projects/linux·blob v4.18 · refs

1230 lines · 33.7 KiB

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 <trace/events/sched.h>
24
25static DEFINE_SPINLOCK(kthread_create_lock);
26static LIST_HEAD(kthread_create_list);
27struct task_struct *kthreadd_task;
28
29struct kthread_create_info
30{
31 /* Information passed to kthread() from kthreadd. */
32 int (*threadfn)(void *data);
33 void *data;
34 int node;
35
36 /* Result passed back to kthread_create() from kthreadd. */
37 struct task_struct *result;
38 struct completion *done;
39
40 struct list_head list;
41};
42
43struct kthread {
44 unsigned long flags;
45 unsigned int cpu;
46 void *data;
47 struct completion parked;
48 struct completion exited;
49#ifdef CONFIG_BLK_CGROUP
50 struct cgroup_subsys_state *blkcg_css;
51#endif
52};
53
54enum KTHREAD_BITS {
55 KTHREAD_IS_PER_CPU = 0,
56 KTHREAD_SHOULD_STOP,
57 KTHREAD_SHOULD_PARK,
58};
59
60static inline void set_kthread_struct(void *kthread)
61{
62 /*
63 * We abuse ->set_child_tid to avoid the new member and because it
64 * can't be wrongly copied by copy_process(). We also rely on fact
65 * that the caller can't exec, so PF_KTHREAD can't be cleared.
66 */
67 current->set_child_tid = (__force void __user *)kthread;
68}
69
70static inline struct kthread *to_kthread(struct task_struct *k)
71{
72 WARN_ON(!(k->flags & PF_KTHREAD));
73 return (__force void *)k->set_child_tid;
74}
75
76void free_kthread_struct(struct task_struct *k)
77{
78 struct kthread *kthread;
79
80 /*
81 * Can be NULL if this kthread was created by kernel_thread()
82 * or if kmalloc() in kthread() failed.
83 */
84 kthread = to_kthread(k);
85#ifdef CONFIG_BLK_CGROUP
86 WARN_ON_ONCE(kthread && kthread->blkcg_css);
87#endif
88 kfree(kthread);
89}
90
91/**
92 * kthread_should_stop - should this kthread return now?
93 *
94 * When someone calls kthread_stop() on your kthread, it will be woken
95 * and this will return true. You should then return, and your return
96 * value will be passed through to kthread_stop().
97 */
98bool kthread_should_stop(void)
99{
100 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
101}
102EXPORT_SYMBOL(kthread_should_stop);
103
104/**
105 * kthread_should_park - should this kthread park now?
106 *
107 * When someone calls kthread_park() on your kthread, it will be woken
108 * and this will return true. You should then do the necessary
109 * cleanup and call kthread_parkme()
110 *
111 * Similar to kthread_should_stop(), but this keeps the thread alive
112 * and in a park position. kthread_unpark() "restarts" the thread and
113 * calls the thread function again.
114 */
115bool kthread_should_park(void)
116{
117 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags);
118}
119EXPORT_SYMBOL_GPL(kthread_should_park);
120
121/**
122 * kthread_freezable_should_stop - should this freezable kthread return now?
123 * @was_frozen: optional out parameter, indicates whether %current was frozen
124 *
125 * kthread_should_stop() for freezable kthreads, which will enter
126 * refrigerator if necessary. This function is safe from kthread_stop() /
127 * freezer deadlock and freezable kthreads should use this function instead
128 * of calling try_to_freeze() directly.
129 */
130bool kthread_freezable_should_stop(bool *was_frozen)
131{
132 bool frozen = false;
133
134 might_sleep();
135
136 if (unlikely(freezing(current)))
137 frozen = __refrigerator(true);
138
139 if (was_frozen)
140 *was_frozen = frozen;
141
142 return kthread_should_stop();
143}
144EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
145
146/**
147 * kthread_data - return data value specified on kthread creation
148 * @task: kthread task in question
149 *
150 * Return the data value specified when kthread @task was created.
151 * The caller is responsible for ensuring the validity of @task when
152 * calling this function.
153 */
154void *kthread_data(struct task_struct *task)
155{
156 return to_kthread(task)->data;
157}
158
159/**
160 * kthread_probe_data - speculative version of kthread_data()
161 * @task: possible kthread task in question
162 *
163 * @task could be a kthread task. Return the data value specified when it
164 * was created if accessible. If @task isn't a kthread task or its data is
165 * inaccessible for any reason, %NULL is returned. This function requires
166 * that @task itself is safe to dereference.
