]>
Commit | Line | Data |
---|---|---|
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 <uapi/linux/sched/types.h> | |
9 | #include <linux/sched.h> | |
10 | #include <linux/sched/task.h> | |
11 | #include <linux/kthread.h> | |
12 | #include <linux/completion.h> | |
13 | #include <linux/err.h> | |
14 | #include <linux/cpuset.h> | |
15 | #include <linux/unistd.h> | |
16 | #include <linux/file.h> | |
17 | #include <linux/export.h> | |
18 | #include <linux/mutex.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/freezer.h> | |
21 | #include <linux/ptrace.h> | |
22 | #include <linux/uaccess.h> | |
23 | #include <linux/cgroup.h> | |
24 | #include <trace/events/sched.h> | |
25 | ||
26 | static DEFINE_SPINLOCK(kthread_create_lock); | |
27 | static LIST_HEAD(kthread_create_list); | |
28 | struct task_struct *kthreadd_task; | |
29 | ||
30 | struct kthread_create_info | |
31 | { | |
32 | /* Information passed to kthread() from kthreadd. */ | |
33 | int (*threadfn)(void *data); | |
34 | void *data; | |
35 | int node; | |
36 | ||
37 | /* Result passed back to kthread_create() from kthreadd. */ | |
38 | struct task_struct *result; | |
39 | struct completion *done; | |
40 | ||
41 | struct list_head list; | |
42 | }; | |
43 | ||
44 | struct kthread { | |
45 | unsigned long flags; | |
46 | unsigned int cpu; | |
47 | void *data; | |
48 | struct completion parked; | |
49 | struct completion exited; | |
50 | }; | |
51 | ||
52 | enum KTHREAD_BITS { | |
53 | KTHREAD_IS_PER_CPU = 0, | |
54 | KTHREAD_SHOULD_STOP, | |
55 | KTHREAD_SHOULD_PARK, | |
56 | KTHREAD_IS_PARKED, | |
57 | }; | |
58 | ||
59 | static inline void set_kthread_struct(void *kthread) | |
60 | { | |
61 | /* | |
62 | * We abuse ->set_child_tid to avoid the new member and because it | |
63 | * can't be wrongly copied by copy_process(). We also rely on fact | |
64 | * that the caller can't exec, so PF_KTHREAD can't be cleared. | |
65 | */ | |
66 | current->set_child_tid = (__force void __user *)kthread; | |
67 | } | |
68 | ||
69 | static inline struct kthread *to_kthread(struct task_struct *k) | |
70 | { | |
71 | WARN_ON(!(k->flags & PF_KTHREAD)); | |
72 | return (__force void *)k->set_child_tid; | |
73 | } | |
74 | ||
75 | void free_kthread_struct(struct task_struct *k) | |
76 | { | |
77 | /* | |
78 | * Can be NULL if this kthread was created by kernel_thread() | |
79 | * or if kmalloc() in kthread() failed. | |
80 | */ | |
81 | kfree(to_kthread(k)); | |
82 | } | |
83 | ||
84 | /** | |
85 | * kthread_should_stop - should this kthread return now? | |
86 | * | |
87 | * When someone calls kthread_stop() on your kthread, it will be woken | |
88 | * and this will return true. You should then return, and your return | |
89 | * value will be passed through to kthread_stop(). | |
90 | */ | |
91 | bool kthread_should_stop(void) | |
92 | { | |
93 | return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags); | |
94 | } | |
95 | EXPORT_SYMBOL(kthread_should_stop); | |
96 | ||
97 | /** | |
98 | * kthread_should_park - should this kthread park now? | |
99 | * | |
100 | * When someone calls kthread_park() on your kthread, it will be woken | |
101 | * and this will return true. You should then do the necessary | |
102 | * cleanup and call kthread_parkme() | |
103 | * | |
104 | * Similar to kthread_should_stop(), but this keeps the thread alive | |
105 | * and in a park position. kthread_unpark() "restarts" the thread and | |
106 | * calls the thread function again. | |
107 | */ | |
108 | bool kthread_should_park(void) | |
109 | { | |
110 | return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags); | |
111 | } | |
112 | EXPORT_SYMBOL_GPL(kthread_should_park); | |
113 | ||
114 | /** | |
115 | * kthread_freezable_should_stop - should this freezable kthread return now? | |
116 | * @was_frozen: optional out parameter, indicates whether %current was frozen | |
117 | * | |
118 | * kthread_should_stop() for freezable kthreads, which will enter | |
119 | * refrigerator if necessary. This function is safe from kthread_stop() / | |
120 | * freezer deadlock and freezable kthreads should use this function instead | |
121 | * of calling try_to_freeze() directly. | |
122 | */ | |
123 | bool kthread_freezable_should_stop(bool *was_frozen) | |
124 | { | |
125 | bool frozen = false; | |
126 | ||
127 | might_sleep(); | |
128 | ||
129 | if (unlikely(freezing(current))) | |
130 | frozen = __refrigerator(true); | |
131 | ||
132 | if (was_frozen) | |
133 | *was_frozen = frozen; | |
134 | ||
135 | return kthread_should_stop(); | |
136 | } | |
137 | EXPORT_SYMBOL_GPL(kthread_freezable_should_stop); | |
138 | ||
139 | /** | |
140 | * kthread_data - return data value specified on kthread creation | |
141 | * @task: kthread task in question | |
142 | * | |
143 | * Return the data value specified when kthread @task was created. | |
144 | * The caller is responsible for ensuring the validity of @task when | |
145 | * calling this function. | |
146 | */ | |
147 | void *kthread_data(struct task_struct *task) | |
148 | { | |
149 | return to_kthread(task)->data; | |
150 | } | |
151 | ||
152 | /** | |
153 | * kthread_probe_data - speculative version of kthread_data() | |
154 | * @task: possible kthread task in question | |
155 | * | |
156 | * @task could be a kthread task. Return the data value specified when it | |
157 | * was created if accessible. If @task isn't a kthread task or its data is | |
