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 void set_kthread_struct(void *kthread)
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.
78 current->set_child_tid = (__force void __user *)kthread;
81 static inline struct kthread *to_kthread(struct task_struct *k)
83 WARN_ON(!(k->flags & PF_KTHREAD));
84 return (__force void *)k->set_child_tid;
87 void free_kthread_struct(struct task_struct *k)
89 struct kthread *kthread;
92 * Can be NULL if this kthread was created by kernel_thread()
93 * or if kmalloc() in kthread() failed.
95 kthread = to_kthread(k);
96 #ifdef CONFIG_BLK_CGROUP
97 WARN_ON_ONCE(kthread && kthread->blkcg_css);
103 * kthread_should_stop - should this kthread return now?
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().
109 bool kthread_should_stop(void)
111 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
113 EXPORT_SYMBOL(kthread_should_stop);
115 bool __kthread_should_park(struct task_struct *k)
117 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(k)->flags);
119 EXPORT_SYMBOL_GPL(__kthread_should_park);
122 * kthread_should_park - should this kthread park now?
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()
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.
132 bool kthread_should_park(void)
134 return __kthread_should_park(current);
136 EXPORT_SYMBOL_GPL(kthread_should_park);
139 * kthread_freezable_should_stop - should this freezable kthread return now?
140 * @was_frozen: optional out parameter, indicates whether %current was frozen
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.
147 bool kthread_freezable_should_stop(bool *was_frozen)
153 if (unlikely(freezing(current)))
154 frozen = __refrigerator(true);
157 *was_frozen = frozen;
159 return kthread_should_stop();
161 EXPORT_SYMBOL_GPL(kthread_freezable_should_stop);
164 * kthread_func - return the function specified on kthread creation
165 * @task: kthread task in question
167 * Returns NULL if the task is not a kthread.
169 void *kthread_func(struct task_struct *task)
171 if (task->flags & PF_KTHREAD)
172 return to_kthread(task)->threadfn;
175 EXPORT_SYMBOL_GPL(kthread_func);
178 * kthread_data - return data value specified on kthread creation
179 * @task: kthread task in question
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.
185 void *kthread_data(struct task_struct *task)
187 return to_kthread(task)->data;
189 EXPORT_SYMBOL_GPL(kthread_data);
192 * kthread_probe_data - speculative version of kthread_data()
193 * @task: possible kthread task in question
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.
200 void *kthread_probe_data(struct task_struct *task)
202 struct kthread *kthread = to_kthread(task);
205 copy_from_kernel_nofault(&data, &kthread->data, sizeof(data));
209 static void __kthread_parkme(struct kthread *self)
213 * TASK_PARKED is a special state; we must serialize against
214 * possible pending wakeups to avoid store-store collisions on
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().
221 set_special_state(TASK_PARKED);
222 if (!test_bit(KTHREAD_SHOULD_PARK, &self->flags))
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().
231 complete(&self->parked);
232 schedule_preempt_disabled();
235 __set_current_state(TASK_RUNNING);
238 void kthread_parkme(void)
240 __kthread_parkme(to_kthread(current));
242 EXPORT_SYMBOL_GPL(kthread_parkme);
244 static int kthread(void *_create)
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;
254 self = kzalloc(sizeof(*self), GFP_KERNEL);
255 set_kthread_struct(self);
257 /* If user was SIGKILLed, I release the structure. */
258 done = xchg(&create->done, NULL);
265 create->result = ERR_PTR(-ENOMEM);
270 self->threadfn = threadfn;
272 init_completion(&self->exited);
273 init_completion(&self->parked);
274 current->vfork_done = &self->exited;
276 /* OK, tell user we're spawned, wait for stop or wakeup */
277 __set_current_state(TASK_UNINTERRUPTIBLE);
278 create->result = current;
280 * Thread is going to call schedule(), do not preempt it,
281 * or the creator may spend more time in wait_task_inactive().