167 */
168void *kthread_probe_data(struct task_struct *task)
169{
170 struct kthread *kthread = to_kthread(task);
171 void *data = NULL;
172
173 probe_kernel_read(&data, &kthread->data, sizeof(data));
174 return data;
175}
176
177static void __kthread_parkme(struct kthread *self)
178{
179 for (;;) {
180 /*
181 * TASK_PARKED is a special state; we must serialize against
182 * possible pending wakeups to avoid store-store collisions on
183 * task->state.
184 *
185 * Such a collision might possibly result in the task state
186 * changin from TASK_PARKED and us failing the
187 * wait_task_inactive() in kthread_park().
188 */
189 set_special_state(TASK_PARKED);
190 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
191 break;
192
193 complete_all(&self->parked);
194 schedule();
195 }
196 __set_current_state(TASK_RUNNING);
197}
198
199void kthread_parkme(void)
200{
201 __kthread_parkme(to_kthread(current));
202}
203EXPORT_SYMBOL_GPL(kthread_parkme);
204
205static int kthread(void *_create)
206{
207 /* Copy data: it's on kthread's stack */
208 struct kthread_create_info *create = _create;
209 int (*threadfn)(void *data) = create->threadfn;
210 void *data = create->data;
211 struct completion *done;
212 struct kthread *self;
213 int ret;
214
215 self = kzalloc(sizeof(*self), GFP_KERNEL);
216 set_kthread_struct(self);
217
218 /* If user was SIGKILLed, I release the structure. */
219 done = xchg(&create->done, NULL);
220 if (!done) {
221 kfree(create);
222 do_exit(-EINTR);
223 }
224
225 if (!self) {
226 create->result = ERR_PTR(-ENOMEM);
227 complete(done);
228 do_exit(-ENOMEM);
229 }
230
231 self->data = data;
232 init_completion(&self->exited);
233 init_completion(&self->parked);
234 current->vfork_done = &self->exited;
235
236 /* OK, tell user we're spawned, wait for stop or wakeup */
237 __set_current_state(TASK_UNINTERRUPTIBLE);
238 create->result = current;
239 complete(done);
240 schedule();
241
242 ret = -EINTR;
243 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
244 cgroup_kthread_ready();
245 __kthread_parkme(self);
246 ret = threadfn(data);
247 }
248 do_exit(ret);
249}
250
251/* called from do_fork() to get node information for about to be created task */
252int tsk_fork_get_node(struct task_struct *tsk)
253{
254#ifdef CONFIG_NUMA
255 if (tsk == kthreadd_task)
256 return tsk->pref_node_fork;
257#endif
258 return NUMA_NO_NODE;
259}
260
261static void create_kthread(struct kthread_create_info *create)
262{
263 int pid;
264
265#ifdef CONFIG_NUMA
266 current->pref_node_fork = create->node;
267#endif
268 /* We want our own signal handler (we take no signals by default). */
269 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
270 if (pid < 0) {
271 /* If user was SIGKILLed, I release the structure. */
272 struct completion *done = xchg(&create->done, NULL);
273
274 if (!done) {
275 kfree(create);
276 return;
277 }
278 create->result = ERR_PTR(pid);
279 complete(done);
280 }
281}
282
283static __printf(4, 0)
284struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
285 void *data, int node,
286 const char namefmt[],
287 va_list args)
288{
289 DECLARE_COMPLETION_ONSTACK(done);
290 struct task_struct *task;
291 struct kthread_create_info *create = kmalloc(sizeof(*create),
292 GFP_KERNEL);
293
294 if (!create)
295 return ERR_PTR(-ENOMEM);
296 create->threadfn = threadfn;
297 create->data = data;
298 create->node = node;
299 create->done = &done;
300
301 spin_lock(&kthread_create_lock);
302 list_add_tail(&create->list, &kthread_create_list);
303 spin_unlock(&kthread_create_lock);
304
305 wake_up_process(kthreadd_task);
306 /*
307 * Wait for completion in killable state, for I might be chosen by
308 * the OOM killer while kthreadd is trying to allocate memory for
309 * new kernel thread.
310 */
311 if (unlikely(wait_for_completion_killable(&done))) {
312 /*
313 * If I was SIGKILLed before kthreadd (or new kernel thread)
314 * calls complete(), leave the cleanup of this structure to
315 * that thread.
316 */
317 if (xchg(&create->done, NULL))
318 return ERR_PTR(-EINTR);
319 /*
320 * kthreadd (or new kernel thread) will call complete()
321 * shortly.
322 */
323 wait_for_completion(&done);
324 }
325 task = create->result;
326 if (!IS_ERR(task)) {
327 static const struct sched_param param = { .sched_priority = 0 };
328 char name[TASK_COMM_LEN];
329
330 /*
331 * task is already visible to other tasks, so updating
332 * COMM must be protected.
