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>
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.
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>
35 #include <linux/lockdep.h>
36 #define CREATE_TRACE_POINTS
37 #include <trace/events/workqueue.h>
40 * The per-CPU workqueue (if single thread, we always use the first
43 struct cpu_workqueue_struct {
47 struct list_head worklist;
48 wait_queue_head_t more_work;
49 struct work_struct *current_work;
51 struct workqueue_struct *wq;
52 struct task_struct *thread;
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 struct lockdep_map lockdep_map;
73 * in_workqueue_context() - in context of specified workqueue?
74 * @wq: the workqueue of interest
76 * Checks lockdep state to see if the current task is executing from
77 * within a workqueue item. This function exists only if lockdep is
80 int in_workqueue_context(struct workqueue_struct *wq)
82 return lock_is_held(&wq->lockdep_map);
86 #ifdef CONFIG_DEBUG_OBJECTS_WORK
88 static struct debug_obj_descr work_debug_descr;
91 * fixup_init is called when:
92 * - an active object is initialized
94 static int work_fixup_init(void *addr, enum debug_obj_state state)
96 struct work_struct *work = addr;
99 case ODEBUG_STATE_ACTIVE:
100 cancel_work_sync(work);
101 debug_object_init(work, &work_debug_descr);
109 * fixup_activate is called when:
110 * - an active object is activated
111 * - an unknown object is activated (might be a statically initialized object)
113 static int work_fixup_activate(void *addr, enum debug_obj_state state)
115 struct work_struct *work = addr;
119 case ODEBUG_STATE_NOTAVAILABLE:
121 * This is not really a fixup. The work struct was
122 * statically initialized. We just make sure that it
123 * is tracked in the object tracker.
125 if (test_bit(WORK_STRUCT_STATIC, work_data_bits(work))) {
126 debug_object_init(work, &work_debug_descr);
127 debug_object_activate(work, &work_debug_descr);
133 case ODEBUG_STATE_ACTIVE:
142 * fixup_free is called when:
143 * - an active object is freed
145 static int work_fixup_free(void *addr, enum debug_obj_state state)
147 struct work_struct *work = addr;
150 case ODEBUG_STATE_ACTIVE:
151 cancel_work_sync(work);
152 debug_object_free(work, &work_debug_descr);
159 static struct debug_obj_descr work_debug_descr = {
160 .name = "work_struct",
161 .fixup_init = work_fixup_init,
162 .fixup_activate = work_fixup_activate,
163 .fixup_free = work_fixup_free,
166 static inline void debug_work_activate(struct work_struct *work)
168 debug_object_activate(work, &work_debug_descr);
171 static inline void debug_work_deactivate(struct work_struct *work)
173 debug_object_deactivate(work, &work_debug_descr);
176 void __init_work(struct work_struct *work, int onstack)
179 debug_object_init_on_stack(work, &work_debug_descr);
181 debug_object_init(work, &work_debug_descr);
183 EXPORT_SYMBOL_GPL(__init_work);
185 void destroy_work_on_stack(struct work_struct *work)
187 debug_object_free(work, &work_debug_descr);
189 EXPORT_SYMBOL_GPL(destroy_work_on_stack);
192 static inline void debug_work_activate(struct work_struct *work) { }
193 static inline void debug_work_deactivate(struct work_struct *work) { }
196 /* Serializes the accesses to the list of workqueues. */
197 static DEFINE_SPINLOCK(workqueue_lock);
198 static LIST_HEAD(workqueues);
200 static int singlethread_cpu __read_mostly;
201 static const struct cpumask *cpu_singlethread_map __read_mostly;
203 * _cpu_down() first removes CPU from cpu_online_map, then CPU_DEAD
204 * flushes cwq->worklist. This means that flush_workqueue/wait_on_work
205 * which comes in between can't use for_each_online_cpu(). We could
206 * use cpu_possible_map, the cpumask below is more a documentation
209 static cpumask_var_t cpu_populated_map __read_mostly;
211 /* If it's single threaded, it isn't in the list of workqueues. */
212 static inline int is_wq_single_threaded(struct workqueue_struct *wq)
214 return wq->singlethread;
217 static const struct cpumask *wq_cpu_map(struct workqueue_struct *wq)
219 return is_wq_single_threaded(wq)
220 ? cpu_singlethread_map : cpu_populated_map;
224 struct cpu_workqueue_struct *wq_per_cpu(struct workqueue_struct *wq, int cpu)
226 if (unlikely(is_wq_single_threaded(wq)))
227 cpu = singlethread_cpu;
228 return per_cpu_ptr(wq->cpu_wq, cpu);
232 * Set the workqueue on which a work item is to be run
233 * - Must *only* be called if the pending flag is set
235 static inline void set_wq_data(struct work_struct *work,
236 struct cpu_workqueue_struct *cwq)
240 BUG_ON(!work_pending(work));
242 new = (unsigned long) cwq | (1UL << WORK_STRUCT_PENDING);
243 new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work);
244 atomic_long_set(&work->data, new);
248 * Clear WORK_STRUCT_PENDING and the workqueue on which it was queued.