285 schedule_preempt_disabled();
289 if (!test_bit(KTHREAD_SHOULD_STOP, &self->flags)) {
290 cgroup_kthread_ready();
291 __kthread_parkme(self);
292 ret = threadfn(data);
297 /* called from kernel_clone() to get node information for about to be created task */
298 int tsk_fork_get_node(struct task_struct *tsk)
301 if (tsk == kthreadd_task)
302 return tsk->pref_node_fork;
307 static void create_kthread(struct kthread_create_info *create)
312 current->pref_node_fork = create->node;
314 /* We want our own signal handler (we take no signals by default). */
315 pid = kernel_thread(kthread, create, CLONE_FS | CLONE_FILES | SIGCHLD);
317 /* If user was SIGKILLed, I release the structure. */
318 struct completion *done = xchg(&create->done, NULL);
324 create->result = ERR_PTR(pid);
329 static __printf(4, 0)
330 struct task_struct *__kthread_create_on_node(int (*threadfn)(void *data),
331 void *data, int node,
332 const char namefmt[],
335 DECLARE_COMPLETION_ONSTACK(done);
336 struct task_struct *task;
337 struct kthread_create_info *create = kmalloc(sizeof(*create),
341 return ERR_PTR(-ENOMEM);
342 create->threadfn = threadfn;
345 create->done = &done;
347 spin_lock(&kthread_create_lock);
348 list_add_tail(&create->list, &kthread_create_list);
349 spin_unlock(&kthread_create_lock);
351 wake_up_process(kthreadd_task);
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
357 if (unlikely(wait_for_completion_killable(&done))) {
359 * If I was SIGKILLed before kthreadd (or new kernel thread)
360 * calls complete(), leave the cleanup of this structure to
363 if (xchg(&create->done, NULL))
364 return ERR_PTR(-EINTR);
366 * kthreadd (or new kernel thread) will call complete()
369 wait_for_completion(&done);
371 task = create->result;
373 static const struct sched_param param = { .sched_priority = 0 };
374 char name[TASK_COMM_LEN];
377 * task is already visible to other tasks, so updating
378 * COMM must be protected.
380 vsnprintf(name, sizeof(name), namefmt, args);
381 set_task_comm(task, name);
383 * root may have changed our (kthreadd's) priority or CPU mask.
384 * The kernel thread should not inherit these properties.
386 sched_setscheduler_nocheck(task, SCHED_NORMAL, ¶m);
387 set_cpus_allowed_ptr(task,
388 housekeeping_cpumask(HK_FLAG_KTHREAD));
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.
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.
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().
415 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
417 struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
418 void *data, int node,
419 const char namefmt[],
422 struct task_struct *task;
425 va_start(args, namefmt);
426 task = __kthread_create_on_node(threadfn, data, node, namefmt, args);
431 EXPORT_SYMBOL(kthread_create_on_node);
433 static void __kthread_bind_mask(struct task_struct *p, const struct cpumask *mask, long state)
437 if (!wait_task_inactive(p, state)) {
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);
449 static void __kthread_bind(struct task_struct *p, unsigned int cpu, long state)
451 __kthread_bind_mask(p, cpumask_of(cpu), state);
454 void kthread_bind_mask(struct task_struct *p, const struct cpumask *mask)
456 __kthread_bind_mask(p, mask, TASK_UNINTERRUPTIBLE);
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.
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()).
468 void kthread_bind(struct task_struct *p, unsigned int cpu)
470 __kthread_bind(p, cpu, TASK_UNINTERRUPTIBLE);
472 EXPORT_SYMBOL(kthread_bind);
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.
482 * Description: This helper function creates and names a kernel thread
484 struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
485 void *data, unsigned int cpu,
488 struct task_struct *p;
490 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
494 kthread_bind(p, cpu);
495 /* CPU hotplug need to bind once again when unparking the thread. */
496 to_kthread(p)->cpu = cpu;
500 void kthread_set_per_cpu(struct task_struct *k, int cpu)
502 struct kthread *kthread = to_kthread(k);
506 WARN_ON_ONCE(!(k->flags & PF_NO_SETAFFINITY));
509 clear_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
514 set_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
517 bool kthread_is_per_cpu(struct task_struct *k)
519 struct kthread *kthread = to_kthread(k);
523 return test_bit(KTHREAD_IS_PER_CPU, &kthread->flags);
527 * kthread_unpark - unpark a thread created by kthread_create().
