Loading...
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 <trace/events/sched.h>
22
23static DEFINE_SPINLOCK(kthread_create_lock);
24static LIST_HEAD(kthread_create_list);
25struct task_struct *kthreadd_task;
26
27struct kthread_create_info
28{
29 /* Information passed to kthread() from kthreadd. */
30 int (*threadfn)(void *data);
31 void *data;
32 int node;
33
34 /* Result passed back to kthread_create() from kthreadd. */
35 struct task_struct *result;
36 struct completion *done;
37
38 struct list_head list;
39};
40
41struct kthread {
42 unsigned long flags;
43 unsigned int cpu;
44 void *data;
45 struct completion parked;
46 struct completion exited;
47};
48
49enum KTHREAD_BITS {
50 KTHREAD_IS_PER_CPU = 0,
51 KTHREAD_SHOULD_STOP,
52 KTHREAD_SHOULD_PARK,
53 KTHREAD_IS_PARKED,
54};
55
56#define __to_kthread(vfork) \
57 container_of(vfork, struct kthread, exited)
58
59static inline struct kthread *to_kthread(struct task_struct *k)
60{
61 return __to_kthread(k->vfork_done);
62}
63
64static struct kthread *to_live_kthread(struct task_struct *k)
65{
66 struct completion *vfork = ACCESS_ONCE(k->vfork_done);
67 if (likely(vfork))
68 return __to_kthread(vfork);
69 return NULL;
70}
71
72/**
73 * kthread_should_stop - should this kthread return now?
74 *
75 * When someone calls kthread_stop() on your kthread, it will be woken
76 * and this will return true. You should then return, and your return
77 * value will be passed through to kthread_stop().
78 */
79bool kthread_should_stop(void)
80{
81 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
82}
83EXPORT_SYMBOL(kthread_should_stop);
84
85/**
86 * kthread_should_park - should this kthread park now?
87 *
88 * When someone calls kthread_park() on your kthread, it will be woken
89 * and this will return true. You should then do the necessary
90 * cleanup and call kthread_parkme()
91 *
92 * Similar to kthread_should_stop(), but this keeps the thread alive
93 * and in a park position. kthread_unpark() "restarts" the thread and
94 * calls the thread function again.
95 */
96bool kthread_should_park(void)
97{
98 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags);
99}
100EXPORT_SYMBOL_GPL(kthread_should_park);
101
102/**
103 * kthread_freezable_should_stop - should this freezable kthread return now?
104 * @was_frozen: optional out parameter, indicates whether %current was frozen
105 *
106 * kthread_should_stop() for freezable kthreads, which will enter
107 * refrigerator if necessary. This function is safe from kthread_stop() /
108 * freezer deadlock and freezable kthreads should use this function instead
109 * of calling try_to_freeze() directly.
110 */
111bool kthread_freezable_should_stop(bool *was_frozen)
112{
113 bool frozen = false;
114
115 might_sleep();
116
117 if (unlikely(freezing(current)))
118 frozen = __refrigerator(true);
119
120 if (was_frozen)
121 *was_frozen = frozen;
122
123 return kthread_should_stop();
124}
125EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
126
127/**
128 * kthread_data - return data value specified on kthread creation
129 * @task: kthread task in question
130 *
131 * Return the data value specified when kthread @task was created.
132 * The caller is responsible for ensuring the validity of @task when
133 * calling this function.
134 */
135void *kthread_data(struct task_struct *task)
136{
137 return to_kthread(task)->data;
138}
139
140/**
141 * probe_kthread_data - speculative version of kthread_data()
142 * @task: possible kthread task in question
143 *
144 * @task could be a kthread task. Return the data value specified when it
145 * was created if accessible. If @task isn't a kthread task or its data is
146 * inaccessible for any reason, %NULL is returned. This function requires
147 * that @task itself is safe to dereference.
148 */
149void *probe_kthread_data(struct task_struct *task)
150{
151 struct kthread *kthread = to_kthread(task);
152 void *data = NULL;
153
154 probe_kernel_read(&data, &kthread->data, sizeof(data));
155 return data;
156}
157
158static void __kthread_parkme(struct kthread *self)
159{
160 __set_current_state(TASK_PARKED);
161 while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) {
162 if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags))
163 complete(&self->parked);
164 schedule();
165 __set_current_state(TASK_PARKED);
166 }
167 clear_bit(KTHREAD_IS_PARKED, &self->flags);
168 __set_current_state(TASK_RUNNING);
169}
170
171void kthread_parkme(void)
172{
173 __kthread_parkme(to_kthread(current));
174}
175EXPORT_SYMBOL_GPL(kthread_parkme);
176
177static int kthread(void *_create)
178{
179 /* Copy data: it's on kthread's stack */
180 struct kthread_create_info *create = _create;
181 int (*threadfn)(void *data) = create->threadfn;
182 void *data = create->data;
183 struct completion *done;
184 struct kthread self;
185 int ret;
186
187 self.flags = 0;
188 self.data = data;
189 init_completion(&self.exited);
190 init_completion(&self.parked);
191 current->vfork_done = &self.exited;
192
193 /* If user was SIGKILLed, I release the structure. */
194 done = xchg(&create->done, NULL);
195 if (!done) {
196 kfree(create);
197 do_exit(-EINTR);
198 }
199 /* OK, tell user we're spawned, wait for stop or wakeup */
200 __set_current_state(TASK_UNINTERRUPTIBLE);
201 create->result = current;
202 complete(done);
203 schedule();
204
205 ret = -EINTR;
206
207 if (!test_bit(KTHREAD_SHOULD_STOP, &self.flags)) {
208 __kthread_parkme(&self);
209 ret = threadfn(data);
210 }
211 /* we can't just return, we must preserve "self" on stack */
212 do_exit(ret);
213}
214
215/* called from do_fork() to get node information for about to be created task */
216int tsk_fork_get_node(struct task_struct *tsk)
217{
218#ifdef CONFIG_NUMA
219 if (tsk == kthreadd_task)
220 return tsk->pref_node_fork;
221#endif
222 return NUMA_NO_NODE;
223}
224
225static void create_kthread(struct kthread_create_info *create)
226{
227 int pid;
228
229#ifdef CONFIG_NUMA
230 current->pref_node_fork = create->node;
231#endif
232 /* We want our own signal handler (we take no signals by default). */
233 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
234 if (pid < 0) {
235 /* If user was SIGKILLed, I release the structure. */
236 struct completion *done = xchg(&create->done, NULL);
237
238 if (!done) {
239 kfree(create);
240 return;
241 }
242 create->result = ERR_PTR(pid);
243 complete(done);
244 }
245}
246
247/**
248 * kthread_create_on_node - create a kthread.
249 * @threadfn: the function to run until signal_pending(current).
250 * @data: data ptr for @threadfn.
251 * @node: task and thread structures for the thread are allocated on this node
252 * @namefmt: printf-style name for the thread.
253 *
254 * Description: This helper function creates and names a kernel
255 * thread. The thread will be stopped: use wake_up_process() to start
256 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
257 * is affine to all CPUs.
258 *
259 * If thread is going to be bound on a particular cpu, give its node
260 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
261 * When woken, the thread will run @threadfn() with @data as its
262 * argument. @threadfn() can either call do_exit() directly if it is a
263 * standalone thread for which no one will call kthread_stop(), or
264 * return when 'kthread_should_stop()' is true (which means
265 * kthread_stop() has been called). The return value should be zero
266 * or a negative error number; it will be passed to kthread_stop().
267 *
268 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
269 */
270struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
271 void *data, int node,
272 const char namefmt[],
273 ...)
274{
275 DECLARE_COMPLETION_ONSTACK(done);
276 struct task_struct *task;
277 struct kthread_create_info *create = kmalloc(sizeof(*create),
278 GFP_KERNEL);
279
280 if (!create)
281 return ERR_PTR(-ENOMEM);
282 create->threadfn = threadfn;
283 create->data = data;
284 create->node = node;
285 create->done = &done;
286
287 spin_lock(&kthread_create_lock);
288 list_add_tail(&create->list, &kthread_create_list);
289 spin_unlock(&kthread_create_lock);
290
291 wake_up_process(kthreadd_task);
292 /*
293 * Wait for completion in killable state, for I might be chosen by
294 * the OOM killer while kthreadd is trying to allocate memory for
295 * new kernel thread.
