Loading...
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 result;
56 int (*threadfn)(void *);
57 void *data;
58 struct completion parked;
59 struct completion exited;
60#ifdef CONFIG_BLK_CGROUP
61 struct cgroup_subsys_state *blkcg_css;
62#endif
63 /* To store the full name if task comm is truncated. */
64 char *full_name;
65};
66
67enum KTHREAD_BITS {
68 KTHREAD_IS_PER_CPU = 0,
69 KTHREAD_SHOULD_STOP,
70 KTHREAD_SHOULD_PARK,
71};
72
73static inline struct kthread *to_kthread(struct task_struct *k)
74{
75 WARN_ON(!(k->flags & PF_KTHREAD));
76 return k->worker_private;
77}
78
79/*
80 * Variant of to_kthread() that doesn't assume @p is a kthread.
81 *
82 * Per construction; when:
83 *
84 * (p->flags & PF_KTHREAD) && p->worker_private
85 *
86 * the task is both a kthread and struct kthread is persistent. However
87 * PF_KTHREAD on it's own is not, kernel_thread() can exec() (See umh.c and
88 * begin_new_exec()).
89 */
90static inline struct kthread *__to_kthread(struct task_struct *p)
91{
92 void *kthread = p->worker_private;
93 if (kthread && !(p->flags & PF_KTHREAD))
94 kthread = NULL;
95 return kthread;
96}
97
98void get_kthread_comm(char *buf, size_t buf_size, struct task_struct *tsk)
99{
100 struct kthread *kthread = to_kthread(tsk);
101
102 if (!kthread || !kthread->full_name) {
103 __get_task_comm(buf, buf_size, tsk);
104 return;
105 }
106
107 strscpy_pad(buf, kthread->full_name, buf_size);
108}
109
110bool set_kthread_struct(struct task_struct *p)
111{
112 struct kthread *kthread;
113
114 if (WARN_ON_ONCE(to_kthread(p)))
115 return false;
116
117 kthread = kzalloc(sizeof(*kthread), GFP_KERNEL);
118 if (!kthread)
119 return false;
120
121 init_completion(&kthread->exited);
122 init_completion(&kthread->parked);
123 p->vfork_done = &kthread->exited;
124
125 p->worker_private = kthread;
126 return true;
127}
128
129void free_kthread_struct(struct task_struct *k)
130{
131 struct kthread *kthread;
132
133 /*
134 * Can be NULL if kmalloc() in set_kthread_struct() failed.
135 */
136 kthread = to_kthread(k);
137 if (!kthread)
138 return;
139
140#ifdef CONFIG_BLK_CGROUP
141 WARN_ON_ONCE(kthread->blkcg_css);
142#endif
143 k->worker_private = NULL;
144 kfree(kthread->full_name);
145 kfree(kthread);
146}
147
148/**
149 * kthread_should_stop - should this kthread return now?
150 *
151 * When someone calls kthread_stop() on your kthread, it will be woken
152 * and this will return true. You should then return, and your return
153 * value will be passed through to kthread_stop().
154 */
155bool kthread_should_stop(void)
156{
157 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
158}
159EXPORT_SYMBOL(kthread_should_stop);
160
161bool __kthread_should_park(struct task_struct *k)
162{
163 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
164}
165EXPORT_SYMBOL_GPL(__kthread_should_park);
166
167/**
168 * kthread_should_park - should this kthread park now?
169 *
170 * When someone calls kthread_park() on your kthread, it will be woken
171 * and this will return true. You should then do the necessary
172 * cleanup and call kthread_parkme()
173 *
174 * Similar to kthread_should_stop(), but this keeps the thread alive
175 * and in a park position. kthread_unpark() "restarts" the thread and
176 * calls the thread function again.
177 */
178bool kthread_should_park(void)
179{
180 return __kthread_should_park(current);
181}
182EXPORT_SYMBOL_GPL(kthread_should_park);
183
184/**
185 * kthread_freezable_should_stop - should this freezable kthread return now?
186 * @was_frozen: optional out parameter, indicates whether %current was frozen
187 *
188 * kthread_should_stop() for freezable kthreads, which will enter
189 * refrigerator if necessary. This function is safe from kthread_stop() /
190 * freezer deadlock and freezable kthreads should use this function instead
191 * of calling try_to_freeze() directly.
192 */
193bool kthread_freezable_should_stop(bool *was_frozen)
194{
195 bool frozen = false;
196
197 might_sleep();
198
199 if (unlikely(freezing(current)))
200 frozen = __refrigerator(true);
201
202 if (was_frozen)
203 *was_frozen = frozen;
204
205 return kthread_should_stop();
206}
207EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
208
209/**
210 * kthread_func - return the function specified on kthread creation
211 * @task: kthread task in question
212 *
213 * Returns NULL if the task is not a kthread.
214 */
215void *kthread_func(struct task_struct *task)
216{
217 struct kthread *kthread = __to_kthread(task);
218 if (kthread)
219 return kthread->threadfn;
220 return NULL;
221}
222EXPORT_SYMBOL_GPL(kthread_func);
223
224/**
225 * kthread_data - return data value specified on kthread creation
226 * @task: kthread task in question
227 *
228 * Return the data value specified when kthread @task was created.
229 * The caller is responsible for ensuring the validity of @task when
230 * calling this function.
231 */
232void *kthread_data(struct task_struct *task)
233{
234 return to_kthread(task)->data;
235}
236EXPORT_SYMBOL_GPL(kthread_data);
237
238/**
239 * kthread_probe_data - speculative version of kthread_data()
240 * @task: possible kthread task in question
241 *
242 * @task could be a kthread task. Return the data value specified when it
243 * was created if accessible. If @task isn't a kthread task or its data is
244 * inaccessible for any reason, %NULL is returned. This function requires
245 * that @task itself is safe to dereference.
246 */
247void *kthread_probe_data(struct task_struct *task)
248{
249 struct kthread *kthread = __to_kthread(task);
250 void *data = NULL;
251
252 if (kthread)
253 copy_from_kernel_nofault(&data, &kthread->data, sizeof(data));
254 return data;
255}
256
257static void __kthread_parkme(struct kthread *self)
258{
259 for (;;) {
260 /*
261 * TASK_PARKED is a special state; we must serialize against
262 * possible pending wakeups to avoid store-store collisions on
263 * task->state.
264 *
265 * Such a collision might possibly result in the task state
266 * changin from TASK_PARKED and us failing the
267 * wait_task_inactive() in kthread_park().
268 */
269 set_special_state(TASK_PARKED);
270 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
271 break;
272
273 /*
274 * Thread is going to call schedule(), do not preempt it,
275 * or the caller of kthread_park() may spend more time in
276 * wait_task_inactive().
277 */
278 preempt_disable();
279 complete(&self->parked);
280 schedule_preempt_disabled();
281 preempt_enable();
282 }
283 __set_current_state(TASK_RUNNING);
284}
285
286void kthread_parkme(void)
287{
288 __kthread_parkme(to_kthread(current));
289}
290EXPORT_SYMBOL_GPL(kthread_parkme);
291
292/**
293 * kthread_exit - Cause the current kthread return @result to kthread_stop().
294 * @result: The integer value to return to kthread_stop().
295 *
296 * While kthread_exit can be called directly, it exists so that
297 * functions which do some additional work in non-modular code such as
298 * module_put_and_kthread_exit can be implemented.
299 *
300 * Does not return.
301 */
302void __noreturn kthread_exit(long result)
303{
304 struct kthread *kthread = to_kthread(current);
305 kthread->result = result;
306 do_exit(0);
307}
308
309/**
310 * kthread_complete_and_exit - Exit the current kthread.
311 * @comp: Completion to complete
312 * @code: The integer value to return to kthread_stop().
313 *
314 * If present complete @comp and the reuturn code to kthread_stop().
315 *
316 * A kernel thread whose module may be removed after the completion of
317 * @comp can use this function exit safely.
318 *
319 * Does not return.
320 */
321void __noreturn kthread_complete_and_exit(struct completion *comp, long code)
322{
323 if (comp)
324 complete(comp);
325
326 kthread_exit(code);
327}
328EXPORT_SYMBOL(kthread_complete_and_exit);
329
330static int kthread(void *_create)
331{
332 static const struct sched_param param = { .sched_priority = 0 };
333 /* Copy data: it's on kthread's stack */
334 struct kthread_create_info *create = _create;
335 int (*threadfn)(void *data) = create->threadfn;
336 void *data = create->data;
337 struct completion *done;
338 struct kthread *self;
339 int ret;
340
341 self = to_kthread(current);
342
343 /* Release the structure when caller killed by a fatal signal. */
344 done = xchg(&create->done, NULL);
345 if (!done) {
346 kfree(create);
347 kthread_exit(-EINTR);
348 }
349
350 self->threadfn = threadfn;
351 self->data = data;
352
353 /*
354 * The new thread inherited kthreadd's priority and CPU mask. Reset
355 * back to default in case they have been changed.
356 */
357 sched_setscheduler_nocheck(current, SCHED_NORMAL, ¶m);
358 set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_TYPE_KTHREAD));
359
360 /* OK, tell user we're spawned, wait for stop or wakeup */
361 __set_current_state(TASK_UNINTERRUPTIBLE);
362 create->result = current;
363 /*
364 * Thread is going to call schedule(), do not preempt it,
365 * or the creator may spend more time in wait_task_inactive().
