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.
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,
10 #include <uapi/linux/sched/types.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>
34 static DEFINE_SPINLOCK(kthread_create_lock);
35 static LIST_HEAD(kthread_create_list);
36 struct task_struct *kthreadd_task;
38 struct kthread_create_info
40 /* Information passed to kthread() from kthreadd. */
41 int (*threadfn)(void *data);
45 /* Result passed back to kthread_create() from kthreadd. */
46 struct task_struct *result;
47 struct completion *done;
49 struct list_head list;
55 int (*threadfn)(void *);
58 struct completion parked;
59 struct completion exited;
60 #ifdef CONFIG_BLK_CGROUP
61 struct cgroup_subsys_state *blkcg_css;
66 KTHREAD_IS_PER_CPU = 0,
71 static inline struct kthread *to_kthread(struct task_struct *k)
73 WARN_ON(!(k->flags & PF_KTHREAD));
74 return (__force void *)k->set_child_tid;
78 * Variant of to_kthread() that doesn't assume @p is a kthread.
80 * Per construction; when:
82 * (p->flags & PF_KTHREAD) && p->set_child_tid
84 * the task is both a kthread and struct kthread is persistent. However
85 * PF_KTHREAD on it's own is not, kernel_thread() can exec() (See umh.c and
88 static inline struct kthread *__to_kthread(struct task_struct *p)
90 void *kthread = (__force void *)p->set_child_tid;
91 if (kthread && !(p->flags & PF_KTHREAD))
96 void set_kthread_struct(struct task_struct *p)
98 struct kthread *kthread;
103 kthread = kzalloc(sizeof(*kthread), GFP_KERNEL);
105 * We abuse ->set_child_tid to avoid the new member and because it
106 * can't be wrongly copied by copy_process(). We also rely on fact
107 * that the caller can't exec, so PF_KTHREAD can't be cleared.
109 p->set_child_tid = (__force void __user *)kthread;
112 void free_kthread_struct(struct task_struct *k)
114 struct kthread *kthread;
117 * Can be NULL if this kthread was created by kernel_thread()
118 * or if kmalloc() in kthread() failed.
120 kthread = to_kthread(k);
121 #ifdef CONFIG_BLK_CGROUP
122 WARN_ON_ONCE(kthread && kthread->blkcg_css);
128 * kthread_should_stop - should this kthread return now?
130 * When someone calls kthread_stop() on your kthread, it will be woken
131 * and this will return true. You should then return, and your return
132 * value will be passed through to kthread_stop().
134 bool kthread_should_stop(void)
136 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
138 EXPORT_SYMBOL(kthread_should_stop);
140 bool __kthread_should_park(struct task_struct *k)
142 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
144 EXPORT_SYMBOL_GPL(__kthread_should_park);
147 * kthread_should_park - should this kthread park now?
149 * When someone calls kthread_park() on your kthread, it will be woken
150 * and this will return true. You should then do the necessary
151 * cleanup and call kthread_parkme()
153 * Similar to kthread_should_stop(), but this keeps the thread alive
154 * and in a park position. kthread_unpark() "restarts" the thread and
155 * calls the thread function again.
157 bool kthread_should_park(void)
159 return __kthread_should_park(current);
161 EXPORT_SYMBOL_GPL(kthread_should_park);
164 * kthread_freezable_should_stop - should this freezable kthread return now?
165 * @was_frozen: optional out parameter, indicates whether %current was frozen
167 * kthread_should_stop() for freezable kthreads, which will enter
168 * refrigerator if necessary. This function is safe from kthread_stop() /
169 * freezer deadlock and freezable kthreads should use this function instead
170 * of calling try_to_freeze() directly.
172 bool kthread_freezable_should_stop(bool *was_frozen)
178 if (unlikely(freezing(current)))
179 frozen = __refrigerator(true);
182 *was_frozen = frozen;
184 return kthread_should_stop();
186 EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
189 * kthread_func - return the function specified on kthread creation
190 * @task: kthread task in question
192 * Returns NULL if the task is not a kthread.
194 void *kthread_func(struct task_struct *task)
196 struct kthread *kthread = __to_kthread(task);
198 return kthread->threadfn;
201 EXPORT_SYMBOL_GPL(kthread_func);
204 * kthread_data - return data value specified on kthread creation
205 * @task: kthread task in question
207 * Return the data value specified when kthread @task was created.
208 * The caller is responsible for ensuring the validity of @task when
209 * calling this function.