158 | * inaccessible for any reason, %NULL is returned. This function requires | |
159 | * that @task itself is safe to dereference. | |
160 | */ | |
161 | void *kthread_probe_data(struct task_struct *task) | |
162 | { | |
163 | struct kthread *kthread = to_kthread(task); | |
164 | void *data = NULL; | |
165 | ||
166 | probe_kernel_read(&data, &kthread->data, sizeof(data)); | |
167 | return data; | |
168 | } | |
169 | ||
170 | static void __kthread_parkme(struct kthread *self) | |
171 | { | |
172 | __set_current_state(TASK_PARKED); | |
173 | while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) { | |
174 | if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags)) | |
175 | complete(&self->parked); | |
176 | schedule(); | |
177 | __set_current_state(TASK_PARKED); | |
178 | } | |
179 | clear_bit(KTHREAD_IS_PARKED, &self->flags); | |
180 | __set_current_state(TASK_RUNNING); | |
181 | } | |
182 | ||
183 | void kthread_parkme(void) | |
184 | { | |
185 | __kthread_parkme(to_kthread(current)); | |
186 | } | |
187 | EXPORT_SYMBOL_GPL(kthread_parkme); | |
188 | ||
189 | static int kthread(void *_create) | |
190 | { | |
191 | /* Copy data: it's on kthread's stack */ | |
192 | struct kthread_create_info *create = _create; | |
193 | int (*threadfn)(void *data) = create->threadfn; | |
194 | void *data = create->data; | |
195 | struct completion *done; | |
196 | struct kthread *self; | |
197 | int ret; | |
198 | ||
199 | self = kmalloc(sizeof(*self), GFP_KERNEL); | |
200 | set_kthread_struct(self); | |
201 | ||
202 | /* If user was SIGKILLed, I release the structure. */ | |
203 | done = xchg(&create->done, NULL); | |
204 | if (!done) { | |
205 | kfree(create); | |
206 | do_exit(-EINTR); | |
207 | } | |
208 | ||
209 | if (!self) { | |
210 | create->result = ERR_PTR(-ENOMEM); | |
211 | complete(done); | |
212 | do_exit(-ENOMEM); | |
213 | } | |
214 | ||
215 | self->flags = 0; | |
216 | self->data = data; | |
217 | init_completion(&self->exited); | |
218 | init_completion(&self->parked); | |
219 | current->vfork_done = &self->exited; | |
220 | ||
221 | /* OK, tell user we're spawned, wait for stop or wakeup */ | |
222 | __set_current_state(TASK_UNINTERRUPTIBLE); | |
223 | create->result = current; | |
224 | complete(done); | |
225 | schedule(); | |
226 | ||
227 | ret = -EINTR; | |
228 | if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) { | |
229 | cgroup_kthread_ready(); | |
230 | __kthread_parkme(self); | |
231 | ret = threadfn(data); | |
232 | } | |
233 | do_exit(ret); | |
234 | } | |
235 | ||
236 | /* called from do_fork() to get node information for about to be created task */ | |
237 | int tsk_fork_get_node(struct task_struct *tsk) | |
238 | { | |
239 | #ifdef CONFIG_NUMA | |
240 | if (tsk == kthreadd_task) | |
241 | return tsk->pref_node_fork; | |
242 | #endif | |
243 | return NUMA_NO_NODE; | |
244 | } | |
245 | ||
246 | static void create_kthread(struct kthread_create_info *create) | |
247 | { | |
248 | int pid; | |
249 | ||
250 | #ifdef CONFIG_NUMA | |
251 | current->pref_node_fork = create->node; | |
252 | #endif | |
253 | /* We want our own signal handler (we take no signals by default). */ | |
254 | pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD); | |
255 | if (pid < 0) { | |
256 | /* If user was SIGKILLed, I release the structure. */ | |
257 | struct completion *done = xchg(&create->done, NULL); | |
258 | ||
259 | if (!done) { | |
260 | kfree(create); | |
261 | return; | |
262 | } | |
263 | create->result = ERR_PTR(pid); | |
264 | complete(done); | |
265 | } | |
266 | } | |
267 | ||
268 | static __printf(4, 0) | |
269 | struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data), | |
270 | void *data, int node, | |
271 | const char namefmt[], | |
272 | va_list args) | |
273 | { | |
274 | DECLARE_COMPLETION_ONSTACK(done); | |
275 | struct task_struct *task; | |
276 | struct kthread_create_info *create = kmalloc(sizeof(*create), | |
277 | GFP_KERNEL); | |
278 | ||
279 | if (!create) | |
280 | return ERR_PTR(-ENOMEM); | |
281 | create->threadfn = threadfn; | |
282 | create->data = data; | |
283 | create->node = node; | |
284 | create->done = &done; | |
285 | ||
286 | spin_lock(&kthread_create_lock); | |
287 | list_add_tail(&create->list, &kthread_create_list); | |
288 | spin_unlock(&kthread_create_lock); | |
289 | ||
290 | wake_up_process(kthreadd_task); | |
291 | /* | |
292 | * Wait for completion in killable state, for I might be chosen by | |
293 | * the OOM killer while kthreadd is trying to allocate memory for | |
294 | * new kernel thread. | |
295 | */ | |
296 | if (unlikely(wait_for_completion_killable(&done))) { | |
297 | int i = 0; | |
298 | ||
299 | /* | |
300 | * I got SIGKILL, but wait for 10 more seconds for completion | |
301 | * unless chosen by the OOM killer. This delay is there as a | |
302 | * workaround for boot failure caused by SIGKILL upon device | |
303 | * driver initialization timeout. | |
304 | */ | |
305 | while (i++ < 10 && !test_tsk_thread_flag(current, TIF_MEMDIE)) | |
306 | if (wait_for_completion_timeout(&done, HZ)) | |
307 | goto ready; | |
308 | /* | |
309 | * If I was SIGKILLed before kthreadd (or new kernel thread) | |
310 | * calls complete(), leave the cleanup of this structure to | |
311 | * that thread. | |
312 | */ | |
313 | if (xchg(&create->done, NULL)) | |
314 | return ERR_PTR(-EINTR); | |