366 */
367 preempt_disable();
368 complete(done);
369 schedule_preempt_disabled();
370 preempt_enable();
371
372 ret = -EINTR;
373 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
374 cgroup_kthread_ready();
375 __kthread_parkme(self);
376 ret = threadfn(data);
377 }
378 kthread_exit(ret);
379}
380
381/* called from kernel_clone() to get node information for about to be created task */
382int tsk_fork_get_node(struct task_struct *tsk)
383{
384#ifdef CONFIG_NUMA
385 if (tsk == kthreadd_task)
386 return tsk->pref_node_fork;
387#endif
388 return NUMA_NO_NODE;
389}
390
391static void create_kthread(struct kthread_create_info *create)
392{
393 int pid;
394
395#ifdef CONFIG_NUMA
396 current->pref_node_fork = create->node;
397#endif
398 /* We want our own signal handler (we take no signals by default). */
399 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
400 if (pid < 0) {
401 /* Release the structure when caller killed by a fatal signal. */
402 struct completion *done = xchg(&create->done, NULL);
403
404 if (!done) {
405 kfree(create);
406 return;
407 }
408 create->result = ERR_PTR(pid);
409 complete(done);
410 }
411}
412
413static __printf(4, 0)
414struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
415 void *data, int node,
416 const char namefmt[],
417 va_list args)
418{
419 DECLARE_COMPLETION_ONSTACK(done);
420 struct task_struct *task;
421 struct kthread_create_info *create = kmalloc(sizeof(*create),
422 GFP_KERNEL);
423
424 if (!create)
425 return ERR_PTR(-ENOMEM);
426 create->threadfn = threadfn;
427 create->data = data;
428 create->node = node;
429 create->done = &done;
430
431 spin_lock(&kthread_create_lock);
432 list_add_tail(&create->list, &kthread_create_list);
433 spin_unlock(&kthread_create_lock);
434
435 wake_up_process(kthreadd_task);
436 /*
437 * Wait for completion in killable state, for I might be chosen by
438 * the OOM killer while kthreadd is trying to allocate memory for
439 * new kernel thread.
440 */
441 if (unlikely(wait_for_completion_killable(&done))) {
442 /*
443 * If I was killed by a fatal signal before kthreadd (or new
444 * kernel thread) calls complete(), leave the cleanup of this
445 * structure to that thread.
446 */
447 if (xchg(&create->done, NULL))
448 return ERR_PTR(-EINTR);
449 /*
450 * kthreadd (or new kernel thread) will call complete()
451 * shortly.
452 */
453 wait_for_completion(&done);
454 }
455 task = create->result;
456 if (!IS_ERR(task)) {
457 char name[TASK_COMM_LEN];
458 va_list aq;
459 int len;
460
461 /*
462 * task is already visible to other tasks, so updating
463 * COMM must be protected.
464 */
465 va_copy(aq, args);
466 len = vsnprintf(name, sizeof(name), namefmt, aq);
467 va_end(aq);
468 if (len >= TASK_COMM_LEN) {
469 struct kthread *kthread = to_kthread(task);
470
471 /* leave it truncated when out of memory. */
472 kthread->full_name = kvasprintf(GFP_KERNEL, namefmt, args);
473 }
474 set_task_comm(task, name);
475 }
476 kfree(create);
477 return task;
478}
479
480/**
481 * kthread_create_on_node - create a kthread.
482 * @threadfn: the function to run until signal_pending(current).
483 * @data: data ptr for @threadfn.
484 * @node: task and thread structures for the thread are allocated on this node
485 * @namefmt: printf-style name for the thread.
486 *
487 * Description: This helper function creates and names a kernel
488 * thread. The thread will be stopped: use wake_up_process() to start
489 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
490 * is affine to all CPUs.
491 *
492 * If thread is going to be bound on a particular cpu, give its node
493 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
494 * When woken, the thread will run @threadfn() with @data as its
495 * argument. @threadfn() can either return directly if it is a
496 * standalone thread for which no one will call kthread_stop(), or
497 * return when 'kthread_should_stop()' is true (which means
498 * kthread_stop() has been called). The return value should be zero
499 * or a negative error number; it will be passed to kthread_stop().
500 *
501 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
502 */
503struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
504 void *data, int node,
505 const char namefmt[],
506 ...)
507{
508 struct task_struct *task;
509 va_list args;
510
511 va_start(args, namefmt);
512 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
513 va_end(args);
514
515 return task;
516}
517EXPORT_SYMBOL(kthread_create_on_node);
518
519static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, unsigned int state)
520{
521 unsigned long flags;
522
523 if (!wait_task_inactive(p, state)) {
524 WARN_ON(1);
525 return;
526 }
527
528 /* It's safe because the task is inactive. */
529 raw_spin_lock_irqsave(&p->pi_lock, flags);
530 do_set_cpus_allowed(p, mask);
531 p->flags |= PF_NO_SETAFFINITY;
532 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
533}
534
535static void __kthread_bind(struct task_struct *p, unsigned int cpu, unsigned int state)
536{
537 __kthread_bind_mask(p, cpumask_of(cpu), state);
538}
539
540void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
541{
542 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
543}
544
545/**
546 * kthread_bind - bind a just-created kthread to a cpu.
547 * @p: thread created by kthread_create().
548 * @cpu: cpu (might not be online, must be possible) for @k to run on.
549 *
550 * Description: This function is equivalent to set_cpus_allowed(),
551 * except that @cpu doesn't need to be online, and the thread must be
552 * stopped (i.e., just returned from kthread_create()).
553 */
554void kthread_bind(struct task_struct *p, unsigned int cpu)
555{
556 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
557}
558EXPORT_SYMBOL(kthread_bind);
559
560/**
561 * kthread_create_on_cpu - Create a cpu bound kthread
562 * @threadfn: the function to run until signal_pending(current).
563 * @data: data ptr for @threadfn.
564 * @cpu: The cpu on which the thread should be bound,
565 * @namefmt: printf-style name for the thread. Format is restricted
566 * to "name.*%u". Code fills in cpu number.
567 *
568 * Description: This helper function creates and names a kernel thread
569 */
570struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
571 void *data, unsigned int cpu,
572 const char *namefmt)
573{
574 struct task_struct *p;
575
576 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
577 cpu);
578 if (IS_ERR(p))
579 return p;
580 kthread_bind(p, cpu);
581 /* CPU hotplug need to bind once again when unparking the thread. */
582 to_kthread(p)->cpu = cpu;
583 return p;
584}
585EXPORT_SYMBOL(kthread_create_on_cpu);
586
587void kthread_set_per_cpu(struct task_struct *k, int cpu)
588{
589 struct kthread *kthread = to_kthread(k);
590 if (!kthread)
591 return;
592
593 WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY));
594
595 if (cpu < 0) {
596 clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
597 return;
598 }
599
600 kthread->cpu = cpu;
601 set_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
602}
603
604bool kthread_is_per_cpu(struct task_struct *p)
605{
606 struct kthread *kthread = __to_kthread(p);
607 if (!kthread)
608 return false;
609
610 return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
611}
612
613/**
614 * kthread_unpark - unpark a thread created by kthread_create().
615 * @k: thread created by kthread_create().
616 *
617 * Sets kthread_should_park() for @k to return false, wakes it, and
618 * waits for it to return. If the thread is marked percpu then its
619 * bound to the cpu again.
620 */
621void kthread_unpark(struct task_struct *k)
622{
623 struct kthread *kthread = to_kthread(k);
624
625 /*
626 * Newly created kthread was parked when the CPU was offline.
627 * The binding was lost and we need to set it again.
628 */
629 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
630 __kthread_bind(k, kthread->cpu, TASK_PARKED);
631
632 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
633 /*
634 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
635 */
636 wake_up_state(k, TASK_PARKED);
637}
638EXPORT_SYMBOL_GPL(kthread_unpark);
639
640/**
641 * kthread_park - park a thread created by kthread_create().
642 * @k: thread created by kthread_create().
643 *
644 * Sets kthread_should_park() for @k to return true, wakes it, and
645 * waits for it to return. This can also be called after kthread_create()
646 * instead of calling wake_up_process(): the thread will park without
647 * calling threadfn().
648 *
649 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
650 * If called by the kthread itself just the park bit is set.
651 */
652int kthread_park(struct task_struct *k)
653{
654 struct kthread *kthread = to_kthread(k);
655
656 if (WARN_ON(k->flags & PF_EXITING))
657 return -ENOSYS;
658
659 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
660 return -EBUSY;
661
662 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
663 if (k != current) {
664 wake_up_process(k);
665 /*
666 * Wait for __kthread_parkme() to complete(), this means we
667 * _will_ have TASK_PARKED and are about to call schedule().
668 */
669 wait_for_completion(&kthread->parked);
670 /*
671 * Now wait for that schedule() to complete and the task to
672 * get scheduled out.
673 */
674 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
675 }
676
677 return 0;
678}
679EXPORT_SYMBOL_GPL(kthread_park);
680
681/**
682 * kthread_stop - stop a thread created by kthread_create().
683 * @k: thread created by kthread_create().
684 *
685 * Sets kthread_should_stop() for @k to return true, wakes it, and
686 * waits for it to exit. This can also be called after kthread_create()
687 * instead of calling wake_up_process(): the thread will exit without
688 * calling threadfn().
689 *
690 * If threadfn() may call kthread_exit() itself, the caller must ensure
691 * task_struct can't go away.
692 *
693 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
694 * was never called.
