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
))) {
388 * I got SIGKILL, but wait for 10 more seconds for completion
389 * unless chosen by the OOM killer. This delay is there as a
390 * workaround for boot failure caused by SIGKILL upon device
391 * driver initialization timeout.
393 while (i
++ < 10 && !test_tsk_thread_flag(current
, TIF_MEMDIE
))
394 if (wait_for_completion_timeout(&done
, HZ
))
397 * If I was SIGKILLed before kthreadd (or new kernel thread)
398 * calls complete(), leave the cleanup of this structure to
401 if (xchg(&create
->done
, NULL
))
402 return ERR_PTR(-EINTR
);
404 * kthreadd (or new kernel thread) will call complete()
407 wait_for_completion(&done
);
410 task
= create
->result
;
412 static const struct sched_param param
= { .sched_priority
= 0 };
413 char name
[TASK_COMM_LEN
];
416 * task is already visible to other tasks, so updating
417 * COMM must be protected.
419 vsnprintf(name
, sizeof(name
), namefmt
, args
);
420 set_task_comm(task
, name
);
422 * root may have changed our (kthreadd's) priority or CPU mask.
423 * The kernel thread should not inherit these properties.
425 sched_setscheduler_nocheck(task
, SCHED_NORMAL
, ¶m
);
426 set_cpus_allowed_ptr(task
,
427 housekeeping_cpumask(HK_FLAG_KTHREAD
));
434 * kthread_create_on_node - create a kthread.
435 * @threadfn: the function to run until signal_pending(current).
436 * @data: data ptr for @threadfn.
437 * @node: task and thread structures for the thread are allocated on this node
438 * @namefmt: printf-style name for the thread.
440 * Description: This helper function creates and names a kernel
441 * thread. The thread will be stopped: use wake_up_process() to start
442 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
443 * is affine to all CPUs.
445 * If thread is going to be bound on a particular cpu, give its node
446 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
447 * When woken, the thread will run @threadfn() with @data as its
448 * argument. @threadfn() can either call do_exit() directly if it is a
449 * standalone thread for which no one will call kthread_stop(), or
450 * return when 'kthread_should_stop()' is true (which means
451 * kthread_stop() has been called). The return value should be zero
452 * or a negative error number; it will be passed to kthread_stop().
454 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
456 struct task_struct
*kthread_create_on_node(int (*threadfn
)(void *data
),
457 void *data
, int node
,
458 const char namefmt
[],
461 struct task_struct
*task
;
464 va_start(args
, namefmt
);
465 task
= __kthread_create_on_node(threadfn
, data
, node
, namefmt
, args
);
470 EXPORT_SYMBOL(kthread_create_on_node
);
472 static void __kthread_bind_mask(struct task_struct
*p
, const struct cpumask
*mask
, unsigned int state
)
476 if (!wait_task_inactive(p
, state
)) {
481 /* It's safe because the task is inactive. */
482 raw_spin_lock_irqsave(&p
->pi_lock
, flags
);
483 do_set_cpus_allowed(p
, mask
);
484 p
->flags
|= PF_NO_SETAFFINITY
;
485 raw_spin_unlock_irqrestore(&p
->pi_lock
, flags
);
488 static void __kthread_bind(struct task_struct
*p
, unsigned int cpu
, unsigned int state
)
490 __kthread_bind_mask(p
, cpumask_of(cpu
), state
);
493 void kthread_bind_mask(struct task_struct
*p
, const struct cpumask
*mask
)
495 __kthread_bind_mask(p
, mask
, TASK_UNINTERRUPTIBLE
);
499 * kthread_bind - bind a just-created kthread to a cpu.
500 * @p: thread created by kthread_create().
501 * @cpu: cpu (might not be online, must be possible) for @k to run on.
503 * Description: This function is equivalent to set_cpus_allowed(),
504 * except that @cpu doesn't need to be online, and the thread must be
505 * stopped (i.e., just returned from kthread_create()).