315 | /* | |
316 | * kthreadd (or new kernel thread) will call complete() | |
317 | * shortly. | |
318 | */ | |
319 | wait_for_completion(&done); | |
320 | } | |
321 | ready: | |
322 | task = create->result; | |
323 | if (!IS_ERR(task)) { | |
324 | static const struct sched_param param = { .sched_priority = 0 }; | |
325 | ||
326 | vsnprintf(task->comm, sizeof(task->comm), namefmt, args); | |
327 | /* | |
328 | * root may have changed our (kthreadd's) priority or CPU mask. | |
329 | * The kernel thread should not inherit these properties. | |
330 | */ | |
331 | sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m); | |
332 | set_cpus_allowed_ptr(task, cpu_all_mask); | |
333 | } | |
334 | kfree(create); | |
335 | return task; | |
336 | } | |
337 | ||
338 | /** | |
339 | * kthread_create_on_node - create a kthread. | |
340 | * @threadfn: the function to run until signal_pending(current). | |
341 | * @data: data ptr for @threadfn. | |
342 | * @node: task and thread structures for the thread are allocated on this node | |
343 | * @namefmt: printf-style name for the thread. | |
344 | * | |
345 | * Description: This helper function creates and names a kernel | |
346 | * thread. The thread will be stopped: use wake_up_process() to start | |
347 | * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and | |
348 | * is affine to all CPUs. | |
349 | * | |
350 | * If thread is going to be bound on a particular cpu, give its node | |
351 | * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE. | |
352 | * When woken, the thread will run @threadfn() with @data as its | |
353 | * argument. @threadfn() can either call do_exit() directly if it is a | |
354 | * standalone thread for which no one will call kthread_stop(), or | |
355 | * return when 'kthread_should_stop()' is true (which means | |
356 | * kthread_stop() has been called). The return value should be zero | |
357 | * or a negative error number; it will be passed to kthread_stop(). | |
358 | * | |
359 | * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR). | |
360 | */ | |
361 | struct task_struct *kthread_create_on_node(int (*threadfn)(void *data), | |
362 | void *data, int node, | |
363 | const char namefmt[], | |
364 | ...) | |
365 | { | |
366 | struct task_struct *task; | |
367 | va_list args; | |
368 | ||
369 | va_start(args, namefmt); | |
370 | task = __kthread_create_on_node(threadfn, data, node, namefmt, args); | |
371 | va_end(args); | |
372 | ||
373 | return task; | |
374 | } | |
375 | EXPORT_SYMBOL(kthread_create_on_node); | |
376 | ||
377 | static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state) | |
378 | { | |
379 | unsigned long flags; | |
380 | ||
381 | if (!wait_task_inactive(p, state)) { | |
382 | WARN_ON(1); | |
383 | return; | |
384 | } | |
385 | ||
386 | /* It's safe because the task is inactive. */ | |
387 | raw_spin_lock_irqsave(&p->pi_lock, flags); | |
388 | do_set_cpus_allowed(p, mask); | |
389 | p->flags |= PF_NO_SETAFFINITY; | |
390 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
391 | } | |
392 | ||
393 | static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state) | |
394 | { | |
395 | __kthread_bind_mask(p, cpumask_of(cpu), state); | |
396 | } | |
397 | ||
398 | void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask) | |
399 | { | |
400 | __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE); | |
401 | } | |
402 | ||
403 | /** | |
404 | * kthread_bind - bind a just-created kthread to a cpu. | |
405 | * @p: thread created by kthread_create(). | |
406 | * @cpu: cpu (might not be online, must be possible) for @k to run on. | |
407 | * | |
408 | * Description: This function is equivalent to set_cpus_allowed(), | |
409 | * except that @cpu doesn't need to be online, and the thread must be | |
410 | * stopped (i.e., just returned from kthread_create()). | |
411 | */ | |
412 | void kthread_bind(struct task_struct *p, unsigned int cpu) | |
413 | { | |
414 | __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE); | |
415 | } | |
416 | EXPORT_SYMBOL(kthread_bind); | |
417 | ||
418 | /** | |
419 | * kthread_create_on_cpu - Create a cpu bound kthread | |
420 | * @threadfn: the function to run until signal_pending(current). | |
421 | * @data: data ptr for @threadfn. | |
422 | * @cpu: The cpu on which the thread should be bound, | |
423 | * @namefmt: printf-style name for the thread. Format is restricted | |
424 | * to "name.*%u". Code fills in cpu number. | |
425 | * | |
426 | * Description: This helper function creates and names a kernel thread | |
427 | * The thread will be woken and put into park mode. | |
428 | */ | |
429 | struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data), | |
430 | void *data, unsigned int cpu, | |
431 | const char *namefmt) | |
432 | { | |
433 | struct task_struct *p; | |
434 | ||
435 | p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt, | |
436 | cpu); | |
437 | if (IS_ERR(p)) | |
438 | return p; | |
439 | kthread_bind(p, cpu); | |
440 | /* CPU hotplug need to bind once again when unparking the thread. */ | |
441 | set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags); | |
442 | to_kthread(p)->cpu = cpu; | |
443 | return p; | |
444 | } | |
445 | ||
446 | /** | |
447 | * kthread_unpark - unpark a thread created by kthread_create(). | |
448 | * @k: thread created by kthread_create(). | |
449 | * | |