296 */
297 if (unlikely(wait_for_completion_killable(&done))) {
298 /*
299 * If I was SIGKILLed before kthreadd (or new kernel thread)
300 * calls complete(), leave the cleanup of this structure to
301 * that thread.
302 */
303 if (xchg(&create->done, NULL))
304 return ERR_PTR(-EINTR);
305 /*
306 * kthreadd (or new kernel thread) will call complete()
307 * shortly.
308 */
309 wait_for_completion(&done);
310 }
311 task = create->result;
312 if (!IS_ERR(task)) {
313 static const struct sched_param param = { .sched_priority = 0 };
314 va_list args;
315
316 va_start(args, namefmt);
317 vsnprintf(task->comm, sizeof(task->comm), namefmt, args);
318 va_end(args);
319 /*
320 * root may have changed our (kthreadd's) priority or CPU mask.
321 * The kernel thread should not inherit these properties.
322 */
323 sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m);
324 set_cpus_allowed_ptr(task, cpu_all_mask);
325 }
326 kfree(create);
327 return task;
328}
329EXPORT_SYMBOL(kthread_create_on_node);
330
331static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
332{
333 unsigned long flags;
334
335 if (!wait_task_inactive(p, state)) {
336 WARN_ON(1);
337 return;
338 }
339
340 /* It's safe because the task is inactive. */
341 raw_spin_lock_irqsave(&p->pi_lock, flags);
342 do_set_cpus_allowed(p, mask);
343 p->flags |= PF_NO_SETAFFINITY;
344 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
345}
346
347static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
348{
349 __kthread_bind_mask(p, cpumask_of(cpu), state);
350}
351
352void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
353{
354 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
355}
356
357/**
358 * kthread_bind - bind a just-created kthread to a cpu.
359 * @p: thread created by kthread_create().
360 * @cpu: cpu (might not be online, must be possible) for @k to run on.
361 *
362 * Description: This function is equivalent to set_cpus_allowed(),
363 * except that @cpu doesn't need to be online, and the thread must be
364 * stopped (i.e., just returned from kthread_create()).
365 */
366void kthread_bind(struct task_struct *p, unsigned int cpu)
367{
368 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
369}
370EXPORT_SYMBOL(kthread_bind);
371
372/**
373 * kthread_create_on_cpu - Create a cpu bound kthread
374 * @threadfn: the function to run until signal_pending(current).
375 * @data: data ptr for @threadfn.
376 * @cpu: The cpu on which the thread should be bound,
377 * @namefmt: printf-style name for the thread. Format is restricted
378 * to "name.*%u". Code fills in cpu number.
379 *
380 * Description: This helper function creates and names a kernel thread
381 * The thread will be woken and put into park mode.
382 */
383struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
384 void *data, unsigned int cpu,
385 const char *namefmt)
386{
387 struct task_struct *p;
388
389 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
390 cpu);
391 if (IS_ERR(p))
392 return p;
393 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
394 to_kthread(p)->cpu = cpu;
395 /* Park the thread to get it out of TASK_UNINTERRUPTIBLE state */
396 kthread_park(p);
397 return p;
398}
399
400static void __kthread_unpark(struct task_struct *k, struct kthread *kthread)
401{
402 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
403 /*
404 * We clear the IS_PARKED bit here as we don't wait
405 * until the task has left the park code. So if we'd
406 * park before that happens we'd see the IS_PARKED bit
407 * which might be about to be cleared.
408 */
409 if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
410 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
411 __kthread_bind(k, kthread->cpu, TASK_PARKED);
412 wake_up_state(k, TASK_PARKED);
413 }
414}
415
416/**
417 * kthread_unpark - unpark a thread created by kthread_create().
418 * @k: thread created by kthread_create().
419 *
420 * Sets kthread_should_park() for @k to return false, wakes it, and
421 * waits for it to return. If the thread is marked percpu then its
422 * bound to the cpu again.
423 */
424void kthread_unpark(struct task_struct *k)
425{
426 struct kthread *kthread = to_live_kthread(k);
427
428 if (kthread)
429 __kthread_unpark(k, kthread);
430}
431EXPORT_SYMBOL_GPL(kthread_unpark);
432
433/**
434 * kthread_park - park a thread created by kthread_create().
435 * @k: thread created by kthread_create().
436 *
437 * Sets kthread_should_park() for @k to return true, wakes it, and
438 * waits for it to return. This can also be called after kthread_create()
439 * instead of calling wake_up_process(): the thread will park without
440 * calling threadfn().
441 *
442 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
443 * If called by the kthread itself just the park bit is set.
444 */
445int kthread_park(struct task_struct *k)
446{
447 struct kthread *kthread = to_live_kthread(k);
448 int ret = -ENOSYS;
449
450 if (kthread) {
451 if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
452 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
453 if (k != current) {
454 wake_up_process(k);
455 wait_for_completion(&kthread->parked);
456 }
457 }
458 ret = 0;
459 }
460 return ret;
461}
462EXPORT_SYMBOL_GPL(kthread_park);
463
464/**
465 * kthread_stop - stop a thread created by kthread_create().
466 * @k: thread created by kthread_create().
467 *
468 * Sets kthread_should_stop() for @k to return true, wakes it, and
469 * waits for it to exit. This can also be called after kthread_create()
470 * instead of calling wake_up_process(): the thread will exit without
471 * calling threadfn().
472 *
473 * If threadfn() may call do_exit() itself, the caller must ensure
474 * task_struct can't go away.
475 *
476 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
477 * was never called.
478 */
479int kthread_stop(struct task_struct *k)
480{
481 struct kthread *kthread;
482 int ret;
483
484 trace_sched_kthread_stop(k);
485
486 get_task_struct(k);
487 kthread = to_live_kthread(k);
488 if (kthread) {
489 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
490 __kthread_unpark(k, kthread);
491 wake_up_process(k);
492 wait_for_completion(&kthread->exited);
493 }
494 ret = k->exit_code;
495 put_task_struct(k);
496
497 trace_sched_kthread_stop_ret(ret);
498 return ret;
499}
500EXPORT_SYMBOL(kthread_stop);
501
502int kthreadd(void *unused)
503{
504 struct task_struct *tsk = current;
505
506 /* Setup a clean context for our children to inherit. */
507 set_task_comm(tsk, "kthreadd");
508 ignore_signals(tsk);
509 set_cpus_allowed_ptr(tsk, cpu_all_mask);
510 set_mems_allowed(node_states[N_MEMORY]);
511
512 current->flags |= PF_NOFREEZE;
513
514 for (;;) {
515 set_current_state(TASK_INTERRUPTIBLE);
516 if (list_empty(&kthread_create_list))
517 schedule();
518 __set_current_state(TASK_RUNNING);
519
520 spin_lock(&kthread_create_lock);
521 while (!list_empty(&kthread_create_list)) {
522 struct kthread_create_info *create;
523
524 create = list_entry(kthread_create_list.next,
525 struct kthread_create_info, list);
526 list_del_init(&create->list);
527 spin_unlock(&kthread_create_lock);
528
529 create_kthread(create);
530
531 spin_lock(&kthread_create_lock);
532 }
533 spin_unlock(&kthread_create_lock);
534 }
535
536 return 0;
537}
538
539void __init_kthread_worker(struct kthread_worker *worker,
540 const char *name,
541 struct lock_class_key *key)
542{
543 spin_lock_init(&worker->lock);
544 lockdep_set_class_and_name(&worker->lock, key, name);
545 INIT_LIST_HEAD(&worker->work_list);
546 worker->task = NULL;
547}
548EXPORT_SYMBOL_GPL(__init_kthread_worker);
549
550/**
551 * kthread_worker_fn - kthread function to process kthread_worker
552 * @worker_ptr: pointer to initialized kthread_worker
553 *
554 * This function can be used as @threadfn to kthread_create() or
555 * kthread_run() with @worker_ptr argument pointing to an initialized
556 * kthread_worker. The started kthread will process work_list until
557 * the it is stopped with kthread_stop(). A kthread can also call
558 * this function directly after extra initialization.
559 *
560 * Different kthreads can be used for the same kthread_worker as long
561 * as there's only one kthread attached to it at any given time. A
562 * kthread_worker without an attached kthread simply collects queued
563 * kthread_works.
