2 * linux/kernel/workqueue.c
4 * Generic mechanism for defining kernel helper threads for running
5 * arbitrary tasks in process context.
7 * Started by Ingo Molnar, Copyright (C) 2002
9 * Derived from the taskqueue/keventd code by:
11 * David Woodhouse <dwmw2@infradead.org>
12 * Andrew Morton <andrewm@uow.edu.au>
13 * Kai Petzke <wpp@marie.physik.tu-berlin.de>
14 * Theodore Ts'o <tytso@mit.edu>
16 * Made to use alloc_percpu by Christoph Lameter <clameter@sgi.com>.
19 #include <linux/module.h>
20 #include <linux/kernel.h>
21 #include <linux/sched.h>
22 #include <linux/init.h>
23 #include <linux/signal.h>
24 #include <linux/completion.h>
25 #include <linux/workqueue.h>
26 #include <linux/slab.h>
27 #include <linux/cpu.h>
28 #include <linux/notifier.h>
29 #include <linux/kthread.h>
30 #include <linux/hardirq.h>
31 #include <linux/mempolicy.h>
32 #include <linux/freezer.h>
33 #include <linux/kallsyms.h>
34 #include <linux/debug_locks.h>
37 * The per-CPU workqueue (if single thread, we always use the first
40 struct cpu_workqueue_struct
{
44 struct list_head worklist
;
45 wait_queue_head_t more_work
;
46 struct work_struct
*current_work
;
48 struct workqueue_struct
*wq
;
49 struct task_struct
*thread
;
52 int run_depth
; /* Detect run_workqueue() recursion depth */
53 } ____cacheline_aligned
;
56 * The externally visible workqueue abstraction is an array of
59 struct workqueue_struct
{
60 struct cpu_workqueue_struct
*cpu_wq
;
61 struct list_head list
;
64 int freezeable
; /* Freeze threads during suspend */
67 /* All the per-cpu workqueues on the system, for hotplug cpu to add/remove
68 threads to each one as cpus come/go. */
69 static DEFINE_MUTEX(workqueue_mutex
);
70 static LIST_HEAD(workqueues
);
72 static int singlethread_cpu __read_mostly
;
73 static cpumask_t cpu_singlethread_map __read_mostly
;
74 /* optimization, we could use cpu_possible_map */
75 static cpumask_t cpu_populated_map __read_mostly
;
77 /* If it's single threaded, it isn't in the list of workqueues. */
78 static inline int is_single_threaded(struct workqueue_struct
*wq
)
80 return wq
->singlethread
;
83 static const cpumask_t
*wq_cpu_map(struct workqueue_struct
*wq
)
85 return is_single_threaded(wq
)
86 ? &cpu_singlethread_map
: &cpu_populated_map
;
90 struct cpu_workqueue_struct
*wq_per_cpu(struct workqueue_struct
*wq
, int cpu
)
92 if (unlikely(is_single_threaded(wq
)))
93 cpu
= singlethread_cpu
;
94 return per_cpu_ptr(wq
->cpu_wq
, cpu
);
98 * Set the workqueue on which a work item is to be run
99 * - Must *only* be called if the pending flag is set
101 static inline void set_wq_data(struct work_struct
*work
,
102 struct cpu_workqueue_struct
*cwq
)
106 BUG_ON(!work_pending(work
));
108 new = (unsigned long) cwq
| (1UL << WORK_STRUCT_PENDING
);
109 new |= WORK_STRUCT_FLAG_MASK
& *work_data_bits(work
);
110 atomic_long_set(&work
->data
, new);
114 struct cpu_workqueue_struct
*get_wq_data(struct work_struct
*work
)
116 return (void *) (atomic_long_read(&work
->data
) & WORK_STRUCT_WQ_DATA_MASK
);
119 static void insert_work(struct cpu_workqueue_struct
*cwq
,
120 struct work_struct
*work
, int tail
)
122 set_wq_data(work
, cwq
);
124 * Ensure that we get the right work->data if we see the
125 * result of list_add() below, see try_to_grab_pending().