507 void kthread_bind(struct task_struct
*p
, unsigned int cpu
)
509 __kthread_bind(p
, cpu
, TASK_UNINTERRUPTIBLE
);
511 EXPORT_SYMBOL(kthread_bind
);
514 * kthread_create_on_cpu - Create a cpu bound kthread
515 * @threadfn: the function to run until signal_pending(current).
516 * @data: data ptr for @threadfn.
517 * @cpu: The cpu on which the thread should be bound,
518 * @namefmt: printf-style name for the thread. Format is restricted
519 * to "name.*%u". Code fills in cpu number.
521 * Description: This helper function creates and names a kernel thread
523 struct task_struct
*kthread_create_on_cpu(int (*threadfn
)(void *data
),
524 void *data
, unsigned int cpu
,
527 struct task_struct
*p
;
529 p
= kthread_create_on_node(threadfn
, data
, cpu_to_node(cpu
), namefmt
,
533 kthread_bind(p
, cpu
);
534 /* CPU hotplug need to bind once again when unparking the thread. */
535 to_kthread(p
)->cpu
= cpu
;
539 void kthread_set_per_cpu(struct task_struct
*k
, int cpu
)
541 struct kthread
*kthread
= to_kthread(k
);
545 WARN_ON_ONCE(!(k
->flags
& PF_NO_SETAFFINITY
));
548 clear_bit(KTHREAD_IS_PER_CPU
, &kthread
->flags
);
553 set_bit(KTHREAD_IS_PER_CPU
, &kthread
->flags
);
556 bool kthread_is_per_cpu(struct task_struct
*p
)
558 struct kthread
*kthread
= __to_kthread(p
);
562 return test_bit(KTHREAD_IS_PER_CPU
, &kthread
->flags
);
566 * kthread_unpark - unpark a thread created by kthread_create().
567 * @k: thread created by kthread_create().
569 * Sets kthread_should_park() for @k to return false, wakes it, and
570 * waits for it to return. If the thread is marked percpu then its
571 * bound to the cpu again.
573 void kthread_unpark(struct task_struct
*k
)
575 struct kthread
*kthread
= to_kthread(k
);
578 * Newly created kthread was parked when the CPU was offline.
579 * The binding was lost and we need to set it again.
581 if (test_bit(KTHREAD_IS_PER_CPU
, &kthread
->flags
))
582 __kthread_bind(k
, kthread
->cpu
, TASK_PARKED
);
584 clear_bit(KTHREAD_SHOULD_PARK
, &kthread
->flags
);
586 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
588 wake_up_state(k
, TASK_PARKED
);
590 EXPORT_SYMBOL_GPL(kthread_unpark
);
593 * kthread_park - park a thread created by kthread_create().
594 * @k: thread created by kthread_create().
596 * Sets kthread_should_park() for @k to return true, wakes it, and
597 * waits for it to return. This can also be called after kthread_create()
598 * instead of calling wake_up_process(): the thread will park without
599 * calling threadfn().
601 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
602 * If called by the kthread itself just the park bit is set.
604 int kthread_park(struct task_struct
*k
)
606 struct kthread
*kthread
= to_kthread(k
);
608 if (WARN_ON(k
->flags
& PF_EXITING
))
611 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK
, &kthread
->flags
)))
614 set_bit(KTHREAD_SHOULD_PARK
, &kthread
->flags
);
618 * Wait for __kthread_parkme() to complete(), this means we
619 * _will_ have TASK_PARKED and are about to call schedule().
621 wait_for_completion(&kthread
->parked
);
623 * Now wait for that schedule() to complete and the task to
626 WARN_ON_ONCE(!wait_task_inactive(k
, TASK_PARKED
));
631 EXPORT_SYMBOL_GPL(kthread_park
);
634 * kthread_stop - stop a thread created by kthread_create().
635 * @k: thread created by kthread_create().
637 * Sets kthread_should_stop() for @k to return true, wakes it, and
638 * waits for it to exit. This can also be called after kthread_create()
639 * instead of calling wake_up_process(): the thread will exit without
640 * calling threadfn().
642 * If threadfn() may call do_exit() itself, the caller must ensure
643 * task_struct can't go away.
