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>
35 * The per-CPU workqueue (if single thread, we always use the first
38 * The sequence counters are for flush_scheduled_work(). It wants to wait
39 * until all currently-scheduled works are completed, but it doesn't
40 * want to be livelocked by new, incoming ones. So it waits until
41 * remove_sequence is >= the insert_sequence which pertained when
42 * flush_scheduled_work() was called.
44 struct cpu_workqueue_struct {
48 long remove_sequence; /* Least-recently added (next to run) */
49 long insert_sequence; /* Next to add */
51 struct list_head worklist;
52 wait_queue_head_t more_work;
53 wait_queue_head_t work_done;
55 struct workqueue_struct *wq;
56 struct task_struct *thread;
58 int run_depth; /* Detect run_workqueue() recursion depth */
60 int freezeable; /* Freeze the thread during suspend */
61 } ____cacheline_aligned;
64 * The externally visible workqueue abstraction is an array of
67 struct workqueue_struct {
68 struct cpu_workqueue_struct *cpu_wq;
70 struct list_head list; /* Empty if single thread */
73 /* All the per-cpu workqueues on the system, for hotplug cpu to add/remove
74 threads to each one as cpus come/go. */
75 static DEFINE_MUTEX(workqueue_mutex);
76 static LIST_HEAD(workqueues);
78 static int singlethread_cpu;
80 /* If it's single threaded, it isn't in the list of workqueues. */
81 static inline int is_single_threaded(struct workqueue_struct *wq)
83 return list_empty(&wq->list);
86 static inline void set_wq_data(struct work_struct *work, void *wq)
88 unsigned long new, old, res;
90 /* assume the pending flag is already set and that the task has already
91 * been queued on this workqueue */
92 new = (unsigned long) wq | (1UL << WORK_STRUCT_PENDING);
93 res = work->management;
97 new = (unsigned long) wq;
98 new |= (old & WORK_STRUCT_FLAG_MASK);
99 res = cmpxchg(&work->management, old, new);
100 } while (res != old);
104 static inline void *get_wq_data(struct work_struct *work)
106 return (void *) (work->management & WORK_STRUCT_WQ_DATA_MASK);
109 /* Preempt must be disabled. */
110 static void __queue_work(struct cpu_workqueue_struct *cwq,
111 struct work_struct *work)
115 spin_lock_irqsave(&cwq->lock, flags);
116 set_wq_data(work, cwq);
117 list_add_tail(&work->entry, &cwq->worklist);
118 cwq->insert_sequence++;
119 wake_up(&cwq->more_work);
120 spin_unlock_irqrestore(&cwq->lock, flags);
124 * queue_work - queue work on a workqueue
125 * @wq: workqueue to use
126 * @work: work to queue
128 * Returns 0 if @work was already on a queue, non-zero otherwise.
130 * We queue the work to the CPU it was submitted, but there is no
131 * guarantee that it will be processed by that CPU.
133 int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
135 int ret = 0, cpu = get_cpu();
137 if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
138 if (unlikely(is_single_threaded(wq)))
139 cpu = singlethread_cpu;
140 BUG_ON(!list_empty(&work->entry));
141 __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
147 EXPORT_SYMBOL_GPL(queue_work);
149 static void delayed_work_timer_fn(unsigned long __data)
151 struct delayed_work *dwork = (struct delayed_work *)__data;
152 struct workqueue_struct *wq = get_wq_data(&dwork->work);
153 int cpu = smp_processor_id();
155 if (unlikely(is_single_threaded(wq)))
156 cpu = singlethread_cpu;
158 __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), &dwork->work);
162 * queue_delayed_work - queue work on a workqueue after delay
163 * @wq: workqueue to use
164 * @work: delayable work to queue
165 * @delay: number of jiffies to wait before queueing
167 * Returns 0 if @work was already on a queue, non-zero otherwise.
169 int fastcall queue_delayed_work(struct workqueue_struct *wq,
170 struct delayed_work *dwork, unsigned long delay)
173 struct timer_list *timer = &dwork->timer;
174 struct work_struct *work = &dwork->work;
177 return queue_work(wq, work);
179 if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
180 BUG_ON(timer_pending(timer));
181 BUG_ON(!list_empty(&work->entry));
183 /* This stores wq for the moment, for the timer_fn */
184 set_wq_data(work, wq);
185 timer->expires = jiffies + delay;
186 timer->data = (unsigned long)dwork;
187 timer->function = delayed_work_timer_fn;
193 EXPORT_SYMBOL_GPL(queue_delayed_work);
196 * queue_delayed_work_on - queue work on specific CPU after delay
197 * @cpu: CPU number to execute work on
198 * @wq: workqueue to use
199 * @work: work to queue
200 * @delay: number of jiffies to wait before queueing
202 * Returns 0 if @work was already on a queue, non-zero otherwise.