528 * @k: thread created by kthread_create().
530 * Sets kthread_should_park() for @k to return false, wakes it, and
531 * waits for it to return. If the thread is marked percpu then its
532 * bound to the cpu again.
534 void kthread_unpark(struct task_struct *k)
536 struct kthread *kthread = to_kthread(k);
539 * Newly created kthread was parked when the CPU was offline.
540 * The binding was lost and we need to set it again.
542 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
543 __kthread_bind(k, kthread->cpu, TASK_PARKED);
545 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
547 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
549 wake_up_state(k, TASK_PARKED);
551 EXPORT_SYMBOL_GPL(kthread_unpark);
554 * kthread_park - park a thread created by kthread_create().
555 * @k: thread created by kthread_create().
557 * Sets kthread_should_park() for @k to return true, wakes it, and
558 * waits for it to return. This can also be called after kthread_create()
559 * instead of calling wake_up_process(): the thread will park without
560 * calling threadfn().
562 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
563 * If called by the kthread itself just the park bit is set.
565 int kthread_park(struct task_struct *k)
567 struct kthread *kthread = to_kthread(k);
569 if (WARN_ON(k->flags & PF_EXITING))
572 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK, &kthread->flags)))
575 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
579 * Wait for __kthread_parkme() to complete(), this means we
580 * _will_ have TASK_PARKED and are about to call schedule().
582 wait_for_completion(&kthread->parked);
584 * Now wait for that schedule() to complete and the task to
587 WARN_ON_ONCE(!wait_task_inactive(k, TASK_PARKED));
592 EXPORT_SYMBOL_GPL(kthread_park);
595 * kthread_stop - stop a thread created by kthread_create().
596 * @k: thread created by kthread_create().
598 * Sets kthread_should_stop() for @k to return true, wakes it, and
599 * waits for it to exit. This can also be called after kthread_create()
600 * instead of calling wake_up_process(): the thread will exit without
601 * calling threadfn().
603 * If threadfn() may call do_exit() itself, the caller must ensure
604 * task_struct can't go away.
606 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
609 int kthread_stop(struct task_struct *k)
611 struct kthread *kthread;
614 trace_sched_kthread_stop(k);
617 kthread = to_kthread(k);
618 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
621 wait_for_completion(&kthread->exited);
625 trace_sched_kthread_stop_ret(ret);
628 EXPORT_SYMBOL(kthread_stop);
630 int kthreadd(void *unused)
632 struct task_struct *tsk = current;
634 /* Setup a clean context for our children to inherit. */
635 set_task_comm(tsk, "kthreadd");
637 set_cpus_allowed_ptr(tsk, housekeeping_cpumask(HK_FLAG_KTHREAD));
638 set_mems_allowed(node_states[N_MEMORY]);
640 current->flags |= PF_NOFREEZE;
641 cgroup_init_kthreadd();
644 set_current_state(TASK_INTERRUPTIBLE);
645 if (list_empty(&kthread_create_list))
647 __set_current_state(TASK_RUNNING);
649 spin_lock(&kthread_create_lock);
650 while (!list_empty(&kthread_create_list)) {
651 struct kthread_create_info *create;
653 create = list_entry(kthread_create_list.next,
654 struct kthread_create_info, list);
655 list_del_init(&create->list);
656 spin_unlock(&kthread_create_lock);
658 create_kthread(create);
660 spin_lock(&kthread_create_lock);
662 spin_unlock(&kthread_create_lock);
668 void __kthread_init_worker(struct kthread_worker *worker,
670 struct lock_class_key *key)
672 memset(worker, 0, sizeof(struct kthread_worker));
673 raw_spin_lock_init(&worker->lock);
674 lockdep_set_class_and_name(&worker->lock, key, name);
675 INIT_LIST_HEAD(&worker->work_list);
676 INIT_LIST_HEAD(&worker->delayed_work_list);
678 EXPORT_SYMBOL_GPL(__kthread_init_worker);
681 * kthread_worker_fn - kthread function to process kthread_worker
682 * @worker_ptr: pointer to initialized kthread_worker
684 * This function implements the main cycle of kthread worker. It processes
685 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
688 * The works are not allowed to keep any locks, disable preemption or interrupts
689 * when they finish. There is defined a safe point for freezing when one work
690 * finishes and before a new one is started.