250 static inline void clear_wq_data(struct work_struct *work)
252 unsigned long flags = *work_data_bits(work) &
253 (1UL << WORK_STRUCT_STATIC);
254 atomic_long_set(&work->data, flags);
258 struct cpu_workqueue_struct *get_wq_data(struct work_struct *work)
260 return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
263 static void insert_work(struct cpu_workqueue_struct *cwq,
264 struct work_struct *work, struct list_head *head)
266 trace_workqueue_insertion(cwq->thread, work);
268 set_wq_data(work, cwq);
270 * Ensure that we get the right work->data if we see the
271 * result of list_add() below, see try_to_grab_pending().
274 list_add_tail(&work->entry, head);
275 wake_up(&cwq->more_work);
278 static void __queue_work(struct cpu_workqueue_struct *cwq,
279 struct work_struct *work)
283 debug_work_activate(work);
284 spin_lock_irqsave(&cwq->lock, flags);
285 insert_work(cwq, work, &cwq->worklist);
286 spin_unlock_irqrestore(&cwq->lock, flags);
290 * queue_work - queue work on a workqueue
291 * @wq: workqueue to use
292 * @work: work to queue
294 * Returns 0 if @work was already on a queue, non-zero otherwise.
296 * We queue the work to the CPU on which it was submitted, but if the CPU dies
297 * it can be processed by another CPU.
299 int queue_work(struct workqueue_struct *wq, struct work_struct *work)
303 ret = queue_work_on(get_cpu(), wq, work);
308 EXPORT_SYMBOL_GPL(queue_work);
311 * queue_work_on - queue work on specific cpu
312 * @cpu: CPU number to execute work on
313 * @wq: workqueue to use
314 * @work: work to queue
316 * Returns 0 if @work was already on a queue, non-zero otherwise.
318 * We queue the work to a specific CPU, the caller must ensure it
322 queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work)
326 if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
327 BUG_ON(!list_empty(&work->entry));
328 __queue_work(wq_per_cpu(wq, cpu), work);
333 EXPORT_SYMBOL_GPL(queue_work_on);
335 static void delayed_work_timer_fn(unsigned long __data)
337 struct delayed_work *dwork = (struct delayed_work *)__data;
338 struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work);
339 struct workqueue_struct *wq = cwq->wq;
341 __queue_work(wq_per_cpu(wq, smp_processor_id()), &dwork->work);
345 * queue_delayed_work - queue work on a workqueue after delay
346 * @wq: workqueue to use
347 * @dwork: delayable work to queue
348 * @delay: number of jiffies to wait before queueing
350 * Returns 0 if @work was already on a queue, non-zero otherwise.
352 int queue_delayed_work(struct workqueue_struct *wq,
353 struct delayed_work *dwork, unsigned long delay)
356 return queue_work(wq, &dwork->work);
358 return queue_delayed_work_on(-1, wq, dwork, delay);
360 EXPORT_SYMBOL_GPL(queue_delayed_work);
363 * queue_delayed_work_on - queue work on specific CPU after delay
364 * @cpu: CPU number to execute work on
365 * @wq: workqueue to use
366 * @dwork: work to queue
367 * @delay: number of jiffies to wait before queueing
369 * Returns 0 if @work was already on a queue, non-zero otherwise.