211 void *kthread_data(struct task_struct *task)
213 return to_kthread(task)->data;
215 EXPORT_SYMBOL_GPL(kthread_data);
218 * kthread_probe_data - speculative version of kthread_data()
219 * @task: possible kthread task in question
221 * @task could be a kthread task. Return the data value specified when it
222 * was created if accessible. If @task isn't a kthread task or its data is
223 * inaccessible for any reason, %NULL is returned. This function requires
224 * that @task itself is safe to dereference.
226 void *kthread_probe_data(struct task_struct *task)
228 struct kthread *kthread = __to_kthread(task);
232 copy_from_kernel_nofault(&data, &kthread->data, sizeof(data));
236 static void __kthread_parkme(struct kthread *self)
240 * TASK_PARKED is a special state; we must serialize against
241 * possible pending wakeups to avoid store-store collisions on
244 * Such a collision might possibly result in the task state
245 * changin from TASK_PARKED and us failing the
246 * wait_task_inactive() in kthread_park().
248 set_special_state(TASK_PARKED);
249 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
253 * Thread is going to call schedule(), do not preempt it,
254 * or the caller of kthread_park() may spend more time in
255 * wait_task_inactive().
258 complete(&self->parked);
259 schedule_preempt_disabled();
262 __set_current_state(TASK_RUNNING);
265 void kthread_parkme(void)
267 __kthread_parkme(to_kthread(current));
269 EXPORT_SYMBOL_GPL(kthread_parkme);
271 static int kthread(void *_create)
273 /* Copy data: it's on kthread's stack */
274 struct kthread_create_info *create = _create;
275 int (*threadfn)(void *data) = create->threadfn;
276 void *data = create->data;
277 struct completion *done;
278 struct kthread *self;
281 set_kthread_struct(current);
282 self = to_kthread(current);
284 /* If user was SIGKILLed, I release the structure. */
285 done = xchg(&create->done, NULL);
292 create->result = ERR_PTR(-ENOMEM);
297 self->threadfn = threadfn;
299 init_completion(&self->exited);
300 init_completion(&self->parked);
301 current->vfork_done = &self->exited;
303 /* OK, tell user we're spawned, wait for stop or wakeup */
304 __set_current_state(TASK_UNINTERRUPTIBLE);
305 create->result = current;
307 * Thread is going to call schedule(), do not preempt it,
308 * or the creator may spend more time in wait_task_inactive().
312 schedule_preempt_disabled();
316 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
317 cgroup_kthread_ready();
318 __kthread_parkme(self);
319 ret = threadfn(data);
324 /* called from kernel_clone() to get node information for about to be created task */
325 int tsk_fork_get_node(struct task_struct *tsk)
328 if (tsk == kthreadd_task)
329 return tsk->pref_node_fork;
334 static void create_kthread(struct kthread_create_info *create)
339 current->pref_node_fork = create->node;
341 /* We want our own signal handler (we take no signals by default). */
342 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
344 /* If user was SIGKILLed, I release the structure. */
345 struct completion *done = xchg(&create->done, NULL);
351 create->result = ERR_PTR(pid);
356 static __printf(4, 0)
357 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
358 void *data, int node,
359 const char namefmt[],
362 DECLARE_COMPLETION_ONSTACK(done);
363 struct task_struct *task;
364 struct kthread_create_info *create = kmalloc(sizeof(*create),
368 return ERR_PTR(-ENOMEM);
369 create->threadfn = threadfn;
372 create->done = &done;
374 spin_lock(&kthread_create_lock);
375 list_add_tail(&create->list, &kthread_create_list);
376 spin_unlock(&kthread_create_lock);
378 wake_up_process(kthreadd_task);
380 * Wait for completion in killable state, for I might be chosen by
381 * the OOM killer while kthreadd is trying to allocate memory for
384 if (unlikely(wait_for_completion_killable(&done))) {
386 * If I was SIGKILLed before kthreadd (or new kernel thread)
387 * calls complete(), leave the cleanup of this structure to
390 if (xchg(&create->done, NULL))
391 return ERR_PTR(-EINTR);
393 * kthreadd (or new kernel thread) will call complete()
396 wait_for_completion(&done);
398 task = create->result;
400 static const struct sched_param param = { .sched_priority = 0 };
401 char name[TASK_COMM_LEN];
404 * task is already visible to other tasks, so updating
405 * COMM must be protected.
407 vsnprintf(name, sizeof(name), namefmt, args);
408 set_task_comm(task, name);
410 * root may have changed our (kthreadd's) priority or CPU mask.