692 * Also the works must not be handled by more than one worker at the same time,
693 * see also kthread_queue_work().
695 int kthread_worker_fn(void *worker_ptr)
697 struct kthread_worker *worker = worker_ptr;
698 struct kthread_work *work;
701 * FIXME: Update the check and remove the assignment when all kthread
702 * worker users are created using kthread_create_worker*() functions.
704 WARN_ON(worker->task && worker->task != current);
705 worker->task = current;
707 if (worker->flags & KTW_FREEZABLE)
711 set_current_state(TASK_INTERRUPTIBLE); /* mb paired w/ kthread_stop */
713 if (kthread_should_stop()) {
714 __set_current_state(TASK_RUNNING);
715 raw_spin_lock_irq(&worker->lock);
717 raw_spin_unlock_irq(&worker->lock);
722 raw_spin_lock_irq(&worker->lock);
723 if (!list_empty(&worker->work_list)) {
724 work = list_first_entry(&worker->work_list,
725 struct kthread_work, node);
726 list_del_init(&work->node);
728 worker->current_work = work;
729 raw_spin_unlock_irq(&worker->lock);
732 kthread_work_func_t func = work->func;
733 __set_current_state(TASK_RUNNING);
734 trace_sched_kthread_work_execute_start(work);
737 * Avoid dereferencing work after this point. The trace
738 * event only cares about the address.
740 trace_sched_kthread_work_execute_end(work, func);
741 } else if (!freezing(current))
748 EXPORT_SYMBOL_GPL(kthread_worker_fn);
750 static __printf(3, 0) struct kthread_worker *
751 __kthread_create_worker(int cpu, unsigned int flags,
752 const char namefmt[], va_list args)
754 struct kthread_worker *worker;
755 struct task_struct *task;
756 int node = NUMA_NO_NODE;
758 worker = kzalloc(sizeof(*worker), GFP_KERNEL);
760 return ERR_PTR(-ENOMEM);
762 kthread_init_worker(worker);
765 node = cpu_to_node(cpu);
767 task = __kthread_create_on_node(kthread_worker_fn, worker,
768 node, namefmt, args);
773 kthread_bind(task, cpu);
775 worker->flags = flags;
777 wake_up_process(task);
782 return ERR_CAST(task);
786 * kthread_create_worker - create a kthread worker
787 * @flags: flags modifying the default behavior of the worker
788 * @namefmt: printf-style name for the kthread worker (task).
790 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
791 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
792 * when the worker was SIGKILLed.
794 struct kthread_worker *
795 kthread_create_worker(unsigned int flags, const char namefmt[], ...)
797 struct kthread_worker *worker;
800 va_start(args, namefmt);
801 worker = __kthread_create_worker(-1, flags, namefmt, args);
806 EXPORT_SYMBOL(kthread_create_worker);
809 * kthread_create_worker_on_cpu - create a kthread worker and bind it
810 * to a given CPU and the associated NUMA node.
812 * @flags: flags modifying the default behavior of the worker
813 * @namefmt: printf-style name for the kthread worker (task).
815 * Use a valid CPU number if you want to bind the kthread worker
816 * to the given CPU and the associated NUMA node.
818 * A good practice is to add the cpu number also into the worker name.
819 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
822 * The kthread worker API is simple and generic. It just provides a way
823 * to create, use, and destroy workers.
825 * It is up to the API user how to handle CPU hotplug. They have to decide
826 * how to handle pending work items, prevent queuing new ones, and
827 * restore the functionality when the CPU goes off and on. There are a
830 * - CPU affinity gets lost when it is scheduled on an offline CPU.