129 list_add_tail(&work
->entry
, &cwq
->worklist
);
131 list_add(&work
->entry
, &cwq
->worklist
);
132 wake_up(&cwq
->more_work
);
135 /* Preempt must be disabled. */
136 static void __queue_work(struct cpu_workqueue_struct
*cwq
,
137 struct work_struct
*work
)
141 spin_lock_irqsave(&cwq
->lock
, flags
);
142 insert_work(cwq
, work
, 1);
143 spin_unlock_irqrestore(&cwq
->lock
, flags
);
147 * queue_work - queue work on a workqueue
148 * @wq: workqueue to use
149 * @work: work to queue
151 * Returns 0 if @work was already on a queue, non-zero otherwise.
153 * We queue the work to the CPU it was submitted, but there is no
154 * guarantee that it will be processed by that CPU.
156 int fastcall
queue_work(struct workqueue_struct
*wq
, struct work_struct
*work
)
160 if (!test_and_set_bit(WORK_STRUCT_PENDING
, work_data_bits(work
))) {
161 BUG_ON(!list_empty(&work
->entry
));
162 __queue_work(wq_per_cpu(wq
, get_cpu()), work
);
168 EXPORT_SYMBOL_GPL(queue_work
);
170 void delayed_work_timer_fn(unsigned long __data
)
172 struct delayed_work
*dwork
= (struct delayed_work
*)__data
;
173 struct cpu_workqueue_struct
*cwq
= get_wq_data(&dwork
->work
);
174 struct workqueue_struct
*wq
= cwq
->wq
;
176 __queue_work(wq_per_cpu(wq
, smp_processor_id()), &dwork
->work
);
180 * queue_delayed_work - queue work on a workqueue after delay
181 * @wq: workqueue to use
182 * @dwork: delayable work to queue
183 * @delay: number of jiffies to wait before queueing
185 * Returns 0 if @work was already on a queue, non-zero otherwise.
187 int fastcall
queue_delayed_work(struct workqueue_struct
*wq
,
188 struct delayed_work
*dwork
, unsigned long delay
)
190 timer_stats_timer_set_start_info(&dwork
->timer
);
192 return queue_work(wq
, &dwork
->work
);
194 return queue_delayed_work_on(-1, wq
, dwork
, delay
);
196 EXPORT_SYMBOL_GPL(queue_delayed_work
);
199 * queue_delayed_work_on - queue work on specific CPU after delay
200 * @cpu: CPU number to execute work on
201 * @wq: workqueue to use
202 * @dwork: work to queue
203 * @delay: number of jiffies to wait before queueing
205 * Returns 0 if @work was already on a queue, non-zero otherwise.
207 int queue_delayed_work_on(int cpu
, struct workqueue_struct
*wq
,
208 struct delayed_work
*dwork
, unsigned long delay
)
211 struct timer_list
*timer
= &dwork
->timer
;
212 struct work_struct
*work
= &dwork
->work
;
214 if (!test_and_set_bit(WORK_STRUCT_PENDING
, work_data_bits(work
))) {
215 BUG_ON(timer_pending(timer
));
216 BUG_ON(!list_empty(&work
->entry
));
218 /* This stores cwq for the moment, for the timer_fn */
219 set_wq_data(work
, wq_per_cpu(wq
, raw_smp_processor_id()));
220 timer
->expires
= jiffies
+ delay
;
221 timer
->data
= (unsigned long)dwork
;
222 timer
->function
= delayed_work_timer_fn
;
224 if (unlikely(cpu
>= 0))
225 add_timer_on(timer
, cpu
);
232 EXPORT_SYMBOL_GPL(queue_delayed_work_on
);
234 static void run_workqueue(struct cpu_workqueue_struct
*cwq
)
236 spin_lock_irq(&cwq
->lock
);
238 if (cwq
->run_depth
> 3) {
239 /* morton gets to eat his hat */
240 printk("%s: recursion depth exceeded: %d\n",
241 __FUNCTION__
, cwq
->run_depth
);
244 while (!list_empty(&cwq
->worklist
)) {
245 struct work_struct
*work
= list_entry(cwq
->worklist
.next
,