695 */
696int kthread_stop(struct task_struct *k)
697{
698 struct kthread *kthread;
699 int ret;
700
701 trace_sched_kthread_stop(k);
702
703 get_task_struct(k);
704 kthread = to_kthread(k);
705 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
706 kthread_unpark(k);
707 set_tsk_thread_flag(k, TIF_NOTIFY_SIGNAL);
708 wake_up_process(k);
709 wait_for_completion(&kthread->exited);
710 ret = kthread->result;
711 put_task_struct(k);
712
713 trace_sched_kthread_stop_ret(ret);
714 return ret;
715}
716EXPORT_SYMBOL(kthread_stop);
717
718int kthreadd(void *unused)
719{
720 struct task_struct *tsk = current;
721
722 /* Setup a clean context for our children to inherit. */
723 set_task_comm(tsk, "kthreadd");
724 ignore_signals(tsk);
725 set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_TYPE_KTHREAD));
726 set_mems_allowed(node_states[N_MEMORY]);
727
728 current->flags |= PF_NOFREEZE;
729 cgroup_init_kthreadd();
730
731 for (;;) {
732 set_current_state(TASK_INTERRUPTIBLE);
733 if (list_empty(&kthread_create_list))
734 schedule();
735 __set_current_state(TASK_RUNNING);
736
737 spin_lock(&kthread_create_lock);
738 while (!list_empty(&kthread_create_list)) {
739 struct kthread_create_info *create;
740
741 create = list_entry(kthread_create_list.next,
742 struct kthread_create_info, list);
743 list_del_init(&create->list);
744 spin_unlock(&kthread_create_lock);
745
746 create_kthread(create);
747
748 spin_lock(&kthread_create_lock);
749 }
750 spin_unlock(&kthread_create_lock);
751 }
752
753 return 0;
754}
755
756void __kthread_init_worker(struct kthread_worker *worker,
757 const char *name,
758 struct lock_class_key *key)
759{
760 memset(worker, 0, sizeof(struct kthread_worker));
761 raw_spin_lock_init(&worker->lock);
762 lockdep_set_class_and_name(&worker->lock, key, name);
763 INIT_LIST_HEAD(&worker->work_list);
764 INIT_LIST_HEAD(&worker->delayed_work_list);
765}
766EXPORT_SYMBOL_GPL(__kthread_init_worker);
767
768/**
769 * kthread_worker_fn - kthread function to process kthread_worker
770 * @worker_ptr: pointer to initialized kthread_worker
771 *
772 * This function implements the main cycle of kthread worker. It processes
773 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
774 * is empty.
775 *
776 * The works are not allowed to keep any locks, disable preemption or interrupts
777 * when they finish. There is defined a safe point for freezing when one work
778 * finishes and before a new one is started.
779 *
780 * Also the works must not be handled by more than one worker at the same time,
781 * see also kthread_queue_work().
782 */
783int kthread_worker_fn(void *worker_ptr)
784{
785 struct kthread_worker *worker = worker_ptr;
786 struct kthread_work *work;
787
788 /*
789 * FIXME: Update the check and remove the assignment when all kthread
790 * worker users are created using kthread_create_worker*() functions.
791 */
792 WARN_ON(worker->task && worker->task != current);
793 worker->task = current;
794
795 if (worker->flags & KTW_FREEZABLE)
796 set_freezable();
797
798repeat:
799 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
800
801 if (kthread_should_stop()) {
802 __set_current_state(TASK_RUNNING);
803 raw_spin_lock_irq(&worker->lock);
804 worker->task = NULL;
805 raw_spin_unlock_irq(&worker->lock);
806 return 0;
807 }
808
809 work = NULL;
810 raw_spin_lock_irq(&worker->lock);
811 if (!list_empty(&worker->work_list)) {
812 work = list_first_entry(&worker->work_list,
813 struct kthread_work, node);
814 list_del_init(&work->node);
815 }
816 worker->current_work = work;
817 raw_spin_unlock_irq(&worker->lock);
818
819 if (work) {
820 kthread_work_func_t func = work->func;
821 __set_current_state(TASK_RUNNING);
822 trace_sched_kthread_work_execute_start(work);
823 work->func(work);
824 /*
825 * Avoid dereferencing work after this point. The trace
826 * event only cares about the address.
827 */
828 trace_sched_kthread_work_execute_end(work, func);
829 } else if (!freezing(current))
830 schedule();
831
832 try_to_freeze();
833 cond_resched();
834 goto repeat;
835}
836EXPORT_SYMBOL_GPL(kthread_worker_fn);
837
838static __printf(3, 0) struct kthread_worker *
839__kthread_create_worker(int cpu, unsigned int flags,
840 const char namefmt[], va_list args)
841{
842 struct kthread_worker *worker;
843 struct task_struct *task;
844 int node = NUMA_NO_NODE;
845
846 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
847 if (!worker)
848 return ERR_PTR(-ENOMEM);
849
850 kthread_init_worker(worker);
851
852 if (cpu >= 0)
853 node = cpu_to_node(cpu);
854
855 task = __kthread_create_on_node(kthread_worker_fn, worker,
856 node, namefmt, args);
857 if (IS_ERR(task))
858 goto fail_task;
859
860 if (cpu >= 0)
861 kthread_bind(task, cpu);
862
863 worker->flags = flags;
864 worker->task = task;
865 wake_up_process(task);
866 return worker;
867
868fail_task:
869 kfree(worker);
870 return ERR_CAST(task);
871}
872
873/**
874 * kthread_create_worker - create a kthread worker
875 * @flags: flags modifying the default behavior of the worker
876 * @namefmt: printf-style name for the kthread worker (task).
877 *
878 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
879 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
880 * when the caller was killed by a fatal signal.
881 */
882struct kthread_worker *
883kthread_create_worker(unsigned int flags, const char namefmt[], ...)
884{
885 struct kthread_worker *worker;
886 va_list args;
887
888 va_start(args, namefmt);
889 worker = __kthread_create_worker(-1, flags, namefmt, args);
890 va_end(args);
891
892 return worker;
893}
894EXPORT_SYMBOL(kthread_create_worker);
895
896/**
897 * kthread_create_worker_on_cpu - create a kthread worker and bind it
898 * to a given CPU and the associated NUMA node.
899 * @cpu: CPU number
900 * @flags: flags modifying the default behavior of the worker
901 * @namefmt: printf-style name for the kthread worker (task).
902 *
903 * Use a valid CPU number if you want to bind the kthread worker
904 * to the given CPU and the associated NUMA node.
905 *
906 * A good practice is to add the cpu number also into the worker name.
907 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
908 *
909 * CPU hotplug:
910 * The kthread worker API is simple and generic. It just provides a way
911 * to create, use, and destroy workers.
912 *
913 * It is up to the API user how to handle CPU hotplug. They have to decide
914 * how to handle pending work items, prevent queuing new ones, and
915 * restore the functionality when the CPU goes off and on. There are a
916 * few catches:
917 *
918 * - CPU affinity gets lost when it is scheduled on an offline CPU.
919 *
920 * - The worker might not exist when the CPU was off when the user
921 * created the workers.
922 *
923 * Good practice is to implement two CPU hotplug callbacks and to
924 * destroy/create the worker when the CPU goes down/up.
925 *
926 * Return:
927 * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
928 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
929 * when the caller was killed by a fatal signal.
930 */
931struct kthread_worker *
932kthread_create_worker_on_cpu(int cpu, unsigned int flags,
933 const char namefmt[], ...)
934{
935 struct kthread_worker *worker;
936 va_list args;
937
938 va_start(args, namefmt);
939 worker = __kthread_create_worker(cpu, flags, namefmt, args);
940 va_end(args);
941
942 return worker;
943}
944EXPORT_SYMBOL(kthread_create_worker_on_cpu);
945
946/*
947 * Returns true when the work could not be queued at the moment.
948 * It happens when it is already pending in a worker list
949 * or when it is being cancelled.
950 */
951static inline bool queuing_blocked(struct kthread_worker *worker,
952 struct kthread_work *work)
953{
954 lockdep_assert_held(&worker->lock);
955
956 return !list_empty(&work->node) || work->canceling;
957}
958
959static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
960 struct kthread_work *work)
961{
962 lockdep_assert_held(&worker->lock);
963 WARN_ON_ONCE(!list_empty(&work->node));
964 /* Do not use a work with >1 worker, see kthread_queue_work() */
965 WARN_ON_ONCE(work->worker && work->worker != worker);
966}
967
968/* insert @work before @pos in @worker */
969static void kthread_insert_work(struct kthread_worker *worker,
970 struct kthread_work *work,
971 struct list_head *pos)
972{
973 kthread_insert_work_sanity_check(worker, work);
974
975 trace_sched_kthread_work_queue_work(worker, work);
976
977 list_add_tail(&work->node, pos);
978 work->worker = worker;
979 if (!worker->current_work && likely(worker->task))
980 wake_up_process(worker->task);
981}
982
983/**
984 * kthread_queue_work - queue a kthread_work
985 * @worker: target kthread_worker
986 * @work: kthread_work to queue
987 *
988 * Queue @work to work processor @task for async execution. @task
989 * must have been created with kthread_worker_create(). Returns %true
990 * if @work was successfully queued, %false if it was already pending.
991 *
992 * Reinitialize the work if it needs to be used by another worker.
993 * For example, when the worker was stopped and started again.
994 */
995bool kthread_queue_work(struct kthread_worker *worker,
996 struct kthread_work *work)
997{
998 bool ret = false;
999 unsigned long flags;
1000
1001 raw_spin_lock_irqsave(&worker->lock, flags);
1002 if (!queuing_blocked(worker, work)) {
1003 kthread_insert_work(worker, work, &worker->work_list);
1004 ret = true;
1005 }
1006 raw_spin_unlock_irqrestore(&worker->lock, flags);
1007 return ret;
1008}
1009EXPORT_SYMBOL_GPL(kthread_queue_work);
1010
1011/**
1012 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
1013 * delayed work when the timer expires.