832 * - The worker might not exist when the CPU was off when the user
833 * created the workers.
835 * Good practice is to implement two CPU hotplug callbacks and to
836 * destroy/create the worker when the CPU goes down/up.
839 * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
840 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
841 * when the worker was SIGKILLed.
843 struct kthread_worker *
844 kthread_create_worker_on_cpu(int cpu, unsigned int flags,
845 const char namefmt[], ...)
847 struct kthread_worker *worker;
850 va_start(args, namefmt);
851 worker = __kthread_create_worker(cpu, flags, namefmt, args);
856 EXPORT_SYMBOL(kthread_create_worker_on_cpu);
859 * Returns true when the work could not be queued at the moment.
860 * It happens when it is already pending in a worker list
861 * or when it is being cancelled.
863 static inline bool queuing_blocked(struct kthread_worker *worker,
864 struct kthread_work *work)
866 lockdep_assert_held(&worker->lock);
868 return !list_empty(&work->node) || work->canceling;
871 static void kthread_insert_work_sanity_check(struct kthread_worker *worker,
872 struct kthread_work *work)
874 lockdep_assert_held(&worker->lock);
875 WARN_ON_ONCE(!list_empty(&work->node));
876 /* Do not use a work with >1 worker, see kthread_queue_work() */
877 WARN_ON_ONCE(work->worker && work->worker != worker);
880 /* insert @work before @pos in @worker */
881 static void kthread_insert_work(struct kthread_worker *worker,
882 struct kthread_work *work,
883 struct list_head *pos)
885 kthread_insert_work_sanity_check(worker, work);
887 trace_sched_kthread_work_queue_work(worker, work);
889 list_add_tail(&work->node, pos);
890 work->worker = worker;
891 if (!worker->current_work && likely(worker->task))
892 wake_up_process(worker->task);
896 * kthread_queue_work - queue a kthread_work
897 * @worker: target kthread_worker
898 * @work: kthread_work to queue
900 * Queue @work to work processor @task for async execution. @task
901 * must have been created with kthread_worker_create(). Returns %true
902 * if @work was successfully queued, %false if it was already pending.
904 * Reinitialize the work if it needs to be used by another worker.
905 * For example, when the worker was stopped and started again.
907 bool kthread_queue_work(struct kthread_worker *worker,
908 struct kthread_work *work)
913 raw_spin_lock_irqsave(&worker->lock, flags);
914 if (!queuing_blocked(worker, work)) {
915 kthread_insert_work(worker, work, &worker->work_list);
918 raw_spin_unlock_irqrestore(&worker->lock, flags);
921 EXPORT_SYMBOL_GPL(kthread_queue_work);
924 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
925 * delayed work when the timer expires.
926 * @t: pointer to the expired timer
928 * The format of the function is defined by struct timer_list.
929 * It should have been called from irqsafe timer with irq already off.
931 void kthread_delayed_work_timer_fn(struct timer_list *t)
933 struct kthread_delayed_work *dwork = from_timer(dwork, t, timer);
934 struct kthread_work *work = &dwork->work;
935 struct kthread_worker *worker = work->worker;
939 * This might happen when a pending work is reinitialized.
940 * It means that it is used a wrong way.
942 if (WARN_ON_ONCE(!worker))
945 raw_spin_lock_irqsave(&worker->lock, flags);
946 /* Work must not be used with >1 worker, see kthread_queue_work(). */
947 WARN_ON_ONCE(work->worker != worker);
949 /* Move the work from worker->delayed_work_list. */
950 WARN_ON_ONCE(list_empty(&work->node));
951 list_del_init(&work->node);
952 if (!work->canceling)
953 kthread_insert_work(worker, work, &worker->work_list);
955 raw_spin_unlock_irqrestore(&worker->lock, flags);
957 EXPORT_SYMBOL(kthread_delayed_work_timer_fn);
959 static void __kthread_queue_delayed_work(struct kthread_worker *worker,
960 struct kthread_delayed_work *dwork,
963 struct timer_list *timer = &dwork->timer;
964 struct kthread_work *work = &dwork->work;
966 WARN_ON_FUNCTION_MISMATCH(timer->function,
967 kthread_delayed_work_timer_fn);
970 * If @delay is 0, queue @dwork->work immediately. This is for
971 * both optimization and correctness. The earliest @timer can
972 * expire is on the closest next tick and delayed_work users depend
973 * on that there's no such delay when @delay is 0.