204 int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
205 struct delayed_work *dwork, unsigned long delay)
208 struct timer_list *timer = &dwork->timer;
209 struct work_struct *work = &dwork->work;
211 if (!test_and_set_bit(WORK_STRUCT_PENDING, &work->management)) {
212 BUG_ON(timer_pending(timer));
213 BUG_ON(!list_empty(&work->entry));
215 /* This stores wq for the moment, for the timer_fn */
216 set_wq_data(work, wq);
217 timer->expires = jiffies + delay;
218 timer->data = (unsigned long)dwork;
219 timer->function = delayed_work_timer_fn;
220 add_timer_on(timer, cpu);
225 EXPORT_SYMBOL_GPL(queue_delayed_work_on);
227 static void run_workqueue(struct cpu_workqueue_struct *cwq)
232 * Keep taking off work from the queue until
235 spin_lock_irqsave(&cwq->lock, flags);
237 if (cwq->run_depth > 3) {
238 /* morton gets to eat his hat */
239 printk("%s: recursion depth exceeded: %d\n",
240 __FUNCTION__, cwq->run_depth);
243 while (!list_empty(&cwq->worklist)) {
244 struct work_struct *work = list_entry(cwq->worklist.next,
245 struct work_struct, entry);
246 work_func_t f = work->func;
248 list_del_init(cwq->worklist.next);
249 spin_unlock_irqrestore(&cwq->lock, flags);
251 BUG_ON(get_wq_data(work) != cwq);
252 if (!test_bit(WORK_STRUCT_NOAUTOREL, &work->management))
256 spin_lock_irqsave(&cwq->lock, flags);
257 cwq->remove_sequence++;
258 wake_up(&cwq->work_done);
261 spin_unlock_irqrestore(&cwq->lock, flags);
264 static int worker_thread(void *__cwq)
266 struct cpu_workqueue_struct *cwq = __cwq;
267 DECLARE_WAITQUEUE(wait, current);
268 struct k_sigaction sa;
271 if (!cwq->freezeable)
272 current->flags |= PF_NOFREEZE;
274 set_user_nice(current, -5);
276 /* Block and flush all signals */
277 sigfillset(&blocked);
278 sigprocmask(SIG_BLOCK, &blocked, NULL);
279 flush_signals(current);
282 * We inherited MPOL_INTERLEAVE from the booting kernel.
283 * Set MPOL_DEFAULT to insure node local allocations.
285 numa_default_policy();
287 /* SIG_IGN makes children autoreap: see do_notify_parent(). */
288 sa.sa.sa_handler = SIG_IGN;
290 siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
291 do_sigaction(SIGCHLD, &sa, (struct k_sigaction *)0);
293 set_current_state(TASK_INTERRUPTIBLE);
294 while (!kthread_should_stop()) {
298 add_wait_queue(&cwq->more_work, &wait);
299 if (list_empty(&cwq->worklist))
302 __set_current_state(TASK_RUNNING);
303 remove_wait_queue(&cwq->more_work, &wait);
305 if (!list_empty(&cwq->worklist))
307 set_current_state(TASK_INTERRUPTIBLE);
309 __set_current_state(TASK_RUNNING);
313 static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
315 if (cwq->thread == current) {
317 * Probably keventd trying to flush its own queue. So simply run
318 * it by hand rather than deadlocking.
323 long sequence_needed;
325 spin_lock_irq(&cwq->lock);
326 sequence_needed = cwq->insert_sequence;
328 while (sequence_needed - cwq->remove_sequence > 0) {
329 prepare_to_wait(&cwq->work_done, &wait,
330 TASK_UNINTERRUPTIBLE);
331 spin_unlock_irq(&cwq->lock);
333 spin_lock_irq(&cwq->lock);
335 finish_wait(&cwq->work_done, &wait);
336 spin_unlock_irq(&cwq->lock);
341 * flush_workqueue - ensure that any scheduled work has run to completion.
342 * @wq: workqueue to flush
344 * Forces execution of the workqueue and blocks until its completion.
345 * This is typically used in driver shutdown handlers.
347 * This function will sample each workqueue's current insert_sequence number and
348 * will sleep until the head sequence is greater than or equal to that. This
349 * means that we sleep until all works which were queued on entry have been
350 * handled, but we are not livelocked by new incoming ones.