450 | * Sets kthread_should_park() for @k to return false, wakes it, and | |
451 | * waits for it to return. If the thread is marked percpu then its | |
452 | * bound to the cpu again. | |
453 | */ | |
454 | void kthread_unpark(struct task_struct *k) | |
455 | { | |
456 | struct kthread *kthread = to_kthread(k); | |
457 | ||
458 | clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags); | |
459 | /* | |
460 | * We clear the IS_PARKED bit here as we don't wait | |
461 | * until the task has left the park code. So if we'd | |
462 | * park before that happens we'd see the IS_PARKED bit | |
463 | * which might be about to be cleared. | |
464 | */ | |
465 | if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) { | |
466 | /* | |
467 | * Newly created kthread was parked when the CPU was offline. | |
468 | * The binding was lost and we need to set it again. | |
469 | */ | |
470 | if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags)) | |
471 | __kthread_bind(k, kthread->cpu, TASK_PARKED); | |
472 | wake_up_state(k, TASK_PARKED); | |
473 | } | |
474 | } | |
475 | EXPORT_SYMBOL_GPL(kthread_unpark); | |
476 | ||
477 | /** | |
478 | * kthread_park - park a thread created by kthread_create(). | |
479 | * @k: thread created by kthread_create(). | |
480 | * | |
481 | * Sets kthread_should_park() for @k to return true, wakes it, and | |
482 | * waits for it to return. This can also be called after kthread_create() | |
483 | * instead of calling wake_up_process(): the thread will park without | |
484 | * calling threadfn(). | |
485 | * | |
486 | * Returns 0 if the thread is parked, -ENOSYS if the thread exited. | |
487 | * If called by the kthread itself just the park bit is set. | |
488 | */ | |
489 | int kthread_park(struct task_struct *k) | |
490 | { | |
491 | struct kthread *kthread = to_kthread(k); | |
492 | ||
493 | if (WARN_ON(k->flags & PF_EXITING)) | |
494 | return -ENOSYS; | |
495 | ||
496 | if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) { | |
497 | set_bit(KTHREAD_SHOULD_PARK, &kthread->flags); | |
498 | if (k != current) { | |
499 | wake_up_process(k); | |
500 | wait_for_completion(&kthread->parked); | |
501 | } | |
502 | } | |
503 | ||
504 | return 0; | |
505 | } | |
506 | EXPORT_SYMBOL_GPL(kthread_park); | |
507 | ||
508 | /** | |
509 | * kthread_stop - stop a thread created by kthread_create(). | |
510 | * @k: thread created by kthread_create(). | |
511 | * | |
512 | * Sets kthread_should_stop() for @k to return true, wakes it, and | |
513 | * waits for it to exit. This can also be called after kthread_create() | |
514 | * instead of calling wake_up_process(): the thread will exit without | |
515 | * calling threadfn(). | |
516 | * | |
517 | * If threadfn() may call do_exit() itself, the caller must ensure | |
518 | * task_struct can't go away. | |
519 | * | |
520 | * Returns the result of threadfn(), or %-EINTR if wake_up_process() | |
521 | * was never called. | |
522 | */ | |
523 | int kthread_stop(struct task_struct *k) | |
524 | { | |
525 | struct kthread *kthread; | |
526 | int ret; | |
527 | ||
528 | trace_sched_kthread_stop(k); | |
529 | ||
530 | get_task_struct(k); | |
531 | kthread = to_kthread(k); | |
532 | set_bit(KTHREAD_SHOULD_STOP, &kthread->flags); | |
533 | kthread_unpark(k); | |
534 | wake_up_process(k); | |
535 | wait_for_completion(&kthread->exited); | |
536 | ret = k->exit_code; | |
537 | put_task_struct(k); | |
538 | ||
539 | trace_sched_kthread_stop_ret(ret); | |
540 | return ret; | |
541 | } | |
542 | EXPORT_SYMBOL(kthread_stop); | |
543 | ||
544 | int kthreadd(void *unused) | |
545 | { | |
546 | struct task_struct *tsk = current; | |
547 | ||
548 | /* Setup a clean context for our children to inherit. */ | |
549 | set_task_comm(tsk, "kthreadd"); | |
550 | ignore_signals(tsk); | |
551 | set_cpus_allowed_ptr(tsk, cpu_all_mask); | |
552 | set_mems_allowed(node_states[N_MEMORY]); | |
553 | ||
554 | current->flags |= PF_NOFREEZE; | |
555 | cgroup_init_kthreadd(); | |
556 | ||
557 | for (;;) { | |
558 | set_current_state(TASK_INTERRUPTIBLE); | |
559 | if (list_empty(&kthread_create_list)) | |
560 | schedule(); | |
561 | __set_current_state(TASK_RUNNING); | |
562 | ||
563 | spin_lock(&kthread_create_lock); | |
564 | while (!list_empty(&kthread_create_list)) { | |
565 | struct kthread_create_info *create; | |
566 | ||
567 | create = list_entry(kthread_create_list.next, | |
568 | struct kthread_create_info, list); | |
569 | list_del_init(&create->list); | |
570 | spin_unlock(&kthread_create_lock); | |
571 | ||
572 | create_kthread(create); | |
573 | ||
574 | spin_lock(&kthread_create_lock); | |
575 | } | |
576 | spin_unlock(&kthread_create_lock); | |
577 | } | |
578 | ||
579 | return 0; | |
580 | } | |
581 | ||
582 | void __kthread_init_worker(struct kthread_worker *worker, | |
583 | const char *name, | |
584 | struct lock_class_key *key) | |
585 | { | |
586 | memset(worker, 0, sizeof(struct kthread_worker)); | |
587 | spin_lock_init(&worker->lock); | |
588 | lockdep_set_class_and_name(&worker->lock, key, name); | |
589 | INIT_LIST_HEAD(&worker->work_list); | |
590 | INIT_LIST_HEAD(&worker->delayed_work_list); | |
591 | } | |
592 | EXPORT_SYMBOL_GPL(__kthread_init_worker); | |
593 | ||
594 | /** | |
595 | * kthread_worker_fn - kthread function to process kthread_worker | |
596 | * @worker_ptr: pointer to initialized kthread_worker | |
597 | * | |