645 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
648 int kthread_stop(struct task_struct
*k
)
650 struct kthread
*kthread
;
653 trace_sched_kthread_stop(k
);
656 kthread
= to_kthread(k
);
657 set_bit(KTHREAD_SHOULD_STOP
, &kthread
->flags
);
660 wait_for_completion(&kthread
->exited
);
664 trace_sched_kthread_stop_ret(ret
);
667 EXPORT_SYMBOL(kthread_stop
);
669 int kthreadd(void *unused
)
671 struct task_struct
*tsk
= current
;
673 /* Setup a clean context for our children to inherit. */
674 set_task_comm(tsk
, "kthreadd");
676 set_cpus_allowed_ptr(tsk
, housekeeping_cpumask(HK_FLAG_KTHREAD
));
677 set_mems_allowed(node_states
[N_MEMORY
]);
679 current
->flags
|= PF_NOFREEZE
;
680 cgroup_init_kthreadd();
683 set_current_state(TASK_INTERRUPTIBLE
);
684 if (list_empty(&kthread_create_list
))
686 __set_current_state(TASK_RUNNING
);
688 spin_lock(&kthread_create_lock
);
689 while (!list_empty(&kthread_create_list
)) {
690 struct kthread_create_info
*create
;
692 create
= list_entry(kthread_create_list
.next
,
693 struct kthread_create_info
, list
);
694 list_del_init(&create
->list
);
695 spin_unlock(&kthread_create_lock
);
697 create_kthread(create
);
699 spin_lock(&kthread_create_lock
);
701 spin_unlock(&kthread_create_lock
);
707 void __kthread_init_worker(struct kthread_worker
*worker
,
709 struct lock_class_key
*key
)
711 memset(worker
, 0, sizeof(struct kthread_worker
));
712 raw_spin_lock_init(&worker
->lock
);
713 lockdep_set_class_and_name(&worker
->lock
, key
, name
);
714 INIT_LIST_HEAD(&worker
->work_list
);
715 INIT_LIST_HEAD(&worker
->delayed_work_list
);
717 EXPORT_SYMBOL_GPL(__kthread_init_worker
);
720 * kthread_worker_fn - kthread function to process kthread_worker
721 * @worker_ptr: pointer to initialized kthread_worker
723 * This function implements the main cycle of kthread worker. It processes
724 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
727 * The works are not allowed to keep any locks, disable preemption or interrupts
728 * when they finish. There is defined a safe point for freezing when one work
729 * finishes and before a new one is started.
731 * Also the works must not be handled by more than one worker at the same time,
732 * see also kthread_queue_work().
734 int kthread_worker_fn(void *worker_ptr
)
736 struct kthread_worker
*worker
= worker_ptr
;
737 struct kthread_work
*work
;
740 * FIXME: Update the check and remove the assignment when all kthread
741 * worker users are created using kthread_create_worker*() functions.
743 WARN_ON(worker
->task
&& worker
->task
!= current
);
744 worker
->task
= current
;
746 if (worker
->flags
& KTW_FREEZABLE
)
750 set_current_state(TASK_INTERRUPTIBLE
); /* mb paired w/ kthread_stop */
752 if (kthread_should_stop()) {
753 __set_current_state(TASK_RUNNING
);
754 raw_spin_lock_irq(&worker
->lock
);
756 raw_spin_unlock_irq(&worker
->lock
);
761 raw_spin_lock_irq(&worker
->lock
);
762 if (!list_empty(&worker
->work_list
)) {
763 work
= list_first_entry(&worker
->work_list
,
764 struct kthread_work
, node
);
765 list_del_init(&work
->node
);
767 worker
->current_work
= work
;
768 raw_spin_unlock_irq(&worker
->lock
);
771 kthread_work_func_t func
= work
->func
;
772 __set_current_state(TASK_RUNNING
);
773 trace_sched_kthread_work_execute_start(work
);
776 * Avoid dereferencing work after this point. The trace
777 * event only cares about the address.