564 */
565int kthread_worker_fn(void *worker_ptr)
566{
567 struct kthread_worker *worker = worker_ptr;
568 struct kthread_work *work;
569
570 WARN_ON(worker->task);
571 worker->task = current;
572repeat:
573 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
574
575 if (kthread_should_stop()) {
576 __set_current_state(TASK_RUNNING);
577 spin_lock_irq(&worker->lock);
578 worker->task = NULL;
579 spin_unlock_irq(&worker->lock);
580 return 0;
581 }
582
583 work = NULL;
584 spin_lock_irq(&worker->lock);
585 if (!list_empty(&worker->work_list)) {
586 work = list_first_entry(&worker->work_list,
587 struct kthread_work, node);
588 list_del_init(&work->node);
589 }
590 worker->current_work = work;
591 spin_unlock_irq(&worker->lock);
592
593 if (work) {
594 __set_current_state(TASK_RUNNING);
595 work->func(work);
596 } else if (!freezing(current))
597 schedule();
598
599 try_to_freeze();
600 goto repeat;
601}
602EXPORT_SYMBOL_GPL(kthread_worker_fn);
603
604/* insert @work before @pos in @worker */
605static void insert_kthread_work(struct kthread_worker *worker,
606 struct kthread_work *work,
607 struct list_head *pos)
608{
609 lockdep_assert_held(&worker->lock);
610
611 list_add_tail(&work->node, pos);
612 work->worker = worker;
613 if (!worker->current_work && likely(worker->task))
614 wake_up_process(worker->task);
615}
616
617/**
618 * queue_kthread_work - queue a kthread_work
619 * @worker: target kthread_worker
620 * @work: kthread_work to queue
621 *
622 * Queue @work to work processor @task for async execution. @task
623 * must have been created with kthread_worker_create(). Returns %true
624 * if @work was successfully queued, %false if it was already pending.
625 */
626bool queue_kthread_work(struct kthread_worker *worker,
627 struct kthread_work *work)
628{
629 bool ret = false;
630 unsigned long flags;
631
632 spin_lock_irqsave(&worker->lock, flags);
633 if (list_empty(&work->node)) {
634 insert_kthread_work(worker, work, &worker->work_list);
635 ret = true;
636 }
637 spin_unlock_irqrestore(&worker->lock, flags);
638 return ret;
639}
640EXPORT_SYMBOL_GPL(queue_kthread_work);
641
642struct kthread_flush_work {
643 struct kthread_work work;
644 struct completion done;
645};
646
647static void kthread_flush_work_fn(struct kthread_work *work)
648{
649 struct kthread_flush_work *fwork =
650 container_of(work, struct kthread_flush_work, work);
651 complete(&fwork->done);
652}
653
654/**
655 * flush_kthread_work - flush a kthread_work
656 * @work: work to flush
657 *
658 * If @work is queued or executing, wait for it to finish execution.
659 */
660void flush_kthread_work(struct kthread_work *work)
661{
662 struct kthread_flush_work fwork = {
663 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
664 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
665 };
666 struct kthread_worker *worker;
667 bool noop = false;
668
669retry:
670 worker = work->worker;
671 if (!worker)
672 return;
673
674 spin_lock_irq(&worker->lock);
675 if (work->worker != worker) {
676 spin_unlock_irq(&worker->lock);
677 goto retry;
678 }
679
680 if (!list_empty(&work->node))
681 insert_kthread_work(worker, &fwork.work, work->node.next);
682 else if (worker->current_work == work)
683 insert_kthread_work(worker, &fwork.work, worker->work_list.next);
684 else
685 noop = true;
686
687 spin_unlock_irq(&worker->lock);
688
689 if (!noop)
690 wait_for_completion(&fwork.done);
691}
692EXPORT_SYMBOL_GPL(flush_kthread_work);
693
694/**
695 * flush_kthread_worker - flush all current works on a kthread_worker
696 * @worker: worker to flush
697 *
698 * Wait until all currently executing or pending works on @worker are
699 * finished.
700 */
701void flush_kthread_worker(struct kthread_worker *worker)
702{
703 struct kthread_flush_work fwork = {
704 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
705 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
706 };
707
708 queue_kthread_work(worker, &fwork.work);
709 wait_for_completion(&fwork.done);
710}
711EXPORT_SYMBOL_GPL(flush_kthread_worker);
1// SPDX-License-Identifier: GPL-2.0-only
2/* Kernel thread helper functions.
3 * Copyright (C) 2004 IBM Corporation, Rusty Russell.
4 * Copyright (C) 2009 Red Hat, Inc.
5 *
6 * Creation is done via kthreadd, so that we get a clean environment
7 * even if we're invoked from userspace (think modprobe, hotplug cpu,
8 * etc.).
9 */
10#include <uapi/linux/sched/types.h>
11#include <linux/mm.h>
12#include <linux/mmu_context.h>
13#include <linux/sched.h>
14#include <linux/sched/mm.h>
15#include <linux/sched/task.h>
16#include <linux/kthread.h>
17#include <linux/completion.h>
18#include <linux/err.h>
19#include <linux/cgroup.h>
20#include <linux/cpuset.h>
21#include <linux/unistd.h>
22#include <linux/file.h>
23#include <linux/export.h>
24#include <linux/mutex.h>
25#include <linux/slab.h>
26#include <linux/freezer.h>
27#include <linux/ptrace.h>
28#include <linux/uaccess.h>
29#include <linux/numa.h>
30#include <linux/sched/isolation.h>
31#include <trace/events/sched.h>
32
33
34static DEFINE_SPINLOCK(kthread_create_lock);
35static LIST_HEAD(kthread_create_list);
36struct task_struct *kthreadd_task;
37
38struct kthread_create_info
39{
40 /* Information passed to kthread() from kthreadd. */
41 int (*threadfn)(void *data);
42 void *data;
43 int node;
44
45 /* Result passed back to kthread_create() from kthreadd. */
46 struct task_struct *result;
47 struct completion *done;
48
49 struct list_head list;
50};
51
52struct kthread {
53 unsigned long flags;
54 unsigned int cpu;
55 int (*threadfn)(void *);
56 void *data;
57 mm_segment_t oldfs;
58 struct completion parked;
59 struct completion exited;
60#ifdef CONFIG_BLK_CGROUP
61 struct cgroup_subsys_state *blkcg_css;
62#endif
63};
64
65enum KTHREAD_BITS {
66 KTHREAD_IS_PER_CPU = 0,
67 KTHREAD_SHOULD_STOP,
68 KTHREAD_SHOULD_PARK,
69};
70
71static inline struct kthread *to_kthread(struct task_struct *k)
72{
73 WARN_ON(!(k->flags & PF_KTHREAD));
74 return (__force void *)k->set_child_tid;
75}
76
77/*
78 * Variant of to_kthread() that doesn't assume @p is a kthread.
79 *
80 * Per construction; when:
81 *
82 * (p->flags & PF_KTHREAD) && p->set_child_tid
83 *
84 * the task is both a kthread and struct kthread is persistent. However
85 * PF_KTHREAD on it's own is not, kernel_thread() can exec() (See umh.c and
86 * begin_new_exec()).
87 */
88static inline struct kthread *__to_kthread(struct task_struct *p)
89{
90 void *kthread = (__force void *)p->set_child_tid;
91 if (kthread && !(p->flags & PF_KTHREAD))
92 kthread = NULL;
93 return kthread;
94}
95
96void set_kthread_struct(struct task_struct *p)
97{
98 struct kthread *kthread;
99
100 if (__to_kthread(p))
101 return;
102
103 kthread = kzalloc(sizeof(*kthread), GFP_KERNEL);
104 /*
105 * We abuse ->set_child_tid to avoid the new member and because it
106 * can't be wrongly copied by copy_process(). We also rely on fact
107 * that the caller can't exec, so PF_KTHREAD can't be cleared.
108 */
109 p->set_child_tid = (__force void __user *)kthread;
110}
111
112void free_kthread_struct(struct task_struct *k)
113{
114 struct kthread *kthread;
115
116 /*
117 * Can be NULL if this kthread was created by kernel_thread()
118 * or if kmalloc() in kthread() failed.
119 */
120 kthread = to_kthread(k);
121#ifdef CONFIG_BLK_CGROUP
122 WARN_ON_ONCE(kthread && kthread->blkcg_css);
123#endif
124 kfree(kthread);
125}
126
127/**
128 * kthread_should_stop - should this kthread return now?