352 * This function used to run the workqueues itself. Now we just wait for the
353 * helper threads to do it.
355 void fastcall flush_workqueue(struct workqueue_struct *wq)
359 if (is_single_threaded(wq)) {
360 /* Always use first cpu's area. */
361 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, singlethread_cpu));
365 mutex_lock(&workqueue_mutex);
366 for_each_online_cpu(cpu)
367 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
368 mutex_unlock(&workqueue_mutex);
371 EXPORT_SYMBOL_GPL(flush_workqueue);
373 static struct task_struct *create_workqueue_thread(struct workqueue_struct *wq,
374 int cpu, int freezeable)
376 struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
377 struct task_struct *p;
379 spin_lock_init(&cwq->lock);
382 cwq->insert_sequence = 0;
383 cwq->remove_sequence = 0;
384 cwq->freezeable = freezeable;
385 INIT_LIST_HEAD(&cwq->worklist);
386 init_waitqueue_head(&cwq->more_work);
387 init_waitqueue_head(&cwq->work_done);
389 if (is_single_threaded(wq))
390 p = kthread_create(worker_thread, cwq, "%s", wq->name);
392 p = kthread_create(worker_thread, cwq, "%s/%d", wq->name, cpu);
399 struct workqueue_struct *__create_workqueue(const char *name,
400 int singlethread, int freezeable)
402 int cpu, destroy = 0;
403 struct workqueue_struct *wq;
404 struct task_struct *p;
406 wq = kzalloc(sizeof(*wq), GFP_KERNEL);
410 wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
417 mutex_lock(&workqueue_mutex);
419 INIT_LIST_HEAD(&wq->list);
420 p = create_workqueue_thread(wq, singlethread_cpu, freezeable);
426 list_add(&wq->list, &workqueues);
427 for_each_online_cpu(cpu) {
428 p = create_workqueue_thread(wq, cpu, freezeable);
430 kthread_bind(p, cpu);
436 mutex_unlock(&workqueue_mutex);
439 * Was there any error during startup? If yes then clean up:
442 destroy_workqueue(wq);
447 EXPORT_SYMBOL_GPL(__create_workqueue);
449 static void cleanup_workqueue_thread(struct workqueue_struct *wq, int cpu)
451 struct cpu_workqueue_struct *cwq;
453 struct task_struct *p;
455 cwq = per_cpu_ptr(wq->cpu_wq, cpu);
456 spin_lock_irqsave(&cwq->lock, flags);
459 spin_unlock_irqrestore(&cwq->lock, flags);
465 * destroy_workqueue - safely terminate a workqueue
466 * @wq: target workqueue
468 * Safely destroy a workqueue. All work currently pending will be done first.
470 void destroy_workqueue(struct workqueue_struct *wq)
476 /* We don't need the distraction of CPUs appearing and vanishing. */
477 mutex_lock(&workqueue_mutex);
478 if (is_single_threaded(wq))
479 cleanup_workqueue_thread(wq, singlethread_cpu);
481 for_each_online_cpu(cpu)
482 cleanup_workqueue_thread(wq, cpu);
485 mutex_unlock(&workqueue_mutex);
486 free_percpu(wq->cpu_wq);
489 EXPORT_SYMBOL_GPL(destroy_workqueue);
491 static struct workqueue_struct *keventd_wq;
494 * schedule_work - put work task in global workqueue
495 * @work: job to be done
497 * This puts a job in the kernel-global workqueue.
499 int fastcall schedule_work(struct work_struct *work)
501 return queue_work(keventd_wq, work);
503 EXPORT_SYMBOL(schedule_work);
506 * schedule_delayed_work - put work task in global workqueue after delay
507 * @dwork: job to be done
508 * @delay: number of jiffies to wait or 0 for immediate execution
510 * After waiting for a given time this puts a job in the kernel-global
513 int fastcall schedule_delayed_work(struct delayed_work *dwork, unsigned long delay)
515 return queue_delayed_work(keventd_wq, dwork, delay);
517 EXPORT_SYMBOL(schedule_delayed_work);
520 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
522 * @dwork: job to be done
523 * @delay: number of jiffies to wait
525 * After waiting for a given time this puts a job in the kernel-global
526 * workqueue on the specified CPU.
528 int schedule_delayed_work_on(int cpu,
529 struct delayed_work *dwork, unsigned long delay)
531 return queue_delayed_work_on(cpu, keventd_wq, dwork, delay);
533 EXPORT_SYMBOL(schedule_delayed_work_on);
536 * schedule_on_each_cpu - call a function on each online CPU from keventd
537 * @func: the function to call
539 * Returns zero on success.