333 */
334 vsnprintf(name, sizeof(name), namefmt, args);
335 set_task_comm(task, name);
336 /*
337 * root may have changed our (kthreadd's) priority or CPU mask.
338 * The kernel thread should not inherit these properties.
339 */
340 sched_setscheduler_nocheck(task, SCHED_NORMAL, &param);
341 set_cpus_allowed_ptr(task, cpu_all_mask);
342 }
343 kfree(create);
344 return task;
345}
346
347/**
348 * kthread_create_on_node - create a kthread.
349 * @threadfn: the function to run until signal_pending(current).
350 * @data: data ptr for @threadfn.
351 * @node: task and thread structures for the thread are allocated on this node
352 * @namefmt: printf-style name for the thread.
353 *
354 * Description: This helper function creates and names a kernel
355 * thread. The thread will be stopped: use wake_up_process() to start
356 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
357 * is affine to all CPUs.
358 *
359 * If thread is going to be bound on a particular cpu, give its node
360 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
361 * When woken, the thread will run @threadfn() with @data as its
362 * argument. @threadfn() can either call do_exit() directly if it is a
363 * standalone thread for which no one will call kthread_stop(), or
364 * return when 'kthread_should_stop()' is true (which means
365 * kthread_stop() has been called). The return value should be zero
366 * or a negative error number; it will be passed to kthread_stop().
367 *
368 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
369 */
370struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
371 void *data, int node,
372 const char namefmt[],
373 ...)
374{
375 struct task_struct *task;
376 va_list args;
377
378 va_start(args, namefmt);
379 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
380 va_end(args);
381
382 return task;
383}
384EXPORT_SYMBOL(kthread_create_on_node);
385
386static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
387{
388 unsigned long flags;
389
390 if (!wait_task_inactive(p, state)) {
391 WARN_ON(1);
392 return;
393 }
394
395 /* It's safe because the task is inactive. */
396 raw_spin_lock_irqsave(&p->pi_lock, flags);
397 do_set_cpus_allowed(p, mask);
398 p->flags |= PF_NO_SETAFFINITY;
399 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
400}
401
402static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
403{
404 __kthread_bind_mask(p, cpumask_of(cpu), state);
405}
406
407void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
408{
409 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
410}
411
412/**
413 * kthread_bind - bind a just-created kthread to a cpu.
414 * @p: thread created by kthread_create().
415 * @cpu: cpu (might not be online, must be possible) for @k to run on.
416 *
417 * Description: This function is equivalent to set_cpus_allowed(),
418 * except that @cpu doesn't need to be online, and the thread must be
419 * stopped (i.e., just returned from kthread_create()).
420 */
421void kthread_bind(struct task_struct *p, unsigned int cpu)
422{
423 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
424}
425EXPORT_SYMBOL(kthread_bind);
426
427/**
428 * kthread_create_on_cpu - Create a cpu bound kthread
429 * @threadfn: the function to run until signal_pending(current).
430 * @data: data ptr for @threadfn.
431 * @cpu: The cpu on which the thread should be bound,
432 * @namefmt: printf-style name for the thread. Format is restricted
433 * to "name.*%u". Code fills in cpu number.
434 *
435 * Description: This helper function creates and names a kernel thread
436 * The thread will be woken and put into park mode.
437 */
438struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
439 void *data, unsigned int cpu,
440 const char *namefmt)
441{
442 struct task_struct *p;
443
444 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
445 cpu);
446 if (IS_ERR(p))
447 return p;
448 kthread_bind(p, cpu);
449 /* CPU hotplug need to bind once again when unparking the thread. */
450 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
451 to_kthread(p)->cpu = cpu;
452 return p;
453}
454
455/**
456 * kthread_unpark - unpark a thread created by kthread_create().
457 * @k: thread created by kthread_create().
458 *
459 * Sets kthread_should_park() for @k to return false, wakes it, and
460 * waits for it to return. If the thread is marked percpu then its
461 * bound to the cpu again.
462 */
463void kthread_unpark(struct task_struct *k)
464{
465 struct kthread *kthread = to_kthread(k);
466
467 /*
468 * Newly created kthread was parked when the CPU was offline.
469 * The binding was lost and we need to set it again.
470 */
471 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
472 __kthread_bind(k, kthread->cpu, TASK_PARKED);
473
474 reinit_completion(&kthread->parked);
475 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
476 /*
477 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
478 */
479 wake_up_state(k, TASK_PARKED);
480}
481EXPORT_SYMBOL_GPL(kthread_unpark);
482
483/**
484 * kthread_park - park a thread created by kthread_create().