779 trace_sched_kthread_work_execute_end(work
, func
);
780 } else if (!freezing(current
))
787 EXPORT_SYMBOL_GPL(kthread_worker_fn
);
789 static __printf(3, 0) struct kthread_worker
*
790 __kthread_create_worker(int cpu
, unsigned int flags
,
791 const char namefmt
[], va_list args
)
793 struct kthread_worker
*worker
;
794 struct task_struct
*task
;
795 int node
= NUMA_NO_NODE
;
797 worker
= kzalloc(sizeof(*worker
), GFP_KERNEL
);
799 return ERR_PTR(-ENOMEM
);
801 kthread_init_worker(worker
);
804 node
= cpu_to_node(cpu
);
806 task
= __kthread_create_on_node(kthread_worker_fn
, worker
,
807 node
, namefmt
, args
);
812 kthread_bind(task
, cpu
);
814 worker
->flags
= flags
;
816 wake_up_process(task
);
821 return ERR_CAST(task
);
825 * kthread_create_worker - create a kthread worker
826 * @flags: flags modifying the default behavior of the worker
827 * @namefmt: printf-style name for the kthread worker (task).
829 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
830 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
831 * when the worker was SIGKILLed.
833 struct kthread_worker
*
834 kthread_create_worker(unsigned int flags
, const char namefmt
[], ...)
836 struct kthread_worker
*worker
;
839 va_start(args
, namefmt
);
840 worker
= __kthread_create_worker(-1, flags
, namefmt
, args
);
845 EXPORT_SYMBOL(kthread_create_worker
);
848 * kthread_create_worker_on_cpu - create a kthread worker and bind it
849 * to a given CPU and the associated NUMA node.
851 * @flags: flags modifying the default behavior of the worker
852 * @namefmt: printf-style name for the kthread worker (task).
854 * Use a valid CPU number if you want to bind the kthread worker
855 * to the given CPU and the associated NUMA node.
857 * A good practice is to add the cpu number also into the worker name.
858 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
861 * The kthread worker API is simple and generic. It just provides a way
862 * to create, use, and destroy workers.
864 * It is up to the API user how to handle CPU hotplug. They have to decide
865 * how to handle pending work items, prevent queuing new ones, and
866 * restore the functionality when the CPU goes off and on. There are a
869 * - CPU affinity gets lost when it is scheduled on an offline CPU.
871 * - The worker might not exist when the CPU was off when the user
872 * created the workers.
874 * Good practice is to implement two CPU hotplug callbacks and to
875 * destroy/create the worker when the CPU goes down/up.
878 * The 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 worker was SIGKILLed.
882 struct kthread_worker
*
883 kthread_create_worker_on_cpu(int cpu
, unsigned int flags
,
884 const char namefmt
[], ...)
886 struct kthread_worker
*worker
;
889 va_start(args
, namefmt
);
890 worker
= __kthread_create_worker(cpu
, flags
, namefmt
, args
);
895 EXPORT_SYMBOL(kthread_create_worker_on_cpu
);
898 * Returns true when the work could not be queued at the moment.
899 * It happens when it is already pending in a worker list
900 * or when it is being cancelled.
902 static inline bool queuing_blocked(struct kthread_worker
*worker
,
903 struct kthread_work
*work
)
905 lockdep_assert_held(&worker
->lock
);
907 return !list_empty(&work
->node
) || work
->canceling
;
910 static void kthread_insert_work_sanity_check(struct kthread_worker
*worker
,
911 struct kthread_work
*work
)
913 lockdep_assert_held(&worker
->lock
);
914 WARN_ON_ONCE(!list_empty(&work
->node
));
915 /* Do not use a work with >1 worker, see kthread_queue_work() */
916 WARN_ON_ONCE(work
->worker
&& work
->worker
!= worker
);
919 /* insert @work before @pos in @worker */
920 static void kthread_insert_work(struct kthread_worker
*worker
,
921 struct kthread_work
*work
,
922 struct list_head
*pos
)
924 kthread_insert_work_sanity_check(worker
, work
);
926 trace_sched_kthread_work_queue_work(worker
, work
);
928 list_add_tail(&work
->node
, pos
);
929 work
->worker
= worker
;
930 if (!worker
->current_work
&& likely(worker
->task
))
931 wake_up_process(worker
->task
);
935 * kthread_queue_work - queue a kthread_work
936 * @worker: target kthread_worker
937 * @work: kthread_work to queue
939 * Queue @work to work processor @task for async execution. @task
940 * must have been created with kthread_worker_create(). Returns %true
941 * if @work was successfully queued, %false if it was already pending.