371 int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
372 struct delayed_work *dwork, unsigned long delay)
375 struct timer_list *timer = &dwork->timer;
376 struct work_struct *work = &dwork->work;
378 if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
379 BUG_ON(timer_pending(timer));
380 BUG_ON(!list_empty(&work->entry));
382 timer_stats_timer_set_start_info(&dwork->timer);
384 /* This stores cwq for the moment, for the timer_fn */
385 set_wq_data(work, wq_per_cpu(wq, raw_smp_processor_id()));
386 timer->expires = jiffies + delay;
387 timer->data = (unsigned long)dwork;
388 timer->function = delayed_work_timer_fn;
390 if (unlikely(cpu >= 0))
391 add_timer_on(timer, cpu);
398 EXPORT_SYMBOL_GPL(queue_delayed_work_on);
400 static void run_workqueue(struct cpu_workqueue_struct *cwq)
402 spin_lock_irq(&cwq->lock);
403 while (!list_empty(&cwq->worklist)) {
404 struct work_struct *work = list_entry(cwq->worklist.next,
405 struct work_struct, entry);
406 work_func_t f = work->func;
407 #ifdef CONFIG_LOCKDEP
409 * It is permissible to free the struct work_struct
410 * from inside the function that is called from it,
411 * this we need to take into account for lockdep too.
412 * To avoid bogus "held lock freed" warnings as well
413 * as problems when looking into work->lockdep_map,
414 * make a copy and use that here.
416 struct lockdep_map lockdep_map = work->lockdep_map;
418 trace_workqueue_execution(cwq->thread, work);
419 debug_work_deactivate(work);
420 cwq->current_work = work;
421 list_del_init(cwq->worklist.next);
422 spin_unlock_irq(&cwq->lock);
424 BUG_ON(get_wq_data(work) != cwq);
425 work_clear_pending(work);
426 lock_map_acquire(&cwq->wq->lockdep_map);
427 lock_map_acquire(&lockdep_map);
429 lock_map_release(&lockdep_map);
430 lock_map_release(&cwq->wq->lockdep_map);
432 if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
433 printk(KERN_ERR "BUG: workqueue leaked lock or atomic: "
435 current->comm, preempt_count(),
436 task_pid_nr(current));
437 printk(KERN_ERR " last function: ");
438 print_symbol("%s\n", (unsigned long)f);
439 debug_show_held_locks(current);
443 spin_lock_irq(&cwq->lock);
444 cwq->current_work = NULL;
446 spin_unlock_irq(&cwq->lock);
449 static int worker_thread(void *__cwq)
451 struct cpu_workqueue_struct *cwq = __cwq;
454 if (cwq->wq->freezeable)
458 prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE);
459 if (!freezing(current) &&
460 !kthread_should_stop() &&
461 list_empty(&cwq->worklist))
463 finish_wait(&cwq->more_work, &wait);
467 if (kthread_should_stop())
477 struct work_struct work;
478 struct completion done;
481 static void wq_barrier_func(struct work_struct *work)
483 struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
484 complete(&barr->done);
487 static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
488 struct wq_barrier *barr, struct list_head *head)
491 * debugobject calls are safe here even with cwq->lock locked
492 * as we know for sure that this will not trigger any of the
493 * checks and call back into the fixup functions where we
496 INIT_WORK_ON_STACK(&barr->work, wq_barrier_func);
497 __set_bit(WORK_STRUCT_PENDING, work_data_bits(&barr->work));
499 init_completion(&barr->done);
501 debug_work_activate(&barr->work);
502 insert_work(cwq, &barr->work, head);
505 static int flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
508 struct wq_barrier barr;
510 WARN_ON(cwq->thread == current);
512 spin_lock_irq(&cwq->lock);
513 if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) {
514 insert_wq_barrier(cwq, &barr, &cwq->worklist);
517 spin_unlock_irq(&cwq->lock);
520 wait_for_completion(&barr.done);
521 destroy_work_on_stack(&barr.work);
528 * flush_workqueue - ensure that any scheduled work has run to completion.
529 * @wq: workqueue to flush
531 * Forces execution of the workqueue and blocks until its completion.
532 * This is typically used in driver shutdown handlers.
534 * We sleep until all works which were queued on entry have been handled,
535 * but we are not livelocked by new incoming ones.
537 * This function used to run the workqueues itself. Now we just wait for the
538 * helper threads to do it.