1014 * @t: pointer to the expired timer
1015 *
1016 * The format of the function is defined by struct timer_list.
1017 * It should have been called from irqsafe timer with irq already off.
1018 */
1019void kthread_delayed_work_timer_fn(struct timer_list *t)
1020{
1021 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
1022 struct kthread_work *work = &dwork->work;
1023 struct kthread_worker *worker = work->worker;
1024 unsigned long flags;
1025
1026 /*
1027 * This might happen when a pending work is reinitialized.
1028 * It means that it is used a wrong way.
1029 */
1030 if (WARN_ON_ONCE(!worker))
1031 return;
1032
1033 raw_spin_lock_irqsave(&worker->lock, flags);
1034 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1035 WARN_ON_ONCE(work->worker != worker);
1036
1037 /* Move the work from worker->delayed_work_list. */
1038 WARN_ON_ONCE(list_empty(&work->node));
1039 list_del_init(&work->node);
1040 if (!work->canceling)
1041 kthread_insert_work(worker, work, &worker->work_list);
1042
1043 raw_spin_unlock_irqrestore(&worker->lock, flags);
1044}
1045EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
1046
1047static void __kthread_queue_delayed_work(struct kthread_worker *worker,
1048 struct kthread_delayed_work *dwork,
1049 unsigned long delay)
1050{
1051 struct timer_list *timer = &dwork->timer;
1052 struct kthread_work *work = &dwork->work;
1053
1054 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
1055
1056 /*
1057 * If @delay is 0, queue @dwork->work immediately. This is for
1058 * both optimization and correctness. The earliest @timer can
1059 * expire is on the closest next tick and delayed_work users depend
1060 * on that there's no such delay when @delay is 0.
1061 */
1062 if (!delay) {
1063 kthread_insert_work(worker, work, &worker->work_list);
1064 return;
1065 }
1066
1067 /* Be paranoid and try to detect possible races already now. */
1068 kthread_insert_work_sanity_check(worker, work);
1069
1070 list_add(&work->node, &worker->delayed_work_list);
1071 work->worker = worker;
1072 timer->expires = jiffies + delay;
1073 add_timer(timer);
1074}
1075
1076/**
1077 * kthread_queue_delayed_work - queue the associated kthread work
1078 * after a delay.
1079 * @worker: target kthread_worker
1080 * @dwork: kthread_delayed_work to queue
1081 * @delay: number of jiffies to wait before queuing
1082 *
1083 * If the work has not been pending it starts a timer that will queue
1084 * the work after the given @delay. If @delay is zero, it queues the
1085 * work immediately.
1086 *
1087 * Return: %false if the @work has already been pending. It means that
1088 * either the timer was running or the work was queued. It returns %true
1089 * otherwise.
1090 */
1091bool kthread_queue_delayed_work(struct kthread_worker *worker,
1092 struct kthread_delayed_work *dwork,
1093 unsigned long delay)
1094{
1095 struct kthread_work *work = &dwork->work;
1096 unsigned long flags;
1097 bool ret = false;
1098
1099 raw_spin_lock_irqsave(&worker->lock, flags);
1100
1101 if (!queuing_blocked(worker, work)) {
1102 __kthread_queue_delayed_work(worker, dwork, delay);
1103 ret = true;
1104 }
1105
1106 raw_spin_unlock_irqrestore(&worker->lock, flags);
1107 return ret;
1108}
1109EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
1110
1111struct kthread_flush_work {
1112 struct kthread_work work;
1113 struct completion done;
1114};
1115
1116static void kthread_flush_work_fn(struct kthread_work *work)
1117{
1118 struct kthread_flush_work *fwork =
1119 container_of(work, struct kthread_flush_work, work);
1120 complete(&fwork->done);
1121}
1122
1123/**
1124 * kthread_flush_work - flush a kthread_work
1125 * @work: work to flush
1126 *
1127 * If @work is queued or executing, wait for it to finish execution.
1128 */
1129void kthread_flush_work(struct kthread_work *work)
1130{
1131 struct kthread_flush_work fwork = {
1132 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1133 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1134 };
1135 struct kthread_worker *worker;
1136 bool noop = false;
1137
1138 worker = work->worker;
1139 if (!worker)
1140 return;
1141
1142 raw_spin_lock_irq(&worker->lock);
1143 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1144 WARN_ON_ONCE(work->worker != worker);
1145
1146 if (!list_empty(&work->node))
1147 kthread_insert_work(worker, &fwork.work, work->node.next);
1148 else if (worker->current_work == work)
1149 kthread_insert_work(worker, &fwork.work,
1150 worker->work_list.next);
1151 else
1152 noop = true;
1153
1154 raw_spin_unlock_irq(&worker->lock);
1155
1156 if (!noop)
1157 wait_for_completion(&fwork.done);
1158}
1159EXPORT_SYMBOL_GPL(kthread_flush_work);
1160
1161/*
1162 * Make sure that the timer is neither set nor running and could
1163 * not manipulate the work list_head any longer.
1164 *
1165 * The function is called under worker->lock. The lock is temporary
1166 * released but the timer can't be set again in the meantime.
1167 */
1168static void kthread_cancel_delayed_work_timer(struct kthread_work *work,
1169 unsigned long *flags)
1170{
1171 struct kthread_delayed_work *dwork =
1172 container_of(work, struct kthread_delayed_work, work);
1173 struct kthread_worker *worker = work->worker;
1174
1175 /*
1176 * del_timer_sync() must be called to make sure that the timer
1177 * callback is not running. The lock must be temporary released
1178 * to avoid a deadlock with the callback. In the meantime,
1179 * any queuing is blocked by setting the canceling counter.
1180 */
1181 work->canceling++;
1182 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1183 del_timer_sync(&dwork->timer);
1184 raw_spin_lock_irqsave(&worker->lock, *flags);
1185 work->canceling--;
1186}
1187
1188/*
1189 * This function removes the work from the worker queue.
1190 *
1191 * It is called under worker->lock. The caller must make sure that
1192 * the timer used by delayed work is not running, e.g. by calling
1193 * kthread_cancel_delayed_work_timer().
1194 *
1195 * The work might still be in use when this function finishes. See the
1196 * current_work proceed by the worker.
1197 *
1198 * Return: %true if @work was pending and successfully canceled,
1199 * %false if @work was not pending
1200 */
1201static bool __kthread_cancel_work(struct kthread_work *work)
1202{
1203 /*
1204 * Try to remove the work from a worker list. It might either
1205 * be from worker->work_list or from worker->delayed_work_list.
1206 */
1207 if (!list_empty(&work->node)) {
1208 list_del_init(&work->node);
1209 return true;
1210 }
1211
1212 return false;
1213}
1214
1215/**
1216 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1217 * @worker: kthread worker to use
1218 * @dwork: kthread delayed work to queue
1219 * @delay: number of jiffies to wait before queuing
1220 *
1221 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1222 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1223 * @work is guaranteed to be queued immediately.
1224 *
1225 * Return: %false if @dwork was idle and queued, %true otherwise.
1226 *
1227 * A special case is when the work is being canceled in parallel.
1228 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1229 * or yet another kthread_mod_delayed_work() call. We let the other command
1230 * win and return %true here. The return value can be used for reference
1231 * counting and the number of queued works stays the same. Anyway, the caller
1232 * is supposed to synchronize these operations a reasonable way.
1233 *
1234 * This function is safe to call from any context including IRQ handler.
1235 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1236 * for details.
1237 */
1238bool kthread_mod_delayed_work(struct kthread_worker *worker,
1239 struct kthread_delayed_work *dwork,
1240 unsigned long delay)
1241{
1242 struct kthread_work *work = &dwork->work;
1243 unsigned long flags;
1244 int ret;
1245
1246 raw_spin_lock_irqsave(&worker->lock, flags);
1247
1248 /* Do not bother with canceling when never queued. */
1249 if (!work->worker) {
1250 ret = false;
1251 goto fast_queue;
1252 }
1253
1254 /* Work must not be used with >1 worker, see kthread_queue_work() */
1255 WARN_ON_ONCE(work->worker != worker);
1256
1257 /*
1258 * Temporary cancel the work but do not fight with another command
1259 * that is canceling the work as well.
1260 *
1261 * It is a bit tricky because of possible races with another
1262 * mod_delayed_work() and cancel_delayed_work() callers.
1263 *
1264 * The timer must be canceled first because worker->lock is released
1265 * when doing so. But the work can be removed from the queue (list)
1266 * only when it can be queued again so that the return value can
1267 * be used for reference counting.
1268 */
1269 kthread_cancel_delayed_work_timer(work, &flags);
1270 if (work->canceling) {
1271 /* The number of works in the queue does not change. */
1272 ret = true;
1273 goto out;
1274 }
1275 ret = __kthread_cancel_work(work);
1276
1277fast_queue:
1278 __kthread_queue_delayed_work(worker, dwork, delay);
1279out:
1280 raw_spin_unlock_irqrestore(&worker->lock, flags);
1281 return ret;
1282}
1283EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1284
1285static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1286{
1287 struct kthread_worker *worker = work->worker;
1288 unsigned long flags;
1289 int ret = false;
1290
1291 if (!worker)
1292 goto out;
1293
1294 raw_spin_lock_irqsave(&worker->lock, flags);
1295 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1296 WARN_ON_ONCE(work->worker != worker);
1297
1298 if (is_dwork)
1299 kthread_cancel_delayed_work_timer(work, &flags);
1300
1301 ret = __kthread_cancel_work(work);
1302
1303 if (worker->current_work != work)
1304 goto out_fast;
1305
1306 /*
1307 * The work is in progress and we need to wait with the lock released.