129 *
130 * When someone calls kthread_stop() on your kthread, it will be woken
131 * and this will return true. You should then return, and your return
132 * value will be passed through to kthread_stop().
133 */
134bool kthread_should_stop(void)
135{
136 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
137}
138EXPORT_SYMBOL(kthread_should_stop);
139
140bool __kthread_should_park(struct task_struct *k)
141{
142 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
143}
144EXPORT_SYMBOL_GPL(__kthread_should_park);
145
146/**
147 * kthread_should_park - should this kthread park now?
148 *
149 * When someone calls kthread_park() on your kthread, it will be woken
150 * and this will return true. You should then do the necessary
151 * cleanup and call kthread_parkme()
152 *
153 * Similar to kthread_should_stop(), but this keeps the thread alive
154 * and in a park position. kthread_unpark() "restarts" the thread and
155 * calls the thread function again.
156 */
157bool kthread_should_park(void)
158{
159 return __kthread_should_park(current);
160}
161EXPORT_SYMBOL_GPL(kthread_should_park);
162
163/**
164 * kthread_freezable_should_stop - should this freezable kthread return now?
165 * @was_frozen: optional out parameter, indicates whether %current was frozen
166 *
167 * kthread_should_stop() for freezable kthreads, which will enter
168 * refrigerator if necessary. This function is safe from kthread_stop() /
169 * freezer deadlock and freezable kthreads should use this function instead
170 * of calling try_to_freeze() directly.
171 */
172bool kthread_freezable_should_stop(bool *was_frozen)
173{
174 bool frozen = false;
175
176 might_sleep();
177
178 if (unlikely(freezing(current)))
179 frozen = __refrigerator(true);
180
181 if (was_frozen)
182 *was_frozen = frozen;
183
184 return kthread_should_stop();
185}
186EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
187
188/**
189 * kthread_func - return the function specified on kthread creation
190 * @task: kthread task in question
191 *
192 * Returns NULL if the task is not a kthread.
193 */
194void *kthread_func(struct task_struct *task)
195{
196 struct kthread *kthread = __to_kthread(task);
197 if (kthread)
198 return kthread->threadfn;
199 return NULL;
200}
201EXPORT_SYMBOL_GPL(kthread_func);
202
203/**
204 * kthread_data - return data value specified on kthread creation
205 * @task: kthread task in question
206 *
207 * Return the data value specified when kthread @task was created.
208 * The caller is responsible for ensuring the validity of @task when
209 * calling this function.
210 */
211void *kthread_data(struct task_struct *task)
212{
213 return to_kthread(task)->data;
214}
215EXPORT_SYMBOL_GPL(kthread_data);
216
217/**
218 * kthread_probe_data - speculative version of kthread_data()
219 * @task: possible kthread task in question
220 *
221 * @task could be a kthread task. Return the data value specified when it
222 * was created if accessible. If @task isn't a kthread task or its data is
223 * inaccessible for any reason, %NULL is returned. This function requires
224 * that @task itself is safe to dereference.
225 */
226void *kthread_probe_data(struct task_struct *task)
227{
228 struct kthread *kthread = __to_kthread(task);
229 void *data = NULL;
230
231 if (kthread)
232 copy_from_kernel_nofault(&data, &kthread->data, sizeof(data));
233 return data;
234}
235
236static void __kthread_parkme(struct kthread *self)
237{
238 for (;;) {
239 /*
240 * TASK_PARKED is a special state; we must serialize against
241 * possible pending wakeups to avoid store-store collisions on
242 * task->state.
243 *
244 * Such a collision might possibly result in the task state
245 * changin from TASK_PARKED and us failing the
246 * wait_task_inactive() in kthread_park().
247 */
248 set_special_state(TASK_PARKED);
249 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
250 break;
251
252 /*
253 * Thread is going to call schedule(), do not preempt it,
254 * or the caller of kthread_park() may spend more time in
255 * wait_task_inactive().
256 */
257 preempt_disable();
258 complete(&self->parked);
259 schedule_preempt_disabled();
260 preempt_enable();
261 }
262 __set_current_state(TASK_RUNNING);
263}
264
265void kthread_parkme(void)
266{
267 __kthread_parkme(to_kthread(current));
268}
269EXPORT_SYMBOL_GPL(kthread_parkme);
270
271static int kthread(void *_create)
272{
273 /* Copy data: it's on kthread's stack */
274 struct kthread_create_info *create = _create;
275 int (*threadfn)(void *data) = create->threadfn;
276 void *data = create->data;
277 struct completion *done;
278 struct kthread *self;
279 int ret;
280
281 set_kthread_struct(current);
282 self = to_kthread(current);
283
284 /* If user was SIGKILLed, I release the structure. */
285 done = xchg(&create->done, NULL);
286 if (!done) {
287 kfree(create);
288 do_exit(-EINTR);
289 }
290
291 if (!self) {
292 create->result = ERR_PTR(-ENOMEM);
293 complete(done);
294 do_exit(-ENOMEM);
295 }
296
297 self->threadfn = threadfn;
298 self->data = data;
299 init_completion(&self->exited);
300 init_completion(&self->parked);
301 current->vfork_done = &self->exited;
302
303 /* OK, tell user we're spawned, wait for stop or wakeup */
304 __set_current_state(TASK_UNINTERRUPTIBLE);
305 create->result = current;
306 /*
307 * Thread is going to call schedule(), do not preempt it,
308 * or the creator may spend more time in wait_task_inactive().
309 */
310 preempt_disable();
311 complete(done);
312 schedule_preempt_disabled();
313 preempt_enable();
314
315 ret = -EINTR;
316 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
317 cgroup_kthread_ready();
318 __kthread_parkme(self);
319 ret = threadfn(data);
320 }
321 do_exit(ret);
322}
323
324/* called from kernel_clone() to get node information for about to be created task */
325int tsk_fork_get_node(struct task_struct *tsk)
326{
327#ifdef CONFIG_NUMA
328 if (tsk == kthreadd_task)
329 return tsk->pref_node_fork;
330#endif
331 return NUMA_NO_NODE;
332}
333
334static void create_kthread(struct kthread_create_info *create)
335{
336 int pid;
337
338#ifdef CONFIG_NUMA
339 current->pref_node_fork = create->node;
340#endif
341 /* We want our own signal handler (we take no signals by default). */
342 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
343 if (pid < 0) {
344 /* If user was SIGKILLed, I release the structure. */
345 struct completion *done = xchg(&create->done, NULL);
346
347 if (!done) {
348 kfree(create);
349 return;
350 }
351 create->result = ERR_PTR(pid);
352 complete(done);
353 }
354}
355
356static __printf(4, 0)
357struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
358 void *data, int node,
359 const char namefmt[],
360 va_list args)
361{
362 DECLARE_COMPLETION_ONSTACK(done);
363 struct task_struct *task;
364 struct kthread_create_info *create = kmalloc(sizeof(*create),
365 GFP_KERNEL);
366
367 if (!create)
368 return ERR_PTR(-ENOMEM);
369 create->threadfn = threadfn;
370 create->data = data;
371 create->node = node;
372 create->done = &done;
373
374 spin_lock(&kthread_create_lock);
375 list_add_tail(&create->list, &kthread_create_list);
376 spin_unlock(&kthread_create_lock);
377
378 wake_up_process(kthreadd_task);
379 /*
380 * Wait for completion in killable state, for I might be chosen by
381 * the OOM killer while kthreadd is trying to allocate memory for
382 * new kernel thread.
383 */
384 if (unlikely(wait_for_completion_killable(&done))) {
385 /*
386 * If I was SIGKILLed before kthreadd (or new kernel thread)
387 * calls complete(), leave the cleanup of this structure to
388 * that thread.
389 */
390 if (xchg(&create->done, NULL))
391 return ERR_PTR(-EINTR);
392 /*
393 * kthreadd (or new kernel thread) will call complete()
394 * shortly.
395 */
396 wait_for_completion(&done);
397 }
398 task = create->result;
399 if (!IS_ERR(task)) {
400 static const struct sched_param param = { .sched_priority = 0 };
401 char name[TASK_COMM_LEN];
402
403 /*
404 * task is already visible to other tasks, so updating
405 * COMM must be protected.
