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
;
88 * Variant of to_kthread() that doesn't assume @p is a kthread.
90 * Per construction; when:
92 * (p->flags & PF_KTHREAD) && p->set_child_tid
94 * the task is both a kthread and struct kthread is persistent. However
95 * PF_KTHREAD on it's own is not, kernel_thread() can exec() (See umh.c and
98 static inline struct kthread
*__to_kthread(struct task_struct
*p
)
100 void *kthread
= (__force
void *)p
->set_child_tid
;
101 if (kthread
&& !(p
->flags
& PF_KTHREAD
))
106 void free_kthread_struct(struct task_struct
*k
)
108 struct kthread
*kthread
;
111 * Can be NULL if this kthread was created by kernel_thread()
112 * or if kmalloc() in kthread() failed.
114 kthread
= to_kthread(k
);
115 #ifdef CONFIG_BLK_CGROUP
116 WARN_ON_ONCE(kthread
&& kthread
->blkcg_css
);
122 * kthread_should_stop - should this kthread return now?
124 * When someone calls kthread_stop() on your kthread, it will be woken
125 * and this will return true. You should then return, and your return
126 * value will be passed through to kthread_stop().
128 bool kthread_should_stop(void)
130 return test_bit(KTHREAD_SHOULD_STOP
, &to_kthread(current
)->flags
);
132 EXPORT_SYMBOL(kthread_should_stop
);
134 bool __kthread_should_park(struct task_struct
*k
)
136 return test_bit(KTHREAD_SHOULD_PARK
, &to_kthread(k
)->flags
);
138 EXPORT_SYMBOL_GPL(__kthread_should_park
);
141 * kthread_should_park - should this kthread park now?
143 * When someone calls kthread_park() on your kthread, it will be woken
144 * and this will return true. You should then do the necessary
145 * cleanup and call kthread_parkme()
147 * Similar to kthread_should_stop(), but this keeps the thread alive
148 * and in a park position. kthread_unpark() "restarts" the thread and
149 * calls the thread function again.
151 bool kthread_should_park(void)
153 return __kthread_should_park(current
);
155 EXPORT_SYMBOL_GPL(kthread_should_park
);
158 * kthread_freezable_should_stop - should this freezable kthread return now?
159 * @was_frozen: optional out parameter, indicates whether %current was frozen
161 * kthread_should_stop() for freezable kthreads, which will enter
162 * refrigerator if necessary. This function is safe from kthread_stop() /
163 * freezer deadlock and freezable kthreads should use this function instead
164 * of calling try_to_freeze() directly.
166 bool kthread_freezable_should_stop(bool *was_frozen
)
172 if (unlikely(freezing(current
)))
173 frozen
= __refrigerator(true);
176 *was_frozen
= frozen
;
178 return kthread_should_stop();
180 EXPORT_SYMBOL_GPL(kthread_freezable_should_stop
);
183 * kthread_func - return the function specified on kthread creation
184 * @task: kthread task in question
186 * Returns NULL if the task is not a kthread.
188 void *kthread_func(struct task_struct
*task
)
190 struct kthread
*kthread
= __to_kthread(task
);
192 return kthread
->threadfn
;
195 EXPORT_SYMBOL_GPL(kthread_func
);
198 * kthread_data - return data value specified on kthread creation
199 * @task: kthread task in question
201 * Return the data value specified when kthread @task was created.
202 * The caller is responsible for ensuring the validity of @task when
203 * calling this function.
205 void *kthread_data(struct task_struct
*task
)
207 return to_kthread(task
)->data
;
209 EXPORT_SYMBOL_GPL(kthread_data
);
212 * kthread_probe_data - speculative version of kthread_data()
213 * @task: possible kthread task in question
215 * @task could be a kthread task. Return the data value specified when it
216 * was created if accessible. If @task isn't a kthread task or its data is
217 * inaccessible for any reason, %NULL is returned. This function requires
218 * that @task itself is safe to dereference.
220 void *kthread_probe_data(struct task_struct
*task
)
222 struct kthread
*kthread
= __to_kthread(task
);
226 copy_from_kernel_nofault(&data
, &kthread
->data
, sizeof(data
));
230 static void __kthread_parkme(struct kthread
*self
)
234 * TASK_PARKED is a special state; we must serialize against
235 * possible pending wakeups to avoid store-store collisions on
238 * Such a collision might possibly result in the task state
239 * changin from TASK_PARKED and us failing the
240 * wait_task_inactive() in kthread_park().
242 set_special_state(TASK_PARKED
);
243 if (!test_bit(KTHREAD_SHOULD_PARK
, &self
->flags
))
247 * Thread is going to call schedule(), do not preempt it,
248 * or the caller of kthread_park() may spend more time in
249 * wait_task_inactive().