485 * @k: thread created by kthread_create().
486 *
487 * Sets kthread_should_park() for @k to return true, wakes it, and
488 * waits for it to return. This can also be called after kthread_create()
489 * instead of calling wake_up_process(): the thread will park without
490 * calling threadfn().
491 *
492 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
493 * If called by the kthread itself just the park bit is set.
494 */
495int kthread_park(struct task_struct *k)
496{
497 struct kthread *kthread = to_kthread(k);
498
499 if (WARN_ON(k->flags & PF_EXITING))
500 return -ENOSYS;
501
502 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
503 if (k != current) {
504 wake_up_process(k);
505 /*
506 * Wait for __kthread_parkme() to complete(), this means we
507 * _will_ have TASK_PARKED and are about to call schedule().
508 */
509 wait_for_completion(&kthread->parked);
510 /*
511 * Now wait for that schedule() to complete and the task to
512 * get scheduled out.
513 */
514 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
515 }
516
517 return 0;
518}
519EXPORT_SYMBOL_GPL(kthread_park);
520
521/**
522 * kthread_stop - stop a thread created by kthread_create().
523 * @k: thread created by kthread_create().
524 *
525 * Sets kthread_should_stop() for @k to return true, wakes it, and
526 * waits for it to exit. This can also be called after kthread_create()
527 * instead of calling wake_up_process(): the thread will exit without
528 * calling threadfn().
529 *
530 * If threadfn() may call do_exit() itself, the caller must ensure
531 * task_struct can't go away.
532 *
533 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
534 * was never called.
535 */
536int kthread_stop(struct task_struct *k)
537{
538 struct kthread *kthread;
539 int ret;
540
541 trace_sched_kthread_stop(k);
542
543 get_task_struct(k);
544 kthread = to_kthread(k);
545 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
546 kthread_unpark(k);
547 wake_up_process(k);
548 wait_for_completion(&kthread->exited);
549 ret = k->exit_code;
550 put_task_struct(k);
551
552 trace_sched_kthread_stop_ret(ret);
553 return ret;
554}
555EXPORT_SYMBOL(kthread_stop);
556
557int kthreadd(void *unused)
558{
559 struct task_struct *tsk = current;
560
561 /* Setup a clean context for our children to inherit. */
562 set_task_comm(tsk, "kthreadd");
563 ignore_signals(tsk);
564 set_cpus_allowed_ptr(tsk, cpu_all_mask);
565 set_mems_allowed(node_states[N_MEMORY]);
566
567 current->flags |= PF_NOFREEZE;
568 cgroup_init_kthreadd();
569
570 for (;;) {
571 set_current_state(TASK_INTERRUPTIBLE);
572 if (list_empty(&kthread_create_list))
573 schedule();
574 __set_current_state(TASK_RUNNING);
575
576 spin_lock(&kthread_create_lock);
577 while (!list_empty(&kthread_create_list)) {
578 struct kthread_create_info *create;
579
580 create = list_entry(kthread_create_list.next,
581 struct kthread_create_info, list);
582 list_del_init(&create->list);
583 spin_unlock(&kthread_create_lock);
584
585 create_kthread(create);
586
587 spin_lock(&kthread_create_lock);
588 }
589 spin_unlock(&kthread_create_lock);
590 }
591
592 return 0;
593}
594
595void __kthread_init_worker(struct kthread_worker *worker,
596 const char *name,
597 struct lock_class_key *key)
598{
599 memset(worker, 0, sizeof(struct kthread_worker));
600 spin_lock_init(&worker->lock);
601 lockdep_set_class_and_name(&worker->lock, key, name);
602 INIT_LIST_HEAD(&worker->work_list);
603 INIT_LIST_HEAD(&worker->delayed_work_list);
604}
605EXPORT_SYMBOL_GPL(__kthread_init_worker);
606
607/**
608 * kthread_worker_fn - kthread function to process kthread_worker
609 * @worker_ptr: pointer to initialized kthread_worker
610 *
611 * This function implements the main cycle of kthread worker. It processes
612 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
613 * is empty.
614 *
615 * The works are not allowed to keep any locks, disable preemption or interrupts
616 * when they finish. There is defined a safe point for freezing when one work
617 * finishes and before a new one is started.
618 *
619 * Also the works must not be handled by more than one worker at the same time,
620 * see also kthread_queue_work().
621 */
622int kthread_worker_fn(void *worker_ptr)
623{
624 struct kthread_worker *worker = worker_ptr;
625 struct kthread_work *work;
626
627 /*
628 * FIXME: Update the check and remove the assignment when all kthread
629 * worker users are created using kthread_create_worker*() functions.