943 * Reinitialize the work if it needs to be used by another worker.
944 * For example, when the worker was stopped and started again.
946 bool kthread_queue_work(struct kthread_worker
*worker
,
947 struct kthread_work
*work
)
952 raw_spin_lock_irqsave(&worker
->lock
, flags
);
953 if (!queuing_blocked(worker
, work
)) {
954 kthread_insert_work(worker
, work
, &worker
->work_list
);
957 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
960 EXPORT_SYMBOL_GPL(kthread_queue_work
);
963 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
964 * delayed work when the timer expires.
965 * @t: pointer to the expired timer
967 * The format of the function is defined by struct timer_list.
968 * It should have been called from irqsafe timer with irq already off.
970 void kthread_delayed_work_timer_fn(struct timer_list
*t
)
972 struct kthread_delayed_work
*dwork
= from_timer(dwork
, t
, timer
);
973 struct kthread_work
*work
= &dwork
->work
;
974 struct kthread_worker
*worker
= work
->worker
;
978 * This might happen when a pending work is reinitialized.
979 * It means that it is used a wrong way.
981 if (WARN_ON_ONCE(!worker
))
984 raw_spin_lock_irqsave(&worker
->lock
, flags
);
985 /* Work must not be used with >1 worker, see kthread_queue_work(). */
986 WARN_ON_ONCE(work
->worker
!= worker
);
988 /* Move the work from worker->delayed_work_list. */
989 WARN_ON_ONCE(list_empty(&work
->node
));
990 list_del_init(&work
->node
);
991 if (!work
->canceling
)
992 kthread_insert_work(worker
, work
, &worker
->work_list
);
994 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
996 EXPORT_SYMBOL(kthread_delayed_work_timer_fn
);
998 static void __kthread_queue_delayed_work(struct kthread_worker
*worker
,
999 struct kthread_delayed_work
*dwork
,
1000 unsigned long delay
)
1002 struct timer_list
*timer
= &dwork
->timer
;
1003 struct kthread_work
*work
= &dwork
->work
;
1005 WARN_ON_FUNCTION_MISMATCH(timer
->function
,
1006 kthread_delayed_work_timer_fn
);
1009 * If @delay is 0, queue @dwork->work immediately. This is for
1010 * both optimization and correctness. The earliest @timer can
1011 * expire is on the closest next tick and delayed_work users depend
1012 * on that there's no such delay when @delay is 0.
1015 kthread_insert_work(worker
, work
, &worker
->work_list
);
1019 /* Be paranoid and try to detect possible races already now. */
1020 kthread_insert_work_sanity_check(worker
, work
);
1022 list_add(&work
->node
, &worker
->delayed_work_list
);
1023 work
->worker
= worker
;
1024 timer
->expires
= jiffies
+ delay
;
1029 * kthread_queue_delayed_work - queue the associated kthread work
1031 * @worker: target kthread_worker
1032 * @dwork: kthread_delayed_work to queue
1033 * @delay: number of jiffies to wait before queuing
1035 * If the work has not been pending it starts a timer that will queue
1036 * the work after the given @delay. If @delay is zero, it queues the
1039 * Return: %false if the @work has already been pending. It means that
1040 * either the timer was running or the work was queued. It returns %true
1043 bool kthread_queue_delayed_work(struct kthread_worker
*worker
,
1044 struct kthread_delayed_work
*dwork
,
1045 unsigned long delay
)
1047 struct kthread_work
*work
= &dwork
->work
;
1048 unsigned long flags
;
1051 raw_spin_lock_irqsave(&worker
->lock
, flags
);
1053 if (!queuing_blocked(worker
, work
)) {
1054 __kthread_queue_delayed_work(worker
, dwork
, delay
);
1058 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
1061 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work
);
1063 struct kthread_flush_work
{
1064 struct kthread_work work
;
1065 struct completion done
;
1068 static void kthread_flush_work_fn(struct kthread_work
*work
)
1070 struct kthread_flush_work
*fwork
=
1071 container_of(work
, struct kthread_flush_work
, work
);
1072 complete(&fwork
->done
);
1076 * kthread_flush_work - flush a kthread_work
1077 * @work: work to flush
1079 * If @work is queued or executing, wait for it to finish execution.