411 * The kernel thread should not inherit these properties.
413 sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m);
414 set_cpus_allowed_ptr(task,
415 housekeeping_cpumask(HK_FLAG_KTHREAD));
422 * kthread_create_on_node - create a kthread.
423 * @threadfn: the function to run until signal_pending(current).
424 * @data: data ptr for @threadfn.
425 * @node: task and thread structures for the thread are allocated on this node
426 * @namefmt: printf-style name for the thread.
428 * Description: This helper function creates and names a kernel
429 * thread. The thread will be stopped: use wake_up_process() to start
430 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
431 * is affine to all CPUs.
433 * If thread is going to be bound on a particular cpu, give its node
434 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
435 * When woken, the thread will run @threadfn() with @data as its
436 * argument. @threadfn() can either call do_exit() directly if it is a
437 * standalone thread for which no one will call kthread_stop(), or
438 * return when 'kthread_should_stop()' is true (which means
439 * kthread_stop() has been called). The return value should be zero
440 * or a negative error number; it will be passed to kthread_stop().
442 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
444 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
445 void *data, int node,
446 const char namefmt[],
449 struct task_struct *task;
452 va_start(args, namefmt);
453 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
458 EXPORT_SYMBOL(kthread_create_on_node);
460 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, unsigned int state)
464 if (!wait_task_inactive(p, state)) {
469 /* It's safe because the task is inactive. */
470 raw_spin_lock_irqsave(&p->pi_lock, flags);
471 do_set_cpus_allowed(p, mask);
472 p->flags |= PF_NO_SETAFFINITY;
473 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
476 static void __kthread_bind(struct task_struct *p, unsigned int cpu, unsigned int state)
478 __kthread_bind_mask(p, cpumask_of(cpu), state);
481 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
483 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
487 * kthread_bind - bind a just-created kthread to a cpu.
488 * @p: thread created by kthread_create().
489 * @cpu: cpu (might not be online, must be possible) for @k to run on.
491 * Description: This function is equivalent to set_cpus_allowed(),
492 * except that @cpu doesn't need to be online, and the thread must be
493 * stopped (i.e., just returned from kthread_create()).
495 void kthread_bind(struct task_struct *p, unsigned int cpu)
497 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
499 EXPORT_SYMBOL(kthread_bind);
502 * kthread_create_on_cpu - Create a cpu bound kthread
503 * @threadfn: the function to run until signal_pending(current).
504 * @data: data ptr for @threadfn.
505 * @cpu: The cpu on which the thread should be bound,
506 * @namefmt: printf-style name for the thread. Format is restricted
507 * to "name.*%u". Code fills in cpu number.
509 * Description: This helper function creates and names a kernel thread
511 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
512 void *data, unsigned int cpu,
515 struct task_struct *p;
517 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
521 kthread_bind(p, cpu);
522 /* CPU hotplug need to bind once again when unparking the thread. */
523 to_kthread(p)->cpu = cpu;
526 EXPORT_SYMBOL(kthread_create_on_cpu);
528 void kthread_set_per_cpu(struct task_struct *k, int cpu)
530 struct kthread *kthread = to_kthread(k);
534 WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY));
537 clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
542 set_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
545 bool kthread_is_per_cpu(struct task_struct *p)
547 struct kthread *kthread = __to_kthread(p);
551 return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
555 * kthread_unpark - unpark a thread created by kthread_create().
556 * @k: thread created by kthread_create().
558 * Sets kthread_should_park() for @k to return false, wakes it, and
559 * waits for it to return. If the thread is marked percpu then its
560 * bound to the cpu again.
562 void kthread_unpark(struct task_struct *k)
564 struct kthread *kthread = to_kthread(k);
567 * Newly created kthread was parked when the CPU was offline.
568 * The binding was lost and we need to set it again.
570 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
571 __kthread_bind(k, kthread->cpu, TASK_PARKED);
573 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
575 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
577 wake_up_state(k, TASK_PARKED);
579 EXPORT_SYMBOL_GPL(kthread_unpark);
582 * kthread_park - park a thread created by kthread_create().
583 * @k: thread created by kthread_create().
585 * Sets kthread_should_park() for @k to return true, wakes it, and
586 * waits for it to return. This can also be called after kthread_create()
587 * instead of calling wake_up_process(): the thread will park without
588 * calling threadfn().
590 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
591 * If called by the kthread itself just the park bit is set.
