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 * Set the workqueue on which a work item is to be run
91 * - Must *only* be called if the pending flag is set
93 static inline void set_wq_data(struct work_struct *work,
94 struct cpu_workqueue_struct *cwq)
98 BUG_ON(!work_pending(work));
100 new = (unsigned long) cwq | (1UL << WORK_STRUCT_PENDING);
101 new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work);
102 atomic_long_set(&work->data, new);
106 struct cpu_workqueue_struct *get_wq_data(struct work_struct *work)
108 return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
111 static void insert_work(struct cpu_workqueue_struct *cwq,
112 struct work_struct *work, int tail)
114 set_wq_data(work, cwq);
116 list_add_tail(&work->entry, &cwq->worklist);
118 list_add(&work->entry, &cwq->worklist);
119 wake_up(&cwq->more_work);
122 /* Preempt must be disabled. */
123 static void __queue_work(struct cpu_workqueue_struct *cwq,
124 struct work_struct *work)
128 spin_lock_irqsave(&cwq->lock, flags);
129 insert_work(cwq, work, 1);
130 spin_unlock_irqrestore(&cwq->lock, flags);
134 * queue_work - queue work on a workqueue
135 * @wq: workqueue to use
136 * @work: work to queue
138 * Returns 0 if @work was already on a queue, non-zero otherwise.
140 * We queue the work to the CPU it was submitted, but there is no
141 * guarantee that it will be processed by that CPU.
143 int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
145 int ret = 0, cpu = get_cpu();
147 if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
148 if (unlikely(is_single_threaded(wq)))
149 cpu = singlethread_cpu;
150 BUG_ON(!list_empty(&work->entry));
151 __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
157 EXPORT_SYMBOL_GPL(queue_work);
159 void delayed_work_timer_fn(unsigned long __data)
161 struct delayed_work *dwork = (struct delayed_work *)__data;
162 struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work);
163 struct workqueue_struct *wq = cwq->wq;
164 int cpu = smp_processor_id();
166 if (unlikely(is_single_threaded(wq)))
167 cpu = singlethread_cpu;
169 __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), &dwork->work);
173 * queue_delayed_work - queue work on a workqueue after delay
174 * @wq: workqueue to use
175 * @dwork: delayable work to queue
176 * @delay: number of jiffies to wait before queueing
178 * Returns 0 if @work was already on a queue, non-zero otherwise.
180 int fastcall queue_delayed_work(struct workqueue_struct *wq,
181 struct delayed_work *dwork, unsigned long delay)
183 timer_stats_timer_set_start_info(&dwork->timer);
185 return queue_work(wq, &dwork->work);
187 return queue_delayed_work_on(-1, wq, dwork, delay);
189 EXPORT_SYMBOL_GPL(queue_delayed_work);
192 * queue_delayed_work_on - queue work on specific CPU after delay
193 * @cpu: CPU number to execute work on
194 * @wq: workqueue to use
195 * @dwork: work to queue
196 * @delay: number of jiffies to wait before queueing
198 * Returns 0 if @work was already on a queue, non-zero otherwise.
200 int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
201 struct delayed_work *dwork, unsigned long delay)
204 struct timer_list *timer = &dwork->timer;
205 struct work_struct *work = &dwork->work;
207 if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
208 BUG_ON(timer_pending(timer));
209 BUG_ON(!list_empty(&work->entry));
211 /* This stores cwq for the moment, for the timer_fn */
213 per_cpu_ptr(wq->cpu_wq, wq->singlethread ?
