c926bab
1 | /* Kernel thread helper functions. |
2 | * Copyright (C) 2004 IBM Corporation, Rusty Russell. |
3 | * |
4 | * Creation is done via kthreadd, so that we get a clean environment |
5 | * even if we're invoked from userspace (think modprobe, hotplug cpu, |
6 | * etc.). |
7 | */ |
8 | #include <linux/sched.h> |
9 | #include <linux/kthread.h> |
10 | #include <linux/completion.h> |
11 | #include <linux/err.h> |
12 | #include <linux/cpuset.h> |
13 | #include <linux/unistd.h> |
14 | #include <linux/file.h> |
15 | #include <linux/export.h> |
16 | #include <linux/mutex.h> |
17 | #include <linux/slab.h> |
18 | #include <linux/freezer.h> |
19 | #include <linux/ptrace.h> |
20 | #include <linux/uaccess.h> |
21 | #include <linux/cgroup.h> |
22 | #include <trace/events/sched.h> |
23 | |
24 | static DEFINE_SPINLOCK(kthread_create_lock); |
25 | static LIST_HEAD(kthread_create_list); |
26 | struct task_struct *kthreadd_task; |
27 | |
28 | struct kthread_create_info |
29 | { |
30 | /* Information passed to kthread() from kthreadd. */ |
31 | int (*threadfn)(void *data); |
32 | void *data; |
33 | int node; |
34 | |
35 | /* Result passed back to kthread_create() from kthreadd. */ |
36 | struct task_struct *result; |
37 | struct completion *done; |
38 | |
39 | struct list_head list; |
40 | }; |
41 | |
42 | struct kthread { |
43 | unsigned long flags; |
44 | unsigned int cpu; |
45 | void *data; |
46 | struct completion parked; |
47 | struct completion exited; |
48 | }; |
49 | |
50 | enum KTHREAD_BITS { |
51 | KTHREAD_IS_PER_CPU = 0, |
52 | KTHREAD_SHOULD_STOP, |
53 | KTHREAD_SHOULD_PARK, |
54 | KTHREAD_IS_PARKED, |
55 | }; |
56 | |
57 | #define __to_kthread(vfork) \ |
58 | container_of(vfork, struct kthread, exited) |
59 | |
60 | static inline struct kthread *to_kthread(struct task_struct *k) |
61 | { |
62 | return __to_kthread(k->vfork_done); |
63 | } |
64 | |
65 | static struct kthread *to_live_kthread(struct task_struct *k) |
66 | { |
67 | struct completion *vfork = ACCESS_ONCE(k->vfork_done); |
68 | if (likely(vfork) && try_get_task_stack(k)) |
69 | return __to_kthread(vfork); |
70 | return NULL; |
71 | } |
72 | |
73 | /** |
74 | * kthread_should_stop - should this kthread return now? |
75 | * |
76 | * When someone calls kthread_stop() on your kthread, it will be woken |
77 | * and this will return true. You should then return, and your return |
78 | * value will be passed through to kthread_stop(). |
79 | */ |
80 | bool kthread_should_stop(void) |
81 | { |
82 | return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags); |
83 | } |
84 | EXPORT_SYMBOL(kthread_should_stop); |
85 | |
86 | /** |
87 | * kthread_should_park - should this kthread park now? |
88 | * |
89 | * When someone calls kthread_park() on your kthread, it will be woken |
90 | * and this will return true. You should then do the necessary |
91 | * cleanup and call kthread_parkme() |
92 | * |
93 | * Similar to kthread_should_stop(), but this keeps the thread alive |
94 | * and in a park position. kthread_unpark() "restarts" the thread and |
95 | * calls the thread function again. |
96 | */ |
97 | bool kthread_should_park(void) |
98 | { |
99 | return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags); |
100 | } |
101 | EXPORT_SYMBOL_GPL(kthread_should_park); |
102 | |
103 | /** |
104 | * kthread_freezable_should_stop - should this freezable kthread return now? |
105 | * @was_frozen: optional out parameter, indicates whether %current was frozen |
106 | * |
107 | * kthread_should_stop() for freezable kthreads, which will enter |
108 | * refrigerator if necessary. This function is safe from kthread_stop() / |
109 | * freezer deadlock and freezable kthreads should use this function instead |
110 | * of calling try_to_freeze() directly. |
111 | */ |
112 | bool kthread_freezable_should_stop(bool *was_frozen) |
113 | { |
114 | bool frozen = false; |
115 | |
116 | might_sleep(); |
117 | |
118 | if (unlikely(freezing(current))) |
119 | frozen = __refrigerator(true); |
120 | |
121 | if (was_frozen) |
122 | *was_frozen = frozen; |
123 | |
124 | return kthread_should_stop(); |
125 | } |
126 | EXPORT_SYMBOL_GPL(kthread_freezable_should_stop); |
127 | |
128 | /** |
129 | * kthread_data - return data value specified on kthread creation |
130 | * @task: kthread task in question |
131 | * |
132 | * Return the data value specified when kthread @task was created. |
133 | * The caller is responsible for ensuring the validity of @task when |
134 | * calling this function. |
135 | */ |
136 | void *kthread_data(struct task_struct *task) |
137 | { |
138 | return to_kthread(task)->data; |
139 | } |
140 | |
141 | /** |
142 | * kthread_probe_data - speculative version of kthread_data() |
143 | * @task: possible kthread task in question |
144 | * |
145 | * @task could be a kthread task. Return the data value specified when it |
146 | * was created if accessible. If @task isn't a kthread task or its data is |
147 | * inaccessible for any reason, %NULL is returned. This function requires |
148 | * that @task itself is safe to dereference. |
149 | */ |