593 int kthread_park(struct task_struct *k)
595 struct kthread *kthread = to_kthread(k);
597 if (WARN_ON(k->flags & PF_EXITING))
600 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
603 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
607 * Wait for __kthread_parkme() to complete(), this means we
608 * _will_ have TASK_PARKED and are about to call schedule().
610 wait_for_completion(&kthread->parked);
612 * Now wait for that schedule() to complete and the task to
615 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
620 EXPORT_SYMBOL_GPL(kthread_park);
623 * kthread_stop - stop a thread created by kthread_create().
624 * @k: thread created by kthread_create().
626 * Sets kthread_should_stop() for @k to return true, wakes it, and
627 * waits for it to exit. This can also be called after kthread_create()
628 * instead of calling wake_up_process(): the thread will exit without
629 * calling threadfn().
631 * If threadfn() may call do_exit() itself, the caller must ensure
632 * task_struct can't go away.
634 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
637 int kthread_stop(struct task_struct *k)
639 struct kthread *kthread;
642 trace_sched_kthread_stop(k);
645 kthread = to_kthread(k);
646 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
649 wait_for_completion(&kthread->exited);
653 trace_sched_kthread_stop_ret(ret);
656 EXPORT_SYMBOL(kthread_stop);
658 int kthreadd(void *unused)
660 struct task_struct *tsk = current;
662 /* Setup a clean context for our children to inherit. */
663 set_task_comm(tsk, "kthreadd");
665 set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD));
666 set_mems_allowed(node_states[N_MEMORY]);
668 current->flags |= PF_NOFREEZE;
669 cgroup_init_kthreadd();
672 set_current_state(TASK_INTERRUPTIBLE);
673 if (list_empty(&kthread_create_list))
675 __set_current_state(TASK_RUNNING);
677 spin_lock(&kthread_create_lock);
678 while (!list_empty(&kthread_create_list)) {
679 struct kthread_create_info *create;
681 create = list_entry(kthread_create_list.next,
682 struct kthread_create_info, list);
683 list_del_init(&create->list);
684 spin_unlock(&kthread_create_lock);
686 create_kthread(create);
688 spin_lock(&kthread_create_lock);
690 spin_unlock(&kthread_create_lock);
696 void __kthread_init_worker(struct kthread_worker *worker,
698 struct lock_class_key *key)
700 memset(worker, 0, sizeof(struct kthread_worker));
701 raw_spin_lock_init(&worker->lock);
702 lockdep_set_class_and_name(&worker->lock, key, name);
703 INIT_LIST_HEAD(&worker->work_list);
704 INIT_LIST_HEAD(&worker->delayed_work_list);
706 EXPORT_SYMBOL_GPL(__kthread_init_worker);
709 * kthread_worker_fn - kthread function to process kthread_worker
710 * @worker_ptr: pointer to initialized kthread_worker
712 * This function implements the main cycle of kthread worker. It processes
713 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
716 * The works are not allowed to keep any locks, disable preemption or interrupts
717 * when they finish. There is defined a safe point for freezing when one work
718 * finishes and before a new one is started.
720 * Also the works must not be handled by more than one worker at the same time,
721 * see also kthread_queue_work().
723 int kthread_worker_fn(void *worker_ptr)
725 struct kthread_worker *worker = worker_ptr;
726 struct kthread_work *work;
729 * FIXME: Update the check and remove the assignment when all kthread
730 * worker users are created using kthread_create_worker*() functions.
732 WARN_ON(worker->task && worker->task != current);
733 worker->task = current;
735 if (worker->flags & KTW_FREEZABLE)
739 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
741 if (kthread_should_stop()) {
742 __set_current_state(TASK_RUNNING);
743 raw_spin_lock_irq(&worker->lock);
745 raw_spin_unlock_irq(&worker->lock);
750 raw_spin_lock_irq(&worker->lock);
751 if (!list_empty(&worker->work_list)) {
752 work = list_first_entry(&worker->work_list,
753 struct kthread_work, node);
754 list_del_init(&work->node);
756 worker->current_work = work;
757 raw_spin_unlock_irq(&worker->lock);
760 kthread_work_func_t func = work->func;
761 __set_current_state(TASK_RUNNING);
762 trace_sched_kthread_work_execute_start(work);
765 * Avoid dereferencing work after this point. The trace
766 * event only cares about the address.