598 | * This function implements the main cycle of kthread worker. It processes | |
599 | * work_list until it is stopped with kthread_stop(). It sleeps when the queue | |
600 | * is empty. | |
601 | * | |
602 | * The works are not allowed to keep any locks, disable preemption or interrupts | |
603 | * when they finish. There is defined a safe point for freezing when one work | |
604 | * finishes and before a new one is started. | |
605 | * | |
606 | * Also the works must not be handled by more than one worker at the same time, | |
607 | * see also kthread_queue_work(). | |
608 | */ | |
609 | int kthread_worker_fn(void *worker_ptr) | |
610 | { | |
611 | struct kthread_worker *worker = worker_ptr; | |
612 | struct kthread_work *work; | |
613 | ||
614 | /* | |
615 | * FIXME: Update the check and remove the assignment when all kthread | |
616 | * worker users are created using kthread_create_worker*() functions. | |
617 | */ | |
618 | WARN_ON(worker->task && worker->task != current); | |
619 | worker->task = current; | |
620 | ||
621 | if (worker->flags & KTW_FREEZABLE) | |
622 | set_freezable(); | |
623 | ||
624 | repeat: | |
625 | set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */ | |
626 | ||
627 | if (kthread_should_stop()) { | |
628 | __set_current_state(TASK_RUNNING); | |
629 | spin_lock_irq(&worker->lock); | |
630 | worker->task = NULL; | |
631 | spin_unlock_irq(&worker->lock); | |
632 | return 0; | |
633 | } | |
634 | ||
635 | work = NULL; | |
636 | spin_lock_irq(&worker->lock); | |
637 | if (!list_empty(&worker->work_list)) { | |
638 | work = list_first_entry(&worker->work_list, | |
639 | struct kthread_work, node); | |
640 | list_del_init(&work->node); | |
641 | } | |
642 | worker->current_work = work; | |
643 | spin_unlock_irq(&worker->lock); | |
644 | ||
645 | if (work) { | |
646 | __set_current_state(TASK_RUNNING); | |
647 | work->func(work); | |
648 | } else if (!freezing(current)) | |
649 | schedule(); | |
650 | ||
651 | try_to_freeze(); | |
652 | cond_resched(); | |
653 | goto repeat; | |
654 | } | |
655 | EXPORT_SYMBOL_GPL(kthread_worker_fn); | |
656 | ||
657 | static __printf(3, 0) struct kthread_worker * | |
658 | __kthread_create_worker(int cpu, unsigned int flags, | |
659 | const char namefmt[], va_list args) | |
660 | { | |
661 | struct kthread_worker *worker; | |
662 | struct task_struct *task; | |
663 | int node = -1; | |
664 | ||
665 | worker = kzalloc(sizeof(*worker), GFP_KERNEL); | |
666 | if (!worker) | |
667 | return ERR_PTR(-ENOMEM); | |
668 | ||
669 | kthread_init_worker(worker); | |
670 | ||
671 | if (cpu >= 0) | |
672 | node = cpu_to_node(cpu); | |
673 | ||
674 | task = __kthread_create_on_node(kthread_worker_fn, worker, | |
675 | node, namefmt, args); | |
676 | if (IS_ERR(task)) | |
677 | goto fail_task; | |
678 | ||
679 | if (cpu >= 0) | |
680 | kthread_bind(task, cpu); | |
681 | ||
682 | worker->flags = flags; | |
683 | worker->task = task; | |
684 | wake_up_process(task); | |
685 | return worker; | |
686 | ||
687 | fail_task: | |
688 | kfree(worker); | |
689 | return ERR_CAST(task); | |
690 | } | |
691 | ||
692 | /** | |
693 | * kthread_create_worker - create a kthread worker | |
694 | * @flags: flags modifying the default behavior of the worker | |
695 | * @namefmt: printf-style name for the kthread worker (task). | |
696 | * | |
697 | * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM) | |
698 | * when the needed structures could not get allocated, and ERR_PTR(-EINTR) | |
699 | * when the worker was SIGKILLed. | |
700 | */ | |
701 | struct kthread_worker * | |
702 | kthread_create_worker(unsigned int flags, const char namefmt[], ...) | |
703 | { | |
704 | struct kthread_worker *worker; | |
705 | va_list args; | |
706 | ||
707 | va_start(args, namefmt); | |
708 | worker = __kthread_create_worker(-1, flags, namefmt, args); | |
709 | va_end(args); | |
710 | ||
711 | return worker; | |
712 | } | |
713 | EXPORT_SYMBOL(kthread_create_worker); | |
714 | ||
715 | /** | |
716 | * kthread_create_worker_on_cpu - create a kthread worker and bind it | |
717 | * it to a given CPU and the associated NUMA node. | |
718 | * @cpu: CPU number | |
719 | * @flags: flags modifying the default behavior of the worker | |
720 | * @namefmt: printf-style name for the kthread worker (task). | |
721 | * | |
722 | * Use a valid CPU number if you want to bind the kthread worker | |
723 | * to the given CPU and the associated NUMA node. | |
724 | * | |
725 | * A good practice is to add the cpu number also into the worker name. | |
726 | * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu). | |
727 | * | |
728 | * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM) | |
729 | * when the needed structures could not get allocated, and ERR_PTR(-EINTR) | |
730 | * when the worker was SIGKILLed. | |
731 | */ | |
732 | struct kthread_worker * | |
733 | kthread_create_worker_on_cpu(int cpu, unsigned int flags, | |
734 | const char namefmt[], ...) | |
735 | { | |
736 | struct kthread_worker *worker; | |
737 | va_list args; | |
738 | ||
739 | va_start(args, namefmt); | |
740 | worker = __kthread_create_worker(cpu, flags, namefmt, args); | |
741 | va_end(args); | |
742 | ||
743 | return worker; | |
744 | } | |
745 | EXPORT_SYMBOL(kthread_create_worker_on_cpu); | |
746 | ||
747 | /* | |
748 | * Returns true when the work could not be queued at the moment. | |