246 struct work_struct
, entry
);
247 work_func_t f
= work
->func
;
249 cwq
->current_work
= work
;
250 list_del_init(cwq
->worklist
.next
);
251 spin_unlock_irq(&cwq
->lock
);
253 BUG_ON(get_wq_data(work
) != cwq
);
254 work_clear_pending(work
);
257 if (unlikely(in_atomic() || lockdep_depth(current
) > 0)) {
258 printk(KERN_ERR
"BUG: workqueue leaked lock or atomic: "
260 current
->comm
, preempt_count(),
262 printk(KERN_ERR
" last function: ");
263 print_symbol("%s\n", (unsigned long)f
);
264 debug_show_held_locks(current
);
268 spin_lock_irq(&cwq
->lock
);
269 cwq
->current_work
= NULL
;
272 spin_unlock_irq(&cwq
->lock
);
276 * NOTE: the caller must not touch *cwq if this func returns true
278 static int cwq_should_stop(struct cpu_workqueue_struct
*cwq
)
280 int should_stop
= cwq
->should_stop
;
282 if (unlikely(should_stop
)) {
283 spin_lock_irq(&cwq
->lock
);
284 should_stop
= cwq
->should_stop
&& list_empty(&cwq
->worklist
);
287 spin_unlock_irq(&cwq
->lock
);
293 static int worker_thread(void *__cwq
)
295 struct cpu_workqueue_struct
*cwq
= __cwq
;
298 if (!cwq
->wq
->freezeable
)
299 current
->flags
|= PF_NOFREEZE
;
301 set_user_nice(current
, -5);
304 prepare_to_wait(&cwq
->more_work
, &wait
, TASK_INTERRUPTIBLE
);
305 if (!freezing(current
) && !cwq
->should_stop
306 && list_empty(&cwq
->worklist
))
308 finish_wait(&cwq
->more_work
, &wait
);
312 if (cwq_should_stop(cwq
))
322 struct work_struct work
;
323 struct completion done
;
326 static void wq_barrier_func(struct work_struct
*work
)
328 struct wq_barrier
*barr
= container_of(work
, struct wq_barrier
, work
);
329 complete(&barr
->done
);
332 static void insert_wq_barrier(struct cpu_workqueue_struct
*cwq
,
333 struct wq_barrier
*barr
, int tail
)
335 INIT_WORK(&barr
->work
, wq_barrier_func
);
336 __set_bit(WORK_STRUCT_PENDING
, work_data_bits(&barr
->work
));
338 init_completion(&barr
->done
);
340 insert_work(cwq
, &barr
->work
, tail
);
343 static void flush_cpu_workqueue(struct cpu_workqueue_struct
*cwq
)
345 if (cwq
->thread
== current
) {
347 * Probably keventd trying to flush its own queue. So simply run
348 * it by hand rather than deadlocking.
352 struct wq_barrier barr
;
355 spin_lock_irq(&cwq
->lock
);
356 if (!list_empty(&cwq
->worklist
) || cwq
->current_work
!= NULL
) {
357 insert_wq_barrier(cwq
, &barr
, 1);
360 spin_unlock_irq(&cwq
->lock
);
363 wait_for_completion(&barr
.done
);
368 * flush_workqueue - ensure that any scheduled work has run to completion.
369 * @wq: workqueue to flush
371 * Forces execution of the workqueue and blocks until its completion.
372 * This is typically used in driver shutdown handlers.
374 * We sleep until all works which were queued on entry have been handled,
375 * but we are not livelocked by new incoming ones.
377 * This function used to run the workqueues itself. Now we just wait for the
378 * helper threads to do it.
380 void fastcall
flush_workqueue(struct workqueue_struct
*wq
)
382 const cpumask_t
*cpu_map
= wq_cpu_map(wq
);
386 for_each_cpu_mask(cpu
, *cpu_map
)
387 flush_cpu_workqueue(per_cpu_ptr(wq
->cpu_wq
, cpu
));
389 EXPORT_SYMBOL_GPL(flush_workqueue
);
392 * Upon a successful return, the caller "owns" WORK_STRUCT_PENDING bit,
393 * so this work can't be re-armed in any way.