150 | void *kthread_probe_data(struct task_struct *task) |
151 | { |
152 | struct kthread *kthread = to_kthread(task); |
153 | void *data = NULL; |
154 | |
155 | probe_kernel_read(&data, &kthread->data, sizeof(data)); |
156 | return data; |
157 | } |
158 | |
159 | static void __kthread_parkme(struct kthread *self) |
160 | { |
161 | __set_current_state(TASK_PARKED); |
162 | while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) { |
163 | if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags)) |
164 | complete(&self->parked); |
165 | schedule(); |
166 | __set_current_state(TASK_PARKED); |
167 | } |
168 | clear_bit(KTHREAD_IS_PARKED, &self->flags); |
169 | __set_current_state(TASK_RUNNING); |
170 | } |
171 | |
172 | void kthread_parkme(void) |
173 | { |
174 | __kthread_parkme(to_kthread(current)); |
175 | } |
176 | EXPORT_SYMBOL_GPL(kthread_parkme); |
177 | |
178 | static int kthread(void *_create) |
179 | { |
180 | /* Copy data: it's on kthread's stack */ |
181 | struct kthread_create_info *create = _create; |
182 | int (*threadfn)(void *data) = create->threadfn; |
183 | void *data = create->data; |
184 | struct completion *done; |
185 | struct kthread self; |
186 | int ret; |
187 | |
188 | self.flags = 0; |
189 | self.data = data; |
190 | init_completion(&self.exited); |
191 | init_completion(&self.parked); |
192 | current->vfork_done = &self.exited; |
193 | |
194 | /* If user was SIGKILLed, I release the structure. */ |
195 | done = xchg(&create->done, NULL); |
196 | if (!done) { |
197 | kfree(create); |
198 | do_exit(-EINTR); |
199 | } |
200 | /* OK, tell user we're spawned, wait for stop or wakeup */ |
201 | __set_current_state(TASK_UNINTERRUPTIBLE); |
202 | create->result = current; |
203 | complete(done); |
204 | schedule(); |
205 | |
206 | ret = -EINTR; |
207 | |
208 | if (!test_bit(KTHREAD_SHOULD_STOP, &self.flags)) { |
209 | cgroup_kthread_ready(); |
210 | __kthread_parkme(&self); |
211 | ret = threadfn(data); |
212 | } |
213 | /* we can't just return, we must preserve "self" on stack */ |
214 | do_exit(ret); |
215 | } |
216 | |
217 | /* called from do_fork() to get node information for about to be created task */ |
218 | int tsk_fork_get_node(struct task_struct *tsk) |
219 | { |
220 | #ifdef CONFIG_NUMA |
221 | if (tsk == kthreadd_task) |
222 | return tsk->pref_node_fork; |
223 | #endif |
224 | return NUMA_NO_NODE; |
225 | } |
226 | |
227 | static void create_kthread(struct kthread_create_info *create) |
228 | { |
229 | int pid; |
230 | |
231 | #ifdef CONFIG_NUMA |
232 | current->pref_node_fork = create->node; |
233 | #endif |
234 | /* We want our own signal handler (we take no signals by default). */ |
235 | pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD); |
236 | if (pid < 0) { |
237 | /* If user was SIGKILLed, I release the structure. */ |
238 | struct completion *done = xchg(&create->done, NULL); |
239 | |
240 | if (!done) { |
241 | kfree(create); |
242 | return; |
243 | } |
244 | create->result = ERR_PTR(pid); |
245 | complete(done); |
246 | } |
247 | } |
248 | |
249 | static struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data), |
250 | void *data, int node, |
251 | const char namefmt[], |
252 | va_list args) |
253 | { |
254 | DECLARE_COMPLETION_ONSTACK(done); |
255 | struct task_struct *task; |
256 | struct kthread_create_info *create = kmalloc(sizeof(*create), |
257 | GFP_KERNEL); |
258 | |
259 | if (!create) |
260 | return ERR_PTR(-ENOMEM); |
261 | create->threadfn = threadfn; |
262 | create->data = data; |
263 | create->node = node; |
264 | create->done = &done; |
265 | |
266 | spin_lock(&kthread_create_lock); |
267 | list_add_tail(&create->list, &kthread_create_list); |
268 | spin_unlock(&kthread_create_lock); |
269 | |
270 | wake_up_process(kthreadd_task); |
271 | /* |
272 | * Wait for completion in killable state, for I might be chosen by |
273 | * the OOM killer while kthreadd is trying to allocate memory for |
274 | * new kernel thread. |
275 | */ |
276 | if (unlikely(wait_for_completion_killable(&done))) { |
277 | /* |
278 | * If I was SIGKILLed before kthreadd (or new kernel thread) |
279 | * calls complete(), leave the cleanup of this structure to |
280 | * that thread. |
281 | */ |
282 | if (xchg(&create->done, NULL)) |
283 | return ERR_PTR(-EINTR); |
284 | /* |
285 | * kthreadd (or new kernel thread) will call complete() |
286 | * shortly. |
287 | */ |
288 | wait_for_completion(&done); |
289 | } |
290 | task = create->result; |
291 | if (!IS_ERR(task)) { |
292 | static const struct sched_param param = { .sched_priority = 0 }; |
293 | char name[TASK_COMM_LEN]; |
294 | |
295 | /* |
296 | * task is already visible to other tasks, so updating |
297 | * COMM must be protected. |
298 | */ |
299 | vsnprintf(name, sizeof(name), namefmt, args); |
300 | set_task_comm(task, name); |
301 | /* |
302 | * root may have changed our (kthreadd's) priority or CPU mask. |
303 | * The kernel thread should not inherit these properties. |
304 | */ |
305 | sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m); |
306 | set_cpus_allowed_ptr(task, cpu_all_mask); |
307 | } |