540 void flush_workqueue(struct workqueue_struct *wq)
542 const struct cpumask *cpu_map = wq_cpu_map(wq);
546 lock_map_acquire(&wq->lockdep_map);
547 lock_map_release(&wq->lockdep_map);
548 for_each_cpu(cpu, cpu_map)
549 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
551 EXPORT_SYMBOL_GPL(flush_workqueue);
554 * flush_work - block until a work_struct's callback has terminated
555 * @work: the work which is to be flushed
557 * Returns false if @work has already terminated.
559 * It is expected that, prior to calling flush_work(), the caller has
560 * arranged for the work to not be requeued, otherwise it doesn't make
561 * sense to use this function.
563 int flush_work(struct work_struct *work)
565 struct cpu_workqueue_struct *cwq;
566 struct list_head *prev;
567 struct wq_barrier barr;
570 cwq = get_wq_data(work);
574 lock_map_acquire(&cwq->wq->lockdep_map);
575 lock_map_release(&cwq->wq->lockdep_map);
578 spin_lock_irq(&cwq->lock);
579 if (!list_empty(&work->entry)) {
581 * See the comment near try_to_grab_pending()->smp_rmb().
582 * If it was re-queued under us we are not going to wait.
585 if (unlikely(cwq != get_wq_data(work)))
589 if (cwq->current_work != work)
591 prev = &cwq->worklist;
593 insert_wq_barrier(cwq, &barr, prev->next);
595 spin_unlock_irq(&cwq->lock);
599 wait_for_completion(&barr.done);
600 destroy_work_on_stack(&barr.work);
603 EXPORT_SYMBOL_GPL(flush_work);
606 * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit,
607 * so this work can't be re-armed in any way.
609 static int try_to_grab_pending(struct work_struct *work)
611 struct cpu_workqueue_struct *cwq;
614 if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work)))
618 * The queueing is in progress, or it is already queued. Try to
619 * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
622 cwq = get_wq_data(work);
626 spin_lock_irq(&cwq->lock);
627 if (!list_empty(&work->entry)) {
629 * This work is queued, but perhaps we locked the wrong cwq.
630 * In that case we must see the new value after rmb(), see
631 * insert_work()->wmb().
634 if (cwq == get_wq_data(work)) {
635 debug_work_deactivate(work);
636 list_del_init(&work->entry);
640 spin_unlock_irq(&cwq->lock);
645 static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq,
646 struct work_struct *work)
648 struct wq_barrier barr;
651 spin_lock_irq(&cwq->lock);
652 if (unlikely(cwq->current_work == work)) {
653 insert_wq_barrier(cwq, &barr, cwq->worklist.next);
656 spin_unlock_irq(&cwq->lock);
658 if (unlikely(running)) {
659 wait_for_completion(&barr.done);
660 destroy_work_on_stack(&barr.work);
664 static void wait_on_work(struct work_struct *work)
666 struct cpu_workqueue_struct *cwq;
667 struct workqueue_struct *wq;
668 const struct cpumask *cpu_map;
673 lock_map_acquire(&work->lockdep_map);
674 lock_map_release(&work->lockdep_map);
676 cwq = get_wq_data(work);
681 cpu_map = wq_cpu_map(wq);
683 for_each_cpu(cpu, cpu_map)
684 wait_on_cpu_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
687 static int __cancel_work_timer(struct work_struct *work,
688 struct timer_list* timer)
693 ret = (timer && likely(del_timer(timer)));
695 ret = try_to_grab_pending(work);
697 } while (unlikely(ret < 0));
704 * cancel_work_sync - block until a work_struct's callback has terminated
705 * @work: the work which is to be flushed
707 * Returns true if @work was pending.
709 * cancel_work_sync() will cancel the work if it is queued. If the work's
710 * callback appears to be running, cancel_work_sync() will block until it
713 * It is possible to use this function if the work re-queues itself. It can
714 * cancel the work even if it migrates to another workqueue, however in that
715 * case it only guarantees that work->func() has completed on the last queued
718 * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not
719 * pending, otherwise it goes into a busy-wait loop until the timer expires.
721 * The caller must ensure that workqueue_struct on which this work was last
722 * queued can't be destroyed before this function returns.
724 int cancel_work_sync(struct work_struct *work)
726 return __cancel_work_timer(work, NULL);
728 EXPORT_SYMBOL_GPL(cancel_work_sync);
731 * cancel_delayed_work_sync - reliably kill off a delayed work.