252 complete(&self
->parked
);
253 schedule_preempt_disabled();
256 __set_current_state(TASK_RUNNING
);
259 void kthread_parkme(void)
261 __kthread_parkme(to_kthread(current
));
263 EXPORT_SYMBOL_GPL(kthread_parkme
);
265 static int kthread(void *_create
)
267 /* Copy data: it's on kthread's stack */
268 struct kthread_create_info
*create
= _create
;
269 int (*threadfn
)(void *data
) = create
->threadfn
;
270 void *data
= create
->data
;
271 struct completion
*done
;
272 struct kthread
*self
;
275 self
= kzalloc(sizeof(*self
), GFP_KERNEL
);
276 set_kthread_struct(self
);
278 /* If user was SIGKILLed, I release the structure. */
279 done
= xchg(&create
->done
, NULL
);
286 create
->result
= ERR_PTR(-ENOMEM
);
291 self
->threadfn
= threadfn
;
293 init_completion(&self
->exited
);
294 init_completion(&self
->parked
);
295 current
->vfork_done
= &self
->exited
;
297 /* OK, tell user we're spawned, wait for stop or wakeup */
298 __set_current_state(TASK_UNINTERRUPTIBLE
);
299 create
->result
= current
;
301 * Thread is going to call schedule(), do not preempt it,
302 * or the creator may spend more time in wait_task_inactive().
306 schedule_preempt_disabled();
310 if (!test_bit(KTHREAD_SHOULD_STOP
, &self
->flags
)) {
311 cgroup_kthread_ready();
312 __kthread_parkme(self
);
313 ret
= threadfn(data
);
318 /* called from kernel_clone() to get node information for about to be created task */
319 int tsk_fork_get_node(struct task_struct
*tsk
)
322 if (tsk
== kthreadd_task
)
323 return tsk
->pref_node_fork
;
328 static void create_kthread(struct kthread_create_info
*create
)
333 current
->pref_node_fork
= create
->node
;
335 /* We want our own signal handler (we take no signals by default). */
336 pid
= kernel_thread(kthread
, create
, CLONE_FS
| CLONE_FILES
| SIGCHLD
);
338 /* If user was SIGKILLed, I release the structure. */
339 struct completion
*done
= xchg(&create
->done
, NULL
);
345 create
->result
= ERR_PTR(pid
);
350 static __printf(4, 0)
351 struct task_struct
*__kthread_create_on_node(int (*threadfn
)(void *data
),
352 void *data
, int node
,
353 const char namefmt
[],
356 DECLARE_COMPLETION_ONSTACK(done
);
357 struct task_struct
*task
;
358 struct kthread_create_info
*create
= kmalloc(sizeof(*create
),
362 return ERR_PTR(-ENOMEM
);
363 create
->threadfn
= threadfn
;
366 create
->done
= &done
;
368 spin_lock(&kthread_create_lock
);
369 list_add_tail(&create
->list
, &kthread_create_list
);
370 spin_unlock(&kthread_create_lock
);
372 wake_up_process(kthreadd_task
);
374 * Wait for completion in killable state, for I might be chosen by
375 * the OOM killer while kthreadd is trying to allocate memory for
378 if (unlikely(wait_for_completion_killable(&done
))) {
380 * If I was SIGKILLed before kthreadd (or new kernel thread)
381 * calls complete(), leave the cleanup of this structure to
384 if (xchg(&create
->done
, NULL
))
385 return ERR_PTR(-EINTR
);
387 * kthreadd (or new kernel thread) will call complete()
390 wait_for_completion(&done
);
392 task
= create
->result
;
394 static const struct sched_param param
= { .sched_priority
= 0 };
395 char name
[TASK_COMM_LEN
];
398 * task is already visible to other tasks, so updating
399 * COMM must be protected.
401 vsnprintf(name
, sizeof(name
), namefmt
, args
);
402 set_task_comm(task
, name
);
404 * root may have changed our (kthreadd's) priority or CPU mask.
405 * The kernel thread should not inherit these properties.
407 sched_setscheduler_nocheck(task
, SCHED_NORMAL
, ¶m
);
408 set_cpus_allowed_ptr(task
,
409 housekeeping_cpumask(HK_FLAG_KTHREAD
));
416 * kthread_create_on_node - create a kthread.
417 * @threadfn: the function to run until signal_pending(current).
418 * @data: data ptr for @threadfn.
419 * @node: task and thread structures for the thread are allocated on this node
420 * @namefmt: printf-style name for the thread.
422 * Description: This helper function creates and names a kernel
423 * thread. The thread will be stopped: use wake_up_process() to start
424 * it. See also kthread_run(). The new thread has SCHED_NORMAL policy and
425 * is affine to all CPUs.
427 * If thread is going to be bound on a particular cpu, give its node
428 * in @node, to get NUMA affinity for kthread stack, or else give NUMA_NO_NODE.
429 * When woken, the thread will run @threadfn() with @data as its
430 * argument. @threadfn() can either call do_exit() directly if it is a
431 * standalone thread for which no one will call kthread_stop(), or
432 * return when 'kthread_should_stop()' is true (which means
433 * kthread_stop() has been called). The return value should be zero
434 * or a negative error number; it will be passed to kthread_stop().
436 * Returns a task_struct or ERR_PTR(-ENOMEM) or ERR_PTR(-EINTR).