1308 * In the meantime, block any queuing by setting the canceling counter.
1309 */
1310 work->canceling++;
1311 raw_spin_unlock_irqrestore(&worker->lock, flags);
1312 kthread_flush_work(work);
1313 raw_spin_lock_irqsave(&worker->lock, flags);
1314 work->canceling--;
1315
1316out_fast:
1317 raw_spin_unlock_irqrestore(&worker->lock, flags);
1318out:
1319 return ret;
1320}
1321
1322/**
1323 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1324 * @work: the kthread work to cancel
1325 *
1326 * Cancel @work and wait for its execution to finish. This function
1327 * can be used even if the work re-queues itself. On return from this
1328 * function, @work is guaranteed to be not pending or executing on any CPU.
1329 *
1330 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1331 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1332 *
1333 * The caller must ensure that the worker on which @work was last
1334 * queued can't be destroyed before this function returns.
1335 *
1336 * Return: %true if @work was pending, %false otherwise.
1337 */
1338bool kthread_cancel_work_sync(struct kthread_work *work)
1339{
1340 return __kthread_cancel_work_sync(work, false);
1341}
1342EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1343
1344/**
1345 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1346 * wait for it to finish.
1347 * @dwork: the kthread delayed work to cancel
1348 *
1349 * This is kthread_cancel_work_sync() for delayed works.
1350 *
1351 * Return: %true if @dwork was pending, %false otherwise.
1352 */
1353bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1354{
1355 return __kthread_cancel_work_sync(&dwork->work, true);
1356}
1357EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1358
1359/**
1360 * kthread_flush_worker - flush all current works on a kthread_worker
1361 * @worker: worker to flush
1362 *
1363 * Wait until all currently executing or pending works on @worker are
1364 * finished.
1365 */
1366void kthread_flush_worker(struct kthread_worker *worker)
1367{
1368 struct kthread_flush_work fwork = {
1369 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1370 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1371 };
1372
1373 kthread_queue_work(worker, &fwork.work);
1374 wait_for_completion(&fwork.done);
1375}
1376EXPORT_SYMBOL_GPL(kthread_flush_worker);
1377
1378/**
1379 * kthread_destroy_worker - destroy a kthread worker
1380 * @worker: worker to be destroyed
1381 *
1382 * Flush and destroy @worker. The simple flush is enough because the kthread
1383 * worker API is used only in trivial scenarios. There are no multi-step state
1384 * machines needed.
1385 */
1386void kthread_destroy_worker(struct kthread_worker *worker)
1387{
1388 struct task_struct *task;
1389
1390 task = worker->task;
1391 if (WARN_ON(!task))
1392 return;
1393
1394 kthread_flush_worker(worker);
1395 kthread_stop(task);
1396 WARN_ON(!list_empty(&worker->work_list));
1397 kfree(worker);
1398}
1399EXPORT_SYMBOL(kthread_destroy_worker);
1400
1401/**
1402 * kthread_use_mm - make the calling kthread operate on an address space
1403 * @mm: address space to operate on
1404 */
1405void kthread_use_mm(struct mm_struct *mm)
1406{
1407 struct mm_struct *active_mm;
1408 struct task_struct *tsk = current;
1409
1410 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1411 WARN_ON_ONCE(tsk->mm);
1412
1413 task_lock(tsk);
1414 /* Hold off tlb flush IPIs while switching mm's */
1415 local_irq_disable();
1416 active_mm = tsk->active_mm;
1417 if (active_mm != mm) {
1418 mmgrab(mm);
1419 tsk->active_mm = mm;
1420 }
1421 tsk->mm = mm;
1422 membarrier_update_current_mm(mm);
1423 switch_mm_irqs_off(active_mm, mm, tsk);
1424 local_irq_enable();
1425 task_unlock(tsk);
1426#ifdef finish_arch_post_lock_switch
1427 finish_arch_post_lock_switch();
1428#endif
1429
1430 /*
1431 * When a kthread starts operating on an address space, the loop
1432 * in membarrier_{private,global}_expedited() may not observe
1433 * that tsk->mm, and not issue an IPI. Membarrier requires a
1434 * memory barrier after storing to tsk->mm, before accessing
1435 * user-space memory. A full memory barrier for membarrier
1436 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1437 * mmdrop(), or explicitly with smp_mb().
1438 */
1439 if (active_mm != mm)
1440 mmdrop(active_mm);
1441 else
1442 smp_mb();
1443}
1444EXPORT_SYMBOL_GPL(kthread_use_mm);
1445
1446/**
1447 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1448 * @mm: address space to operate on
1449 */
1450void kthread_unuse_mm(struct mm_struct *mm)
1451{
1452 struct task_struct *tsk = current;
1453
1454 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1455 WARN_ON_ONCE(!tsk->mm);
1456
1457 task_lock(tsk);
1458 /*
1459 * When a kthread stops operating on an address space, the loop
1460 * in membarrier_{private,global}_expedited() may not observe
1461 * that tsk->mm, and not issue an IPI. Membarrier requires a
1462 * memory barrier after accessing user-space memory, before
1463 * clearing tsk->mm.
1464 */
1465 smp_mb__after_spinlock();
1466 sync_mm_rss(mm);
1467 local_irq_disable();
1468 tsk->mm = NULL;
1469 membarrier_update_current_mm(NULL);
1470 /* active_mm is still 'mm' */
1471 enter_lazy_tlb(mm, tsk);
1472 local_irq_enable();
1473 task_unlock(tsk);
1474}
1475EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1476
1477#ifdef CONFIG_BLK_CGROUP
1478/**
1479 * kthread_associate_blkcg - associate blkcg to current kthread
1480 * @css: the cgroup info
1481 *
1482 * Current thread must be a kthread. The thread is running jobs on behalf of
1483 * other threads. In some cases, we expect the jobs attach cgroup info of
1484 * original threads instead of that of current thread. This function stores
1485 * original thread's cgroup info in current kthread context for later
1486 * retrieval.
1487 */
1488void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1489{
1490 struct kthread *kthread;
1491
1492 if (!(current->flags & PF_KTHREAD))
1493 return;
1494 kthread = to_kthread(current);
1495 if (!kthread)
1496 return;
1497
1498 if (kthread->blkcg_css) {
1499 css_put(kthread->blkcg_css);
1500 kthread->blkcg_css = NULL;
1501 }
1502 if (css) {
1503 css_get(css);
1504 kthread->blkcg_css = css;
1505 }
1506}
1507EXPORT_SYMBOL(kthread_associate_blkcg);
1508
1509/**
1510 * kthread_blkcg - get associated blkcg css of current kthread
1511 *
1512 * Current thread must be a kthread.
1513 */
1514struct cgroup_subsys_state *kthread_blkcg(void)
1515{
1516 struct kthread *kthread;
1517
1518 if (current->flags & PF_KTHREAD) {
1519 kthread = to_kthread(current);
1520 if (kthread)
1521 return kthread->blkcg_css;
1522 }
1523 return NULL;
1524}
1525#endif
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 set_current_state(TASK_PARKED);
181 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
182 break;
183 schedule();
184 }
185 __set_current_state(TASK_RUNNING);
186}
187
188void kthread_parkme(void)
189{
190 __kthread_parkme(to_kthread(current));
191}
192EXPORT_SYMBOL_GPL(kthread_parkme);
193
194void kthread_park_complete(struct task_struct *k)
195{
196 complete_all(&to_kthread(k)->parked);
197}
198
199static int kthread(void *_create)
200{
201 /* Copy data: it's on kthread's stack */
202 struct kthread_create_info *create = _create;
203 int (*threadfn)(void *data) = create->threadfn;
204 void *data = create->data;
205 struct completion *done;
206 struct kthread *self;
207 int ret;
208
209 self = kzalloc(sizeof(*self), GFP_KERNEL);
210 set_kthread_struct(self);
211
212 /* If user was SIGKILLed, I release the structure. */
213 done = xchg(&create->done, NULL);
214 if (!done) {
215 kfree(create);
216 do_exit(-EINTR);
217 }
218
219 if (!self) {
220 create->result = ERR_PTR(-ENOMEM);
221 complete(done);
222 do_exit(-ENOMEM);
223 }
224
225 self->data = data;
226 init_completion(&self->exited);
227 init_completion(&self->parked);
228 current->vfork_done = &self->exited;
229
230 /* OK, tell user we're spawned, wait for stop or wakeup */
231 __set_current_state(TASK_UNINTERRUPTIBLE);
232 create->result = current;
233 complete(done);
234 schedule();
235
236 ret = -EINTR;
237 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
238 cgroup_kthread_ready();
239 __kthread_parkme(self);
240 ret = threadfn(data);
241 }
242 do_exit(ret);
243}
244
245/* called from do_fork() to get node information for about to be created task */
246int tsk_fork_get_node(struct task_struct *tsk)
247{
248#ifdef CONFIG_NUMA
249 if (tsk == kthreadd_task)
250 return tsk->pref_node_fork;
251#endif
252 return NUMA_NO_NODE;
253}
254
255static void create_kthread(struct kthread_create_info *create)
256{
257 int pid;
258
259#ifdef CONFIG_NUMA
260 current->pref_node_fork = create->node;
261#endif
262 /* We want our own signal handler (we take no signals by default). */
263 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
264 if (pid < 0) {
265 /* If user was SIGKILLed, I release the structure. */
266 struct completion *done = xchg(&create->done, NULL);
267
268 if (!done) {
269 kfree(create);
270 return;
271 }
272 create->result = ERR_PTR(pid);
273 complete(done);
274 }
275}
276
277static __printf(4, 0)
278struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
279 void *data, int node,
280 const char namefmt[],
281 va_list args)
282{
283 DECLARE_COMPLETION_ONSTACK(done);
284 struct task_struct *task;
285 struct kthread_create_info *create = kmalloc(sizeof(*create),
286 GFP_KERNEL);
287
288 if (!create)
289 return ERR_PTR(-ENOMEM);
290 create->threadfn = threadfn;
291 create->data = data;
292 create->node = node;
293 create->done = &done;
294
295 spin_lock(&kthread_create_lock);
296 list_add_tail(&create->list, &kthread_create_list);
297 spin_unlock(&kthread_create_lock);
298
299 wake_up_process(kthreadd_task);
300 /*
301 * Wait for completion in killable state, for I might be chosen by
302 * the OOM killer while kthreadd is trying to allocate memory for
303 * new kernel thread.