308 | kfree(create); |
309 | return task; |
310 | } |
311 | |
312 | /** |
313 | * kthread_create_on_node - create a kthread. |
314 | * @threadfn: the function to run until signal_pending(current). |
315 | * @data: data ptr for @threadfn. |
316 | * @node: task and thread structures for the thread are allocated on this node |
317 | * @namefmt: printf-style name for the thread. |
318 | * |
319 | * Description: This helper function creates and names a kernel |
320 | * thread. The thread will be stopped: use wake_up_process() to start |
321 | * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and |
322 | * is affine to all CPUs. |
323 | * |
324 | * If thread is going to be bound on a particular cpu, give its node |
325 | * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE. |
326 | * When woken, the thread will run @threadfn() with @data as its |
327 | * argument. @threadfn() can either call do_exit() directly if it is a |
328 | * standalone thread for which no one will call kthread_stop(), or |
329 | * return when 'kthread_should_stop()' is true (which means |
330 | * kthread_stop() has been called). The return value should be zero |
331 | * or a negative error number; it will be passed to kthread_stop(). |
332 | * |
333 | * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR). |
334 | */ |
335 | struct task_struct *kthread_create_on_node(int (*threadfn)(void *data), |
336 | void *data, int node, |
337 | const char namefmt[], |
338 | ...) |
339 | { |
340 | struct task_struct *task; |
341 | va_list args; |
342 | |
343 | va_start(args, namefmt); |
344 | task = __kthread_create_on_node(threadfn, data, node, namefmt, args); |
345 | va_end(args); |
346 | |
347 | return task; |
348 | } |
349 | EXPORT_SYMBOL(kthread_create_on_node); |
350 | |
351 | static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state) |
352 | { |
353 | unsigned long flags; |
354 | |
355 | if (!wait_task_inactive(p, state)) { |
356 | WARN_ON(1); |
357 | return; |
358 | } |
359 | |
360 | /* It's safe because the task is inactive. */ |
361 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
362 | do_set_cpus_allowed(p, mask); |
363 | p->flags |= PF_NO_SETAFFINITY; |
364 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
365 | } |
366 | |
367 | static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state) |
368 | { |
369 | __kthread_bind_mask(p, cpumask_of(cpu), state); |
370 | } |
371 | |
372 | void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask) |
373 | { |
374 | __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE); |
375 | } |
376 | |
377 | /** |
378 | * kthread_bind - bind a just-created kthread to a cpu. |
379 | * @p: thread created by kthread_create(). |
380 | * @cpu: cpu (might not be online, must be possible) for @k to run on. |
381 | * |
382 | * Description: This function is equivalent to set_cpus_allowed(), |
383 | * except that @cpu doesn't need to be online, and the thread must be |
384 | * stopped (i.e., just returned from kthread_create()). |
385 | */ |
386 | void kthread_bind(struct task_struct *p, unsigned int cpu) |
387 | { |
388 | __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE); |
389 | } |
390 | EXPORT_SYMBOL(kthread_bind); |
391 | |
392 | /** |
393 | * kthread_create_on_cpu - Create a cpu bound kthread |
394 | * @threadfn: the function to run until signal_pending(current). |
395 | * @data: data ptr for @threadfn. |
396 | * @cpu: The cpu on which the thread should be bound, |
397 | * @namefmt: printf-style name for the thread. Format is restricted |
398 | * to "name.*%u". Code fills in cpu number. |
399 | * |
400 | * Description: This helper function creates and names a kernel thread |
401 | * The thread will be woken and put into park mode. |
402 | */ |
403 | struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data), |
404 | void *data, unsigned int cpu, |
405 | const char *namefmt) |
406 | { |
407 | struct task_struct *p; |
408 | |
409 | p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt, |
410 | cpu); |
411 | if (IS_ERR(p)) |
412 | return p; |
413 | kthread_bind(p, cpu); |
414 | /* CPU hotplug need to bind once again when unparking the thread. */ |
415 | set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags); |
416 | to_kthread(p)->cpu = cpu; |
417 | return p; |
418 | } |
419 | |
420 | static void __kthread_unpark(struct task_struct *k, struct kthread *kthread) |
421 | { |
422 | clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags); |
423 | /* |
424 | * We clear the IS_PARKED bit here as we don't wait |
425 | * until the task has left the park code. So if we'd |
426 | * park before that happens we'd see the IS_PARKED bit |
427 | * which might be about to be cleared. |
428 | */ |
429 | if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) { |
430 | /* |
431 | * Newly created kthread was parked when the CPU was offline. |
432 | * The binding was lost and we need to set it again. |
433 | */ |
434 | if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags)) |
435 | __kthread_bind(k, kthread->cpu, TASK_PARKED); |
436 | wake_up_state(k, TASK_PARKED); |
437 | } |