438 struct task_struct
*kthread_create_on_node(int (*threadfn
)(void *data
),
439 void *data
, int node
,
440 const char namefmt
[],
443 struct task_struct
*task
;
446 va_start(args
, namefmt
);
447 task
= __kthread_create_on_node(threadfn
, data
, node
, namefmt
, args
);
452 EXPORT_SYMBOL(kthread_create_on_node
);
454 static void __kthread_bind_mask(struct task_struct
*p
, const struct cpumask
*mask
, long state
)
458 if (!wait_task_inactive(p
, state
)) {
463 /* It's safe because the task is inactive. */
464 raw_spin_lock_irqsave(&p
->pi_lock
, flags
);
465 do_set_cpus_allowed(p
, mask
);
466 p
->flags
|= PF_NO_SETAFFINITY
;
467 raw_spin_unlock_irqrestore(&p
->pi_lock
, flags
);
470 static void __kthread_bind(struct task_struct
*p
, unsigned int cpu
, long state
)
472 __kthread_bind_mask(p
, cpumask_of(cpu
), state
);
475 void kthread_bind_mask(struct task_struct
*p
, const struct cpumask
*mask
)
477 __kthread_bind_mask(p
, mask
, TASK_UNINTERRUPTIBLE
);
481 * kthread_bind - bind a just-created kthread to a cpu.
482 * @p: thread created by kthread_create().
483 * @cpu: cpu (might not be online, must be possible) for @k to run on.
485 * Description: This function is equivalent to set_cpus_allowed(),
486 * except that @cpu doesn't need to be online, and the thread must be
487 * stopped (i.e., just returned from kthread_create()).
489 void kthread_bind(struct task_struct
*p
, unsigned int cpu
)
491 __kthread_bind(p
, cpu
, TASK_UNINTERRUPTIBLE
);
493 EXPORT_SYMBOL(kthread_bind
);
496 * kthread_create_on_cpu - Create a cpu bound kthread
497 * @threadfn: the function to run until signal_pending(current).
498 * @data: data ptr for @threadfn.
499 * @cpu: The cpu on which the thread should be bound,
500 * @namefmt: printf-style name for the thread. Format is restricted
501 * to "name.*%u". Code fills in cpu number.
503 * Description: This helper function creates and names a kernel thread
505 struct task_struct
*kthread_create_on_cpu(int (*threadfn
)(void *data
),
506 void *data
, unsigned int cpu
,
509 struct task_struct
*p
;
511 p
= kthread_create_on_node(threadfn
, data
, cpu_to_node(cpu
), namefmt
,
515 kthread_bind(p
, cpu
);
516 /* CPU hotplug need to bind once again when unparking the thread. */
517 to_kthread(p
)->cpu
= cpu
;
521 void kthread_set_per_cpu(struct task_struct
*k
, int cpu
)
523 struct kthread
*kthread
= to_kthread(k
);
527 WARN_ON_ONCE(!(k
->flags
& PF_NO_SETAFFINITY
));
530 clear_bit(KTHREAD_IS_PER_CPU
, &kthread
->flags
);
535 set_bit(KTHREAD_IS_PER_CPU
, &kthread
->flags
);
538 bool kthread_is_per_cpu(struct task_struct
*p
)
540 struct kthread
*kthread
= __to_kthread(p
);
544 return test_bit(KTHREAD_IS_PER_CPU
, &kthread
->flags
);
548 * kthread_unpark - unpark a thread created by kthread_create().
549 * @k: thread created by kthread_create().
551 * Sets kthread_should_park() for @k to return false, wakes it, and
552 * waits for it to return. If the thread is marked percpu then its
553 * bound to the cpu again.
555 void kthread_unpark(struct task_struct
*k
)
557 struct kthread
*kthread
= to_kthread(k
);
560 * Newly created kthread was parked when the CPU was offline.
561 * The binding was lost and we need to set it again.
563 if (test_bit(KTHREAD_IS_PER_CPU
, &kthread
->flags
))
564 __kthread_bind(k
, kthread
->cpu
, TASK_PARKED
);
566 clear_bit(KTHREAD_SHOULD_PARK
, &kthread
->flags
);
568 * __kthread_parkme() will either see !SHOULD_PARK or get the wakeup.
570 wake_up_state(k
, TASK_PARKED
);
572 EXPORT_SYMBOL_GPL(kthread_unpark
);
575 * kthread_park - park a thread created by kthread_create().
576 * @k: thread created by kthread_create().
578 * Sets kthread_should_park() for @k to return true, wakes it, and
579 * waits for it to return. This can also be called after kthread_create()
580 * instead of calling wake_up_process(): the thread will park without
581 * calling threadfn().
583 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
584 * If called by the kthread itself just the park bit is set.
586 int kthread_park(struct task_struct
*k
)
588 struct kthread
*kthread
= to_kthread(k
);
590 if (WARN_ON(k
->flags
& PF_EXITING
))
593 if (WARN_ON_ONCE(test_bit(KTHREAD_SHOULD_PARK
, &kthread
->flags
)))
596 set_bit(KTHREAD_SHOULD_PARK
, &kthread
->flags
);
600 * Wait for __kthread_parkme() to complete(), this means we
601 * _will_ have TASK_PARKED and are about to call schedule().