406 */
407 vsnprintf(name, sizeof(name), namefmt, args);
408 set_task_comm(task, name);
409 /*
410 * root may have changed our (kthreadd's) priority or CPU mask.
411 * The kernel thread should not inherit these properties.
412 */
413 sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m);
414 set_cpus_allowed_ptr(task,
415 housekeeping_cpumask(HK_FLAG_KTHREAD));
416 }
417 kfree(create);
418 return task;
419}
420
421/**
422 * kthread_create_on_node - create a kthread.
423 * @threadfn: the function to run until signal_pending(current).
424 * @data: data ptr for @threadfn.
425 * @node: task and thread structures for the thread are allocated on this node
426 * @namefmt: printf-style name for the thread.
427 *
428 * Description: This helper function creates and names a kernel
429 * thread. The thread will be stopped: use wake_up_process() to start
430 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
431 * is affine to all CPUs.
432 *
433 * If thread is going to be bound on a particular cpu, give its node
434 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
435 * When woken, the thread will run @threadfn() with @data as its
436 * argument. @threadfn() can either call do_exit() directly if it is a
437 * standalone thread for which no one will call kthread_stop(), or
438 * return when 'kthread_should_stop()' is true (which means
439 * kthread_stop() has been called). The return value should be zero
440 * or a negative error number; it will be passed to kthread_stop().
441 *
442 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
443 */
444struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
445 void *data, int node,
446 const char namefmt[],
447 ...)
448{
449 struct task_struct *task;
450 va_list args;
451
452 va_start(args, namefmt);
453 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
454 va_end(args);
455
456 return task;
457}
458EXPORT_SYMBOL(kthread_create_on_node);
459
460static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, unsigned int state)
461{
462 unsigned long flags;
463
464 if (!wait_task_inactive(p, state)) {
465 WARN_ON(1);
466 return;
467 }
468
469 /* It's safe because the task is inactive. */
470 raw_spin_lock_irqsave(&p->pi_lock, flags);
471 do_set_cpus_allowed(p, mask);
472 p->flags |= PF_NO_SETAFFINITY;
473 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
474}
475
476static void __kthread_bind(struct task_struct *p, unsigned int cpu, unsigned int state)
477{
478 __kthread_bind_mask(p, cpumask_of(cpu), state);
479}
480
481void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
482{
483 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
484}
485
486/**
487 * kthread_bind - bind a just-created kthread to a cpu.
488 * @p: thread created by kthread_create().
489 * @cpu: cpu (might not be online, must be possible) for @k to run on.
490 *
491 * Description: This function is equivalent to set_cpus_allowed(),
492 * except that @cpu doesn't need to be online, and the thread must be
493 * stopped (i.e., just returned from kthread_create()).
494 */
495void kthread_bind(struct task_struct *p, unsigned int cpu)
496{
497 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
498}
499EXPORT_SYMBOL(kthread_bind);
500
501/**
502 * kthread_create_on_cpu - Create a cpu bound kthread
503 * @threadfn: the function to run until signal_pending(current).
504 * @data: data ptr for @threadfn.
505 * @cpu: The cpu on which the thread should be bound,
506 * @namefmt: printf-style name for the thread. Format is restricted
507 * to "name.*%u". Code fills in cpu number.
508 *
509 * Description: This helper function creates and names a kernel thread
510 */
511struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
512 void *data, unsigned int cpu,
513 const char *namefmt)
514{
515 struct task_struct *p;
516
517 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
518 cpu);
519 if (IS_ERR(p))
520 return p;
521 kthread_bind(p, cpu);
522 /* CPU hotplug need to bind once again when unparking the thread. */
523 to_kthread(p)->cpu = cpu;
524 return p;
525}
526
527void kthread_set_per_cpu(struct task_struct *k, int cpu)
528{
529 struct kthread *kthread = to_kthread(k);
530 if (!kthread)
531 return;
532
533 WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY));
534
535 if (cpu < 0) {
536 clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
537 return;
538 }
539
540 kthread->cpu = cpu;
541 set_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
542}
543
544bool kthread_is_per_cpu(struct task_struct *p)
545{
546 struct kthread *kthread = __to_kthread(p);
547 if (!kthread)
548 return false;
549
550 return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
551}
552
553/**
554 * kthread_unpark - unpark a thread created by kthread_create().
555 * @k: thread created by kthread_create().
556 *
557 * Sets kthread_should_park() for @k to return false, wakes it, and
558 * waits for it to return. If the thread is marked percpu then its
559 * bound to the cpu again.
560 */
561void kthread_unpark(struct task_struct *k)
562{
563 struct kthread *kthread = to_kthread(k);
564
565 /*
566 * Newly created kthread was parked when the CPU was offline.
567 * The binding was lost and we need to set it again.
568 */
569 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
570 __kthread_bind(k, kthread->cpu, TASK_PARKED);
571
572 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
573 /*
574 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
575 */
576 wake_up_state(k, TASK_PARKED);
577}
578EXPORT_SYMBOL_GPL(kthread_unpark);
579
580/**
581 * kthread_park - park a thread created by kthread_create().
582 * @k: thread created by kthread_create().
583 *
584 * Sets kthread_should_park() for @k to return true, wakes it, and
585 * waits for it to return. This can also be called after kthread_create()
586 * instead of calling wake_up_process(): the thread will park without
587 * calling threadfn().
588 *
589 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
590 * If called by the kthread itself just the park bit is set.
591 */
592int kthread_park(struct task_struct *k)
593{
594 struct kthread *kthread = to_kthread(k);
595
596 if (WARN_ON(k->flags & PF_EXITING))
597 return -ENOSYS;
598
599 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
600 return -EBUSY;
601
602 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
603 if (k != current) {
604 wake_up_process(k);
605 /*
606 * Wait for __kthread_parkme() to complete(), this means we
607 * _will_ have TASK_PARKED and are about to call schedule().
608 */
609 wait_for_completion(&kthread->parked);
610 /*
611 * Now wait for that schedule() to complete and the task to
612 * get scheduled out.
613 */
614 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
615 }
616
617 return 0;
618}
619EXPORT_SYMBOL_GPL(kthread_park);
620
621/**
622 * kthread_stop - stop a thread created by kthread_create().
623 * @k: thread created by kthread_create().
624 *
625 * Sets kthread_should_stop() for @k to return true, wakes it, and
626 * waits for it to exit. This can also be called after kthread_create()
627 * instead of calling wake_up_process(): the thread will exit without
628 * calling threadfn().
629 *
630 * If threadfn() may call do_exit() itself, the caller must ensure
631 * task_struct can't go away.
632 *
633 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
634 * was never called.
635 */
636int kthread_stop(struct task_struct *k)
637{
638 struct kthread *kthread;
639 int ret;
640
641 trace_sched_kthread_stop(k);
642
643 get_task_struct(k);
644 kthread = to_kthread(k);
645 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
646 kthread_unpark(k);
647 wake_up_process(k);
648 wait_for_completion(&kthread->exited);
649 ret = k->exit_code;
650 put_task_struct(k);
651
652 trace_sched_kthread_stop_ret(ret);
653 return ret;
654}
655EXPORT_SYMBOL(kthread_stop);
656
657int kthreadd(void *unused)
658{
659 struct task_struct *tsk = current;
660
661 /* Setup a clean context for our children to inherit. */
662 set_task_comm(tsk, "kthreadd");
663 ignore_signals(tsk);
664 set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD));
665 set_mems_allowed(node_states[N_MEMORY]);
666
667 current->flags |= PF_NOFREEZE;
668 cgroup_init_kthreadd();
669
670 for (;;) {
671 set_current_state(TASK_INTERRUPTIBLE);
672 if (list_empty(&kthread_create_list))
673 schedule();
674 __set_current_state(TASK_RUNNING);
675
676 spin_lock(&kthread_create_lock);
677 while (!list_empty(&kthread_create_list)) {
678 struct kthread_create_info *create;
679
680 create = list_entry(kthread_create_list.next,
681 struct kthread_create_info, list);
682 list_del_init(&create->list);
683 spin_unlock(&kthread_create_lock);
684
685 create_kthread(create);
686
687 spin_lock(&kthread_create_lock);
688 }
689 spin_unlock(&kthread_create_lock);
690 }
691
692 return 0;
693}
694
695void __kthread_init_worker(struct kthread_worker *worker,
696 const char *name,
697 struct lock_class_key *key)
698{
699 memset(worker, 0, sizeof(struct kthread_worker));
700 raw_spin_lock_init(&worker->lock);
701 lockdep_set_class_and_name(&worker->lock, key, name);
702 INIT_LIST_HEAD(&worker->work_list);
703 INIT_LIST_HEAD(&worker->delayed_work_list);
704}
705EXPORT_SYMBOL_GPL(__kthread_init_worker);
706
707/**
708 * kthread_worker_fn - kthread function to process kthread_worker
709 * @worker_ptr: pointer to initialized kthread_worker
710 *
711 * This function implements the main cycle of kthread worker. It processes
712 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
713 * is empty.