438 | } |
439 | |
440 | /** |
441 | * kthread_unpark - unpark a thread created by kthread_create(). |
442 | * @k: thread created by kthread_create(). |
443 | * |
444 | * Sets kthread_should_park() for @k to return false, wakes it, and |
445 | * waits for it to return. If the thread is marked percpu then its |
446 | * bound to the cpu again. |
447 | */ |
448 | void kthread_unpark(struct task_struct *k) |
449 | { |
450 | struct kthread *kthread = to_live_kthread(k); |
451 | |
452 | if (kthread) { |
453 | __kthread_unpark(k, kthread); |
454 | put_task_stack(k); |
455 | } |
456 | } |
457 | EXPORT_SYMBOL_GPL(kthread_unpark); |
458 | |
459 | /** |
460 | * kthread_park - park a thread created by kthread_create(). |
461 | * @k: thread created by kthread_create(). |
462 | * |
463 | * Sets kthread_should_park() for @k to return true, wakes it, and |
464 | * waits for it to return. This can also be called after kthread_create() |
465 | * instead of calling wake_up_process(): the thread will park without |
466 | * calling threadfn(). |
467 | * |
468 | * Returns 0 if the thread is parked, -ENOSYS if the thread exited. |
469 | * If called by the kthread itself just the park bit is set. |
470 | */ |
471 | int kthread_park(struct task_struct *k) |
472 | { |
473 | struct kthread *kthread = to_live_kthread(k); |
474 | int ret = -ENOSYS; |
475 | |
476 | if (kthread) { |
477 | if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) { |
478 | set_bit(KTHREAD_SHOULD_PARK, &kthread->flags); |
479 | if (k != current) { |
480 | wake_up_process(k); |
481 | wait_for_completion(&kthread->parked); |
482 | } |
483 | } |
484 | put_task_stack(k); |
485 | ret = 0; |
486 | } |
487 | return ret; |
488 | } |
489 | EXPORT_SYMBOL_GPL(kthread_park); |
490 | |
491 | /** |
492 | * kthread_stop - stop a thread created by kthread_create(). |
493 | * @k: thread created by kthread_create(). |
494 | * |
495 | * Sets kthread_should_stop() for @k to return true, wakes it, and |
496 | * waits for it to exit. This can also be called after kthread_create() |
497 | * instead of calling wake_up_process(): the thread will exit without |
498 | * calling threadfn(). |
499 | * |
500 | * If threadfn() may call do_exit() itself, the caller must ensure |
501 | * task_struct can't go away. |
502 | * |
503 | * Returns the result of threadfn(), or %-EINTR if wake_up_process() |
504 | * was never called. |
505 | */ |
506 | int kthread_stop(struct task_struct *k) |
507 | { |
508 | struct kthread *kthread; |
509 | int ret; |
510 | |
511 | trace_sched_kthread_stop(k); |
512 | |
513 | get_task_struct(k); |
514 | kthread = to_live_kthread(k); |
515 | if (kthread) { |
516 | set_bit(KTHREAD_SHOULD_STOP, &kthread->flags); |
517 | __kthread_unpark(k, kthread); |
518 | wake_up_process(k); |
519 | wait_for_completion(&kthread->exited); |
520 | put_task_stack(k); |
521 | } |
522 | ret = k->exit_code; |
523 | put_task_struct(k); |
524 | |
525 | trace_sched_kthread_stop_ret(ret); |
526 | return ret; |
527 | } |
528 | EXPORT_SYMBOL(kthread_stop); |
529 | |
530 | int kthreadd(void *unused) |
531 | { |
532 | struct task_struct *tsk = current; |
533 | |
534 | /* Setup a clean context for our children to inherit. */ |
535 | set_task_comm(tsk, "kthreadd"); |
536 | ignore_signals(tsk); |
537 | set_cpus_allowed_ptr(tsk, cpu_all_mask); |
538 | set_mems_allowed(node_states[N_MEMORY]); |
539 | |
540 | current->flags |= PF_NOFREEZE; |
541 | cgroup_init_kthreadd(); |
542 | |
543 | for (;;) { |
544 | set_current_state(TASK_INTERRUPTIBLE); |
545 | if (list_empty(&kthread_create_list)) |
546 | schedule(); |
547 | __set_current_state(TASK_RUNNING); |
548 | |
549 | spin_lock(&kthread_create_lock); |
550 | while (!list_empty(&kthread_create_list)) { |
551 | struct kthread_create_info *create; |
552 | |
553 | create = list_entry(kthread_create_list.next, |
554 | struct kthread_create_info, list); |
555 | list_del_init(&create->list); |
556 | spin_unlock(&kthread_create_lock); |
557 | |
558 | create_kthread(create); |
559 | |
560 | spin_lock(&kthread_create_lock); |
561 | } |
562 | spin_unlock(&kthread_create_lock); |
563 | } |
564 | |
565 | return 0; |
566 | } |
567 | |
568 | void __kthread_init_worker(struct kthread_worker *worker, |
569 | const char *name, |
570 | struct lock_class_key *key) |
571 | { |
572 | memset(worker, 0, sizeof(struct kthread_worker)); |
573 | spin_lock_init(&worker->lock); |
574 | lockdep_set_class_and_name(&worker->lock, key, name); |
575 | INIT_LIST_HEAD(&worker->work_list); |
576 | INIT_LIST_HEAD(&worker->delayed_work_list); |
577 | } |
578 | EXPORT_SYMBOL_GPL(__kthread_init_worker); |
579 | |
580 | /** |
581 | * kthread_worker_fn - kthread function to process kthread_worker |
582 | * @worker_ptr: pointer to initialized kthread_worker |
583 | * |
584 | * This function implements the main cycle of kthread worker. It processes |
585 | * work_list until it is stopped with kthread_stop(). It sleeps when the queue |
586 | * is empty. |
587 | * |
588 | * The works are not allowed to keep any locks, disable preemption or interrupts |
589 | * when they finish. There is defined a safe point for freezing when one work |