630 */
631 WARN_ON(worker->task && worker->task != current);
632 worker->task = current;
633
634 if (worker->flags & KTW_FREEZABLE)
635 set_freezable();
636
637repeat:
638 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
639
640 if (kthread_should_stop()) {
641 __set_current_state(TASK_RUNNING);
642 spin_lock_irq(&worker->lock);
643 worker->task = NULL;
644 spin_unlock_irq(&worker->lock);
645 return 0;
646 }
647
648 work = NULL;
649 spin_lock_irq(&worker->lock);
650 if (!list_empty(&worker->work_list)) {
651 work = list_first_entry(&worker->work_list,
652 struct kthread_work, node);
653 list_del_init(&work->node);
654 }
655 worker->current_work = work;
656 spin_unlock_irq(&worker->lock);
657
658 if (work) {
659 __set_current_state(TASK_RUNNING);
660 work->func(work);
661 } else if (!freezing(current))
662 schedule();
663
664 try_to_freeze();
665 cond_resched();
666 goto repeat;
667}
668EXPORT_SYMBOL_GPL(kthread_worker_fn);
669
670static __printf(3, 0) struct kthread_worker *
671__kthread_create_worker(int cpu, unsigned int flags,
672 const char namefmt[], va_list args)
673{
674 struct kthread_worker *worker;
675 struct task_struct *task;
676 int node = -1;
677
678 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
679 if (!worker)
680 return ERR_PTR(-ENOMEM);
681
682 kthread_init_worker(worker);
683
684 if (cpu >= 0)
685 node = cpu_to_node(cpu);
686
687 task = __kthread_create_on_node(kthread_worker_fn, worker,
688 node, namefmt, args);
689 if (IS_ERR(task))
690 goto fail_task;
691
692 if (cpu >= 0)
693 kthread_bind(task, cpu);
694
695 worker->flags = flags;
696 worker->task = task;
697 wake_up_process(task);
698 return worker;
699
700fail_task:
701 kfree(worker);
702 return ERR_CAST(task);
703}
704
705/**
706 * kthread_create_worker - create a kthread worker
707 * @flags: flags modifying the default behavior of the worker
708 * @namefmt: printf-style name for the kthread worker (task).
709 *
710 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
711 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
712 * when the worker was SIGKILLed.
713 */
714struct kthread_worker *
715kthread_create_worker(unsigned int flags, const char namefmt[], ...)
716{
717 struct kthread_worker *worker;
718 va_list args;
719
720 va_start(args, namefmt);
721 worker = __kthread_create_worker(-1, flags, namefmt, args);
722 va_end(args);
723
724 return worker;
725}
726EXPORT_SYMBOL(kthread_create_worker);
727
728/**
729 * kthread_create_worker_on_cpu - create a kthread worker and bind it
730 * it to a given CPU and the associated NUMA node.
731 * @cpu: CPU number
732 * @flags: flags modifying the default behavior of the worker
733 * @namefmt: printf-style name for the kthread worker (task).
734 *
735 * Use a valid CPU number if you want to bind the kthread worker
736 * to the given CPU and the associated NUMA node.
737 *
738 * A good practice is to add the cpu number also into the worker name.
739 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
740 *
741 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
742 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
743 * when the worker was SIGKILLed.
744 */
745struct kthread_worker *
746kthread_create_worker_on_cpu(int cpu, unsigned int flags,
747 const char namefmt[], ...)
748{
749 struct kthread_worker *worker;
750 va_list args;
751
752 va_start(args, namefmt);
753 worker = __kthread_create_worker(cpu, flags, namefmt, args);
754 va_end(args);
755
756 return worker;
757}
758EXPORT_SYMBOL(kthread_create_worker_on_cpu);
759
760/*
761 * Returns true when the work could not be queued at the moment.
762 * It happens when it is already pending in a worker list
763 * or when it is being cancelled.
764 */
765static inline bool queuing_blocked(struct kthread_worker *worker,
766 struct kthread_work *work)
767{
768 lockdep_assert_held(&worker->lock);
769
770 return !list_empty(&work->node) || work->canceling;
771}
772
773static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
774 struct kthread_work *work)
775{
776 lockdep_assert_held(&worker->lock);
777 WARN_ON_ONCE(!list_empty(&work->node));
778 /* Do not use a work with >1 worker, see kthread_queue_work() */
779 WARN_ON_ONCE(work->worker && work->worker != worker);
780}
781
782/* insert @work before @pos in @worker */
783static void kthread_insert_work(struct kthread_worker *worker,
784 struct kthread_work *work,
785 struct list_head *pos)
786{
787 kthread_insert_work_sanity_check(worker, work);
788
789 list_add_tail(&work->node, pos);
790 work->worker = worker;
791 if (!worker->current_work && likely(worker->task))
792 wake_up_process(worker->task);
793}
794
795/**
796 * kthread_queue_work - queue a kthread_work
797 * @worker: target kthread_worker
798 * @work: kthread_work to queue
799 *
800 * Queue @work to work processor @task for async execution. @task
801 * must have been created with kthread_worker_create(). Returns %true
802 * if @work was successfully queued, %false if it was already pending.