714 *
715 * The works are not allowed to keep any locks, disable preemption or interrupts
716 * when they finish. There is defined a safe point for freezing when one work
717 * finishes and before a new one is started.
718 *
719 * Also the works must not be handled by more than one worker at the same time,
720 * see also kthread_queue_work().
721 */
722int kthread_worker_fn(void *worker_ptr)
723{
724 struct kthread_worker *worker = worker_ptr;
725 struct kthread_work *work;
726
727 /*
728 * FIXME: Update the check and remove the assignment when all kthread
729 * worker users are created using kthread_create_worker*() functions.
730 */
731 WARN_ON(worker->task && worker->task != current);
732 worker->task = current;
733
734 if (worker->flags & KTW_FREEZABLE)
735 set_freezable();
736
737repeat:
738 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
739
740 if (kthread_should_stop()) {
741 __set_current_state(TASK_RUNNING);
742 raw_spin_lock_irq(&worker->lock);
743 worker->task = NULL;
744 raw_spin_unlock_irq(&worker->lock);
745 return 0;
746 }
747
748 work = NULL;
749 raw_spin_lock_irq(&worker->lock);
750 if (!list_empty(&worker->work_list)) {
751 work = list_first_entry(&worker->work_list,
752 struct kthread_work, node);
753 list_del_init(&work->node);
754 }
755 worker->current_work = work;
756 raw_spin_unlock_irq(&worker->lock);
757
758 if (work) {
759 kthread_work_func_t func = work->func;
760 __set_current_state(TASK_RUNNING);
761 trace_sched_kthread_work_execute_start(work);
762 work->func(work);
763 /*
764 * Avoid dereferencing work after this point. The trace
765 * event only cares about the address.
766 */
767 trace_sched_kthread_work_execute_end(work, func);
768 } else if (!freezing(current))
769 schedule();
770
771 try_to_freeze();
772 cond_resched();
773 goto repeat;
774}
775EXPORT_SYMBOL_GPL(kthread_worker_fn);
776
777static __printf(3, 0) struct kthread_worker *
778__kthread_create_worker(int cpu, unsigned int flags,
779 const char namefmt[], va_list args)
780{
781 struct kthread_worker *worker;
782 struct task_struct *task;
783 int node = NUMA_NO_NODE;
784
785 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
786 if (!worker)
787 return ERR_PTR(-ENOMEM);
788
789 kthread_init_worker(worker);
790
791 if (cpu >= 0)
792 node = cpu_to_node(cpu);
793
794 task = __kthread_create_on_node(kthread_worker_fn, worker,
795 node, namefmt, args);
796 if (IS_ERR(task))
797 goto fail_task;
798
799 if (cpu >= 0)
800 kthread_bind(task, cpu);
801
802 worker->flags = flags;
803 worker->task = task;
804 wake_up_process(task);
805 return worker;
806
807fail_task:
808 kfree(worker);
809 return ERR_CAST(task);
810}
811
812/**
813 * kthread_create_worker - create a kthread worker
814 * @flags: flags modifying the default behavior of the worker
815 * @namefmt: printf-style name for the kthread worker (task).
816 *
817 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
818 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
819 * when the worker was SIGKILLed.
820 */
821struct kthread_worker *
822kthread_create_worker(unsigned int flags, const char namefmt[], ...)
823{
824 struct kthread_worker *worker;
825 va_list args;
826
827 va_start(args, namefmt);
828 worker = __kthread_create_worker(-1, flags, namefmt, args);
829 va_end(args);
830
831 return worker;
832}
833EXPORT_SYMBOL(kthread_create_worker);
834
835/**
836 * kthread_create_worker_on_cpu - create a kthread worker and bind it
837 * to a given CPU and the associated NUMA node.
838 * @cpu: CPU number
839 * @flags: flags modifying the default behavior of the worker
840 * @namefmt: printf-style name for the kthread worker (task).
841 *
842 * Use a valid CPU number if you want to bind the kthread worker
843 * to the given CPU and the associated NUMA node.
844 *
845 * A good practice is to add the cpu number also into the worker name.
846 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
847 *
848 * CPU hotplug:
849 * The kthread worker API is simple and generic. It just provides a way
850 * to create, use, and destroy workers.
851 *
852 * It is up to the API user how to handle CPU hotplug. They have to decide
853 * how to handle pending work items, prevent queuing new ones, and
854 * restore the functionality when the CPU goes off and on. There are a
855 * few catches:
856 *
857 * - CPU affinity gets lost when it is scheduled on an offline CPU.
858 *
859 * - The worker might not exist when the CPU was off when the user
860 * created the workers.
861 *
862 * Good practice is to implement two CPU hotplug callbacks and to
863 * destroy/create the worker when the CPU goes down/up.
864 *
865 * Return:
866 * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
867 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
868 * when the worker was SIGKILLed.
869 */
870struct kthread_worker *
871kthread_create_worker_on_cpu(int cpu, unsigned int flags,
872 const char namefmt[], ...)
873{
874 struct kthread_worker *worker;
875 va_list args;
876
877 va_start(args, namefmt);
878 worker = __kthread_create_worker(cpu, flags, namefmt, args);
879 va_end(args);
880
881 return worker;
882}
883EXPORT_SYMBOL(kthread_create_worker_on_cpu);
884
885/*
886 * Returns true when the work could not be queued at the moment.
887 * It happens when it is already pending in a worker list
888 * or when it is being cancelled.
889 */
890static inline bool queuing_blocked(struct kthread_worker *worker,
891 struct kthread_work *work)
892{
893 lockdep_assert_held(&worker->lock);
894
895 return !list_empty(&work->node) || work->canceling;
896}
897
898static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
899 struct kthread_work *work)
900{
901 lockdep_assert_held(&worker->lock);
902 WARN_ON_ONCE(!list_empty(&work->node));
903 /* Do not use a work with >1 worker, see kthread_queue_work() */
904 WARN_ON_ONCE(work->worker && work->worker != worker);
905}
906
907/* insert @work before @pos in @worker */
908static void kthread_insert_work(struct kthread_worker *worker,
909 struct kthread_work *work,
910 struct list_head *pos)
911{
912 kthread_insert_work_sanity_check(worker, work);
913
914 trace_sched_kthread_work_queue_work(worker, work);
915
916 list_add_tail(&work->node, pos);
917 work->worker = worker;
918 if (!worker->current_work && likely(worker->task))
919 wake_up_process(worker->task);
920}
921
922/**
923 * kthread_queue_work - queue a kthread_work
924 * @worker: target kthread_worker
925 * @work: kthread_work to queue
926 *
927 * Queue @work to work processor @task for async execution. @task
928 * must have been created with kthread_worker_create(). Returns %true
929 * if @work was successfully queued, %false if it was already pending.
930 *
931 * Reinitialize the work if it needs to be used by another worker.
932 * For example, when the worker was stopped and started again.
933 */
934bool kthread_queue_work(struct kthread_worker *worker,
935 struct kthread_work *work)
936{
937 bool ret = false;
938 unsigned long flags;
939
940 raw_spin_lock_irqsave(&worker->lock, flags);
941 if (!queuing_blocked(worker, work)) {
942 kthread_insert_work(worker, work, &worker->work_list);
943 ret = true;
944 }
945 raw_spin_unlock_irqrestore(&worker->lock, flags);
946 return ret;
947}
948EXPORT_SYMBOL_GPL(kthread_queue_work);
949
950/**
951 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
952 * delayed work when the timer expires.
953 * @t: pointer to the expired timer
954 *
955 * The format of the function is defined by struct timer_list.
956 * It should have been called from irqsafe timer with irq already off.