603 wait_for_completion(&kthread
->parked
);
605 * Now wait for that schedule() to complete and the task to
608 WARN_ON_ONCE(!wait_task_inactive(k
, TASK_PARKED
));
613 EXPORT_SYMBOL_GPL(kthread_park
);
616 * kthread_stop - stop a thread created by kthread_create().
617 * @k: thread created by kthread_create().
619 * Sets kthread_should_stop() for @k to return true, wakes it, and
620 * waits for it to exit. This can also be called after kthread_create()
621 * instead of calling wake_up_process(): the thread will exit without
622 * calling threadfn().
624 * If threadfn() may call do_exit() itself, the caller must ensure
625 * task_struct can't go away.
627 * Returns the result of threadfn(), or %-EINTR if wake_up_process()
630 int kthread_stop(struct task_struct
*k
)
632 struct kthread
*kthread
;
635 trace_sched_kthread_stop(k
);
638 kthread
= to_kthread(k
);
639 set_bit(KTHREAD_SHOULD_STOP
, &kthread
->flags
);
642 wait_for_completion(&kthread
->exited
);
646 trace_sched_kthread_stop_ret(ret
);
649 EXPORT_SYMBOL(kthread_stop
);
651 int kthreadd(void *unused
)
653 struct task_struct
*tsk
= current
;
655 /* Setup a clean context for our children to inherit. */
656 set_task_comm(tsk
, "kthreadd");
658 set_cpus_allowed_ptr(tsk
, housekeeping_cpumask(HK_FLAG_KTHREAD
));
659 set_mems_allowed(node_states
[N_MEMORY
]);
661 current
->flags
|= PF_NOFREEZE
;
662 cgroup_init_kthreadd();
665 set_current_state(TASK_INTERRUPTIBLE
);
666 if (list_empty(&kthread_create_list
))
668 __set_current_state(TASK_RUNNING
);
670 spin_lock(&kthread_create_lock
);
671 while (!list_empty(&kthread_create_list
)) {
672 struct kthread_create_info
*create
;
674 create
= list_entry(kthread_create_list
.next
,
675 struct kthread_create_info
, list
);
676 list_del_init(&create
->list
);
677 spin_unlock(&kthread_create_lock
);
679 create_kthread(create
);
681 spin_lock(&kthread_create_lock
);
683 spin_unlock(&kthread_create_lock
);
689 void __kthread_init_worker(struct kthread_worker
*worker
,
691 struct lock_class_key
*key
)
693 memset(worker
, 0, sizeof(struct kthread_worker
));
694 raw_spin_lock_init(&worker
->lock
);
695 lockdep_set_class_and_name(&worker
->lock
, key
, name
);
696 INIT_LIST_HEAD(&worker
->work_list
);
697 INIT_LIST_HEAD(&worker
->delayed_work_list
);
699 EXPORT_SYMBOL_GPL(__kthread_init_worker
);
702 * kthread_worker_fn - kthread function to process kthread_worker
703 * @worker_ptr: pointer to initialized kthread_worker
705 * This function implements the main cycle of kthread worker. It processes
706 * work_list until it is stopped with kthread_stop(). It sleeps when the queue
709 * The works are not allowed to keep any locks, disable preemption or interrupts
710 * when they finish. There is defined a safe point for freezing when one work
711 * finishes and before a new one is started.
713 * Also the works must not be handled by more than one worker at the same time,
714 * see also kthread_queue_work().
716 int kthread_worker_fn(void *worker_ptr
)
718 struct kthread_worker
*worker
= worker_ptr
;
719 struct kthread_work
*work
;
722 * FIXME: Update the check and remove the assignment when all kthread
723 * worker users are created using kthread_create_worker*() functions.
725 WARN_ON(worker
->task
&& worker
->task
!= current
);
726 worker
->task
= current
;
728 if (worker
->flags
& KTW_FREEZABLE
)
732 set_current_state(TASK_INTERRUPTIBLE
); /* mb paired w/ kthread_stop */
734 if (kthread_should_stop()) {
735 __set_current_state(TASK_RUNNING
);
736 raw_spin_lock_irq(&worker
->lock
);
738 raw_spin_unlock_irq(&worker
->lock
);
743 raw_spin_lock_irq(&worker
->lock
);
744 if (!list_empty(&worker
->work_list
)) {
745 work
= list_first_entry(&worker
->work_list
,
746 struct kthread_work
, node
);
747 list_del_init(&work
->node
);
749 worker
->current_work
= work
;
750 raw_spin_unlock_irq(&worker
->lock
);
753 kthread_work_func_t func
= work
->func
;
754 __set_current_state(TASK_RUNNING
);
755 trace_sched_kthread_work_execute_start(work
);
758 * Avoid dereferencing work after this point. The trace
759 * event only cares about the address.