803 *
804 * Reinitialize the work if it needs to be used by another worker.
805 * For example, when the worker was stopped and started again.
806 */
807bool kthread_queue_work(struct kthread_worker *worker,
808 struct kthread_work *work)
809{
810 bool ret = false;
811 unsigned long flags;
812
813 spin_lock_irqsave(&worker->lock, flags);
814 if (!queuing_blocked(worker, work)) {
815 kthread_insert_work(worker, work, &worker->work_list);
816 ret = true;
817 }
818 spin_unlock_irqrestore(&worker->lock, flags);
819 return ret;
820}
821EXPORT_SYMBOL_GPL(kthread_queue_work);
822
823/**
824 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
825 * delayed work when the timer expires.
826 * @t: pointer to the expired timer
827 *
828 * The format of the function is defined by struct timer_list.
829 * It should have been called from irqsafe timer with irq already off.
830 */
831void kthread_delayed_work_timer_fn(struct timer_list *t)
832{
833 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
834 struct kthread_work *work = &dwork->work;
835 struct kthread_worker *worker = work->worker;
836
837 /*
838 * This might happen when a pending work is reinitialized.
839 * It means that it is used a wrong way.
840 */
841 if (WARN_ON_ONCE(!worker))
842 return;
843
844 spin_lock(&worker->lock);
845 /* Work must not be used with >1 worker, see kthread_queue_work(). */
846 WARN_ON_ONCE(work->worker != worker);
847
848 /* Move the work from worker->delayed_work_list. */
849 WARN_ON_ONCE(list_empty(&work->node));
850 list_del_init(&work->node);
851 kthread_insert_work(worker, work, &worker->work_list);
852
853 spin_unlock(&worker->lock);
854}
855EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
856
857void __kthread_queue_delayed_work(struct kthread_worker *worker,
858 struct kthread_delayed_work *dwork,
859 unsigned long delay)
860{
861 struct timer_list *timer = &dwork->timer;
862 struct kthread_work *work = &dwork->work;
863
864 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
865
866 /*
867 * If @delay is 0, queue @dwork->work immediately. This is for
868 * both optimization and correctness. The earliest @timer can
869 * expire is on the closest next tick and delayed_work users depend
870 * on that there's no such delay when @delay is 0.
871 */
872 if (!delay) {
873 kthread_insert_work(worker, work, &worker->work_list);
874 return;
875 }
876
877 /* Be paranoid and try to detect possible races already now. */
878 kthread_insert_work_sanity_check(worker, work);
879
880 list_add(&work->node, &worker->delayed_work_list);
881 work->worker = worker;
882 timer->expires = jiffies + delay;
883 add_timer(timer);
884}
885
886/**
887 * kthread_queue_delayed_work - queue the associated kthread work
888 * after a delay.
889 * @worker: target kthread_worker
890 * @dwork: kthread_delayed_work to queue
891 * @delay: number of jiffies to wait before queuing
892 *
893 * If the work has not been pending it starts a timer that will queue
894 * the work after the given @delay. If @delay is zero, it queues the
895 * work immediately.
896 *
897 * Return: %false if the @work has already been pending. It means that
898 * either the timer was running or the work was queued. It returns %true
899 * otherwise.
900 */
901bool kthread_queue_delayed_work(struct kthread_worker *worker,
902 struct kthread_delayed_work *dwork,
903 unsigned long delay)
904{
905 struct kthread_work *work = &dwork->work;
906 unsigned long flags;
907 bool ret = false;
908
909 spin_lock_irqsave(&worker->lock, flags);
910
911 if (!queuing_blocked(worker, work)) {
912 __kthread_queue_delayed_work(worker, dwork, delay);
913 ret = true;
914 }
915
916 spin_unlock_irqrestore(&worker->lock, flags);
917 return ret;
918}
919EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
920
921struct kthread_flush_work {
922 struct kthread_work work;
923 struct completion done;
924};
925
926static void kthread_flush_work_fn(struct kthread_work *work)
927{
928 struct kthread_flush_work *fwork =
929 container_of(work, struct kthread_flush_work, work);
930 complete(&fwork->done);
931}
932
933/**
934 * kthread_flush_work - flush a kthread_work
935 * @work: work to flush
936 *
937 * If @work is queued or executing, wait for it to finish execution.