976 kthread_insert_work(worker, work, &worker->work_list);
980 /* Be paranoid and try to detect possible races already now. */
981 kthread_insert_work_sanity_check(worker, work);
983 list_add(&work->node, &worker->delayed_work_list);
984 work->worker = worker;
985 timer->expires = jiffies + delay;
990 * kthread_queue_delayed_work - queue the associated kthread work
992 * @worker: target kthread_worker
993 * @dwork: kthread_delayed_work to queue
994 * @delay: number of jiffies to wait before queuing
996 * If the work has not been pending it starts a timer that will queue
997 * the work after the given @delay. If @delay is zero, it queues the
1000 * Return: %false if the @work has already been pending. It means that
1001 * either the timer was running or the work was queued. It returns %true
1004 bool kthread_queue_delayed_work(struct kthread_worker *worker,
1005 struct kthread_delayed_work *dwork,
1006 unsigned long delay)
1008 struct kthread_work *work = &dwork->work;
1009 unsigned long flags;
1012 raw_spin_lock_irqsave(&worker->lock, flags);
1014 if (!queuing_blocked(worker, work)) {
1015 __kthread_queue_delayed_work(worker, dwork, delay);
1019 raw_spin_unlock_irqrestore(&worker->lock, flags);
1022 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work);
1024 struct kthread_flush_work {
1025 struct kthread_work work;
1026 struct completion done;
1029 static void kthread_flush_work_fn(struct kthread_work *work)
1031 struct kthread_flush_work *fwork =
1032 container_of(work, struct kthread_flush_work, work);
1033 complete(&fwork->done);
1037 * kthread_flush_work - flush a kthread_work
1038 * @work: work to flush
1040 * If @work is queued or executing, wait for it to finish execution.
1042 void kthread_flush_work(struct kthread_work *work)
1044 struct kthread_flush_work fwork = {
1045 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1046 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1048 struct kthread_worker *worker;
1051 worker = work->worker;
1055 raw_spin_lock_irq(&worker->lock);
1056 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1057 WARN_ON_ONCE(work->worker != worker);
1059 if (!list_empty(&work->node))
1060 kthread_insert_work(worker, &fwork.work, work->node.next);
1061 else if (worker->current_work == work)
1062 kthread_insert_work(worker, &fwork.work,
1063 worker->work_list.next);
1067 raw_spin_unlock_irq(&worker->lock);
1070 wait_for_completion(&fwork.done);
1072 EXPORT_SYMBOL_GPL(kthread_flush_work);
1075 * This function removes the work from the worker queue. Also it makes sure
1076 * that it won't get queued later via the delayed work's timer.
1078 * The work might still be in use when this function finishes. See the
1079 * current_work proceed by the worker.
1081 * Return: %true if @work was pending and successfully canceled,
1082 * %false if @work was not pending
1084 static bool __kthread_cancel_work(struct kthread_work *work, bool is_dwork,
1085 unsigned long *flags)
1087 /* Try to cancel the timer if exists. */
1089 struct kthread_delayed_work *dwork =
1090 container_of(work, struct kthread_delayed_work, work);
1091 struct kthread_worker *worker = work->worker;
1094 * del_timer_sync() must be called to make sure that the timer
1095 * callback is not running. The lock must be temporary released
1096 * to avoid a deadlock with the callback. In the meantime,
1097 * any queuing is blocked by setting the canceling counter.
1100 raw_spin_unlock_irqrestore(&worker->lock, *flags);
1101 del_timer_sync(&dwork->timer);
1102 raw_spin_lock_irqsave(&worker->lock, *flags);
1107 * Try to remove the work from a worker list. It might either
1108 * be from worker->work_list or from worker->delayed_work_list.