590 | * finishes and before a new one is started. |
591 | * |
592 | * Also the works must not be handled by more than one worker at the same time, |
593 | * see also kthread_queue_work(). |
594 | */ |
595 | int kthread_worker_fn(void *worker_ptr) |
596 | { |
597 | struct kthread_worker *worker = worker_ptr; |
598 | struct kthread_work *work; |
599 | |
600 | /* |
601 | * FIXME: Update the check and remove the assignment when all kthread |
602 | * worker users are created using kthread_create_worker*() functions. |
603 | */ |
604 | WARN_ON(worker->task && worker->task != current); |
605 | worker->task = current; |
606 | |
607 | if (worker->flags & KTW_FREEZABLE) |
608 | set_freezable(); |
609 | |
610 | repeat: |
611 | set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */ |
612 | |
613 | if (kthread_should_stop()) { |
614 | __set_current_state(TASK_RUNNING); |
615 | spin_lock_irq(&worker->lock); |
616 | worker->task = NULL; |
617 | spin_unlock_irq(&worker->lock); |
618 | return 0; |
619 | } |
620 | |
621 | work = NULL; |
622 | spin_lock_irq(&worker->lock); |
623 | if (!list_empty(&worker->work_list)) { |
624 | work = list_first_entry(&worker->work_list, |
625 | struct kthread_work, node); |
626 | list_del_init(&work->node); |
627 | } |
628 | worker->current_work = work; |
629 | spin_unlock_irq(&worker->lock); |
630 | |
631 | if (work) { |
632 | __set_current_state(TASK_RUNNING); |
633 | work->func(work); |
634 | } else if (!freezing(current)) |
635 | schedule(); |
636 | |
637 | try_to_freeze(); |
638 | goto repeat; |
639 | } |
640 | EXPORT_SYMBOL_GPL(kthread_worker_fn); |
641 | |
642 | static struct kthread_worker * |
643 | __kthread_create_worker(int cpu, unsigned int flags, |
644 | const char namefmt[], va_list args) |
645 | { |
646 | struct kthread_worker *worker; |
647 | struct task_struct *task; |
648 | |
649 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); |
650 | if (!worker) |
651 | return ERR_PTR(-ENOMEM); |
652 | |
653 | kthread_init_worker(worker); |
654 | |
655 | if (cpu >= 0) { |
656 | char name[TASK_COMM_LEN]; |
657 | |
658 | /* |
659 | * kthread_create_worker_on_cpu() allows to pass a generic |
660 | * namefmt in compare with kthread_create_on_cpu. We need |
661 | * to format it here. |
662 | */ |
663 | vsnprintf(name, sizeof(name), namefmt, args); |
664 | task = kthread_create_on_cpu(kthread_worker_fn, worker, |
665 | cpu, name); |
666 | } else { |
667 | task = __kthread_create_on_node(kthread_worker_fn, worker, |
668 | -1, namefmt, args); |
669 | } |
670 | |
671 | if (IS_ERR(task)) |
672 | goto fail_task; |
673 | |
674 | worker->flags = flags; |
675 | worker->task = task; |
676 | wake_up_process(task); |
677 | return worker; |
678 | |
679 | fail_task: |
680 | kfree(worker); |
681 | return ERR_CAST(task); |
682 | } |
683 | |
684 | /** |
685 | * kthread_create_worker - create a kthread worker |
686 | * @flags: flags modifying the default behavior of the worker |
687 | * @namefmt: printf-style name for the kthread worker (task). |
688 | * |
689 | * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM) |
690 | * when the needed structures could not get allocated, and ERR_PTR(-EINTR) |
691 | * when the worker was SIGKILLed. |
692 | */ |
693 | struct kthread_worker * |
694 | kthread_create_worker(unsigned int flags, const char namefmt[], ...) |
695 | { |
696 | struct kthread_worker *worker; |
697 | va_list args; |
698 | |
699 | va_start(args, namefmt); |
700 | worker = __kthread_create_worker(-1, flags, namefmt, args); |
701 | va_end(args); |
702 | |
703 | return worker; |
704 | } |
705 | EXPORT_SYMBOL(kthread_create_worker); |
706 | |
707 | /** |
708 | * kthread_create_worker_on_cpu - create a kthread worker and bind it |
709 | * it to a given CPU and the associated NUMA node. |
710 | * @cpu: CPU number |
711 | * @flags: flags modifying the default behavior of the worker |
712 | * @namefmt: printf-style name for the kthread worker (task). |
713 | * |
714 | * Use a valid CPU number if you want to bind the kthread worker |
715 | * to the given CPU and the associated NUMA node. |
716 | * |
717 | * A good practice is to add the cpu number also into the worker name. |
718 | * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu). |
719 | * |
720 | * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM) |
721 | * when the needed structures could not get allocated, and ERR_PTR(-EINTR) |
722 | * when the worker was SIGKILLed. |
723 | */ |
724 | struct kthread_worker * |
725 | kthread_create_worker_on_cpu(int cpu, unsigned int flags, |
726 | const char namefmt[], ...) |
727 | { |
728 | struct kthread_worker *worker; |
729 | va_list args; |
730 | |
731 | va_start(args, namefmt); |
732 | worker = __kthread_create_worker(cpu, flags, namefmt, args); |
733 | va_end(args); |
734 | |
735 | return worker; |
736 | } |
737 | EXPORT_SYMBOL(kthread_create_worker_on_cpu); |
738 | |
739 | /* |
740 | * Returns true when the work could not be queued at the moment. |
741 | * It happens when it is already pending in a worker list |
742 | * or when it is being cancelled. |