749 | * It happens when it is already pending in a worker list | |
750 | * or when it is being cancelled. | |
751 | */ | |
752 | static inline bool queuing_blocked(struct kthread_worker *worker, | |
753 | struct kthread_work *work) | |
754 | { | |
755 | lockdep_assert_held(&worker->lock); | |
756 | ||
757 | return !list_empty(&work->node) || work->canceling; | |
758 | } | |
759 | ||
760 | static void kthread_insert_work_sanity_check(struct kthread_worker *worker, | |
761 | struct kthread_work *work) | |
762 | { | |
763 | lockdep_assert_held(&worker->lock); | |
764 | WARN_ON_ONCE(!list_empty(&work->node)); | |
765 | /* Do not use a work with >1 worker, see kthread_queue_work() */ | |
766 | WARN_ON_ONCE(work->worker && work->worker != worker); | |
767 | } | |
768 | ||
769 | /* insert @work before @pos in @worker */ | |
770 | static void kthread_insert_work(struct kthread_worker *worker, | |
771 | struct kthread_work *work, | |
772 | struct list_head *pos) | |
773 | { | |
774 | kthread_insert_work_sanity_check(worker, work); | |
775 | ||
776 | list_add_tail(&work->node, pos); | |
777 | work->worker = worker; | |
778 | if (!worker->current_work && likely(worker->task)) | |
779 | wake_up_process(worker->task); | |
780 | } | |
781 | ||
782 | /** | |
783 | * kthread_queue_work - queue a kthread_work | |
784 | * @worker: target kthread_worker | |
785 | * @work: kthread_work to queue | |
786 | * | |
787 | * Queue @work to work processor @task for async execution. @task | |
788 | * must have been created with kthread_worker_create(). Returns %true | |
789 | * if @work was successfully queued, %false if it was already pending. | |
790 | * | |
791 | * Reinitialize the work if it needs to be used by another worker. | |
792 | * For example, when the worker was stopped and started again. | |
793 | */ | |
794 | bool kthread_queue_work(struct kthread_worker *worker, | |
795 | struct kthread_work *work) | |
796 | { | |
797 | bool ret = false; | |
798 | unsigned long flags; | |
799 | ||
800 | spin_lock_irqsave(&worker->lock, flags); | |
801 | if (!queuing_blocked(worker, work)) { | |
802 | kthread_insert_work(worker, work, &worker->work_list); | |
803 | ret = true; | |
804 | } | |
805 | spin_unlock_irqrestore(&worker->lock, flags); | |
806 | return ret; | |
807 | } | |
808 | EXPORT_SYMBOL_GPL(kthread_queue_work); | |
809 | ||
810 | /** | |
811 | * kthread_delayed_work_timer_fn - callback that queues the associated kthread | |
812 | * delayed work when the timer expires. | |
813 | * @__data: pointer to the data associated with the timer | |
814 | * | |
815 | * The format of the function is defined by struct timer_list. | |
816 | * It should have been called from irqsafe timer with irq already off. | |
817 | */ | |
818 | void kthread_delayed_work_timer_fn(unsigned long __data) | |
819 | { | |
820 | struct kthread_delayed_work *dwork = | |
821 | (struct kthread_delayed_work *)__data; | |
822 | struct kthread_work *work = &dwork->work; | |
823 | struct kthread_worker *worker = work->worker; | |
824 | ||
825 | /* | |
826 | * This might happen when a pending work is reinitialized. | |
827 | * It means that it is used a wrong way. | |
828 | */ | |
829 | if (WARN_ON_ONCE(!worker)) | |
830 | return; | |
831 | ||
832 | spin_lock(&worker->lock); | |
833 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ | |
834 | WARN_ON_ONCE(work->worker != worker); | |
835 | ||
836 | /* Move the work from worker->delayed_work_list. */ | |
837 | WARN_ON_ONCE(list_empty(&work->node)); | |
838 | list_del_init(&work->node); | |
839 | kthread_insert_work(worker, work, &worker->work_list); | |
840 | ||
841 | spin_unlock(&worker->lock); | |
842 | } | |
843 | EXPORT_SYMBOL(kthread_delayed_work_timer_fn); | |
844 | ||
845 | void __kthread_queue_delayed_work(struct kthread_worker *worker, | |
846 | struct kthread_delayed_work *dwork, | |
847 | unsigned long delay) | |
848 | { | |
849 | struct timer_list *timer = &dwork->timer; | |
850 | struct kthread_work *work = &dwork->work; | |
851 | ||
852 | WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn || | |
853 | timer->data != (unsigned long)dwork); | |
854 | ||
855 | /* | |
856 | * If @delay is 0, queue @dwork->work immediately. This is for | |
857 | * both optimization and correctness. The earliest @timer can | |
858 | * expire is on the closest next tick and delayed_work users depend | |
859 | * on that there's no such delay when @delay is 0. | |
860 | */ | |
861 | if (!delay) { | |
862 | kthread_insert_work(worker, work, &worker->work_list); | |
863 | return; | |
864 | } | |
865 | ||
866 | /* Be paranoid and try to detect possible races already now. */ | |
867 | kthread_insert_work_sanity_check(worker, work); | |
868 | ||
869 | list_add(&work->node, &worker->delayed_work_list); | |
870 | work->worker = worker; | |
871 | timer->expires = jiffies + delay; | |
872 | add_timer(timer); | |
873 | } | |
874 | ||
875 | /** | |
876 | * kthread_queue_delayed_work - queue the associated kthread work | |
877 | * after a delay. | |
878 | * @worker: target kthread_worker | |
879 | * @dwork: kthread_delayed_work to queue | |
880 | * @delay: number of jiffies to wait before queuing | |
881 | * | |
882 | * If the work has not been pending it starts a timer that will queue | |
883 | * the work after the given @delay. If @delay is zero, it queues the | |
884 | * work immediately. | |
885 | * | |