761 trace_sched_kthread_work_execute_end(work
, func
);
762 } else if (!freezing(current
))
769 EXPORT_SYMBOL_GPL(kthread_worker_fn
);
771 static __printf(3, 0) struct kthread_worker
*
772 __kthread_create_worker(int cpu
, unsigned int flags
,
773 const char namefmt
[], va_list args
)
775 struct kthread_worker
*worker
;
776 struct task_struct
*task
;
777 int node
= NUMA_NO_NODE
;
779 worker
= kzalloc(sizeof(*worker
), GFP_KERNEL
);
781 return ERR_PTR(-ENOMEM
);
783 kthread_init_worker(worker
);
786 node
= cpu_to_node(cpu
);
788 task
= __kthread_create_on_node(kthread_worker_fn
, worker
,
789 node
, namefmt
, args
);
794 kthread_bind(task
, cpu
);
796 worker
->flags
= flags
;
798 wake_up_process(task
);
803 return ERR_CAST(task
);
807 * kthread_create_worker - create a kthread worker
808 * @flags: flags modifying the default behavior of the worker
809 * @namefmt: printf-style name for the kthread worker (task).
811 * Returns a pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
812 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
813 * when the worker was SIGKILLed.
815 struct kthread_worker
*
816 kthread_create_worker(unsigned int flags
, const char namefmt
[], ...)
818 struct kthread_worker
*worker
;
821 va_start(args
, namefmt
);
822 worker
= __kthread_create_worker(-1, flags
, namefmt
, args
);
827 EXPORT_SYMBOL(kthread_create_worker
);
830 * kthread_create_worker_on_cpu - create a kthread worker and bind it
831 * to a given CPU and the associated NUMA node.
833 * @flags: flags modifying the default behavior of the worker
834 * @namefmt: printf-style name for the kthread worker (task).
836 * Use a valid CPU number if you want to bind the kthread worker
837 * to the given CPU and the associated NUMA node.
839 * A good practice is to add the cpu number also into the worker name.
840 * For example, use kthread_create_worker_on_cpu(cpu, "helper/%d", cpu).
843 * The kthread worker API is simple and generic. It just provides a way
844 * to create, use, and destroy workers.
846 * It is up to the API user how to handle CPU hotplug. They have to decide
847 * how to handle pending work items, prevent queuing new ones, and
848 * restore the functionality when the CPU goes off and on. There are a
851 * - CPU affinity gets lost when it is scheduled on an offline CPU.
853 * - The worker might not exist when the CPU was off when the user
854 * created the workers.
856 * Good practice is to implement two CPU hotplug callbacks and to
857 * destroy/create the worker when the CPU goes down/up.
860 * The pointer to the allocated worker on success, ERR_PTR(-ENOMEM)
861 * when the needed structures could not get allocated, and ERR_PTR(-EINTR)
862 * when the worker was SIGKILLed.
864 struct kthread_worker
*
865 kthread_create_worker_on_cpu(int cpu
, unsigned int flags
,
866 const char namefmt
[], ...)
868 struct kthread_worker
*worker
;
871 va_start(args
, namefmt
);
872 worker
= __kthread_create_worker(cpu
, flags
, namefmt
, args
);
877 EXPORT_SYMBOL(kthread_create_worker_on_cpu
);
880 * Returns true when the work could not be queued at the moment.
881 * It happens when it is already pending in a worker list
882 * or when it is being cancelled.
884 static inline bool queuing_blocked(struct kthread_worker
*worker
,
885 struct kthread_work
*work
)
887 lockdep_assert_held(&worker
->lock
);
889 return !list_empty(&work
->node
) || work
->canceling
;
892 static void kthread_insert_work_sanity_check(struct kthread_worker
*worker
,
893 struct kthread_work
*work
)
895 lockdep_assert_held(&worker
->lock
);
896 WARN_ON_ONCE(!list_empty(&work
->node
));
897 /* Do not use a work with >1 worker, see kthread_queue_work() */
898 WARN_ON_ONCE(work
->worker
&& work
->worker
!= worker
);
901 /* insert @work before @pos in @worker */
902 static void kthread_insert_work(struct kthread_worker
*worker
,
903 struct kthread_work
*work
,
904 struct list_head
*pos
)
906 kthread_insert_work_sanity_check(worker
, work
);
908 trace_sched_kthread_work_queue_work(worker
, work
);
910 list_add_tail(&work
->node
, pos
);
911 work
->worker
= worker
;
912 if (!worker
->current_work
&& likely(worker
->task
))
913 wake_up_process(worker
->task
);
917 * kthread_queue_work - queue a kthread_work
918 * @worker: target kthread_worker
919 * @work: kthread_work to queue
921 * Queue @work to work processor @task for async execution. @task
922 * must have been created with kthread_worker_create(). Returns %true
923 * if @work was successfully queued, %false if it was already pending.
925 * Reinitialize the work if it needs to be used by another worker.
926 * For example, when the worker was stopped and started again.
928 bool kthread_queue_work(struct kthread_worker
*worker
,
929 struct kthread_work
*work
)
934 raw_spin_lock_irqsave(&worker
->lock
, flags
);
935 if (!queuing_blocked(worker
, work
)) {
936 kthread_insert_work(worker
, work
, &worker
->work_list
);
939 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
942 EXPORT_SYMBOL_GPL(kthread_queue_work
);
945 * kthread_delayed_work_timer_fn - callback that queues the associated kthread
946 * delayed work when the timer expires.
947 * @t: pointer to the expired timer
949 * The format of the function is defined by struct timer_list.