732 * @dwork: the delayed work struct
734 * Returns true if @dwork was pending.
736 * It is possible to use this function if @dwork rearms itself via queue_work()
737 * or queue_delayed_work(). See also the comment for cancel_work_sync().
739 int cancel_delayed_work_sync(struct delayed_work *dwork)
741 return __cancel_work_timer(&dwork->work, &dwork->timer);
743 EXPORT_SYMBOL(cancel_delayed_work_sync);
745 static struct workqueue_struct *keventd_wq __read_mostly;
748 * schedule_work - put work task in global workqueue
749 * @work: job to be done
751 * Returns zero if @work was already on the kernel-global workqueue and
752 * non-zero otherwise.
754 * This puts a job in the kernel-global workqueue if it was not already
755 * queued and leaves it in the same position on the kernel-global
756 * workqueue otherwise.
758 int schedule_work(struct work_struct *work)
760 return queue_work(keventd_wq, work);
762 EXPORT_SYMBOL(schedule_work);
765 * schedule_work_on - put work task on a specific cpu
766 * @cpu: cpu to put the work task on
767 * @work: job to be done
769 * This puts a job on a specific cpu
771 int schedule_work_on(int cpu, struct work_struct *work)
773 return queue_work_on(cpu, keventd_wq, work);
775 EXPORT_SYMBOL(schedule_work_on);
778 * schedule_delayed_work - put work task in global workqueue after delay
779 * @dwork: job to be done
780 * @delay: number of jiffies to wait or 0 for immediate execution
782 * After waiting for a given time this puts a job in the kernel-global
785 int schedule_delayed_work(struct delayed_work *dwork,
788 return queue_delayed_work(keventd_wq, dwork, delay);
790 EXPORT_SYMBOL(schedule_delayed_work);
793 * flush_delayed_work - block until a dwork_struct's callback has terminated
794 * @dwork: the delayed work which is to be flushed
796 * Any timeout is cancelled, and any pending work is run immediately.
798 void flush_delayed_work(struct delayed_work *dwork)
800 if (del_timer_sync(&dwork->timer)) {
801 struct cpu_workqueue_struct *cwq;
802 cwq = wq_per_cpu(get_wq_data(&dwork->work)->wq, get_cpu());
803 __queue_work(cwq, &dwork->work);
806 flush_work(&dwork->work);
808 EXPORT_SYMBOL(flush_delayed_work);
811 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
813 * @dwork: job to be done
814 * @delay: number of jiffies to wait
816 * After waiting for a given time this puts a job in the kernel-global
817 * workqueue on the specified CPU.
819 int schedule_delayed_work_on(int cpu,
820 struct delayed_work *dwork, unsigned long delay)
822 return queue_delayed_work_on(cpu, keventd_wq, dwork, delay);
824 EXPORT_SYMBOL(schedule_delayed_work_on);
827 * schedule_on_each_cpu - call a function on each online CPU from keventd
828 * @func: the function to call
830 * Returns zero on success.
831 * Returns -ve errno on failure.
833 * schedule_on_each_cpu() is very slow.
835 int schedule_on_each_cpu(work_func_t func)
839 struct work_struct *works;
841 works = alloc_percpu(struct work_struct);
848 * When running in keventd don't schedule a work item on
849 * itself. Can just call directly because the work queue is
850 * already bound. This also is faster.
852 if (current_is_keventd())
853 orig = raw_smp_processor_id();
855 for_each_online_cpu(cpu) {
856 struct work_struct *work = per_cpu_ptr(works, cpu);
858 INIT_WORK(work, func);
860 schedule_work_on(cpu, work);
863 func(per_cpu_ptr(works, orig));
865 for_each_online_cpu(cpu)
866 flush_work(per_cpu_ptr(works, cpu));
874 * flush_scheduled_work - ensure that any scheduled work has run to completion.
876 * Forces execution of the kernel-global workqueue and blocks until its
879 * Think twice before calling this function! It's very easy to get into
880 * trouble if you don't take great care. Either of the following situations
881 * will lead to deadlock:
883 * One of the work items currently on the workqueue needs to acquire
884 * a lock held by your code or its caller.
886 * Your code is running in the context of a work routine.