886 | * Return: %false if the @work has already been pending. It means that | |
887 | * either the timer was running or the work was queued. It returns %true | |
888 | * otherwise. | |
889 | */ | |
890 | bool kthread_queue_delayed_work(struct kthread_worker *worker, | |
891 | struct kthread_delayed_work *dwork, | |
892 | unsigned long delay) | |
893 | { | |
894 | struct kthread_work *work = &dwork->work; | |
895 | unsigned long flags; | |
896 | bool ret = false; | |
897 | ||
898 | spin_lock_irqsave(&worker->lock, flags); | |
899 | ||
900 | if (!queuing_blocked(worker, work)) { | |
901 | __kthread_queue_delayed_work(worker, dwork, delay); | |
902 | ret = true; | |
903 | } | |
904 | ||
905 | spin_unlock_irqrestore(&worker->lock, flags); | |
906 | return ret; | |
907 | } | |
908 | EXPORT_SYMBOL_GPL(kthread_queue_delayed_work); | |
909 | ||
910 | struct kthread_flush_work { | |
911 | struct kthread_work work; | |
912 | struct completion done; | |
913 | }; | |
914 | ||
915 | static void kthread_flush_work_fn(struct kthread_work *work) | |
916 | { | |
917 | struct kthread_flush_work *fwork = | |
918 | container_of(work, struct kthread_flush_work, work); | |
919 | complete(&fwork->done); | |
920 | } | |
921 | ||
922 | /** | |
923 | * kthread_flush_work - flush a kthread_work | |
924 | * @work: work to flush | |
925 | * | |
926 | * If @work is queued or executing, wait for it to finish execution. | |
927 | */ | |
928 | void kthread_flush_work(struct kthread_work *work) | |
929 | { | |
930 | struct kthread_flush_work fwork = { | |
931 | KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn), | |
932 | COMPLETION_INITIALIZER_ONSTACK(fwork.done), | |
933 | }; | |
934 | struct kthread_worker *worker; | |
935 | bool noop = false; | |
936 | ||
937 | worker = work->worker; | |
938 | if (!worker) | |
939 | return; | |
940 | ||
941 | spin_lock_irq(&worker->lock); | |
942 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ | |
943 | WARN_ON_ONCE(work->worker != worker); | |
944 | ||
945 | if (!list_empty(&work->node)) | |
946 | kthread_insert_work(worker, &fwork.work, work->node.next); | |
947 | else if (worker->current_work == work) | |
948 | kthread_insert_work(worker, &fwork.work, | |
949 | worker->work_list.next); | |
950 | else | |
951 | noop = true; | |
952 | ||
953 | spin_unlock_irq(&worker->lock); | |
954 | ||
955 | if (!noop) | |
956 | wait_for_completion(&fwork.done); | |
957 | } | |
958 | EXPORT_SYMBOL_GPL(kthread_flush_work); | |
959 | ||
960 | /* | |
961 | * This function removes the work from the worker queue. Also it makes sure | |
962 | * that it won't get queued later via the delayed work's timer. | |
963 | * | |
964 | * The work might still be in use when this function finishes. See the | |
965 | * current_work proceed by the worker. | |
966 | * | |
967 | * Return: %true if @work was pending and successfully canceled, | |
968 | * %false if @work was not pending | |
969 | */ | |
970 | static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork, | |
971 | unsigned long *flags) | |
972 | { | |
973 | /* Try to cancel the timer if exists. */ | |
974 | if (is_dwork) { | |
975 | struct kthread_delayed_work *dwork = | |
976 | container_of(work, struct kthread_delayed_work, work); | |
977 | struct kthread_worker *worker = work->worker; | |
978 | ||
979 | /* | |
980 | * del_timer_sync() must be called to make sure that the timer | |
981 | * callback is not running. The lock must be temporary released | |
982 | * to avoid a deadlock with the callback. In the meantime, | |
983 | * any queuing is blocked by setting the canceling counter. | |
984 | */ | |
985 | work->canceling++; | |
986 | spin_unlock_irqrestore(&worker->lock, *flags); | |
987 | del_timer_sync(&dwork->timer); | |
988 | spin_lock_irqsave(&worker->lock, *flags); | |
989 | work->canceling--; | |
990 | } | |
991 | ||
992 | /* | |
993 | * Try to remove the work from a worker list. It might either | |
994 | * be from worker->work_list or from worker->delayed_work_list. | |
995 | */ | |
996 | if (!list_empty(&work->node)) { | |
997 | list_del_init(&work->node); | |
998 | return true; | |
999 | } | |
1000 | ||
1001 | return false; | |
1002 | } | |
1003 | ||
1004 | /** | |
1005 | * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work | |
1006 | * @worker: kthread worker to use | |
1007 | * @dwork: kthread delayed work to queue | |
1008 | * @delay: number of jiffies to wait before queuing | |
1009 | * | |
1010 | * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise, | |
1011 | * modify @dwork's timer so that it expires after @delay. If @delay is zero, | |
1012 | * @work is guaranteed to be queued immediately. | |
1013 | * | |
1014 | * Return: %true if @dwork was pending and its timer was modified, | |
1015 | * %false otherwise. | |
1016 | * | |
1017 | * A special case is when the work is being canceled in parallel. | |
1018 | * It might be caused either by the real kthread_cancel_delayed_work_sync() | |
1019 | * or yet another kthread_mod_delayed_work() call. We let the other command | |
1020 | * win and return %false here. The caller is supposed to synchronize these | |
1021 | * operations a reasonable way. | |
1022 | * | |
1023 | * This function is safe to call from any context including IRQ handler. | |
1024 | * See __kthread_cancel_work() and kthread_delayed_work_timer_fn() | |
1025 | * for details. | |
1026 | */ | |