540 * Returns -ve errno on failure.
542 * Appears to be racy against CPU hotplug.
544 * schedule_on_each_cpu() is very slow.
546 int schedule_on_each_cpu(work_func_t func)
549 struct work_struct *works;
551 works = alloc_percpu(struct work_struct);
555 mutex_lock(&workqueue_mutex);
556 for_each_online_cpu(cpu) {
557 INIT_WORK(per_cpu_ptr(works, cpu), func);
558 __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu),
559 per_cpu_ptr(works, cpu));
561 mutex_unlock(&workqueue_mutex);
562 flush_workqueue(keventd_wq);
567 void flush_scheduled_work(void)
569 flush_workqueue(keventd_wq);
571 EXPORT_SYMBOL(flush_scheduled_work);
574 * cancel_rearming_delayed_workqueue - reliably kill off a delayed
575 * work whose handler rearms the delayed work.
576 * @wq: the controlling workqueue structure
577 * @dwork: the delayed work struct
579 void cancel_rearming_delayed_workqueue(struct workqueue_struct *wq,
580 struct delayed_work *dwork)
582 while (!cancel_delayed_work(dwork))
585 EXPORT_SYMBOL(cancel_rearming_delayed_workqueue);
588 * cancel_rearming_delayed_work - reliably kill off a delayed keventd
589 * work whose handler rearms the delayed work.
590 * @dwork: the delayed work struct
592 void cancel_rearming_delayed_work(struct delayed_work *dwork)
594 cancel_rearming_delayed_workqueue(keventd_wq, dwork);
596 EXPORT_SYMBOL(cancel_rearming_delayed_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 #ifdef CONFIG_HOTPLUG_CPU
646 /* Take the work from this (downed) CPU. */
647 static void take_over_work(struct workqueue_struct *wq, unsigned int cpu)
649 struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
650 struct list_head list;
651 struct work_struct *work;
653 spin_lock_irq(&cwq->lock);
654 list_replace_init(&cwq->worklist, &list);
656 while (!list_empty(&list)) {
657 printk("Taking work for %s\n", wq->name);
658 work = list_entry(list.next,struct work_struct,entry);
659 list_del(&work->entry);
660 __queue_work(per_cpu_ptr(wq->cpu_wq, smp_processor_id()), work);
662 spin_unlock_irq(&cwq->lock);
665 /* We're holding the cpucontrol mutex here */
666 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
667 unsigned long action,
670 unsigned int hotcpu = (unsigned long)hcpu;
671 struct workqueue_struct *wq;
675 mutex_lock(&workqueue_mutex);
676 /* Create a new workqueue thread for it. */
677 list_for_each_entry(wq, &workqueues, list) {
678 if (!create_workqueue_thread(wq, hotcpu, 0)) {
679 printk("workqueue for %i failed\n", hotcpu);
686 /* Kick off worker threads. */
687 list_for_each_entry(wq, &workqueues, list) {
688 struct cpu_workqueue_struct *cwq;
690 cwq = per_cpu_ptr(wq->cpu_wq, hotcpu);
691 kthread_bind(cwq->thread, hotcpu);
692 wake_up_process(cwq->thread);
694 mutex_unlock(&workqueue_mutex);
697 case CPU_UP_CANCELED:
698 list_for_each_entry(wq, &workqueues, list) {
699 if (!per_cpu_ptr(wq->cpu_wq, hotcpu)->thread)
701 /* Unbind so it can run. */
702 kthread_bind(per_cpu_ptr(wq->cpu_wq, hotcpu)->thread,
703 any_online_cpu(cpu_online_map));
704 cleanup_workqueue_thread(wq, hotcpu);
706 mutex_unlock(&workqueue_mutex);
709 case CPU_DOWN_PREPARE:
710 mutex_lock(&workqueue_mutex);
713 case CPU_DOWN_FAILED:
714 mutex_unlock(&workqueue_mutex);
718 list_for_each_entry(wq, &workqueues, list)
719 cleanup_workqueue_thread(wq, hotcpu);
720 list_for_each_entry(wq, &workqueues, list)
721 take_over_work(wq, hotcpu);
722 mutex_unlock(&workqueue_mutex);
730 void init_workqueues(void)
732 singlethread_cpu = first_cpu(cpu_possible_map);
733 hotcpu_notifier(workqueue_cpu_callback, 0);
734 keventd_wq = create_workqueue("events");