888 * They will be detected by lockdep when they occur, but the first might not
889 * occur very often. It depends on what work items are on the workqueue and
890 * what locks they need, which you have no control over.
892 * In most situations flushing the entire workqueue is overkill; you merely
893 * need to know that a particular work item isn't queued and isn't running.
894 * In such cases you should use cancel_delayed_work_sync() or
895 * cancel_work_sync() instead.
897 void flush_scheduled_work(void)
899 flush_workqueue(keventd_wq);
901 EXPORT_SYMBOL(flush_scheduled_work);
904 * execute_in_process_context - reliably execute the routine with user context
905 * @fn: the function to execute
906 * @ew: guaranteed storage for the execute work structure (must
907 * be available when the work executes)
909 * Executes the function immediately if process context is available,
910 * otherwise schedules the function for delayed execution.
912 * Returns: 0 - function was executed
913 * 1 - function was scheduled for execution
915 int execute_in_process_context(work_func_t fn, struct execute_work *ew)
917 if (!in_interrupt()) {
922 INIT_WORK(&ew->work, fn);
923 schedule_work(&ew->work);
927 EXPORT_SYMBOL_GPL(execute_in_process_context);
931 return keventd_wq != NULL;
934 int current_is_keventd(void)
936 struct cpu_workqueue_struct *cwq;
937 int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */
942 cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
943 if (current == cwq->thread)
950 static struct cpu_workqueue_struct *
951 init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
953 struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
956 spin_lock_init(&cwq->lock);
957 INIT_LIST_HEAD(&cwq->worklist);
958 init_waitqueue_head(&cwq->more_work);
963 static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
965 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
966 struct workqueue_struct *wq = cwq->wq;
967 const char *fmt = is_wq_single_threaded(wq) ? "%s" : "%s/%d";
968 struct task_struct *p;
970 p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu);
972 * Nobody can add the work_struct to this cwq,
973 * if (caller is __create_workqueue)
974 * nobody should see this wq
975 * else // caller is CPU_UP_PREPARE
976 * cpu is not on cpu_online_map
977 * so we can abort safely.
982 sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m);
985 trace_workqueue_creation(cwq->thread, cpu);
990 static void start_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
992 struct task_struct *p = cwq->thread;
996 kthread_bind(p, cpu);
1001 struct workqueue_struct *__create_workqueue_key(const char *name,
1005 struct lock_class_key *key,
1006 const char *lock_name)
1008 struct workqueue_struct *wq;
1009 struct cpu_workqueue_struct *cwq;
1012 wq = kzalloc(sizeof(*wq), GFP_KERNEL);
1016 wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
1023 lockdep_init_map(&wq->lockdep_map, lock_name, key, 0);
1024 wq->singlethread = singlethread;
1025 wq->freezeable = freezeable;
1027 INIT_LIST_HEAD(&wq->list);
1030 cwq = init_cpu_workqueue(wq, singlethread_cpu);
1031 err = create_workqueue_thread(cwq, singlethread_cpu);
1032 start_workqueue_thread(cwq, -1);
1034 cpu_maps_update_begin();
1036 * We must place this wq on list even if the code below fails.
1037 * cpu_down(cpu) can remove cpu from cpu_populated_map before
1038 * destroy_workqueue() takes the lock, in that case we leak
1041 spin_lock(&workqueue_lock);
1042 list_add(&wq->list, &workqueues);
1043 spin_unlock(&workqueue_lock);
1045 * We must initialize cwqs for each possible cpu even if we
1046 * are going to call destroy_workqueue() finally. Otherwise
1047 * cpu_up() can hit the uninitialized cwq once we drop the
1050 for_each_possible_cpu(cpu) {
1051 cwq = init_cpu_workqueue(wq, cpu);
1052 if (err || !cpu_online(cpu))
1054 err = create_workqueue_thread(cwq, cpu);
1055 start_workqueue_thread(cwq, cpu);
1057 cpu_maps_update_done();
1061 destroy_workqueue(wq);
1066 EXPORT_SYMBOL_GPL(__create_workqueue_key);
1068 static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq)
1071 * Our caller is either destroy_workqueue() or CPU_POST_DEAD,
1072 * cpu_add_remove_lock protects cwq->thread.