395 static int try_to_grab_pending(struct work_struct
*work
)
397 struct cpu_workqueue_struct
*cwq
;
400 if (!test_and_set_bit(WORK_STRUCT_PENDING
, work_data_bits(work
)))
404 * The queueing is in progress, or it is already queued. Try to
405 * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
408 cwq
= get_wq_data(work
);
412 spin_lock_irq(&cwq
->lock
);
413 if (!list_empty(&work
->entry
)) {
415 * This work is queued, but perhaps we locked the wrong cwq.
416 * In that case we must see the new value after rmb(), see
417 * insert_work()->wmb().
420 if (cwq
== get_wq_data(work
)) {
421 list_del_init(&work
->entry
);
425 spin_unlock_irq(&cwq
->lock
);
430 static void wait_on_cpu_work(struct cpu_workqueue_struct
*cwq
,
431 struct work_struct
*work
)
433 struct wq_barrier barr
;
436 spin_lock_irq(&cwq
->lock
);
437 if (unlikely(cwq
->current_work
== work
)) {
438 insert_wq_barrier(cwq
, &barr
, 0);
441 spin_unlock_irq(&cwq
->lock
);
443 if (unlikely(running
))
444 wait_for_completion(&barr
.done
);
447 static void wait_on_work(struct work_struct
*work
)
449 struct cpu_workqueue_struct
*cwq
;
450 struct workqueue_struct
*wq
;
451 const cpumask_t
*cpu_map
;
456 cwq
= get_wq_data(work
);
461 cpu_map
= wq_cpu_map(wq
);
463 for_each_cpu_mask(cpu
, *cpu_map
)
464 wait_on_cpu_work(per_cpu_ptr(wq
->cpu_wq
, cpu
), work
);
468 * cancel_work_sync - block until a work_struct's callback has terminated
469 * @work: the work which is to be flushed
471 * cancel_work_sync() will cancel the work if it is queued. If the work's
472 * callback appears to be running, cancel_work_sync() will block until it
475 * It is possible to use this function if the work re-queues itself. It can
476 * cancel the work even if it migrates to another workqueue, however in that
477 * case it only guarantees that work->func() has completed on the last queued
480 * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not
481 * pending, otherwise it goes into a busy-wait loop until the timer expires.
483 * The caller must ensure that workqueue_struct on which this work was last
484 * queued can't be destroyed before this function returns.
486 void cancel_work_sync(struct work_struct
*work
)
488 while (!try_to_grab_pending(work
))
491 work_clear_pending(work
);
493 EXPORT_SYMBOL_GPL(cancel_work_sync
);
496 * cancel_rearming_delayed_work - reliably kill off a delayed work.
497 * @dwork: the delayed work struct
499 * It is possible to use this function if @dwork rearms itself via queue_work()
500 * or queue_delayed_work(). See also the comment for cancel_work_sync().
502 void cancel_rearming_delayed_work(struct delayed_work
*dwork
)
504 while (!del_timer(&dwork
->timer
) &&
505 !try_to_grab_pending(&dwork
->work
))
507 wait_on_work(&dwork
->work
);
508 work_clear_pending(&dwork
->work
);
510 EXPORT_SYMBOL(cancel_rearming_delayed_work
);
512 static struct workqueue_struct
*keventd_wq __read_mostly
;
515 * schedule_work - put work task in global workqueue
516 * @work: job to be done
518 * This puts a job in the kernel-global workqueue.
520 int fastcall
schedule_work(struct work_struct
*work
)
522 return queue_work(keventd_wq
, work
);
524 EXPORT_SYMBOL(schedule_work
);
527 * schedule_delayed_work - put work task in global workqueue after delay
528 * @dwork: job to be done
529 * @delay: number of jiffies to wait or 0 for immediate execution
531 * After waiting for a given time this puts a job in the kernel-global
534 int fastcall
schedule_delayed_work(struct delayed_work
*dwork
,
537 timer_stats_timer_set_start_info(&dwork
->timer
);
538 return queue_delayed_work(keventd_wq
, dwork
, delay
);
540 EXPORT_SYMBOL(schedule_delayed_work
);
543 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
545 * @dwork: job to be done
546 * @delay: number of jiffies to wait
548 * After waiting for a given time this puts a job in the kernel-global
549 * workqueue on the specified CPU.