1081 void kthread_flush_work(struct kthread_work
*work
)
1083 struct kthread_flush_work fwork
= {
1084 KTHREAD_WORK_INIT(fwork
.work
, kthread_flush_work_fn
),
1085 COMPLETION_INITIALIZER_ONSTACK(fwork
.done
),
1087 struct kthread_worker
*worker
;
1090 worker
= work
->worker
;
1094 raw_spin_lock_irq(&worker
->lock
);
1095 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1096 WARN_ON_ONCE(work
->worker
!= worker
);
1098 if (!list_empty(&work
->node
))
1099 kthread_insert_work(worker
, &fwork
.work
, work
->node
.next
);
1100 else if (worker
->current_work
== work
)
1101 kthread_insert_work(worker
, &fwork
.work
,
1102 worker
->work_list
.next
);
1106 raw_spin_unlock_irq(&worker
->lock
);
1109 wait_for_completion(&fwork
.done
);
1111 EXPORT_SYMBOL_GPL(kthread_flush_work
);
1114 * Make sure that the timer is neither set nor running and could
1115 * not manipulate the work list_head any longer.
1117 * The function is called under worker->lock. The lock is temporary
1118 * released but the timer can't be set again in the meantime.
1120 static void kthread_cancel_delayed_work_timer(struct kthread_work
*work
,
1121 unsigned long *flags
)
1123 struct kthread_delayed_work
*dwork
=
1124 container_of(work
, struct kthread_delayed_work
, work
);
1125 struct kthread_worker
*worker
= work
->worker
;
1128 * del_timer_sync() must be called to make sure that the timer
1129 * callback is not running. The lock must be temporary released
1130 * to avoid a deadlock with the callback. In the meantime,
1131 * any queuing is blocked by setting the canceling counter.
1134 raw_spin_unlock_irqrestore(&worker
->lock
, *flags
);
1135 del_timer_sync(&dwork
->timer
);
1136 raw_spin_lock_irqsave(&worker
->lock
, *flags
);
1141 * This function removes the work from the worker queue.
1143 * It is called under worker->lock. The caller must make sure that
1144 * the timer used by delayed work is not running, e.g. by calling
1145 * kthread_cancel_delayed_work_timer().
1147 * The work might still be in use when this function finishes. See the
1148 * current_work proceed by the worker.
1150 * Return: %true if @work was pending and successfully canceled,
1151 * %false if @work was not pending
1153 static bool __kthread_cancel_work(struct kthread_work
*work
)
1156 * Try to remove the work from a worker list. It might either
1157 * be from worker->work_list or from worker->delayed_work_list.
1159 if (!list_empty(&work
->node
)) {
1160 list_del_init(&work
->node
);
1168 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1169 * @worker: kthread worker to use
1170 * @dwork: kthread delayed work to queue
1171 * @delay: number of jiffies to wait before queuing
1173 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1174 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1175 * @work is guaranteed to be queued immediately.
1177 * Return: %false if @dwork was idle and queued, %true otherwise.
1179 * A special case is when the work is being canceled in parallel.
1180 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1181 * or yet another kthread_mod_delayed_work() call. We let the other command
1182 * win and return %true here. The return value can be used for reference
1183 * counting and the number of queued works stays the same. Anyway, the caller
1184 * is supposed to synchronize these operations a reasonable way.
1186 * This function is safe to call from any context including IRQ handler.