1074 if (cwq->thread == NULL)
1077 lock_map_acquire(&cwq->wq->lockdep_map);
1078 lock_map_release(&cwq->wq->lockdep_map);
1080 flush_cpu_workqueue(cwq);
1082 * If the caller is CPU_POST_DEAD and cwq->worklist was not empty,
1083 * a concurrent flush_workqueue() can insert a barrier after us.
1084 * However, in that case run_workqueue() won't return and check
1085 * kthread_should_stop() until it flushes all work_struct's.
1086 * When ->worklist becomes empty it is safe to exit because no
1087 * more work_structs can be queued on this cwq: flush_workqueue
1088 * checks list_empty(), and a "normal" queue_work() can't use
1091 trace_workqueue_destruction(cwq->thread);
1092 kthread_stop(cwq->thread);
1097 * destroy_workqueue - safely terminate a workqueue
1098 * @wq: target workqueue
1100 * Safely destroy a workqueue. All work currently pending will be done first.
1102 void destroy_workqueue(struct workqueue_struct *wq)
1104 const struct cpumask *cpu_map = wq_cpu_map(wq);
1107 cpu_maps_update_begin();
1108 spin_lock(&workqueue_lock);
1109 list_del(&wq->list);
1110 spin_unlock(&workqueue_lock);
1112 for_each_cpu(cpu, cpu_map)
1113 cleanup_workqueue_thread(per_cpu_ptr(wq->cpu_wq, cpu));
1114 cpu_maps_update_done();
1116 free_percpu(wq->cpu_wq);
1119 EXPORT_SYMBOL_GPL(destroy_workqueue);
1121 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
1122 unsigned long action,
1125 unsigned int cpu = (unsigned long)hcpu;
1126 struct cpu_workqueue_struct *cwq;
1127 struct workqueue_struct *wq;
1130 action &= ~CPU_TASKS_FROZEN;
1133 case CPU_UP_PREPARE:
1134 cpumask_set_cpu(cpu, cpu_populated_map);
1137 list_for_each_entry(wq, &workqueues, list) {
1138 cwq = per_cpu_ptr(wq->cpu_wq, cpu);
1141 case CPU_UP_PREPARE:
1142 err = create_workqueue_thread(cwq, cpu);
1145 printk(KERN_ERR "workqueue [%s] for %i failed\n",
1147 action = CPU_UP_CANCELED;
1152 start_workqueue_thread(cwq, cpu);
1155 case CPU_UP_CANCELED:
1156 start_workqueue_thread(cwq, -1);
1158 cleanup_workqueue_thread(cwq);
1164 case CPU_UP_CANCELED:
1166 cpumask_clear_cpu(cpu, cpu_populated_map);
1169 return notifier_from_errno(err);
1174 struct work_for_cpu {
1175 struct completion completion;
1181 static int do_work_for_cpu(void *_wfc)
1183 struct work_for_cpu *wfc = _wfc;
1184 wfc->ret = wfc->fn(wfc->arg);
1185 complete(&wfc->completion);
1190 * work_on_cpu - run a function in user context on a particular cpu
1191 * @cpu: the cpu to run on
1192 * @fn: the function to run
1193 * @arg: the function arg
1195 * This will return the value @fn returns.
1196 * It is up to the caller to ensure that the cpu doesn't go offline.
1197 * The caller must not hold any locks which would prevent @fn from completing.
1199 long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg)
1201 struct task_struct *sub_thread;
1202 struct work_for_cpu wfc = {
1203 .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion),
1208 sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu");
1209 if (IS_ERR(sub_thread))
1210 return PTR_ERR(sub_thread);
1211 kthread_bind(sub_thread, cpu);
1212 wake_up_process(sub_thread);
1213 wait_for_completion(&wfc.completion);
1216 EXPORT_SYMBOL_GPL(work_on_cpu);
1217 #endif /* CONFIG_SMP */
1219 void __init init_workqueues(void)
1221 alloc_cpumask_var(&cpu_populated_map, GFP_KERNEL);
1223 cpumask_copy(cpu_populated_map, cpu_online_mask);
1224 singlethread_cpu = cpumask_first(cpu_possible_mask);
1225 cpu_singlethread_map = cpumask_of(singlethread_cpu);
1226 hotcpu_notifier(workqueue_cpu_callback, 0);
1227 keventd_wq = create_workqueue("events");
1228 BUG_ON(!keventd_wq);