551 int schedule_delayed_work_on(int cpu
,
552 struct delayed_work
*dwork
, unsigned long delay
)
554 return queue_delayed_work_on(cpu
, keventd_wq
, dwork
, delay
);
556 EXPORT_SYMBOL(schedule_delayed_work_on
);
559 * schedule_on_each_cpu - call a function on each online CPU from keventd
560 * @func: the function to call
562 * Returns zero on success.
563 * Returns -ve errno on failure.
565 * Appears to be racy against CPU hotplug.
567 * schedule_on_each_cpu() is very slow.
569 int schedule_on_each_cpu(work_func_t func
)
572 struct work_struct
*works
;
574 works
= alloc_percpu(struct work_struct
);
578 preempt_disable(); /* CPU hotplug */
579 for_each_online_cpu(cpu
) {
580 struct work_struct
*work
= per_cpu_ptr(works
, cpu
);
582 INIT_WORK(work
, func
);
583 set_bit(WORK_STRUCT_PENDING
, work_data_bits(work
));
584 __queue_work(per_cpu_ptr(keventd_wq
->cpu_wq
, cpu
), work
);
587 flush_workqueue(keventd_wq
);
592 void flush_scheduled_work(void)
594 flush_workqueue(keventd_wq
);
596 EXPORT_SYMBOL(flush_scheduled_work
);
599 * execute_in_process_context - reliably execute the routine with user context
600 * @fn: the function to execute
601 * @ew: guaranteed storage for the execute work structure (must
602 * be available when the work executes)
604 * Executes the function immediately if process context is available,
605 * otherwise schedules the function for delayed execution.
607 * Returns: 0 - function was executed
608 * 1 - function was scheduled for execution
610 int execute_in_process_context(work_func_t fn
, struct execute_work
*ew
)
612 if (!in_interrupt()) {
617 INIT_WORK(&ew
->work
, fn
);
618 schedule_work(&ew
->work
);
622 EXPORT_SYMBOL_GPL(execute_in_process_context
);
626 return keventd_wq
!= NULL
;
629 int current_is_keventd(void)
631 struct cpu_workqueue_struct
*cwq
;
632 int cpu
= smp_processor_id(); /* preempt-safe: keventd is per-cpu */
637 cwq
= per_cpu_ptr(keventd_wq
->cpu_wq
, cpu
);
638 if (current
== cwq
->thread
)
645 static struct cpu_workqueue_struct
*
646 init_cpu_workqueue(struct workqueue_struct
*wq
, int cpu
)
648 struct cpu_workqueue_struct
*cwq
= per_cpu_ptr(wq
->cpu_wq
, cpu
);
651 spin_lock_init(&cwq
->lock
);
652 INIT_LIST_HEAD(&cwq
->worklist
);
653 init_waitqueue_head(&cwq
->more_work
);
658 static int create_workqueue_thread(struct cpu_workqueue_struct
*cwq
, int cpu
)
660 struct workqueue_struct
*wq
= cwq
->wq
;
661 const char *fmt
= is_single_threaded(wq
) ? "%s" : "%s/%d";
662 struct task_struct
*p
;
664 p
= kthread_create(worker_thread
, cwq
, fmt
, wq
->name
, cpu
);
666 * Nobody can add the work_struct to this cwq,
667 * if (caller is __create_workqueue)
668 * nobody should see this wq
669 * else // caller is CPU_UP_PREPARE
670 * cpu is not on cpu_online_map
671 * so we can abort safely.