950 * It should have been called from irqsafe timer with irq already off.
952 void kthread_delayed_work_timer_fn(struct timer_list
*t
)
954 struct kthread_delayed_work
*dwork
= from_timer(dwork
, t
, timer
);
955 struct kthread_work
*work
= &dwork
->work
;
956 struct kthread_worker
*worker
= work
->worker
;
960 * This might happen when a pending work is reinitialized.
961 * It means that it is used a wrong way.
963 if (WARN_ON_ONCE(!worker
))
966 raw_spin_lock_irqsave(&worker
->lock
, flags
);
967 /* Work must not be used with >1 worker, see kthread_queue_work(). */
968 WARN_ON_ONCE(work
->worker
!= worker
);
970 /* Move the work from worker->delayed_work_list. */
971 WARN_ON_ONCE(list_empty(&work
->node
));
972 list_del_init(&work
->node
);
973 if (!work
->canceling
)
974 kthread_insert_work(worker
, work
, &worker
->work_list
);
976 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
978 EXPORT_SYMBOL(kthread_delayed_work_timer_fn
);
980 static void __kthread_queue_delayed_work(struct kthread_worker
*worker
,
981 struct kthread_delayed_work
*dwork
,
984 struct timer_list
*timer
= &dwork
->timer
;
985 struct kthread_work
*work
= &dwork
->work
;
987 WARN_ON_FUNCTION_MISMATCH(timer
->function
,
988 kthread_delayed_work_timer_fn
);
991 * If @delay is 0, queue @dwork->work immediately. This is for
992 * both optimization and correctness. The earliest @timer can
993 * expire is on the closest next tick and delayed_work users depend
994 * on that there's no such delay when @delay is 0.
997 kthread_insert_work(worker
, work
, &worker
->work_list
);
1001 /* Be paranoid and try to detect possible races already now. */
1002 kthread_insert_work_sanity_check(worker
, work
);
1004 list_add(&work
->node
, &worker
->delayed_work_list
);
1005 work
->worker
= worker
;
1006 timer
->expires
= jiffies
+ delay
;
1011 * kthread_queue_delayed_work - queue the associated kthread work
1013 * @worker: target kthread_worker
1014 * @dwork: kthread_delayed_work to queue
1015 * @delay: number of jiffies to wait before queuing
1017 * If the work has not been pending it starts a timer that will queue
1018 * the work after the given @delay. If @delay is zero, it queues the
1021 * Return: %false if the @work has already been pending. It means that
1022 * either the timer was running or the work was queued. It returns %true
1025 bool kthread_queue_delayed_work(struct kthread_worker
*worker
,
1026 struct kthread_delayed_work
*dwork
,
1027 unsigned long delay
)
1029 struct kthread_work
*work
= &dwork
->work
;
1030 unsigned long flags
;
1033 raw_spin_lock_irqsave(&worker
->lock
, flags
);
1035 if (!queuing_blocked(worker
, work
)) {
1036 __kthread_queue_delayed_work(worker
, dwork
, delay
);
1040 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
1043 EXPORT_SYMBOL_GPL(kthread_queue_delayed_work
);
1045 struct kthread_flush_work
{
1046 struct kthread_work work
;
1047 struct completion done
;
1050 static void kthread_flush_work_fn(struct kthread_work
*work
)
1052 struct kthread_flush_work
*fwork
=
1053 container_of(work
, struct kthread_flush_work
, work
);
1054 complete(&fwork
->done
);
1058 * kthread_flush_work - flush a kthread_work
1059 * @work: work to flush
1061 * If @work is queued or executing, wait for it to finish execution.
1063 void kthread_flush_work(struct kthread_work
*work
)
1065 struct kthread_flush_work fwork
= {
1066 KTHREAD_WORK_INIT(fwork
.work
, kthread_flush_work_fn
),
1067 COMPLETION_INITIALIZER_ONSTACK(fwork
.done
),
1069 struct kthread_worker
*worker
;
1072 worker
= work
->worker
;
1076 raw_spin_lock_irq(&worker
->lock
);
1077 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1078 WARN_ON_ONCE(work
->worker
!= worker
);
1080 if (!list_empty(&work
->node
))
1081 kthread_insert_work(worker
, &fwork
.work
, work
->node
.next
);
1082 else if (worker
->current_work
== work
)
1083 kthread_insert_work(worker
, &fwork
.work
,
1084 worker
->work_list
.next
);
1088 raw_spin_unlock_irq(&worker
->lock
);
1091 wait_for_completion(&fwork
.done
);
1093 EXPORT_SYMBOL_GPL(kthread_flush_work
);
1096 * This function removes the work from the worker queue. Also it makes sure
1097 * that it won't get queued later via the delayed work's timer.
1099 * The work might still be in use when this function finishes. See the
1100 * current_work proceed by the worker.
