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