938 */
939void kthread_flush_work(struct kthread_work *work)
940{
941 struct kthread_flush_work fwork = {
942 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
943 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
944 };
945 struct kthread_worker *worker;
946 bool noop = false;
947
948 worker = work->worker;
949 if (!worker)
950 return;
951
952 spin_lock_irq(&worker->lock);
953 /* Work must not be used with >1 worker, see kthread_queue_work(). */
954 WARN_ON_ONCE(work->worker != worker);
955
956 if (!list_empty(&work->node))
957 kthread_insert_work(worker, &fwork.work, work->node.next);
958 else if (worker->current_work == work)
959 kthread_insert_work(worker, &fwork.work,
960 worker->work_list.next);
961 else
962 noop = true;
963
964 spin_unlock_irq(&worker->lock);
965
966 if (!noop)
967 wait_for_completion(&fwork.done);
968}
969EXPORT_SYMBOL_GPL(kthread_flush_work);
970
971/*
972 * This function removes the work from the worker queue. Also it makes sure
973 * that it won't get queued later via the delayed work's timer.
974 *
975 * The work might still be in use when this function finishes. See the
976 * current_work proceed by the worker.
977 *
978 * Return: %true if @work was pending and successfully canceled,
979 * %false if @work was not pending
980 */
981static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
982 unsigned long *flags)
983{
984 /* Try to cancel the timer if exists. */
985 if (is_dwork) {
986 struct kthread_delayed_work *dwork =
987 container_of(work, struct kthread_delayed_work, work);
988 struct kthread_worker *worker = work->worker;
989
990 /*
991 * del_timer_sync() must be called to make sure that the timer
992 * callback is not running. The lock must be temporary released
993 * to avoid a deadlock with the callback. In the meantime,
994 * any queuing is blocked by setting the canceling counter.
995 */
996 work->canceling++;
997 spin_unlock_irqrestore(&worker->lock, *flags);
998 del_timer_sync(&dwork->timer);
999 spin_lock_irqsave(&worker->lock, *flags);
1000 work->canceling--;
1001 }
1002
1003 /*
1004 * Try to remove the work from a worker list. It might either
1005 * be from worker->work_list or from worker->delayed_work_list.
1006 */
1007 if (!list_empty(&work->node)) {
1008 list_del_init(&work->node);
1009 return true;
1010 }
1011
1012 return false;
1013}
1014
1015/**
1016 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1017 * @worker: kthread worker to use
1018 * @dwork: kthread delayed work to queue
1019 * @delay: number of jiffies to wait before queuing
1020 *
1021 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1022 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1023 * @work is guaranteed to be queued immediately.
1024 *
1025 * Return: %true if @dwork was pending and its timer was modified,
1026 * %false otherwise.
1027 *
1028 * A special case is when the work is being canceled in parallel.
1029 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1030 * or yet another kthread_mod_delayed_work() call. We let the other command
1031 * win and return %false here. The caller is supposed to synchronize these
1032 * operations a reasonable way.
1033 *
1034 * This function is safe to call from any context including IRQ handler.
1035 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1036 * for details.
1037 */
1038bool kthread_mod_delayed_work(struct kthread_worker *worker,
1039 struct kthread_delayed_work *dwork,
1040 unsigned long delay)
1041{
1042 struct kthread_work *work = &dwork->work;
1043 unsigned long flags;
1044 int ret = false;
1045
1046 spin_lock_irqsave(&worker->lock, flags);
1047
1048 /* Do not bother with canceling when never queued. */
1049 if (!work->worker)
1050 goto fast_queue;
1051
1052 /* Work must not be used with >1 worker, see kthread_queue_work() */
1053 WARN_ON_ONCE(work->worker != worker);
1054
1055 /* Do not fight with another command that is canceling this work. */
1056 if (work->canceling)
1057 goto out;
1058
1059 ret = __kthread_cancel_work(work, true, &flags);
1060fast_queue:
1061 __kthread_queue_delayed_work(worker, dwork, delay);
1062out:
1063 spin_unlock_irqrestore(&worker->lock, flags);
1064 return ret;
1065}
1066EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1067
1068static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1069{
1070 struct kthread_worker *worker = work->worker;
1071 unsigned long flags;
1072 int ret = false;
1073
1074 if (!worker)
1075 goto out;
1076
1077 spin_lock_irqsave(&worker->lock, flags);
1078 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1079 WARN_ON_ONCE(work->worker != worker);
1080
1081 ret = __kthread_cancel_work(work, is_dwork, &flags);
1082
1083 if (worker->current_work != work)
1084 goto out_fast;
1085
1086 /*
1087 * The work is in progress and we need to wait with the lock released.