677 cwq
->should_stop
= 0;
682 static void start_workqueue_thread(struct cpu_workqueue_struct
*cwq
, int cpu
)
684 struct task_struct
*p
= cwq
->thread
;
688 kthread_bind(p
, cpu
);
693 struct workqueue_struct
*__create_workqueue(const char *name
,
694 int singlethread
, int freezeable
)
696 struct workqueue_struct
*wq
;
697 struct cpu_workqueue_struct
*cwq
;
700 wq
= kzalloc(sizeof(*wq
), GFP_KERNEL
);
704 wq
->cpu_wq
= alloc_percpu(struct cpu_workqueue_struct
);
711 wq
->singlethread
= singlethread
;
712 wq
->freezeable
= freezeable
;
713 INIT_LIST_HEAD(&wq
->list
);
716 cwq
= init_cpu_workqueue(wq
, singlethread_cpu
);
717 err
= create_workqueue_thread(cwq
, singlethread_cpu
);
718 start_workqueue_thread(cwq
, -1);
720 mutex_lock(&workqueue_mutex
);
721 list_add(&wq
->list
, &workqueues
);
723 for_each_possible_cpu(cpu
) {
724 cwq
= init_cpu_workqueue(wq
, cpu
);
725 if (err
|| !cpu_online(cpu
))
727 err
= create_workqueue_thread(cwq
, cpu
);
728 start_workqueue_thread(cwq
, cpu
);
730 mutex_unlock(&workqueue_mutex
);
734 destroy_workqueue(wq
);
739 EXPORT_SYMBOL_GPL(__create_workqueue
);
741 static void cleanup_workqueue_thread(struct cpu_workqueue_struct
*cwq
, int cpu
)
743 struct wq_barrier barr
;
746 spin_lock_irq(&cwq
->lock
);
747 if (cwq
->thread
!= NULL
) {
748 insert_wq_barrier(cwq
, &barr
, 1);
749 cwq
->should_stop
= 1;
752 spin_unlock_irq(&cwq
->lock
);
755 wait_for_completion(&barr
.done
);
757 while (unlikely(cwq
->thread
!= NULL
))
760 * Wait until cwq->thread unlocks cwq->lock,
761 * it won't touch *cwq after that.
764 spin_unlock_wait(&cwq
->lock
);
769 * destroy_workqueue - safely terminate a workqueue
770 * @wq: target workqueue
772 * Safely destroy a workqueue. All work currently pending will be done first.
774 void destroy_workqueue(struct workqueue_struct
*wq
)
776 const cpumask_t
*cpu_map
= wq_cpu_map(wq
);
777 struct cpu_workqueue_struct
*cwq
;
780 mutex_lock(&workqueue_mutex
);
782 mutex_unlock(&workqueue_mutex
);
784 for_each_cpu_mask(cpu
, *cpu_map
) {
785 cwq
= per_cpu_ptr(wq
->cpu_wq
, cpu
);
786 cleanup_workqueue_thread(cwq
, cpu
);
789 free_percpu(wq
->cpu_wq
);
792 EXPORT_SYMBOL_GPL(destroy_workqueue
);
794 static int __devinit
workqueue_cpu_callback(struct notifier_block
*nfb
,
795 unsigned long action
,
798 unsigned int cpu
= (unsigned long)hcpu
;
799 struct cpu_workqueue_struct
*cwq
;
800 struct workqueue_struct
*wq
;
802 action
&= ~CPU_TASKS_FROZEN
;
805 case CPU_LOCK_ACQUIRE
:
806 mutex_lock(&workqueue_mutex
);
809 case CPU_LOCK_RELEASE
:
810 mutex_unlock(&workqueue_mutex
);
814 cpu_set(cpu
, cpu_populated_map
);
817 list_for_each_entry(wq
, &workqueues
, list
) {
818 cwq
= per_cpu_ptr(wq
->cpu_wq
, cpu
);
822 if (!create_workqueue_thread(cwq
, cpu
))
824 printk(KERN_ERR
"workqueue for %i failed\n", cpu
);
828 start_workqueue_thread(cwq
, cpu
);
831 case CPU_UP_CANCELED
:
832 start_workqueue_thread(cwq
, -1);
834 cleanup_workqueue_thread(cwq
, cpu
);
842 void __init
init_workqueues(void)
844 cpu_populated_map
= cpu_online_map
;
845 singlethread_cpu
= first_cpu(cpu_possible_map
);
846 cpu_singlethread_map
= cpumask_of_cpu(singlethread_cpu
);
847 hotcpu_notifier(workqueue_cpu_callback
, 0);
848 keventd_wq
= create_workqueue("events");