304 */
305 if (unlikely(wait_for_completion_killable(&done))) {
306 /*
307 * If I was SIGKILLed before kthreadd (or new kernel thread)
308 * calls complete(), leave the cleanup of this structure to
309 * that thread.
310 */
311 if (xchg(&create->done, NULL))
312 return ERR_PTR(-EINTR);
313 /*
314 * kthreadd (or new kernel thread) will call complete()
315 * shortly.
316 */
317 wait_for_completion(&done);
318 }
319 task = create->result;
320 if (!IS_ERR(task)) {
321 static const struct sched_param param = { .sched_priority = 0 };
322
323 vsnprintf(task->comm, sizeof(task->comm), namefmt, args);
324 /*
325 * root may have changed our (kthreadd's) priority or CPU mask.
326 * The kernel thread should not inherit these properties.
327 */
328 sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m);
329 set_cpus_allowed_ptr(task, cpu_all_mask);
330 }
331 kfree(create);
332 return task;
333}
334
335/**
336 * kthread_create_on_node - create a kthread.
337 * @threadfn: the function to run until signal_pending(current).
338 * @data: data ptr for @threadfn.
339 * @node: task and thread structures for the thread are allocated on this node
340 * @namefmt: printf-style name for the thread.
341 *
342 * Description: This helper function creates and names a kernel
343 * thread. The thread will be stopped: use wake_up_process() to start
344 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
345 * is affine to all CPUs.
346 *
347 * If thread is going to be bound on a particular cpu, give its node
348 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
349 * When woken, the thread will run @threadfn() with @data as its
350 * argument. @threadfn() can either call do_exit() directly if it is a
351 * standalone thread for which no one will call kthread_stop(), or
352 * return when 'kthread_should_stop()' is true (which means
353 * kthread_stop() has been called). The return value should be zero
354 * or a negative error number; it will be passed to kthread_stop().
355 *
356 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
357 */
358struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
359 void *data, int node,
360 const char namefmt[],
361 ...)
362{
363 struct task_struct *task;
364 va_list args;
365
366 va_start(args, namefmt);
367 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
368 va_end(args);
369
370 return task;
371}
372EXPORT_SYMBOL(kthread_create_on_node);
373
374static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
375{
376 unsigned long flags;
377
378 if (!wait_task_inactive(p, state)) {
379 WARN_ON(1);
380 return;
381 }
382
383 /* It's safe because the task is inactive. */
384 raw_spin_lock_irqsave(&p->pi_lock, flags);
385 do_set_cpus_allowed(p, mask);
386 p->flags |= PF_NO_SETAFFINITY;
387 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
388}
389
390static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
391{
392 __kthread_bind_mask(p, cpumask_of(cpu), state);
393}
394
395void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
396{
397 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
398}
399
400/**
401 * kthread_bind - bind a just-created kthread to a cpu.
402 * @p: thread created by kthread_create().
403 * @cpu: cpu (might not be online, must be possible) for @k to run on.
404 *
405 * Description: This function is equivalent to set_cpus_allowed(),
406 * except that @cpu doesn't need to be online, and the thread must be
407 * stopped (i.e., just returned from kthread_create()).
408 */
409void kthread_bind(struct task_struct *p, unsigned int cpu)
410{
411 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
412}
413EXPORT_SYMBOL(kthread_bind);
414
415/**
416 * kthread_create_on_cpu - Create a cpu bound kthread
417 * @threadfn: the function to run until signal_pending(current).
418 * @data: data ptr for @threadfn.
419 * @cpu: The cpu on which the thread should be bound,
420 * @namefmt: printf-style name for the thread. Format is restricted
421 * to "name.*%u". Code fills in cpu number.
422 *
423 * Description: This helper function creates and names a kernel thread
424 * The thread will be woken and put into park mode.
425 */
426struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
427 void *data, unsigned int cpu,
428 const char *namefmt)
429{
430 struct task_struct *p;
431
432 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
433 cpu);
434 if (IS_ERR(p))
435 return p;
436 kthread_bind(p, cpu);
437 /* CPU hotplug need to bind once again when unparking the thread. */
438 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
439 to_kthread(p)->cpu = cpu;
440 return p;
441}
442
443/**
444 * kthread_unpark - unpark a thread created by kthread_create().
445 * @k: thread created by kthread_create().
446 *
447 * Sets kthread_should_park() for @k to return false, wakes it, and
448 * waits for it to return. If the thread is marked percpu then its
449 * bound to the cpu again.
450 */
451void kthread_unpark(struct task_struct *k)
452{
453 struct kthread *kthread = to_kthread(k);
454
455 /*
456 * Newly created kthread was parked when the CPU was offline.
457 * The binding was lost and we need to set it again.
458 */
459 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
460 __kthread_bind(k, kthread->cpu, TASK_PARKED);
461
462 reinit_completion(&kthread->parked);
463 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
464 wake_up_state(k, TASK_PARKED);
465}
466EXPORT_SYMBOL_GPL(kthread_unpark);
467
468/**
469 * kthread_park - park a thread created by kthread_create().
470 * @k: thread created by kthread_create().
471 *
472 * Sets kthread_should_park() for @k to return true, wakes it, and
473 * waits for it to return. This can also be called after kthread_create()
474 * instead of calling wake_up_process(): the thread will park without
475 * calling threadfn().
476 *
477 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
478 * If called by the kthread itself just the park bit is set.
479 */
480int kthread_park(struct task_struct *k)
481{
482 struct kthread *kthread = to_kthread(k);
483
484 if (WARN_ON(k->flags & PF_EXITING))
485 return -ENOSYS;
486
487 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
488 if (k != current) {
489 wake_up_process(k);
490 wait_for_completion(&kthread->parked);
491 }
492
493 return 0;
494}
495EXPORT_SYMBOL_GPL(kthread_park);
496
497/**
498 * kthread_stop - stop a thread created by kthread_create().
499 * @k: thread created by kthread_create().
500 *
501 * Sets kthread_should_stop() for @k to return true, wakes it, and
502 * waits for it to exit. This can also be called after kthread_create()
503 * instead of calling wake_up_process(): the thread will exit without
504 * calling threadfn().
505 *
506 * If threadfn() may call do_exit() itself, the caller must ensure
507 * task_struct can't go away.
508 *
509 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
510 * was never called.
511 */
512int kthread_stop(struct task_struct *k)
513{
514 struct kthread *kthread;
515 int ret;
516
517 trace_sched_kthread_stop(k);
518
519 get_task_struct(k);
520 kthread = to_kthread(k);
521 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
522 kthread_unpark(k);
523 wake_up_process(k);
524 wait_for_completion(&kthread->exited);
525 ret = k->exit_code;
526 put_task_struct(k);
527
528 trace_sched_kthread_stop_ret(ret);
529 return ret;
530}
531EXPORT_SYMBOL(kthread_stop);
532
533int kthreadd(void *unused)
534{
535 struct task_struct *tsk = current;
536
537 /* Setup a clean context for our children to inherit. */
538 set_task_comm(tsk, "kthreadd");
539 ignore_signals(tsk);
540 set_cpus_allowed_ptr(tsk, cpu_all_mask);
541 set_mems_allowed(node_states[N_MEMORY]);
542
543 current->flags |= PF_NOFREEZE;
544 cgroup_init_kthreadd();
545
546 for (;;) {
547 set_current_state(TASK_INTERRUPTIBLE);
548 if (list_empty(&kthread_create_list))
549 schedule();
550 __set_current_state(TASK_RUNNING);
551
552 spin_lock(&kthread_create_lock);
553 while (!list_empty(&kthread_create_list)) {
554 struct kthread_create_info *create;
555
556 create = list_entry(kthread_create_list.next,
557 struct kthread_create_info, list);
558 list_del_init(&create->list);
559 spin_unlock(&kthread_create_lock);
560
561 create_kthread(create);
562
563 spin_lock(&kthread_create_lock);
564 }
565 spin_unlock(&kthread_create_lock);
566 }
567
568 return 0;
569}
570
571void __kthread_init_worker(struct kthread_worker *worker,
572 const char *name,
573 struct lock_class_key *key)
574{
575 memset(worker, 0, sizeof(struct kthread_worker));
576 spin_lock_init(&worker->lock);
577 lockdep_set_class_and_name(&worker->lock, key, name);
578 INIT_LIST_HEAD(&worker->work_list);
579 INIT_LIST_HEAD(&worker->delayed_work_list);
580}
581EXPORT_SYMBOL_GPL(__kthread_init_worker);
582
583/**
584 * kthread_worker_fn - kthread function to process kthread_worker
585 * @worker_ptr: pointer to initialized kthread_worker
586 *
587 * This function implements the main cycle of kthread worker. It processes
588 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
589 * is empty.