957 */
958void kthread_delayed_work_timer_fn(struct timer_list *t)
959{
960 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
961 struct kthread_work *work = &dwork->work;
962 struct kthread_worker *worker = work->worker;
963 unsigned long flags;
964
965 /*
966 * This might happen when a pending work is reinitialized.
967 * It means that it is used a wrong way.
968 */
969 if (WARN_ON_ONCE(!worker))
970 return;
971
972 raw_spin_lock_irqsave(&worker->lock, flags);
973 /* Work must not be used with >1 worker, see kthread_queue_work(). */
974 WARN_ON_ONCE(work->worker != worker);
975
976 /* Move the work from worker->delayed_work_list. */
977 WARN_ON_ONCE(list_empty(&work->node));
978 list_del_init(&work->node);
979 if (!work->canceling)
980 kthread_insert_work(worker, work, &worker->work_list);
981
982 raw_spin_unlock_irqrestore(&worker->lock, flags);
983}
984EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
985
986static void __kthread_queue_delayed_work(struct kthread_worker *worker,
987 struct kthread_delayed_work *dwork,
988 unsigned long delay)
989{
990 struct timer_list *timer = &dwork->timer;
991 struct kthread_work *work = &dwork->work;
992
993 WARN_ON_FUNCTION_MISMATCH(timer->function,
994 kthread_delayed_work_timer_fn);
995
996 /*
997 * If @delay is 0, queue @dwork->work immediately. This is for
998 * both optimization and correctness. The earliest @timer can
999 * expire is on the closest next tick and delayed_work users depend
1000 * on that there's no such delay when @delay is 0.
1001 */
1002 if (!delay) {
1003 kthread_insert_work(worker, work, &worker->work_list);
1004 return;
1005 }
1006
1007 /* Be paranoid and try to detect possible races already now. */
1008 kthread_insert_work_sanity_check(worker, work);
1009
1010 list_add(&work->node, &worker->delayed_work_list);
1011 work->worker = worker;
1012 timer->expires = jiffies + delay;
1013 add_timer(timer);
1014}
1015
1016/**
1017 * kthread_queue_delayed_work - queue the associated kthread work
1018 * after a delay.
1019 * @worker: target kthread_worker
1020 * @dwork: kthread_delayed_work to queue
1021 * @delay: number of jiffies to wait before queuing
1022 *
1023 * If the work has not been pending it starts a timer that will queue
1024 * the work after the given @delay. If @delay is zero, it queues the
1025 * work immediately.
1026 *
1027 * Return: %false if the @work has already been pending. It means that
1028 * either the timer was running or the work was queued. It returns %true
1029 * otherwise.
1030 */
1031bool kthread_queue_delayed_work(struct kthread_worker *worker,
1032 struct kthread_delayed_work *dwork,
1033 unsigned long delay)
1034{
1035 struct kthread_work *work = &dwork->work;
1036 unsigned long flags;
1037 bool ret = false;
1038
1039 raw_spin_lock_irqsave(&worker->lock, flags);
1040
1041 if (!queuing_blocked(worker, work)) {
1042 __kthread_queue_delayed_work(worker, dwork, delay);
1043 ret = true;
1044 }
1045
1046 raw_spin_unlock_irqrestore(&worker->lock, flags);
1047 return ret;
1048}
1049EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
1050
1051struct kthread_flush_work {
1052 struct kthread_work work;
1053 struct completion done;
1054};
1055
1056static void kthread_flush_work_fn(struct kthread_work *work)
1057{
1058 struct kthread_flush_work *fwork =
1059 container_of(work, struct kthread_flush_work, work);
1060 complete(&fwork->done);
1061}
1062
1063/**
1064 * kthread_flush_work - flush a kthread_work
1065 * @work: work to flush
1066 *
1067 * If @work is queued or executing, wait for it to finish execution.
1068 */
1069void kthread_flush_work(struct kthread_work *work)
1070{
1071 struct kthread_flush_work fwork = {
1072 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1073 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1074 };
1075 struct kthread_worker *worker;
1076 bool noop = false;
1077
1078 worker = work->worker;
1079 if (!worker)
1080 return;
1081
1082 raw_spin_lock_irq(&worker->lock);
1083 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1084 WARN_ON_ONCE(work->worker != worker);
1085
1086 if (!list_empty(&work->node))
1087 kthread_insert_work(worker, &fwork.work, work->node.next);
1088 else if (worker->current_work == work)
1089 kthread_insert_work(worker, &fwork.work,
1090 worker->work_list.next);
1091 else
1092 noop = true;
1093
1094 raw_spin_unlock_irq(&worker->lock);
1095
1096 if (!noop)
1097 wait_for_completion(&fwork.done);
1098}
1099EXPORT_SYMBOL_GPL(kthread_flush_work);
1100
1101/*
1102 * Make sure that the timer is neither set nor running and could
1103 * not manipulate the work list_head any longer.
1104 *
1105 * The function is called under worker->lock. The lock is temporary
1106 * released but the timer can't be set again in the meantime.
1107 */
1108static void kthread_cancel_delayed_work_timer(struct kthread_work *work,
1109 unsigned long *flags)
1110{
1111 struct kthread_delayed_work *dwork =
1112 container_of(work, struct kthread_delayed_work, work);
1113 struct kthread_worker *worker = work->worker;
1114
1115 /*
1116 * del_timer_sync() must be called to make sure that the timer
1117 * callback is not running. The lock must be temporary released
1118 * to avoid a deadlock with the callback. In the meantime,
1119 * any queuing is blocked by setting the canceling counter.
1120 */
1121 work->canceling++;
1122 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1123 del_timer_sync(&dwork->timer);
1124 raw_spin_lock_irqsave(&worker->lock, *flags);
1125 work->canceling--;
1126}
1127
1128/*
1129 * This function removes the work from the worker queue.
1130 *
1131 * It is called under worker->lock. The caller must make sure that
1132 * the timer used by delayed work is not running, e.g. by calling
1133 * kthread_cancel_delayed_work_timer().
1134 *
1135 * The work might still be in use when this function finishes. See the
1136 * current_work proceed by the worker.
1137 *
1138 * Return: %true if @work was pending and successfully canceled,
1139 * %false if @work was not pending
1140 */
1141static bool __kthread_cancel_work(struct kthread_work *work)
1142{
1143 /*
1144 * Try to remove the work from a worker list. It might either
1145 * be from worker->work_list or from worker->delayed_work_list.
1146 */
1147 if (!list_empty(&work->node)) {
1148 list_del_init(&work->node);
1149 return true;
1150 }
1151
1152 return false;
1153}
1154
1155/**
1156 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1157 * @worker: kthread worker to use
1158 * @dwork: kthread delayed work to queue
1159 * @delay: number of jiffies to wait before queuing
1160 *
1161 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1162 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1163 * @work is guaranteed to be queued immediately.
1164 *
1165 * Return: %false if @dwork was idle and queued, %true otherwise.
1166 *
1167 * A special case is when the work is being canceled in parallel.
1168 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1169 * or yet another kthread_mod_delayed_work() call. We let the other command
1170 * win and return %true here. The return value can be used for reference
1171 * counting and the number of queued works stays the same. Anyway, the caller
1172 * is supposed to synchronize these operations a reasonable way.
1173 *
1174 * This function is safe to call from any context including IRQ handler.
1175 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1176 * for details.
1177 */
1178bool kthread_mod_delayed_work(struct kthread_worker *worker,
1179 struct kthread_delayed_work *dwork,
1180 unsigned long delay)
1181{
1182 struct kthread_work *work = &dwork->work;
1183 unsigned long flags;
1184 int ret;
1185
1186 raw_spin_lock_irqsave(&worker->lock, flags);
1187
1188 /* Do not bother with canceling when never queued. */
1189 if (!work->worker) {
1190 ret = false;
1191 goto fast_queue;
1192 }
1193
1194 /* Work must not be used with >1 worker, see kthread_queue_work() */
1195 WARN_ON_ONCE(work->worker != worker);
1196
1197 /*
1198 * Temporary cancel the work but do not fight with another command
1199 * that is canceling the work as well.
1200 *
1201 * It is a bit tricky because of possible races with another
1202 * mod_delayed_work() and cancel_delayed_work() callers.
1203 *
1204 * The timer must be canceled first because worker->lock is released
1205 * when doing so. But the work can be removed from the queue (list)
1206 * only when it can be queued again so that the return value can
1207 * be used for reference counting.