768 trace_sched_kthread_work_execute_end(work, func);
769 } else if (!freezing(current))
776 EXPORT_SYMBOL_GPL(kthread_worker_fn);
778 static __printf(3, 0) struct kthread_worker *
779 __kthread_create_worker(int cpu, unsigned int flags,
780 const char namefmt[], va_list args)
782 struct kthread_worker *worker;
783 struct task_struct *task;
784 int node = NUMA_NO_NODE;
786 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
788 return ERR_PTR(-ENOMEM);
790 kthread_init_worker(worker);
793 node = cpu_to_node(cpu);
795 task = __kthread_create_on_node(kthread_worker_fn, worker,
796 node, namefmt, args);
801 kthread_bind(task, cpu);
803 worker->flags = flags;
805 wake_up_process(task);
810 return ERR_CAST(task);
814 * kthread_create_worker - create a kthread worker
815 * @flags: flags modifying the default behavior of the worker
816 * @namefmt: printf-style name for the kthread worker (task).
818 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
819 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
820 * when the worker was SIGKILLed.
822 struct kthread_worker *
823 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
825 struct kthread_worker *worker;
828 va_start(args, namefmt);
829 worker = __kthread_create_worker(-1, flags, namefmt, args);
834 EXPORT_SYMBOL(kthread_create_worker);
837 * kthread_create_worker_on_cpu - create a kthread worker and bind it
838 * to a given CPU and the associated NUMA node.
840 * @flags: flags modifying the default behavior of the worker
841 * @namefmt: printf-style name for the kthread worker (task).
843 * Use a valid CPU number if you want to bind the kthread worker
844 * to the given CPU and the associated NUMA node.
846 * A good practice is to add the cpu number also into the worker name.
847 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
850 * The kthread worker API is simple and generic. It just provides a way
851 * to create, use, and destroy workers.
853 * It is up to the API user how to handle CPU hotplug. They have to decide
854 * how to handle pending work items, prevent queuing new ones, and
855 * restore the functionality when the CPU goes off and on. There are a
858 * - CPU affinity gets lost when it is scheduled on an offline CPU.
860 * - The worker might not exist when the CPU was off when the user
861 * created the workers.
863 * Good practice is to implement two CPU hotplug callbacks and to
864 * destroy/create the worker when the CPU goes down/up.
867 * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
868 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
869 * when the worker was SIGKILLed.
871 struct kthread_worker *
872 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
873 const char namefmt[], ...)
875 struct kthread_worker *worker;
878 va_start(args, namefmt);
879 worker = __kthread_create_worker(cpu, flags, namefmt, args);
884 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
887 * Returns true when the work could not be queued at the moment.
888 * It happens when it is already pending in a worker list
889 * or when it is being cancelled.
891 static inline bool queuing_blocked(struct kthread_worker *worker,
892 struct kthread_work *work)
894 lockdep_assert_held(&worker->lock);
896 return !list_empty(&work->node) || work->canceling;
899 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
900 struct kthread_work *work)
902 lockdep_assert_held(&worker->lock);
903 WARN_ON_ONCE(!list_empty(&work->node));
904 /* Do not use a work with >1 worker, see kthread_queue_work() */
905 WARN_ON_ONCE(work->worker && work->worker != worker);
908 /* insert @work before @pos in @worker */
909 static void kthread_insert_work(struct kthread_worker *worker,
910 struct kthread_work *work,
911 struct list_head *pos)
913 kthread_insert_work_sanity_check(worker, work);
915 trace_sched_kthread_work_queue_work(worker, work);
917 list_add_tail(&work->node, pos);
918 work->worker = worker;
919 if (!worker->current_work && likely(worker->task))
920 wake_up_process(worker->task);
924 * kthread_queue_work - queue a kthread_work
925 * @worker: target kthread_worker
926 * @work: kthread_work to queue
928 * Queue @work to work processor @task for async execution. @task
929 * must have been created with kthread_worker_create(). Returns %true
930 * if @work was successfully queued, %false if it was already pending.
932 * Reinitialize the work if it needs to be used by another worker.
933 * For example, when the worker was stopped and started again.
935 bool kthread_queue_work(struct kthread_worker *worker,
936 struct kthread_work *work)
941 raw_spin_lock_irqsave(&worker->lock, flags);
942 if (!queuing_blocked(worker, work)) {
943 kthread_insert_work(worker, work, &worker->work_list);
946 raw_spin_unlock_irqrestore(&worker->lock, flags);
949 EXPORT_SYMBOL_GPL(kthread_queue_work);
952 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
953 * delayed work when the timer expires.
954 * @t: pointer to the expired timer
956 * The format of the function is defined by struct timer_list.