743 | */ |
744 | static inline bool queuing_blocked(struct kthread_worker *worker, |
745 | struct kthread_work *work) |
746 | { |
747 | lockdep_assert_held(&worker->lock); |
748 | |
749 | return !list_empty(&work->node) || work->canceling; |
750 | } |
751 | |
752 | static void kthread_insert_work_sanity_check(struct kthread_worker *worker, |
753 | struct kthread_work *work) |
754 | { |
755 | lockdep_assert_held(&worker->lock); |
756 | WARN_ON_ONCE(!list_empty(&work->node)); |
757 | /* Do not use a work with >1 worker, see kthread_queue_work() */ |
758 | WARN_ON_ONCE(work->worker && work->worker != worker); |
759 | } |
760 | |
761 | /* insert @work before @pos in @worker */ |
762 | static void kthread_insert_work(struct kthread_worker *worker, |
763 | struct kthread_work *work, |
764 | struct list_head *pos) |
765 | { |
766 | kthread_insert_work_sanity_check(worker, work); |
767 | |
768 | list_add_tail(&work->node, pos); |
769 | work->worker = worker; |
770 | if (!worker->current_work && likely(worker->task)) |
771 | wake_up_process(worker->task); |
772 | } |
773 | |
774 | /** |
775 | * kthread_queue_work - queue a kthread_work |
776 | * @worker: target kthread_worker |
777 | * @work: kthread_work to queue |
778 | * |
779 | * Queue @work to work processor @task for async execution. @task |
780 | * must have been created with kthread_worker_create(). Returns %true |
781 | * if @work was successfully queued, %false if it was already pending. |
782 | * |
783 | * Reinitialize the work if it needs to be used by another worker. |
784 | * For example, when the worker was stopped and started again. |
785 | */ |
786 | bool kthread_queue_work(struct kthread_worker *worker, |
787 | struct kthread_work *work) |
788 | { |
789 | bool ret = false; |
790 | unsigned long flags; |
791 | |
792 | spin_lock_irqsave(&worker->lock, flags); |
793 | if (!queuing_blocked(worker, work)) { |
794 | kthread_insert_work(worker, work, &worker->work_list); |
795 | ret = true; |
796 | } |
797 | spin_unlock_irqrestore(&worker->lock, flags); |
798 | return ret; |
799 | } |
800 | EXPORT_SYMBOL_GPL(kthread_queue_work); |
801 | |
802 | /** |
803 | * kthread_delayed_work_timer_fn - callback that queues the associated kthread |
804 | * delayed work when the timer expires. |
805 | * @__data: pointer to the data associated with the timer |
806 | * |
807 | * The format of the function is defined by struct timer_list. |
808 | * It should have been called from irqsafe timer with irq already off. |
809 | */ |
810 | void kthread_delayed_work_timer_fn(unsigned long __data) |
811 | { |
812 | struct kthread_delayed_work *dwork = |
813 | (struct kthread_delayed_work *)__data; |
814 | struct kthread_work *work = &dwork->work; |
815 | struct kthread_worker *worker = work->worker; |
816 | |
817 | /* |
818 | * This might happen when a pending work is reinitialized. |
819 | * It means that it is used a wrong way. |
820 | */ |
821 | if (WARN_ON_ONCE(!worker)) |
822 | return; |
823 | |
824 | spin_lock(&worker->lock); |
825 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
826 | WARN_ON_ONCE(work->worker != worker); |
827 | |
828 | /* Move the work from worker->delayed_work_list. */ |
829 | WARN_ON_ONCE(list_empty(&work->node)); |
830 | list_del_init(&work->node); |
831 | kthread_insert_work(worker, work, &worker->work_list); |
832 | |
833 | spin_unlock(&worker->lock); |
834 | } |
835 | EXPORT_SYMBOL(kthread_delayed_work_timer_fn); |
836 | |
837 | void __kthread_queue_delayed_work(struct kthread_worker *worker, |
838 | struct kthread_delayed_work *dwork, |
839 | unsigned long delay) |
840 | { |
841 | struct timer_list *timer = &dwork->timer; |
842 | struct kthread_work *work = &dwork->work; |
843 | |
844 | WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn || |
845 | timer->data != (unsigned long)dwork); |
846 | |
847 | /* |
848 | * If @delay is 0, queue @dwork->work immediately. This is for |
849 | * both optimization and correctness. The earliest @timer can |
850 | * expire is on the closest next tick and delayed_work users depend |
851 | * on that there's no such delay when @delay is 0. |
852 | */ |
853 | if (!delay) { |
854 | kthread_insert_work(worker, work, &worker->work_list); |
855 | return; |
856 | } |
857 | |
858 | /* Be paranoid and try to detect possible races already now. */ |
859 | kthread_insert_work_sanity_check(worker, work); |
860 | |
861 | list_add(&work->node, &worker->delayed_work_list); |
862 | work->worker = worker; |
863 | timer_stats_timer_set_start_info(&dwork->timer); |
864 | timer->expires = jiffies + delay; |
865 | add_timer(timer); |
866 | } |
867 | |
868 | /** |
869 | * kthread_queue_delayed_work - queue the associated kthread work |
870 | * after a delay. |
871 | * @worker: target kthread_worker |
872 | * @dwork: kthread_delayed_work to queue |
873 | * @delay: number of jiffies to wait before queuing |
874 | * |
875 | * If the work has not been pending it starts a timer that will queue |
876 | * the work after the given @delay. If @delay is zero, it queues the |
877 | * work immediately. |
878 | * |