590 *
591 * The works are not allowed to keep any locks, disable preemption or interrupts
592 * when they finish. There is defined a safe point for freezing when one work
593 * finishes and before a new one is started.
594 *
595 * Also the works must not be handled by more than one worker at the same time,
596 * see also kthread_queue_work().
597 */
598int kthread_worker_fn(void *worker_ptr)
599{
600 struct kthread_worker *worker = worker_ptr;
601 struct kthread_work *work;
602
603 /*
604 * FIXME: Update the check and remove the assignment when all kthread
605 * worker users are created using kthread_create_worker*() functions.
606 */
607 WARN_ON(worker->task && worker->task != current);
608 worker->task = current;
609
610 if (worker->flags & KTW_FREEZABLE)
611 set_freezable();
612
613repeat:
614 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
615
616 if (kthread_should_stop()) {
617 __set_current_state(TASK_RUNNING);
618 spin_lock_irq(&worker->lock);
619 worker->task = NULL;
620 spin_unlock_irq(&worker->lock);
621 return 0;
622 }
623
624 work = NULL;
625 spin_lock_irq(&worker->lock);
626 if (!list_empty(&worker->work_list)) {
627 work = list_first_entry(&worker->work_list,
628 struct kthread_work, node);
629 list_del_init(&work->node);
630 }
631 worker->current_work = work;
632 spin_unlock_irq(&worker->lock);
633
634 if (work) {
635 __set_current_state(TASK_RUNNING);
636 work->func(work);
637 } else if (!freezing(current))
638 schedule();
639
640 try_to_freeze();
641 cond_resched();
642 goto repeat;
643}
644EXPORT_SYMBOL_GPL(kthread_worker_fn);
645
646static __printf(3, 0) struct kthread_worker *
647__kthread_create_worker(int cpu, unsigned int flags,
648 const char namefmt[], va_list args)
649{
650 struct kthread_worker *worker;
651 struct task_struct *task;
652 int node = -1;
653
654 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
655 if (!worker)
656 return ERR_PTR(-ENOMEM);
657
658 kthread_init_worker(worker);
659
660 if (cpu >= 0)
661 node = cpu_to_node(cpu);
662
663 task = __kthread_create_on_node(kthread_worker_fn, worker,
664 node, namefmt, args);
665 if (IS_ERR(task))
666 goto fail_task;
667
668 if (cpu >= 0)
669 kthread_bind(task, cpu);
670
671 worker->flags = flags;
672 worker->task = task;
673 wake_up_process(task);
674 return worker;
675
676fail_task:
677 kfree(worker);
678 return ERR_CAST(task);
679}
680
681/**
682 * kthread_create_worker - create a kthread worker
683 * @flags: flags modifying the default behavior of the worker
684 * @namefmt: printf-style name for the kthread worker (task).
685 *
686 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
687 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
688 * when the worker was SIGKILLed.
689 */
690struct kthread_worker *
691kthread_create_worker(unsigned int flags, const char namefmt[], ...)
692{
693 struct kthread_worker *worker;
694 va_list args;
695
696 va_start(args, namefmt);
697 worker = __kthread_create_worker(-1, flags, namefmt, args);
698 va_end(args);
699
700 return worker;
701}
702EXPORT_SYMBOL(kthread_create_worker);
703
704/**
705 * kthread_create_worker_on_cpu - create a kthread worker and bind it
706 * it to a given CPU and the associated NUMA node.
707 * @cpu: CPU number
708 * @flags: flags modifying the default behavior of the worker
709 * @namefmt: printf-style name for the kthread worker (task).
710 *
711 * Use a valid CPU number if you want to bind the kthread worker
712 * to the given CPU and the associated NUMA node.
713 *
714 * A good practice is to add the cpu number also into the worker name.
715 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
716 *
717 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
718 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
719 * when the worker was SIGKILLed.
720 */
721struct kthread_worker *
722kthread_create_worker_on_cpu(int cpu, unsigned int flags,
723 const char namefmt[], ...)
724{
725 struct kthread_worker *worker;
726 va_list args;
727
728 va_start(args, namefmt);
729 worker = __kthread_create_worker(cpu, flags, namefmt, args);
730 va_end(args);
731
732 return worker;
733}
734EXPORT_SYMBOL(kthread_create_worker_on_cpu);
735
736/*
737 * Returns true when the work could not be queued at the moment.
738 * It happens when it is already pending in a worker list
739 * or when it is being cancelled.
740 */
741static inline bool queuing_blocked(struct kthread_worker *worker,
742 struct kthread_work *work)
743{
744 lockdep_assert_held(&worker->lock);
745
746 return !list_empty(&work->node) || work->canceling;
747}
748
749static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
750 struct kthread_work *work)
751{
752 lockdep_assert_held(&worker->lock);
753 WARN_ON_ONCE(!list_empty(&work->node));
754 /* Do not use a work with >1 worker, see kthread_queue_work() */
755 WARN_ON_ONCE(work->worker && work->worker != worker);
756}
757
758/* insert @work before @pos in @worker */
759static void kthread_insert_work(struct kthread_worker *worker,
760 struct kthread_work *work,
761 struct list_head *pos)
762{
763 kthread_insert_work_sanity_check(worker, work);
764
765 list_add_tail(&work->node, pos);
766 work->worker = worker;
767 if (!worker->current_work && likely(worker->task))
768 wake_up_process(worker->task);
769}
770
771/**
772 * kthread_queue_work - queue a kthread_work
773 * @worker: target kthread_worker
774 * @work: kthread_work to queue
775 *
776 * Queue @work to work processor @task for async execution. @task
777 * must have been created with kthread_worker_create(). Returns %true
778 * if @work was successfully queued, %false if it was already pending.
779 *
780 * Reinitialize the work if it needs to be used by another worker.
781 * For example, when the worker was stopped and started again.
782 */
783bool kthread_queue_work(struct kthread_worker *worker,
784 struct kthread_work *work)
785{
786 bool ret = false;
787 unsigned long flags;
788
789 spin_lock_irqsave(&worker->lock, flags);
790 if (!queuing_blocked(worker, work)) {
791 kthread_insert_work(worker, work, &worker->work_list);
792 ret = true;
793 }
794 spin_unlock_irqrestore(&worker->lock, flags);
795 return ret;
796}
797EXPORT_SYMBOL_GPL(kthread_queue_work);
798
799/**
800 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
801 * delayed work when the timer expires.
802 * @t: pointer to the expired timer
803 *
804 * The format of the function is defined by struct timer_list.
805 * It should have been called from irqsafe timer with irq already off.
806 */
807void kthread_delayed_work_timer_fn(struct timer_list *t)
808{
809 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
810 struct kthread_work *work = &dwork->work;
811 struct kthread_worker *worker = work->worker;
812
813 /*
814 * This might happen when a pending work is reinitialized.
815 * It means that it is used a wrong way.
816 */
817 if (WARN_ON_ONCE(!worker))
818 return;
819
820 spin_lock(&worker->lock);
821 /* Work must not be used with >1 worker, see kthread_queue_work(). */
822 WARN_ON_ONCE(work->worker != worker);
823
824 /* Move the work from worker->delayed_work_list. */
825 WARN_ON_ONCE(list_empty(&work->node));
826 list_del_init(&work->node);
827 kthread_insert_work(worker, work, &worker->work_list);
828
829 spin_unlock(&worker->lock);
830}
831EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
832
833void __kthread_queue_delayed_work(struct kthread_worker *worker,
834 struct kthread_delayed_work *dwork,
835 unsigned long delay)
836{
837 struct timer_list *timer = &dwork->timer;
838 struct kthread_work *work = &dwork->work;
839
840 WARN_ON_ONCE(timer->function != kthread_delayed_work_timer_fn);
841
842 /*
843 * If @delay is 0, queue @dwork->work immediately. This is for
844 * both optimization and correctness. The earliest @timer can
845 * expire is on the closest next tick and delayed_work users depend
846 * on that there's no such delay when @delay is 0.
847 */
848 if (!delay) {
849 kthread_insert_work(worker, work, &worker->work_list);
850 return;
851 }
852
853 /* Be paranoid and try to detect possible races already now. */
854 kthread_insert_work_sanity_check(worker, work);
855
856 list_add(&work->node, &worker->delayed_work_list);
857 work->worker = worker;
858 timer->expires = jiffies + delay;
859 add_timer(timer);
860}
861
862/**
863 * kthread_queue_delayed_work - queue the associated kthread work
864 * after a delay.
865 * @worker: target kthread_worker
866 * @dwork: kthread_delayed_work to queue
867 * @delay: number of jiffies to wait before queuing
868 *
869 * If the work has not been pending it starts a timer that will queue
870 * the work after the given @delay. If @delay is zero, it queues the
871 * work immediately.
872 *
873 * Return: %false if the @work has already been pending. It means that
874 * either the timer was running or the work was queued. It returns %true
875 * otherwise.