957 * It should have been called from irqsafe timer with irq already off.
959 void kthread_delayed_work_timer_fn(struct timer_list *t)
961 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
962 struct kthread_work *work = &dwork->work;
963 struct kthread_worker *worker = work->worker;
967 * This might happen when a pending work is reinitialized.
968 * It means that it is used a wrong way.
970 if (WARN_ON_ONCE(!worker))
973 raw_spin_lock_irqsave(&worker->lock, flags);
974 /* Work must not be used with >1 worker, see kthread_queue_work(). */
975 WARN_ON_ONCE(work->worker != worker);
977 /* Move the work from worker->delayed_work_list. */
978 WARN_ON_ONCE(list_empty(&work->node));
979 list_del_init(&work->node);
980 if (!work->canceling)
981 kthread_insert_work(worker, work, &worker->work_list);
983 raw_spin_unlock_irqrestore(&worker->lock, flags);
985 EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
987 static void __kthread_queue_delayed_work(struct kthread_worker *worker,
988 struct kthread_delayed_work *dwork,
991 struct timer_list *timer = &dwork->timer;
992 struct kthread_work *work = &dwork->work;
994 WARN_ON_FUNCTION_MISMATCH(timer->function,
995 kthread_delayed_work_timer_fn);
998 * If @delay is 0, queue @dwork->work immediately. This is for
999 * both optimization and correctness. The earliest @timer can
1000 * expire is on the closest next tick and delayed_work users depend
1001 * on that there's no such delay when @delay is 0.
1004 kthread_insert_work(worker, work, &worker->work_list);
1008 /* Be paranoid and try to detect possible races already now. */
1009 kthread_insert_work_sanity_check(worker, work);
1011 list_add(&work->node, &worker->delayed_work_list);
1012 work->worker = worker;
1013 timer->expires = jiffies + delay;
1018 * kthread_queue_delayed_work - queue the associated kthread work
1020 * @worker: target kthread_worker
1021 * @dwork: kthread_delayed_work to queue
1022 * @delay: number of jiffies to wait before queuing
1024 * If the work has not been pending it starts a timer that will queue
1025 * the work after the given @delay. If @delay is zero, it queues the
1028 * Return: %false if the @work has already been pending. It means that
1029 * either the timer was running or the work was queued. It returns %true
1032 bool kthread_queue_delayed_work(struct kthread_worker *worker,
1033 struct kthread_delayed_work *dwork,
1034 unsigned long delay)
1036 struct kthread_work *work = &dwork->work;
1037 unsigned long flags;
1040 raw_spin_lock_irqsave(&worker->lock, flags);
1042 if (!queuing_blocked(worker, work)) {
1043 __kthread_queue_delayed_work(worker, dwork, delay);
1047 raw_spin_unlock_irqrestore(&worker->lock, flags);
1050 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
1052 struct kthread_flush_work {
1053 struct kthread_work work;
1054 struct completion done;
1057 static void kthread_flush_work_fn(struct kthread_work *work)
1059 struct kthread_flush_work *fwork =
1060 container_of(work, struct kthread_flush_work, work);
1061 complete(&fwork->done);
1065 * kthread_flush_work - flush a kthread_work
1066 * @work: work to flush
1068 * If @work is queued or executing, wait for it to finish execution.
1070 void kthread_flush_work(struct kthread_work *work)
1072 struct kthread_flush_work fwork = {
1073 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1074 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1076 struct kthread_worker *worker;
1079 worker = work->worker;
1083 raw_spin_lock_irq(&worker->lock);
1084 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1085 WARN_ON_ONCE(work->worker != worker);
1087 if (!list_empty(&work->node))
1088 kthread_insert_work(worker, &fwork.work, work->node.next);
1089 else if (worker->current_work == work)
1090 kthread_insert_work(worker, &fwork.work,
1091 worker->work_list.next);
1095 raw_spin_unlock_irq(&worker->lock);
1098 wait_for_completion(&fwork.done);
1100 EXPORT_SYMBOL_GPL(kthread_flush_work);
1103 * Make sure that the timer is neither set nor running and could
1104 * not manipulate the work list_head any longer.
1106 * The function is called under worker->lock. The lock is temporary
1107 * released but the timer can't be set again in the meantime.
1109 static void kthread_cancel_delayed_work_timer(struct kthread_work *work,
1110 unsigned long *flags)
1112 struct kthread_delayed_work *dwork =
1113 container_of(work, struct kthread_delayed_work, work);
1114 struct kthread_worker *worker = work->worker;
1117 * del_timer_sync() must be called to make sure that the timer
1118 * callback is not running. The lock must be temporary released
1119 * to avoid a deadlock with the callback. In the meantime,
1120 * any queuing is blocked by setting the canceling counter.