1110 if (!list_empty(&work->node)) {
1111 list_del_init(&work->node);
1119 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1120 * @worker: kthread worker to use
1121 * @dwork: kthread delayed work to queue
1122 * @delay: number of jiffies to wait before queuing
1124 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1125 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1126 * @work is guaranteed to be queued immediately.
1128 * Return: %true if @dwork was pending and its timer was modified,
1131 * A special case is when the work is being canceled in parallel.
1132 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1133 * or yet another kthread_mod_delayed_work() call. We let the other command
1134 * win and return %false here. The caller is supposed to synchronize these
1135 * operations a reasonable way.
1137 * This function is safe to call from any context including IRQ handler.
1138 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1141 bool kthread_mod_delayed_work(struct kthread_worker *worker,
1142 struct kthread_delayed_work *dwork,
1143 unsigned long delay)
1145 struct kthread_work *work = &dwork->work;
1146 unsigned long flags;
1149 raw_spin_lock_irqsave(&worker->lock, flags);
1151 /* Do not bother with canceling when never queued. */
1155 /* Work must not be used with >1 worker, see kthread_queue_work() */
1156 WARN_ON_ONCE(work->worker != worker);
1158 /* Do not fight with another command that is canceling this work. */
1159 if (work->canceling)
1162 ret = __kthread_cancel_work(work, true, &flags);
1164 __kthread_queue_delayed_work(worker, dwork, delay);
1166 raw_spin_unlock_irqrestore(&worker->lock, flags);
1169 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work);
1171 static bool __kthread_cancel_work_sync(struct kthread_work *work, bool is_dwork)
1173 struct kthread_worker *worker = work->worker;
1174 unsigned long flags;
1180 raw_spin_lock_irqsave(&worker->lock, flags);
1181 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1182 WARN_ON_ONCE(work->worker != worker);
1184 ret = __kthread_cancel_work(work, is_dwork, &flags);
1186 if (worker->current_work != work)
1190 * The work is in progress and we need to wait with the lock released.
1191 * In the meantime, block any queuing by setting the canceling counter.
1194 raw_spin_unlock_irqrestore(&worker->lock, flags);
1195 kthread_flush_work(work);
1196 raw_spin_lock_irqsave(&worker->lock, flags);
1200 raw_spin_unlock_irqrestore(&worker->lock, flags);
1206 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1207 * @work: the kthread work to cancel
1209 * Cancel @work and wait for its execution to finish. This function
1210 * can be used even if the work re-queues itself. On return from this
1211 * function, @work is guaranteed to be not pending or executing on any CPU.
1213 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1214 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1216 * The caller must ensure that the worker on which @work was last
1217 * queued can't be destroyed before this function returns.
1219 * Return: %true if @work was pending, %false otherwise.
1221 bool kthread_cancel_work_sync(struct kthread_work *work)
1223 return __kthread_cancel_work_sync(work, false);
1225 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync);
1228 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1229 * wait for it to finish.
1230 * @dwork: the kthread delayed work to cancel
1232 * This is kthread_cancel_work_sync() for delayed works.
1234 * Return: %true if @dwork was pending, %false otherwise.