1088 * In the meantime, block any queuing by setting the canceling counter.
1089 */
1090 work->canceling++;
1091 spin_unlock_irqrestore(&worker->lock, flags);
1092 kthread_flush_work(work);
1093 spin_lock_irqsave(&worker->lock, flags);
1094 work->canceling--;
1095
1096out_fast:
1097 spin_unlock_irqrestore(&worker->lock, flags);
1098out:
1099 return ret;
1100}
1101
1102/**
1103 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1104 * @work: the kthread work to cancel
1105 *
1106 * Cancel @work and wait for its execution to finish. This function
1107 * can be used even if the work re-queues itself. On return from this
1108 * function, @work is guaranteed to be not pending or executing on any CPU.
1109 *
1110 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1111 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1112 *
1113 * The caller must ensure that the worker on which @work was last
1114 * queued can't be destroyed before this function returns.
1115 *
1116 * Return: %true if @work was pending, %false otherwise.
1117 */
1118bool kthread_cancel_work_sync(struct kthread_work *work)
1119{
1120 return __kthread_cancel_work_sync(work, false);
1121}
1122EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1123
1124/**
1125 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1126 * wait for it to finish.
1127 * @dwork: the kthread delayed work to cancel
1128 *
1129 * This is kthread_cancel_work_sync() for delayed works.
1130 *
1131 * Return: %true if @dwork was pending, %false otherwise.
1132 */
1133bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1134{
1135 return __kthread_cancel_work_sync(&dwork->work, true);
1136}
1137EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1138
1139/**
1140 * kthread_flush_worker - flush all current works on a kthread_worker
1141 * @worker: worker to flush
1142 *
1143 * Wait until all currently executing or pending works on @worker are
1144 * finished.
1145 */
1146void kthread_flush_worker(struct kthread_worker *worker)
1147{
1148 struct kthread_flush_work fwork = {
1149 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1150 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1151 };
1152
1153 kthread_queue_work(worker, &fwork.work);
1154 wait_for_completion(&fwork.done);
1155}
1156EXPORT_SYMBOL_GPL(kthread_flush_worker);
1157
1158/**
1159 * kthread_destroy_worker - destroy a kthread worker
1160 * @worker: worker to be destroyed
1161 *
1162 * Flush and destroy @worker. The simple flush is enough because the kthread
1163 * worker API is used only in trivial scenarios. There are no multi-step state
1164 * machines needed.
1165 */
1166void kthread_destroy_worker(struct kthread_worker *worker)
1167{
1168 struct task_struct *task;
1169
1170 task = worker->task;
1171 if (WARN_ON(!task))
1172 return;
1173
1174 kthread_flush_worker(worker);
1175 kthread_stop(task);
1176 WARN_ON(!list_empty(&worker->work_list));
1177 kfree(worker);
1178}
1179EXPORT_SYMBOL(kthread_destroy_worker);
1180
1181#ifdef CONFIG_BLK_CGROUP
1182/**
1183 * kthread_associate_blkcg - associate blkcg to current kthread
1184 * @css: the cgroup info
1185 *
1186 * Current thread must be a kthread. The thread is running jobs on behalf of
1187 * other threads. In some cases, we expect the jobs attach cgroup info of
1188 * original threads instead of that of current thread. This function stores
1189 * original thread's cgroup info in current kthread context for later
1190 * retrieval.
1191 */
1192void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1193{
1194 struct kthread *kthread;
1195
1196 if (!(current->flags & PF_KTHREAD))
1197 return;
1198 kthread = to_kthread(current);
1199 if (!kthread)
1200 return;
1201
1202 if (kthread->blkcg_css) {
1203 css_put(kthread->blkcg_css);
1204 kthread->blkcg_css = NULL;
1205 }
1206 if (css) {
1207 css_get(css);
1208 kthread->blkcg_css = css;
1209 }
1210}
1211EXPORT_SYMBOL(kthread_associate_blkcg);
1212
1213/**
1214 * kthread_blkcg - get associated blkcg css of current kthread
1215 *
1216 * Current thread must be a kthread.
1217 */
1218struct cgroup_subsys_state *kthread_blkcg(void)
1219{
1220 struct kthread *kthread;
1221
1222 if (current->flags & PF_KTHREAD) {
1223 kthread = to_kthread(current);
1224 if (kthread)
1225 return kthread->blkcg_css;
1226 }
1227 return NULL;
1228}
1229EXPORT_SYMBOL(kthread_blkcg);
1230#endif