879 | * Return: %false if the @work has already been pending. It means that |
880 | * either the timer was running or the work was queued. It returns %true |
881 | * otherwise. |
882 | */ |
883 | bool kthread_queue_delayed_work(struct kthread_worker *worker, |
884 | struct kthread_delayed_work *dwork, |
885 | unsigned long delay) |
886 | { |
887 | struct kthread_work *work = &dwork->work; |
888 | unsigned long flags; |
889 | bool ret = false; |
890 | |
891 | spin_lock_irqsave(&worker->lock, flags); |
892 | |
893 | if (!queuing_blocked(worker, work)) { |
894 | __kthread_queue_delayed_work(worker, dwork, delay); |
895 | ret = true; |
896 | } |
897 | |
898 | spin_unlock_irqrestore(&worker->lock, flags); |
899 | return ret; |
900 | } |
901 | EXPORT_SYMBOL_GPL(kthread_queue_delayed_work); |
902 | |
903 | struct kthread_flush_work { |
904 | struct kthread_work work; |
905 | struct completion done; |
906 | }; |
907 | |
908 | static void kthread_flush_work_fn(struct kthread_work *work) |
909 | { |
910 | struct kthread_flush_work *fwork = |
911 | container_of(work, struct kthread_flush_work, work); |
912 | complete(&fwork->done); |
913 | } |
914 | |
915 | /** |
916 | * kthread_flush_work - flush a kthread_work |
917 | * @work: work to flush |
918 | * |
919 | * If @work is queued or executing, wait for it to finish execution. |
920 | */ |
921 | void kthread_flush_work(struct kthread_work *work) |
922 | { |
923 | struct kthread_flush_work fwork = { |
924 | KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn), |
925 | COMPLETION_INITIALIZER_ONSTACK(fwork.done), |
926 | }; |
927 | struct kthread_worker *worker; |
928 | bool noop = false; |
929 | |
930 | worker = work->worker; |
931 | if (!worker) |
932 | return; |
933 | |
934 | spin_lock_irq(&worker->lock); |
935 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
936 | WARN_ON_ONCE(work->worker != worker); |
937 | |
938 | if (!list_empty(&work->node)) |
939 | kthread_insert_work(worker, &fwork.work, work->node.next); |
940 | else if (worker->current_work == work) |
941 | kthread_insert_work(worker, &fwork.work, |
942 | worker->work_list.next); |
943 | else |
944 | noop = true; |
945 | |
946 | spin_unlock_irq(&worker->lock); |
947 | |
948 | if (!noop) |
949 | wait_for_completion(&fwork.done); |
950 | } |
951 | EXPORT_SYMBOL_GPL(kthread_flush_work); |
952 | |
953 | /* |
954 | * This function removes the work from the worker queue. Also it makes sure |
955 | * that it won't get queued later via the delayed work's timer. |
956 | * |
957 | * The work might still be in use when this function finishes. See the |
958 | * current_work proceed by the worker. |
959 | * |
960 | * Return: %true if @work was pending and successfully canceled, |
961 | * %false if @work was not pending |
962 | */ |
963 | static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork, |
964 | unsigned long *flags) |
965 | { |
966 | /* Try to cancel the timer if exists. */ |
967 | if (is_dwork) { |
968 | struct kthread_delayed_work *dwork = |
969 | container_of(work, struct kthread_delayed_work, work); |
970 | struct kthread_worker *worker = work->worker; |
971 | |
972 | /* |
973 | * del_timer_sync() must be called to make sure that the timer |
974 | * callback is not running. The lock must be temporary released |
975 | * to avoid a deadlock with the callback. In the meantime, |
976 | * any queuing is blocked by setting the canceling counter. |
977 | */ |
978 | work->canceling++; |
979 | spin_unlock_irqrestore(&worker->lock, *flags); |
980 | del_timer_sync(&dwork->timer); |
981 | spin_lock_irqsave(&worker->lock, *flags); |
982 | work->canceling--; |
983 | } |
984 | |
985 | /* |
986 | * Try to remove the work from a worker list. It might either |
987 | * be from worker->work_list or from worker->delayed_work_list. |
988 | */ |
989 | if (!list_empty(&work->node)) { |
990 | list_del_init(&work->node); |
991 | return true; |
992 | } |
993 | |
994 | return false; |
995 | } |
996 | |
997 | /** |
998 | * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work |
999 | * @worker: kthread worker to use |
1000 | * @dwork: kthread delayed work to queue |
1001 | * @delay: number of jiffies to wait before queuing |
1002 | * |
1003 | * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise, |
1004 | * modify @dwork's timer so that it expires after @delay. If @delay is zero, |
1005 | * @work is guaranteed to be queued immediately. |
1006 | * |
1007 | * Return: %true if @dwork was pending and its timer was modified, |
1008 | * %false otherwise. |
1009 | * |
1010 | * A special case is when the work is being canceled in parallel. |
1011 | * It might be caused either by the real kthread_cancel_delayed_work_sync() |
1012 | * or yet another kthread_mod_delayed_work() call. We let the other command |
1013 | * win and return %false here. The caller is supposed to synchronize these |
1014 | * operations a reasonable way. |
1015 | * |
1016 | * This function is safe to call from any context including IRQ handler. |
1017 | * See __kthread_cancel_work() and kthread_delayed_work_timer_fn() |
1018 | * for details. |
1019 | */ |