876 */
877bool kthread_queue_delayed_work(struct kthread_worker *worker,
878 struct kthread_delayed_work *dwork,
879 unsigned long delay)
880{
881 struct kthread_work *work = &dwork->work;
882 unsigned long flags;
883 bool ret = false;
884
885 spin_lock_irqsave(&worker->lock, flags);
886
887 if (!queuing_blocked(worker, work)) {
888 __kthread_queue_delayed_work(worker, dwork, delay);
889 ret = true;
890 }
891
892 spin_unlock_irqrestore(&worker->lock, flags);
893 return ret;
894}
895EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
896
897struct kthread_flush_work {
898 struct kthread_work work;
899 struct completion done;
900};
901
902static void kthread_flush_work_fn(struct kthread_work *work)
903{
904 struct kthread_flush_work *fwork =
905 container_of(work, struct kthread_flush_work, work);
906 complete(&fwork->done);
907}
908
909/**
910 * kthread_flush_work - flush a kthread_work
911 * @work: work to flush
912 *
913 * If @work is queued or executing, wait for it to finish execution.
914 */
915void kthread_flush_work(struct kthread_work *work)
916{
917 struct kthread_flush_work fwork = {
918 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
919 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
920 };
921 struct kthread_worker *worker;
922 bool noop = false;
923
924 worker = work->worker;
925 if (!worker)
926 return;
927
928 spin_lock_irq(&worker->lock);
929 /* Work must not be used with >1 worker, see kthread_queue_work(). */
930 WARN_ON_ONCE(work->worker != worker);
931
932 if (!list_empty(&work->node))
933 kthread_insert_work(worker, &fwork.work, work->node.next);
934 else if (worker->current_work == work)
935 kthread_insert_work(worker, &fwork.work,
936 worker->work_list.next);
937 else
938 noop = true;
939
940 spin_unlock_irq(&worker->lock);
941
942 if (!noop)
943 wait_for_completion(&fwork.done);
944}
945EXPORT_SYMBOL_GPL(kthread_flush_work);
946
947/*
948 * This function removes the work from the worker queue. Also it makes sure
949 * that it won't get queued later via the delayed work's timer.
950 *
951 * The work might still be in use when this function finishes. See the
952 * current_work proceed by the worker.
953 *
954 * Return: %true if @work was pending and successfully canceled,
955 * %false if @work was not pending
956 */
957static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
958 unsigned long *flags)
959{
960 /* Try to cancel the timer if exists. */
961 if (is_dwork) {
962 struct kthread_delayed_work *dwork =
963 container_of(work, struct kthread_delayed_work, work);
964 struct kthread_worker *worker = work->worker;
965
966 /*
967 * del_timer_sync() must be called to make sure that the timer
968 * callback is not running. The lock must be temporary released
969 * to avoid a deadlock with the callback. In the meantime,
970 * any queuing is blocked by setting the canceling counter.
971 */
972 work->canceling++;
973 spin_unlock_irqrestore(&worker->lock, *flags);
974 del_timer_sync(&dwork->timer);
975 spin_lock_irqsave(&worker->lock, *flags);
976 work->canceling--;
977 }
978
979 /*
980 * Try to remove the work from a worker list. It might either
981 * be from worker->work_list or from worker->delayed_work_list.
982 */
983 if (!list_empty(&work->node)) {
984 list_del_init(&work->node);
985 return true;
986 }
987
988 return false;
989}
990
991/**
992 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
993 * @worker: kthread worker to use
994 * @dwork: kthread delayed work to queue
995 * @delay: number of jiffies to wait before queuing
996 *
997 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
998 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
999 * @work is guaranteed to be queued immediately.
1000 *
1001 * Return: %true if @dwork was pending and its timer was modified,
1002 * %false otherwise.
1003 *
1004 * A special case is when the work is being canceled in parallel.
1005 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1006 * or yet another kthread_mod_delayed_work() call. We let the other command
1007 * win and return %false here. The caller is supposed to synchronize these
1008 * operations a reasonable way.
1009 *
1010 * This function is safe to call from any context including IRQ handler.
1011 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1012 * for details.
1013 */
1014bool kthread_mod_delayed_work(struct kthread_worker *worker,
1015 struct kthread_delayed_work *dwork,
1016 unsigned long delay)
1017{
1018 struct kthread_work *work = &dwork->work;
1019 unsigned long flags;
1020 int ret = false;
1021
1022 spin_lock_irqsave(&worker->lock, flags);
1023
1024 /* Do not bother with canceling when never queued. */
1025 if (!work->worker)
1026 goto fast_queue;
1027
1028 /* Work must not be used with >1 worker, see kthread_queue_work() */
1029 WARN_ON_ONCE(work->worker != worker);
1030
1031 /* Do not fight with another command that is canceling this work. */
1032 if (work->canceling)
1033 goto out;
1034
1035 ret = __kthread_cancel_work(work, true, &flags);
1036fast_queue:
1037 __kthread_queue_delayed_work(worker, dwork, delay);
1038out:
1039 spin_unlock_irqrestore(&worker->lock, flags);
1040 return ret;
1041}
1042EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1043
1044static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1045{
1046 struct kthread_worker *worker = work->worker;
1047 unsigned long flags;
1048 int ret = false;
1049
1050 if (!worker)
1051 goto out;
1052
1053 spin_lock_irqsave(&worker->lock, flags);
1054 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1055 WARN_ON_ONCE(work->worker != worker);
1056
1057 ret = __kthread_cancel_work(work, is_dwork, &flags);
1058
1059 if (worker->current_work != work)
1060 goto out_fast;
1061
1062 /*
1063 * The work is in progress and we need to wait with the lock released.
1064 * In the meantime, block any queuing by setting the canceling counter.
1065 */
1066 work->canceling++;
1067 spin_unlock_irqrestore(&worker->lock, flags);
1068 kthread_flush_work(work);
1069 spin_lock_irqsave(&worker->lock, flags);
1070 work->canceling--;
1071
1072out_fast:
1073 spin_unlock_irqrestore(&worker->lock, flags);
1074out:
1075 return ret;
1076}
1077
1078/**
1079 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1080 * @work: the kthread work to cancel
1081 *
1082 * Cancel @work and wait for its execution to finish. This function
1083 * can be used even if the work re-queues itself. On return from this
1084 * function, @work is guaranteed to be not pending or executing on any CPU.
1085 *
1086 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1087 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1088 *
1089 * The caller must ensure that the worker on which @work was last
1090 * queued can't be destroyed before this function returns.
1091 *
1092 * Return: %true if @work was pending, %false otherwise.
1093 */
1094bool kthread_cancel_work_sync(struct kthread_work *work)
1095{
1096 return __kthread_cancel_work_sync(work, false);
1097}
1098EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1099
1100/**
1101 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1102 * wait for it to finish.
1103 * @dwork: the kthread delayed work to cancel
1104 *
1105 * This is kthread_cancel_work_sync() for delayed works.
1106 *
1107 * Return: %true if @dwork was pending, %false otherwise.
1108 */
1109bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1110{
1111 return __kthread_cancel_work_sync(&dwork->work, true);
1112}
1113EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1114
1115/**
1116 * kthread_flush_worker - flush all current works on a kthread_worker
1117 * @worker: worker to flush
1118 *
1119 * Wait until all currently executing or pending works on @worker are
1120 * finished.
1121 */
1122void kthread_flush_worker(struct kthread_worker *worker)
1123{
1124 struct kthread_flush_work fwork = {
1125 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1126 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1127 };
1128
1129 kthread_queue_work(worker, &fwork.work);
1130 wait_for_completion(&fwork.done);
1131}
1132EXPORT_SYMBOL_GPL(kthread_flush_worker);
1133
1134/**
1135 * kthread_destroy_worker - destroy a kthread worker
1136 * @worker: worker to be destroyed
1137 *
1138 * Flush and destroy @worker. The simple flush is enough because the kthread
1139 * worker API is used only in trivial scenarios. There are no multi-step state
1140 * machines needed.
1141 */
1142void kthread_destroy_worker(struct kthread_worker *worker)
1143{
1144 struct task_struct *task;
1145
1146 task = worker->task;
1147 if (WARN_ON(!task))
1148 return;
1149
1150 kthread_flush_worker(worker);
1151 kthread_stop(task);
1152 WARN_ON(!list_empty(&worker->work_list));
1153 kfree(worker);
1154}
1155EXPORT_SYMBOL(kthread_destroy_worker);
1156
1157#ifdef CONFIG_BLK_CGROUP
1158/**
1159 * kthread_associate_blkcg - associate blkcg to current kthread
1160 * @css: the cgroup info
1161 *
1162 * Current thread must be a kthread. The thread is running jobs on behalf of
1163 * other threads. In some cases, we expect the jobs attach cgroup info of
1164 * original threads instead of that of current thread. This function stores
1165 * original thread's cgroup info in current kthread context for later
1166 * retrieval.
1167 */
1168void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1169{
1170 struct kthread *kthread;
1171
1172 if (!(current->flags & PF_KTHREAD))
1173 return;
1174 kthread = to_kthread(current);
1175 if (!kthread)
1176 return;
1177
1178 if (kthread->blkcg_css) {
1179 css_put(kthread->blkcg_css);
1180 kthread->blkcg_css = NULL;
1181 }
1182 if (css) {
1183 css_get(css);
1184 kthread->blkcg_css = css;
1185 }
1186}
1187EXPORT_SYMBOL(kthread_associate_blkcg);
1188
1189/**
1190 * kthread_blkcg - get associated blkcg css of current kthread
1191 *
1192 * Current thread must be a kthread.
1193 */
1194struct cgroup_subsys_state *kthread_blkcg(void)
1195{
1196 struct kthread *kthread;
1197
1198 if (current->flags & PF_KTHREAD) {
1199 kthread = to_kthread(current);
1200 if (kthread)
1201 return kthread->blkcg_css;
1202 }
1203 return NULL;
1204}
1205EXPORT_SYMBOL(kthread_blkcg);
1206#endif