1123 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1124 del_timer_sync(&dwork->timer);
1125 raw_spin_lock_irqsave(&worker->lock, *flags);
1130 * This function removes the work from the worker queue.
1132 * It is called under worker->lock. The caller must make sure that
1133 * the timer used by delayed work is not running, e.g. by calling
1134 * kthread_cancel_delayed_work_timer().
1136 * The work might still be in use when this function finishes. See the
1137 * current_work proceed by the worker.
1139 * Return: %true if @work was pending and successfully canceled,
1140 * %false if @work was not pending
1142 static bool __kthread_cancel_work(struct kthread_work *work)
1145 * Try to remove the work from a worker list. It might either
1146 * be from worker->work_list or from worker->delayed_work_list.
1148 if (!list_empty(&work->node)) {
1149 list_del_init(&work->node);
1157 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1158 * @worker: kthread worker to use
1159 * @dwork: kthread delayed work to queue
1160 * @delay: number of jiffies to wait before queuing
1162 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1163 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1164 * @work is guaranteed to be queued immediately.
1166 * Return: %false if @dwork was idle and queued, %true otherwise.
1168 * A special case is when the work is being canceled in parallel.
1169 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1170 * or yet another kthread_mod_delayed_work() call. We let the other command
1171 * win and return %true here. The return value can be used for reference
1172 * counting and the number of queued works stays the same. Anyway, the caller
1173 * is supposed to synchronize these operations a reasonable way.
1175 * This function is safe to call from any context including IRQ handler.
1176 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1179 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1180 struct kthread_delayed_work *dwork,
1181 unsigned long delay)
1183 struct kthread_work *work = &dwork->work;
1184 unsigned long flags;
1187 raw_spin_lock_irqsave(&worker->lock, flags);
1189 /* Do not bother with canceling when never queued. */
1190 if (!work->worker) {
1195 /* Work must not be used with >1 worker, see kthread_queue_work() */
1196 WARN_ON_ONCE(work->worker != worker);
1199 * Temporary cancel the work but do not fight with another command
1200 * that is canceling the work as well.
1202 * It is a bit tricky because of possible races with another
1203 * mod_delayed_work() and cancel_delayed_work() callers.
1205 * The timer must be canceled first because worker->lock is released
1206 * when doing so. But the work can be removed from the queue (list)
1207 * only when it can be queued again so that the return value can
1208 * be used for reference counting.
1210 kthread_cancel_delayed_work_timer(work, &flags);
1211 if (work->canceling) {
1212 /* The number of works in the queue does not change. */
1216 ret = __kthread_cancel_work(work);
1219 __kthread_queue_delayed_work(worker, dwork, delay);
1221 raw_spin_unlock_irqrestore(&worker->lock, flags);
1224 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1226 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1228 struct kthread_worker *worker = work->worker;
1229 unsigned long flags;
1235 raw_spin_lock_irqsave(&worker->lock, flags);
1236 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1237 WARN_ON_ONCE(work->worker != worker);
1240 kthread_cancel_delayed_work_timer(work, &flags);
1242 ret = __kthread_cancel_work(work);
1244 if (worker->current_work != work)
1248 * The work is in progress and we need to wait with the lock released.
1249 * In the meantime, block any queuing by setting the canceling counter.
1252 raw_spin_unlock_irqrestore(&worker->lock, flags);
1253 kthread_flush_work(work);
1254 raw_spin_lock_irqsave(&worker->lock, flags);
1258 raw_spin_unlock_irqrestore(&worker->lock, flags);
1264 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1265 * @work: the kthread work to cancel
1267 * Cancel @work and wait for its execution to finish. This function
1268 * can be used even if the work re-queues itself. On return from this
1269 * function, @work is guaranteed to be not pending or executing on any CPU.
1271 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1272 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1274 * The caller must ensure that the worker on which @work was last
1275 * queued can't be destroyed before this function returns.
1277 * Return: %true if @work was pending, %false otherwise.
1279 bool kthread_cancel_work_sync(struct kthread_work *work)
1281 return __kthread_cancel_work_sync(work, false);
1283 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1286 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1287 * wait for it to finish.
1288 * @dwork: the kthread delayed work to cancel
1290 * This is kthread_cancel_work_sync() for delayed works.