1020 | bool kthread_mod_delayed_work(struct kthread_worker *worker, |
1021 | struct kthread_delayed_work *dwork, |
1022 | unsigned long delay) |
1023 | { |
1024 | struct kthread_work *work = &dwork->work; |
1025 | unsigned long flags; |
1026 | int ret = false; |
1027 | |
1028 | spin_lock_irqsave(&worker->lock, flags); |
1029 | |
1030 | /* Do not bother with canceling when never queued. */ |
1031 | if (!work->worker) |
1032 | goto fast_queue; |
1033 | |
1034 | /* Work must not be used with >1 worker, see kthread_queue_work() */ |
1035 | WARN_ON_ONCE(work->worker != worker); |
1036 | |
1037 | /* Do not fight with another command that is canceling this work. */ |
1038 | if (work->canceling) |
1039 | goto out; |
1040 | |
1041 | ret = __kthread_cancel_work(work, true, &flags); |
1042 | fast_queue: |
1043 | __kthread_queue_delayed_work(worker, dwork, delay); |
1044 | out: |
1045 | spin_unlock_irqrestore(&worker->lock, flags); |
1046 | return ret; |
1047 | } |
1048 | EXPORT_SYMBOL_GPL(kthread_mod_delayed_work); |
1049 | |
1050 | static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork) |
1051 | { |
1052 | struct kthread_worker *worker = work->worker; |
1053 | unsigned long flags; |
1054 | int ret = false; |
1055 | |
1056 | if (!worker) |
1057 | goto out; |
1058 | |
1059 | spin_lock_irqsave(&worker->lock, flags); |
1060 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ |
1061 | WARN_ON_ONCE(work->worker != worker); |
1062 | |
1063 | ret = __kthread_cancel_work(work, is_dwork, &flags); |
1064 | |
1065 | if (worker->current_work != work) |
1066 | goto out_fast; |
1067 | |
1068 | /* |
1069 | * The work is in progress and we need to wait with the lock released. |
1070 | * In the meantime, block any queuing by setting the canceling counter. |
1071 | */ |
1072 | work->canceling++; |
1073 | spin_unlock_irqrestore(&worker->lock, flags); |
1074 | kthread_flush_work(work); |
1075 | spin_lock_irqsave(&worker->lock, flags); |
1076 | work->canceling--; |
1077 | |
1078 | out_fast: |
1079 | spin_unlock_irqrestore(&worker->lock, flags); |
1080 | out: |
1081 | return ret; |
1082 | } |
1083 | |
1084 | /** |
1085 | * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish |
1086 | * @work: the kthread work to cancel |
1087 | * |
1088 | * Cancel @work and wait for its execution to finish. This function |
1089 | * can be used even if the work re-queues itself. On return from this |
1090 | * function, @work is guaranteed to be not pending or executing on any CPU. |
1091 | * |
1092 | * kthread_cancel_work_sync(&delayed_work->work) must not be used for |
1093 | * delayed_work's. Use kthread_cancel_delayed_work_sync() instead. |
1094 | * |
1095 | * The caller must ensure that the worker on which @work was last |
1096 | * queued can't be destroyed before this function returns. |
1097 | * |
1098 | * Return: %true if @work was pending, %false otherwise. |
1099 | */ |
1100 | bool kthread_cancel_work_sync(struct kthread_work *work) |
1101 | { |
1102 | return __kthread_cancel_work_sync(work, false); |
1103 | } |
1104 | EXPORT_SYMBOL_GPL(kthread_cancel_work_sync); |
1105 | |
1106 | /** |
1107 | * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and |
1108 | * wait for it to finish. |
1109 | * @dwork: the kthread delayed work to cancel |
1110 | * |
1111 | * This is kthread_cancel_work_sync() for delayed works. |
1112 | * |
1113 | * Return: %true if @dwork was pending, %false otherwise. |
1114 | */ |
1115 | bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork) |
1116 | { |
1117 | return __kthread_cancel_work_sync(&dwork->work, true); |
1118 | } |
1119 | EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync); |
1120 | |
1121 | /** |
1122 | * kthread_flush_worker - flush all current works on a kthread_worker |
1123 | * @worker: worker to flush |
1124 | * |
1125 | * Wait until all currently executing or pending works on @worker are |
1126 | * finished. |
1127 | */ |
1128 | void kthread_flush_worker(struct kthread_worker *worker) |
1129 | { |
1130 | struct kthread_flush_work fwork = { |
1131 | KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn), |
1132 | COMPLETION_INITIALIZER_ONSTACK(fwork.done), |
1133 | }; |
1134 | |
1135 | kthread_queue_work(worker, &fwork.work); |
1136 | wait_for_completion(&fwork.done); |
1137 | } |
1138 | EXPORT_SYMBOL_GPL(kthread_flush_worker); |
1139 | |
1140 | /** |
1141 | * kthread_destroy_worker - destroy a kthread worker |
1142 | * @worker: worker to be destroyed |
1143 | * |
1144 | * Flush and destroy @worker. The simple flush is enough because the kthread |
1145 | * worker API is used only in trivial scenarios. There are no multi-step state |
1146 | * machines needed. |
1147 | */ |
1148 | void kthread_destroy_worker(struct kthread_worker *worker) |
1149 | { |
1150 | struct task_struct *task; |
1151 | |
1152 | task = worker->task; |
1153 | if (WARN_ON(!task)) |
1154 | return; |
1155 | |
1156 | kthread_flush_worker(worker); |
1157 | kthread_stop(task); |
1158 | WARN_ON(!list_empty(&worker->work_list)); |
1159 | kfree(worker); |
1160 | } |
1161 | EXPORT_SYMBOL(kthread_destroy_worker); |
1162 |