1102 * Return: %true if @work was pending and successfully canceled,
1103 * %false if @work was not pending
1105 static bool __kthread_cancel_work(struct kthread_work
*work
, bool is_dwork
,
1106 unsigned long *flags
)
1108 /* Try to cancel the timer if exists. */
1110 struct kthread_delayed_work
*dwork
=
1111 container_of(work
, struct kthread_delayed_work
, work
);
1112 struct kthread_worker
*worker
= work
->worker
;
1115 * del_timer_sync() must be called to make sure that the timer
1116 * callback is not running. The lock must be temporary released
1117 * to avoid a deadlock with the callback. In the meantime,
1118 * any queuing is blocked by setting the canceling counter.
1121 raw_spin_unlock_irqrestore(&worker
->lock
, *flags
);
1122 del_timer_sync(&dwork
->timer
);
1123 raw_spin_lock_irqsave(&worker
->lock
, *flags
);
1128 * Try to remove the work from a worker list. It might either
1129 * be from worker->work_list or from worker->delayed_work_list.
1131 if (!list_empty(&work
->node
)) {
1132 list_del_init(&work
->node
);
1140 * kthread_mod_delayed_work - modify delay of or queue a kthread delayed work
1141 * @worker: kthread worker to use
1142 * @dwork: kthread delayed work to queue
1143 * @delay: number of jiffies to wait before queuing
1145 * If @dwork is idle, equivalent to kthread_queue_delayed_work(). Otherwise,
1146 * modify @dwork's timer so that it expires after @delay. If @delay is zero,
1147 * @work is guaranteed to be queued immediately.
1149 * Return: %true if @dwork was pending and its timer was modified,
1152 * A special case is when the work is being canceled in parallel.
1153 * It might be caused either by the real kthread_cancel_delayed_work_sync()
1154 * or yet another kthread_mod_delayed_work() call. We let the other command
1155 * win and return %false here. The caller is supposed to synchronize these
1156 * operations a reasonable way.
1158 * This function is safe to call from any context including IRQ handler.
1159 * See __kthread_cancel_work() and kthread_delayed_work_timer_fn()
1162 bool kthread_mod_delayed_work(struct kthread_worker
*worker
,
1163 struct kthread_delayed_work
*dwork
,
1164 unsigned long delay
)
1166 struct kthread_work
*work
= &dwork
->work
;
1167 unsigned long flags
;
1170 raw_spin_lock_irqsave(&worker
->lock
, flags
);
1172 /* Do not bother with canceling when never queued. */
1176 /* Work must not be used with >1 worker, see kthread_queue_work() */
1177 WARN_ON_ONCE(work
->worker
!= worker
);
1179 /* Do not fight with another command that is canceling this work. */
1180 if (work
->canceling
)
1183 ret
= __kthread_cancel_work(work
, true, &flags
);
1185 __kthread_queue_delayed_work(worker
, dwork
, delay
);
1187 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
1190 EXPORT_SYMBOL_GPL(kthread_mod_delayed_work
);
1192 static bool __kthread_cancel_work_sync(struct kthread_work
*work
, bool is_dwork
)
1194 struct kthread_worker
*worker
= work
->worker
;
1195 unsigned long flags
;
1201 raw_spin_lock_irqsave(&worker
->lock
, flags
);
1202 /* Work must not be used with >1 worker, see kthread_queue_work(). */
1203 WARN_ON_ONCE(work
->worker
!= worker
);
1205 ret
= __kthread_cancel_work(work
, is_dwork
, &flags
);
1207 if (worker
->current_work
!= work
)
1211 * The work is in progress and we need to wait with the lock released.
1212 * In the meantime, block any queuing by setting the canceling counter.
1215 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
1216 kthread_flush_work(work
);
1217 raw_spin_lock_irqsave(&worker
->lock
, flags
);
1221 raw_spin_unlock_irqrestore(&worker
->lock
, flags
);
1227 * kthread_cancel_work_sync - cancel a kthread work and wait for it to finish
1228 * @work: the kthread work to cancel
1230 * Cancel @work and wait for its execution to finish. This function
1231 * can be used even if the work re-queues itself. On return from this
1232 * function, @work is guaranteed to be not pending or executing on any CPU.
1234 * kthread_cancel_work_sync(&delayed_work->work) must not be used for
1235 * delayed_work's. Use kthread_cancel_delayed_work_sync() instead.
1237 * The caller must ensure that the worker on which @work was last
1238 * queued can't be destroyed before this function returns.
1240 * Return: %true if @work was pending, %false otherwise.
1242 bool kthread_cancel_work_sync(struct kthread_work
*work
)
1244 return __kthread_cancel_work_sync(work
, false);
1246 EXPORT_SYMBOL_GPL(kthread_cancel_work_sync
);
1249 * kthread_cancel_delayed_work_sync - cancel a kthread delayed work and
1250 * wait for it to finish.
1251 * @dwork: the kthread delayed work to cancel
1253 * This is kthread_cancel_work_sync() for delayed works.
1255 * Return: %true if @dwork was pending, %false otherwise.