1027 | bool kthread_mod_delayed_work(struct kthread_worker *worker, | |
1028 | struct kthread_delayed_work *dwork, | |
1029 | unsigned long delay) | |
1030 | { | |
1031 | struct kthread_work *work = &dwork->work; | |
1032 | unsigned long flags; | |
1033 | int ret = false; | |
1034 | ||
1035 | spin_lock_irqsave(&worker->lock, flags); | |
1036 | ||
1037 | /* Do not bother with canceling when never queued. */ | |
1038 | if (!work->worker) | |
1039 | goto fast_queue; | |
1040 | ||
1041 | /* Work must not be used with >1 worker, see kthread_queue_work() */ | |
1042 | WARN_ON_ONCE(work->worker != worker); | |
1043 | ||
1044 | /* Do not fight with another command that is canceling this work. */ | |
1045 | if (work->canceling) | |
1046 | goto out; | |
1047 | ||
1048 | ret = __kthread_cancel_work(work, true, &flags); | |
1049 | fast_queue: | |
1050 | __kthread_queue_delayed_work(worker, dwork, delay); | |
1051 | out: | |
1052 | spin_unlock_irqrestore(&worker->lock, flags); | |
1053 | return ret; | |
1054 | } | |
1055 | EXPORT_SYMBOL_GPL(kthread_mod_delayed_work); | |
1056 | ||
1057 | static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork) | |
1058 | { | |
1059 | struct kthread_worker *worker = work->worker; | |
1060 | unsigned long flags; | |
1061 | int ret = false; | |
1062 | ||
1063 | if (!worker) | |
1064 | goto out; | |
1065 | ||
1066 | spin_lock_irqsave(&worker->lock, flags); | |
1067 | /* Work must not be used with >1 worker, see kthread_queue_work(). */ | |
1068 | WARN_ON_ONCE(work->worker != worker); | |
1069 | ||
1070 | ret = __kthread_cancel_work(work, is_dwork, &flags); | |
1071 | ||
1072 | if (worker->current_work != work) | |
1073 | goto out_fast; | |
1074 | ||
1075 | /* | |
1076 | * The work is in progress and we need to wait with the lock released. | |
1077 | * In the meantime, block any queuing by setting the canceling counter. | |
1078 | */ | |
1079 | work->canceling++; | |
1080 | spin_unlock_irqrestore(&worker->lock, flags); | |
1081 | kthread_flush_work(work); | |
1082 | spin_lock_irqsave(&worker->lock, flags); | |
1083 | work->canceling--; | |
1084 | ||
1085 | out_fast: | |
1086 | spin_unlock_irqrestore(&worker->lock, flags); | |
1087 | out: | |
1088 | return ret; | |
1089 | } | |
1090 | ||
1091 | /** | |
1092 | * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish | |
1093 | * @work: the kthread work to cancel | |
1094 | * | |
1095 | * Cancel @work and wait for its execution to finish. This function | |
1096 | * can be used even if the work re-queues itself. On return from this | |
1097 | * function, @work is guaranteed to be not pending or executing on any CPU. | |
1098 | * | |
1099 | * kthread_cancel_work_sync(&delayed_work->work) must not be used for | |
1100 | * delayed_work's. Use kthread_cancel_delayed_work_sync() instead. | |
1101 | * | |
1102 | * The caller must ensure that the worker on which @work was last | |
1103 | * queued can't be destroyed before this function returns. | |
1104 | * | |
1105 | * Return: %true if @work was pending, %false otherwise. | |
1106 | */ | |
1107 | bool kthread_cancel_work_sync(struct kthread_work *work) | |
1108 | { | |
1109 | return __kthread_cancel_work_sync(work, false); | |
1110 | } | |
1111 | EXPORT_SYMBOL_GPL(kthread_cancel_work_sync); | |
1112 | ||
1113 | /** | |
1114 | * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and | |
1115 | * wait for it to finish. | |
1116 | * @dwork: the kthread delayed work to cancel | |
1117 | * | |
1118 | * This is kthread_cancel_work_sync() for delayed works. | |
1119 | * | |
1120 | * Return: %true if @dwork was pending, %false otherwise. | |
1121 | */ | |
1122 | bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork) | |
1123 | { | |
1124 | return __kthread_cancel_work_sync(&dwork->work, true); | |
1125 | } | |
1126 | EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync); | |
1127 | ||
1128 | /** | |
1129 | * kthread_flush_worker - flush all current works on a kthread_worker | |
1130 | * @worker: worker to flush | |
1131 | * | |
1132 | * Wait until all currently executing or pending works on @worker are | |
1133 | * finished. | |
1134 | */ | |
1135 | void kthread_flush_worker(struct kthread_worker *worker) | |
1136 | { | |
1137 | struct kthread_flush_work fwork = { | |
1138 | KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn), | |
1139 | COMPLETION_INITIALIZER_ONSTACK(fwork.done), | |
1140 | }; | |
1141 | ||
1142 | kthread_queue_work(worker, &fwork.work); | |
1143 | wait_for_completion(&fwork.done); | |
1144 | } | |
1145 | EXPORT_SYMBOL_GPL(kthread_flush_worker); | |
1146 | ||
1147 | /** | |
1148 | * kthread_destroy_worker - destroy a kthread worker | |
1149 | * @worker: worker to be destroyed | |
1150 | * | |
1151 | * Flush and destroy @worker. The simple flush is enough because the kthread | |
1152 | * worker API is used only in trivial scenarios. There are no multi-step state | |
1153 | * machines needed. | |
1154 | */ | |
1155 | void kthread_destroy_worker(struct kthread_worker *worker) | |
1156 | { | |
1157 | struct task_struct *task; | |
1158 | ||
1159 | task = worker->task; | |
1160 | if (WARN_ON(!task)) | |
1161 | return; | |
1162 | ||
1163 | kthread_flush_worker(worker); | |
1164 | kthread_stop(task); | |
1165 | WARN_ON(!list_empty(&worker->work_list)); | |
1166 | kfree(worker); | |
1167 | } | |
1168 | EXPORT_SYMBOL(kthread_destroy_worker); |