1236 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work *dwork)
1238 return __kthread_cancel_work_sync(&dwork->work, true);
1240 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync);
1243 * kthread_flush_worker - flush all current works on a kthread_worker
1244 * @worker: worker to flush
1246 * Wait until all currently executing or pending works on @worker are
1249 void kthread_flush_worker(struct kthread_worker *worker)
1251 struct kthread_flush_work fwork = {
1252 KTHREAD_WORK_INIT(fwork.work, kthread_flush_work_fn),
1253 COMPLETION_INITIALIZER_ONSTACK(fwork.done),
1256 kthread_queue_work(worker, &fwork.work);
1257 wait_for_completion(&fwork.done);
1259 EXPORT_SYMBOL_GPL(kthread_flush_worker);
1262 * kthread_destroy_worker - destroy a kthread worker
1263 * @worker: worker to be destroyed
1265 * Flush and destroy @worker. The simple flush is enough because the kthread
1266 * worker API is used only in trivial scenarios. There are no multi-step state
1269 void kthread_destroy_worker(struct kthread_worker *worker)
1271 struct task_struct *task;
1273 task = worker->task;
1277 kthread_flush_worker(worker);
1279 WARN_ON(!list_empty(&worker->work_list));
1282 EXPORT_SYMBOL(kthread_destroy_worker);
1285 * kthread_use_mm - make the calling kthread operate on an address space
1286 * @mm: address space to operate on
1288 void kthread_use_mm(struct mm_struct *mm)
1290 struct mm_struct *active_mm;
1291 struct task_struct *tsk = current;
1293 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1294 WARN_ON_ONCE(tsk->mm);
1297 /* Hold off tlb flush IPIs while switching mm's */
1298 local_irq_disable();
1299 active_mm = tsk->active_mm;
1300 if (active_mm != mm) {
1302 tsk->active_mm = mm;
1305 membarrier_update_current_mm(mm);
1306 switch_mm_irqs_off(active_mm, mm, tsk);
1309 #ifdef finish_arch_post_lock_switch
1310 finish_arch_post_lock_switch();
1314 * When a kthread starts operating on an address space, the loop
1315 * in membarrier_{private,global}_expedited() may not observe
1316 * that tsk->mm, and not issue an IPI. Membarrier requires a
1317 * memory barrier after storing to tsk->mm, before accessing
1318 * user-space memory. A full memory barrier for membarrier
1319 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1320 * mmdrop(), or explicitly with smp_mb().
1322 if (active_mm != mm)
1327 to_kthread(tsk)->oldfs = force_uaccess_begin();
1329 EXPORT_SYMBOL_GPL(kthread_use_mm);
1332 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1333 * @mm: address space to operate on
1335 void kthread_unuse_mm(struct mm_struct *mm)
1337 struct task_struct *tsk = current;
1339 WARN_ON_ONCE(!(tsk->flags & PF_KTHREAD));
1340 WARN_ON_ONCE(!tsk->mm);
1342 force_uaccess_end(to_kthread(tsk)->oldfs);
1346 * When a kthread stops operating on an address space, the loop
1347 * in membarrier_{private,global}_expedited() may not observe
1348 * that tsk->mm, and not issue an IPI. Membarrier requires a
1349 * memory barrier after accessing user-space memory, before
1352 smp_mb__after_spinlock();
1354 local_irq_disable();
1356 membarrier_update_current_mm(NULL);
1357 /* active_mm is still 'mm' */
1358 enter_lazy_tlb(mm, tsk);
1362 EXPORT_SYMBOL_GPL(kthread_unuse_mm);
1364 #ifdef CONFIG_BLK_CGROUP
1366 * kthread_associate_blkcg - associate blkcg to current kthread
1367 * @css: the cgroup info
1369 * Current thread must be a kthread. The thread is running jobs on behalf of
1370 * other threads. In some cases, we expect the jobs attach cgroup info of
1371 * original threads instead of that of current thread. This function stores
1372 * original thread's cgroup info in current kthread context for later
1375 void kthread_associate_blkcg(struct cgroup_subsys_state *css)
1377 struct kthread *kthread;
1379 if (!(current->flags & PF_KTHREAD))
1381 kthread = to_kthread(current);
1385 if (kthread->blkcg_css) {
1386 css_put(kthread->blkcg_css);
1387 kthread->blkcg_css = NULL;
1391 kthread->blkcg_css = css;
1394 EXPORT_SYMBOL(kthread_associate_blkcg);
1397 * kthread_blkcg - get associated blkcg css of current kthread
1399 * Current thread must be a kthread.
1401 struct cgroup_subsys_state *kthread_blkcg(void)
1403 struct kthread *kthread;
1405 if (current->flags & PF_KTHREAD) {
1406 kthread = to_kthread(current);
1408 return kthread->blkcg_css;
1412 EXPORT_SYMBOL(kthread_blkcg);