1208 */
1209 kthread_cancel_delayed_work_timer(work, &flags);
1210 if (work->canceling) {
1211 /* The number of works in the queue does not change. */
1212 ret = true;
1213 goto out;
1214 }
1215 ret = __kthread_cancel_work(work);
1216
1217fast_queue:
1218 __kthread_queue_delayed_work(worker, dwork, delay);
1219out:
1220 raw_spin_unlock_irqrestore(&worker->lock, flags);
1221 return ret;
1222}
1223EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1224
1225static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1226{
1227 struct kthread_worker *worker = work->worker;
1228 unsigned long flags;
1229 int ret = false;
1230
1231 if (!worker)
1232 goto out;
1233
1234 raw_spin_lock_irqsave(&worker->lock, flags);
1235 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1236 WARN_ON_ONCE(work->worker != worker);
1237
1238 if (is_dwork)
1239 kthread_cancel_delayed_work_timer(work, &flags);
1240
1241 ret = __kthread_cancel_work(work);
1242
1243 if (worker->current_work != work)
1244 goto out_fast;
1245
1246 /*
1247 * The work is in progress and we need to wait with the lock released.
1248 * In the meantime, block any queuing by setting the canceling counter.
1249 */
1250 work->canceling++;
1251 raw_spin_unlock_irqrestore(&worker->lock, flags);
1252 kthread_flush_work(work);
1253 raw_spin_lock_irqsave(&worker->lock, flags);
1254 work->canceling--;
1255
1256out_fast:
1257 raw_spin_unlock_irqrestore(&worker->lock, flags);
1258out:
1259 return ret;
1260}
1261
1262/**
1263 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1264 * @work: the kthread work to cancel
1265 *
1266 * Cancel @work and wait for its execution to finish. This function
1267 * can be used even if the work re-queues itself. On return from this
1268 * function, @work is guaranteed to be not pending or executing on any CPU.
1269 *
1270 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1271 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1272 *
1273 * The caller must ensure that the worker on which @work was last
1274 * queued can't be destroyed before this function returns.
1275 *
1276 * Return: %true if @work was pending, %false otherwise.
1277 */
1278bool kthread_cancel_work_sync(struct kthread_work *work)
1279{
1280 return __kthread_cancel_work_sync(work, false);
1281}
1282EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1283
1284/**
1285 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1286 * wait for it to finish.
1287 * @dwork: the kthread delayed work to cancel
1288 *
1289 * This is kthread_cancel_work_sync() for delayed works.
1290 *
1291 * Return: %true if @dwork was pending, %false otherwise.
1292 */
1293bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1294{
1295 return __kthread_cancel_work_sync(&dwork->work, true);
1296}
1297EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1298
1299/**
1300 * kthread_flush_worker - flush all current works on a kthread_worker
1301 * @worker: worker to flush
1302 *
1303 * Wait until all currently executing or pending works on @worker are
1304 * finished.
1305 */
1306void kthread_flush_worker(struct kthread_worker *worker)
1307{
1308 struct kthread_flush_work fwork = {
1309 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1310 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1311 };
1312
1313 kthread_queue_work(worker, &fwork.work);
1314 wait_for_completion(&fwork.done);
1315}
1316EXPORT_SYMBOL_GPL(kthread_flush_worker);
1317
1318/**
1319 * kthread_destroy_worker - destroy a kthread worker
1320 * @worker: worker to be destroyed
1321 *
1322 * Flush and destroy @worker. The simple flush is enough because the kthread
1323 * worker API is used only in trivial scenarios. There are no multi-step state
1324 * machines needed.
1325 */
1326void kthread_destroy_worker(struct kthread_worker *worker)
1327{
1328 struct task_struct *task;
1329
1330 task = worker->task;
1331 if (WARN_ON(!task))
1332 return;
1333
1334 kthread_flush_worker(worker);
1335 kthread_stop(task);
1336 WARN_ON(!list_empty(&worker->work_list));
1337 kfree(worker);
1338}
1339EXPORT_SYMBOL(kthread_destroy_worker);
1340
1341/**
1342 * kthread_use_mm - make the calling kthread operate on an address space
1343 * @mm: address space to operate on
1344 */
1345void kthread_use_mm(struct mm_struct *mm)
1346{
1347 struct mm_struct *active_mm;
1348 struct task_struct *tsk = current;
1349
1350 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1351 WARN_ON_ONCE(tsk->mm);
1352
1353 task_lock(tsk);
1354 /* Hold off tlb flush IPIs while switching mm's */
1355 local_irq_disable();
1356 active_mm = tsk->active_mm;
1357 if (active_mm != mm) {
1358 mmgrab(mm);
1359 tsk->active_mm = mm;
1360 }
1361 tsk->mm = mm;
1362 membarrier_update_current_mm(mm);
1363 switch_mm_irqs_off(active_mm, mm, tsk);
1364 local_irq_enable();
1365 task_unlock(tsk);
1366#ifdef finish_arch_post_lock_switch
1367 finish_arch_post_lock_switch();
1368#endif
1369
1370 /*
1371 * When a kthread starts operating on an address space, the loop
1372 * in membarrier_{private,global}_expedited() may not observe
1373 * that tsk->mm, and not issue an IPI. Membarrier requires a
1374 * memory barrier after storing to tsk->mm, before accessing
1375 * user-space memory. A full memory barrier for membarrier
1376 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1377 * mmdrop(), or explicitly with smp_mb().
1378 */
1379 if (active_mm != mm)
1380 mmdrop(active_mm);
1381 else
1382 smp_mb();
1383
1384 to_kthread(tsk)->oldfs = force_uaccess_begin();
1385}
1386EXPORT_SYMBOL_GPL(kthread_use_mm);
1387
1388/**
1389 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1390 * @mm: address space to operate on
1391 */
1392void kthread_unuse_mm(struct mm_struct *mm)
1393{
1394 struct task_struct *tsk = current;
1395
1396 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1397 WARN_ON_ONCE(!tsk->mm);
1398
1399 force_uaccess_end(to_kthread(tsk)->oldfs);
1400
1401 task_lock(tsk);
1402 /*
1403 * When a kthread stops operating on an address space, the loop
1404 * in membarrier_{private,global}_expedited() may not observe
1405 * that tsk->mm, and not issue an IPI. Membarrier requires a
1406 * memory barrier after accessing user-space memory, before
1407 * clearing tsk->mm.
1408 */
1409 smp_mb__after_spinlock();
1410 sync_mm_rss(mm);
1411 local_irq_disable();
1412 tsk->mm = NULL;
1413 membarrier_update_current_mm(NULL);
1414 /* active_mm is still 'mm' */
1415 enter_lazy_tlb(mm, tsk);
1416 local_irq_enable();
1417 task_unlock(tsk);
1418}
1419EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1420
1421#ifdef CONFIG_BLK_CGROUP
1422/**
1423 * kthread_associate_blkcg - associate blkcg to current kthread
1424 * @css: the cgroup info
1425 *
1426 * Current thread must be a kthread. The thread is running jobs on behalf of
1427 * other threads. In some cases, we expect the jobs attach cgroup info of
1428 * original threads instead of that of current thread. This function stores
1429 * original thread's cgroup info in current kthread context for later
1430 * retrieval.
1431 */
1432void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1433{
1434 struct kthread *kthread;
1435
1436 if (!(current->flags & PF_KTHREAD))
1437 return;
1438 kthread = to_kthread(current);
1439 if (!kthread)
1440 return;
1441
1442 if (kthread->blkcg_css) {
1443 css_put(kthread->blkcg_css);
1444 kthread->blkcg_css = NULL;
1445 }
1446 if (css) {
1447 css_get(css);
1448 kthread->blkcg_css = css;
1449 }
1450}
1451EXPORT_SYMBOL(kthread_associate_blkcg);
1452
1453/**
1454 * kthread_blkcg - get associated blkcg css of current kthread
1455 *
1456 * Current thread must be a kthread.
1457 */
1458struct cgroup_subsys_state *kthread_blkcg(void)
1459{
1460 struct kthread *kthread;
1461
1462 if (current->flags & PF_KTHREAD) {
1463 kthread = to_kthread(current);
1464 if (kthread)
1465 return kthread->blkcg_css;
1466 }
1467 return NULL;
1468}
1469EXPORT_SYMBOL(kthread_blkcg);
1470#endif