1187 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1190 bool kthread_mod_delayed_work(struct kthread_worker
*worker
,
1191 struct kthread_delayed_work
*dwork
,
1192 unsigned long delay
)
1194 struct kthread_work
*work
= &dwork
->work
;
1195 unsigned long flags
;
1198 raw_spin_lock_irqsave(&worker
->lock
, flags
);
1200 /* Do not bother with canceling when never queued. */
1201 if (!work
->worker
) {
1206 /* Work must not be used with >1 worker, see kthread_queue_work() */
1207 WARN_ON_ONCE(work
->worker
!= worker
);
1210 * Temporary cancel the work but do not fight with another command
1211 * that is canceling the work as well.
1213 * It is a bit tricky because of possible races with another
1214 * mod_delayed_work() and cancel_delayed_work() callers.
1216 * The timer must be canceled first because worker->lock is released
1217 * when doing so. But the work can be removed from the queue (list)
1218 * only when it can be queued again so that the return value can
1219 * be used for reference counting.
1221 kthread_cancel_delayed_work_timer(work
, &flags
);
1222 if (work
->canceling
) {
1223 /* The number of works in the queue does not change. */
1227 ret
= __kthread_cancel_work(work
);
1230 __kthread_queue_delayed_work(worker
, dwork
, delay
);
1232 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
1235 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work
);
1237 static bool __kthread_cancel_work_sync(struct kthread_work
*work
, bool is_dwork
)
1239 struct kthread_worker
*worker
= work
->worker
;
1240 unsigned long flags
;
1246 raw_spin_lock_irqsave(&worker
->lock
, flags
);
1247 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1248 WARN_ON_ONCE(work
->worker
!= worker
);
1251 kthread_cancel_delayed_work_timer(work
, &flags
);
1253 ret
= __kthread_cancel_work(work
);
1255 if (worker
->current_work
!= work
)
1259 * The work is in progress and we need to wait with the lock released.
1260 * In the meantime, block any queuing by setting the canceling counter.
1263 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
1264 kthread_flush_work(work
);
1265 raw_spin_lock_irqsave(&worker
->lock
, flags
);
1269 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
1275 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1276 * @work: the kthread work to cancel
1278 * Cancel @work and wait for its execution to finish. This function
1279 * can be used even if the work re-queues itself. On return from this
1280 * function, @work is guaranteed to be not pending or executing on any CPU.
1282 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1283 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1285 * The caller must ensure that the worker on which @work was last
1286 * queued can't be destroyed before this function returns.
1288 * Return: %true if @work was pending, %false otherwise.
1290 bool kthread_cancel_work_sync(struct kthread_work
*work
)
1292 return __kthread_cancel_work_sync(work
, false);
1294 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync
);
1297 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1298 * wait for it to finish.
1299 * @dwork: the kthread delayed work to cancel
1301 * This is kthread_cancel_work_sync() for delayed works.
1303 * Return: %true if @dwork was pending, %false otherwise.