1257 bool kthread_cancel_delayed_work_sync(struct kthread_delayed_work
*dwork
)
1259 return __kthread_cancel_work_sync(&dwork
->work
, true);
1261 EXPORT_SYMBOL_GPL(kthread_cancel_delayed_work_sync
);
1264 * kthread_flush_worker - flush all current works on a kthread_worker
1265 * @worker: worker to flush
1267 * Wait until all currently executing or pending works on @worker are
1270 void kthread_flush_worker(struct kthread_worker
*worker
)
1272 struct kthread_flush_work fwork
= {
1273 KTHREAD_WORK_INIT(fwork
.work
, kthread_flush_work_fn
),
1274 COMPLETION_INITIALIZER_ONSTACK(fwork
.done
),
1277 kthread_queue_work(worker
, &fwork
.work
);
1278 wait_for_completion(&fwork
.done
);
1280 EXPORT_SYMBOL_GPL(kthread_flush_worker
);
1283 * kthread_destroy_worker - destroy a kthread worker
1284 * @worker: worker to be destroyed
1286 * Flush and destroy @worker. The simple flush is enough because the kthread
1287 * worker API is used only in trivial scenarios. There are no multi-step state
1290 void kthread_destroy_worker(struct kthread_worker
*worker
)
1292 struct task_struct
*task
;
1294 task
= worker
->task
;
1298 kthread_flush_worker(worker
);
1300 WARN_ON(!list_empty(&worker
->work_list
));
1303 EXPORT_SYMBOL(kthread_destroy_worker
);
1306 * kthread_use_mm - make the calling kthread operate on an address space
1307 * @mm: address space to operate on
1309 void kthread_use_mm(struct mm_struct
*mm
)
1311 struct mm_struct
*active_mm
;
1312 struct task_struct
*tsk
= current
;
1314 WARN_ON_ONCE(!(tsk
->flags
& PF_KTHREAD
));
1315 WARN_ON_ONCE(tsk
->mm
);
1318 /* Hold off tlb flush IPIs while switching mm's */
1319 local_irq_disable();
1320 active_mm
= tsk
->active_mm
;
1321 if (active_mm
!= mm
) {
1323 tsk
->active_mm
= mm
;
1326 membarrier_update_current_mm(mm
);
1327 switch_mm_irqs_off(active_mm
, mm
, tsk
);
1330 #ifdef finish_arch_post_lock_switch
1331 finish_arch_post_lock_switch();
1335 * When a kthread starts operating on an address space, the loop
1336 * in membarrier_{private,global}_expedited() may not observe
1337 * that tsk->mm, and not issue an IPI. Membarrier requires a
1338 * memory barrier after storing to tsk->mm, before accessing
1339 * user-space memory. A full memory barrier for membarrier
1340 * {PRIVATE,GLOBAL}_EXPEDITED is implicitly provided by
1341 * mmdrop(), or explicitly with smp_mb().
1343 if (active_mm
!= mm
)
1348 to_kthread(tsk
)->oldfs
= force_uaccess_begin();
1350 EXPORT_SYMBOL_GPL(kthread_use_mm
);
1353 * kthread_unuse_mm - reverse the effect of kthread_use_mm()
1354 * @mm: address space to operate on
1356 void kthread_unuse_mm(struct mm_struct
*mm
)
1358 struct task_struct
*tsk
= current
;
1360 WARN_ON_ONCE(!(tsk
->flags
& PF_KTHREAD
));
1361 WARN_ON_ONCE(!tsk
->mm
);
1363 force_uaccess_end(to_kthread(tsk
)->oldfs
);
1367 * When a kthread stops operating on an address space, the loop
1368 * in membarrier_{private,global}_expedited() may not observe
1369 * that tsk->mm, and not issue an IPI. Membarrier requires a
1370 * memory barrier after accessing user-space memory, before
1373 smp_mb__after_spinlock();
1375 local_irq_disable();
1377 membarrier_update_current_mm(NULL
);
1378 /* active_mm is still 'mm' */
1379 enter_lazy_tlb(mm
, tsk
);
1383 EXPORT_SYMBOL_GPL(kthread_unuse_mm
);
1385 #ifdef CONFIG_BLK_CGROUP
1387 * kthread_associate_blkcg - associate blkcg to current kthread
1388 * @css: the cgroup info
1390 * Current thread must be a kthread. The thread is running jobs on behalf of
1391 * other threads. In some cases, we expect the jobs attach cgroup info of
1392 * original threads instead of that of current thread. This function stores
1393 * original thread's cgroup info in current kthread context for later
1396 void kthread_associate_blkcg(struct cgroup_subsys_state
*css
)
1398 struct kthread
*kthread
;
1400 if (!(current
->flags
& PF_KTHREAD
))
1402 kthread
= to_kthread(current
);
1406 if (kthread
->blkcg_css
) {
1407 css_put(kthread
->blkcg_css
);
1408 kthread
->blkcg_css
= NULL
;
1412 kthread
->blkcg_css
= css
;
1415 EXPORT_SYMBOL(kthread_associate_blkcg
);
1418 * kthread_blkcg - get associated blkcg css of current kthread
1420 * Current thread must be a kthread.
1422 struct cgroup_subsys_state
*kthread_blkcg(void)
1424 struct kthread
*kthread
;
1426 if (current
->flags
& PF_KTHREAD
) {
1427 kthread
= to_kthread(current
);
1429 return kthread
->blkcg_css
;
1433 EXPORT_SYMBOL(kthread_blkcg
);