1292 * Return: %true if @dwork was pending, %false otherwise.
1294 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1296 return __kthread_cancel_work_sync(&dwork->work, true);
1298 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1301 * kthread_flush_worker - flush all current works on a kthread_worker
1302 * @worker: worker to flush
1304 * Wait until all currently executing or pending works on @worker are
1307 void kthread_flush_worker(struct kthread_worker *worker)
1309 struct kthread_flush_work fwork = {
1310 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1311 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1314 kthread_queue_work(worker, &fwork.work);
1315 wait_for_completion(&fwork.done);
1317 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1320 * kthread_destroy_worker - destroy a kthread worker
1321 * @worker: worker to be destroyed
1323 * Flush and destroy @worker. The simple flush is enough because the kthread
1324 * worker API is used only in trivial scenarios. There are no multi-step state
1327 void kthread_destroy_worker(struct kthread_worker *worker)
1329 struct task_struct *task;
1331 task = worker->task;
1335 kthread_flush_worker(worker);
1337 WARN_ON(!list_empty(&worker->work_list));
1340 EXPORT_SYMBOL(kthread_destroy_worker);
1343 * kthread_use_mm - make the calling kthread operate on an address space
1344 * @mm: address space to operate on
1346 void kthread_use_mm(struct mm_struct *mm)
1348 struct mm_struct *active_mm;
1349 struct task_struct *tsk = current;
1351 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1352 WARN_ON_ONCE(tsk->mm);
1355 /* Hold off tlb flush IPIs while switching mm's */
1356 local_irq_disable();
1357 active_mm = tsk->active_mm;
1358 if (active_mm != mm) {
1360 tsk->active_mm = mm;
1363 membarrier_update_current_mm(mm);
1364 switch_mm_irqs_off(active_mm, mm, tsk);
1367 #ifdef finish_arch_post_lock_switch
1368 finish_arch_post_lock_switch();
1372 * When a kthread starts operating on an address space, the loop
1373 * in membarrier_{private,global}_expedited() may not observe
1374 * that tsk->mm, and not issue an IPI. Membarrier requires a
1375 * memory barrier after storing to tsk->mm, before accessing
1376 * user-space memory. A full memory barrier for membarrier
1377 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1378 * mmdrop(), or explicitly with smp_mb().
1380 if (active_mm != mm)
1385 to_kthread(tsk)->oldfs = force_uaccess_begin();
1387 EXPORT_SYMBOL_GPL(kthread_use_mm);
1390 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1391 * @mm: address space to operate on
1393 void kthread_unuse_mm(struct mm_struct *mm)
1395 struct task_struct *tsk = current;
1397 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1398 WARN_ON_ONCE(!tsk->mm);
1400 force_uaccess_end(to_kthread(tsk)->oldfs);
1404 * When a kthread stops operating on an address space, the loop
1405 * in membarrier_{private,global}_expedited() may not observe
1406 * that tsk->mm, and not issue an IPI. Membarrier requires a
1407 * memory barrier after accessing user-space memory, before
1410 smp_mb__after_spinlock();
1412 local_irq_disable();
1414 membarrier_update_current_mm(NULL);
1415 /* active_mm is still 'mm' */
1416 enter_lazy_tlb(mm, tsk);
1420 EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1422 #ifdef CONFIG_BLK_CGROUP
1424 * kthread_associate_blkcg - associate blkcg to current kthread
1425 * @css: the cgroup info
1427 * Current thread must be a kthread. The thread is running jobs on behalf of
1428 * other threads. In some cases, we expect the jobs attach cgroup info of
1429 * original threads instead of that of current thread. This function stores
1430 * original thread's cgroup info in current kthread context for later
1433 void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1435 struct kthread *kthread;
1437 if (!(current->flags & PF_KTHREAD))
1439 kthread = to_kthread(current);
1443 if (kthread->blkcg_css) {
1444 css_put(kthread->blkcg_css);
1445 kthread->blkcg_css = NULL;
1449 kthread->blkcg_css = css;
1452 EXPORT_SYMBOL(kthread_associate_blkcg);
1455 * kthread_blkcg - get associated blkcg css of current kthread
1457 * Current thread must be a kthread.
1459 struct cgroup_subsys_state *kthread_blkcg(void)
1461 struct kthread *kthread;
1463 if (current->flags & PF_KTHREAD) {
1464 kthread = to_kthread(current);
1466 return kthread->blkcg_css;
1470 EXPORT_SYMBOL(kthread_blkcg);