1305 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work
*dwork
)
1307 return __kthread_cancel_work_sync(&dwork
->work
, true);
1309 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync
);
1312 * kthread_flush_worker - flush all current works on a kthread_worker
1313 * @worker: worker to flush
1315 * Wait until all currently executing or pending works on @worker are
1318 void kthread_flush_worker(struct kthread_worker
*worker
)
1320 struct kthread_flush_work fwork
= {
1321 KTHREAD_WORK_INIT(fwork
.work
, kthread_flush_work_fn
),
1322 COMPLETION_INITIALIZER_ONSTACK(fwork
.done
),
1325 kthread_queue_work(worker
, &fwork
.work
);
1326 wait_for_completion(&fwork
.done
);
1328 EXPORT_SYMBOL_GPL(kthread_flush_worker
);
1331 * kthread_destroy_worker - destroy a kthread worker
1332 * @worker: worker to be destroyed
1334 * Flush and destroy @worker. The simple flush is enough because the kthread
1335 * worker API is used only in trivial scenarios. There are no multi-step state
1338 void kthread_destroy_worker(struct kthread_worker
*worker
)
1340 struct task_struct
*task
;
1342 task
= worker
->task
;
1346 kthread_flush_worker(worker
);
1348 WARN_ON(!list_empty(&worker
->work_list
));
1351 EXPORT_SYMBOL(kthread_destroy_worker
);
1354 * kthread_use_mm - make the calling kthread operate on an address space
1355 * @mm: address space to operate on
1357 void kthread_use_mm(struct mm_struct
*mm
)
1359 struct mm_struct
*active_mm
;
1360 struct task_struct
*tsk
= current
;
1362 WARN_ON_ONCE(!(tsk
->flags
& PF_KTHREAD
));
1363 WARN_ON_ONCE(tsk
->mm
);
1366 /* Hold off tlb flush IPIs while switching mm's */
1367 local_irq_disable();
1368 active_mm
= tsk
->active_mm
;
1369 if (active_mm
!= mm
) {
1371 tsk
->active_mm
= mm
;
1374 membarrier_update_current_mm(mm
);
1375 switch_mm_irqs_off(active_mm
, mm
, tsk
);
1378 #ifdef finish_arch_post_lock_switch
1379 finish_arch_post_lock_switch();
1383 * When a kthread starts operating on an address space, the loop
1384 * in membarrier_{private,global}_expedited() may not observe
1385 * that tsk->mm, and not issue an IPI. Membarrier requires a
1386 * memory barrier after storing to tsk->mm, before accessing
1387 * user-space memory. A full memory barrier for membarrier
1388 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1389 * mmdrop(), or explicitly with smp_mb().
1391 if (active_mm
!= mm
)
1396 to_kthread(tsk
)->oldfs
= force_uaccess_begin();
1398 EXPORT_SYMBOL_GPL(kthread_use_mm
);
1401 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1402 * @mm: address space to operate on
1404 void kthread_unuse_mm(struct mm_struct
*mm
)
1406 struct task_struct
*tsk
= current
;
1408 WARN_ON_ONCE(!(tsk
->flags
& PF_KTHREAD
));
1409 WARN_ON_ONCE(!tsk
->mm
);
1411 force_uaccess_end(to_kthread(tsk
)->oldfs
);
1415 * When a kthread stops operating on an address space, the loop
1416 * in membarrier_{private,global}_expedited() may not observe
1417 * that tsk->mm, and not issue an IPI. Membarrier requires a
1418 * memory barrier after accessing user-space memory, before
1421 smp_mb__after_spinlock();
1423 local_irq_disable();
1425 membarrier_update_current_mm(NULL
);
1426 /* active_mm is still 'mm' */
1427 enter_lazy_tlb(mm
, tsk
);
1431 EXPORT_SYMBOL_GPL(kthread_unuse_mm
);
1433 #ifdef CONFIG_BLK_CGROUP
1435 * kthread_associate_blkcg - associate blkcg to current kthread
1436 * @css: the cgroup info
1438 * Current thread must be a kthread. The thread is running jobs on behalf of
1439 * other threads. In some cases, we expect the jobs attach cgroup info of
1440 * original threads instead of that of current thread. This function stores
1441 * original thread's cgroup info in current kthread context for later
1444 void kthread_associate_blkcg(struct cgroup_subsys_state
*css
)
1446 struct kthread
*kthread
;
1448 if (!(current
->flags
& PF_KTHREAD
))
1450 kthread
= to_kthread(current
);
1454 if (kthread
->blkcg_css
) {
1455 css_put(kthread
->blkcg_css
);
1456 kthread
->blkcg_css
= NULL
;
1460 kthread
->blkcg_css
= css
;
1463 EXPORT_SYMBOL(kthread_associate_blkcg
);
1466 * kthread_blkcg - get associated blkcg css of current kthread
1468 * Current thread must be a kthread.
1470 struct cgroup_subsys_state
*kthread_blkcg(void)
1472 struct kthread
*kthread
;
1474 if (current
->flags
& PF_KTHREAD
) {
1475 kthread
= to_kthread(current
);
1477 return kthread
->blkcg_css
;
1481 EXPORT_SYMBOL(kthread_blkcg
);