214 singlethread_cpu : raw_smp_processor_id()));
215 timer->expires = jiffies + delay;
216 timer->data = (unsigned long)dwork;
217 timer->function = delayed_work_timer_fn;
219 if (unlikely(cpu >= 0))
220 add_timer_on(timer, cpu);
227 EXPORT_SYMBOL_GPL(queue_delayed_work_on);
229 static void run_workqueue(struct cpu_workqueue_struct *cwq)
231 spin_lock_irq(&cwq->lock);
233 if (cwq->run_depth > 3) {
234 /* morton gets to eat his hat */
235 printk("%s: recursion depth exceeded: %d\n",
236 __FUNCTION__, cwq->run_depth);
239 while (!list_empty(&cwq->worklist)) {
240 struct work_struct *work = list_entry(cwq->worklist.next,
241 struct work_struct, entry);
242 work_func_t f = work->func;
244 cwq->current_work = work;
245 list_del_init(cwq->worklist.next);
246 spin_unlock_irq(&cwq->lock);
248 BUG_ON(get_wq_data(work) != cwq);
249 if (!test_bit(WORK_STRUCT_NOAUTOREL, work_data_bits(work)))
253 if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
254 printk(KERN_ERR "BUG: workqueue leaked lock or atomic: "
256 current->comm, preempt_count(),
258 printk(KERN_ERR " last function: ");
259 print_symbol("%s\n", (unsigned long)f);
260 debug_show_held_locks(current);
264 spin_lock_irq(&cwq->lock);
265 cwq->current_work = NULL;
268 spin_unlock_irq(&cwq->lock);
272 * NOTE: the caller must not touch *cwq if this func returns true
274 static int cwq_should_stop(struct cpu_workqueue_struct *cwq)
276 int should_stop = cwq->should_stop;
278 if (unlikely(should_stop)) {
279 spin_lock_irq(&cwq->lock);
280 should_stop = cwq->should_stop && list_empty(&cwq->worklist);
283 spin_unlock_irq(&cwq->lock);
289 static int worker_thread(void *__cwq)
291 struct cpu_workqueue_struct *cwq = __cwq;
293 struct k_sigaction sa;
296 if (!cwq->wq->freezeable)
297 current->flags |= PF_NOFREEZE;
299 set_user_nice(current, -5);
301 /* Block and flush all signals */
302 sigfillset(&blocked);
303 sigprocmask(SIG_BLOCK, &blocked, NULL);
304 flush_signals(current);
307 * We inherited MPOL_INTERLEAVE from the booting kernel.
308 * Set MPOL_DEFAULT to insure node local allocations.
310 numa_default_policy();
312 /* SIG_IGN makes children autoreap: see do_notify_parent(). */
313 sa.sa.sa_handler = SIG_IGN;
315 siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
316 do_sigaction(SIGCHLD, &sa, (struct k_sigaction *)0);
319 if (cwq->wq->freezeable)
322 prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE);
323 if (!cwq->should_stop && list_empty(&cwq->worklist))
325 finish_wait(&cwq->more_work, &wait);
327 if (cwq_should_stop(cwq))
337 struct work_struct work;
338 struct completion done;
341 static void wq_barrier_func(struct work_struct *work)
343 struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
344 complete(&barr->done);
347 static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
348 struct wq_barrier *barr, int tail)
350 INIT_WORK(&barr->work, wq_barrier_func);
351 __set_bit(WORK_STRUCT_PENDING, work_data_bits(&barr->work));
353 init_completion(&barr->done);
355 insert_work(cwq, &barr->work, tail);
358 static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
360 if (cwq->thread == current) {
362 * Probably keventd trying to flush its own queue. So simply run
363 * it by hand rather than deadlocking.
367 struct wq_barrier barr;
370 spin_lock_irq(&cwq->lock);
371 if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) {
372 insert_wq_barrier(cwq, &barr, 1);
375 spin_unlock_irq(&cwq->lock);
378 wait_for_completion(&barr.done);
383 * flush_workqueue - ensure that any scheduled work has run to completion.
384 * @wq: workqueue to flush
386 * Forces execution of the workqueue and blocks until its completion.
387 * This is typically used in driver shutdown handlers.
389 * We sleep until all works which were queued on entry have been handled,
390 * but we are not livelocked by new incoming ones.
392 * This function used to run the workqueues itself. Now we just wait for the
393 * helper threads to do it.
395 void fastcall flush_workqueue(struct workqueue_struct *wq)
397 const cpumask_t *cpu_map = wq_cpu_map(wq);
401 for_each_cpu_mask(cpu, *cpu_map)
402 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
404 EXPORT_SYMBOL_GPL(flush_workqueue);
406 static void wait_on_work(struct cpu_workqueue_struct *cwq,
407 struct work_struct *work)
409 struct wq_barrier barr;
412 spin_lock_irq(&cwq->lock);
413 if (unlikely(cwq->current_work == work)) {
414 insert_wq_barrier(cwq, &barr, 0);
417 spin_unlock_irq(&cwq->lock);
419 if (unlikely(running))
420 wait_for_completion(&barr.done);
424 * flush_work - block until a work_struct's callback has terminated
425 * @wq: the workqueue on which the work is queued
426 * @work: the work which is to be flushed
428 * flush_work() will attempt to cancel the work if it is queued. If the work's
429 * callback appears to be running, flush_work() will block until it has
432 * flush_work() is designed to be used when the caller is tearing down data
433 * structures which the callback function operates upon. It is expected that,
434 * prior to calling flush_work(), the caller has arranged for the work to not
437 void flush_work(struct workqueue_struct *wq, struct work_struct *work)
439 const cpumask_t *cpu_map = wq_cpu_map(wq);
440 struct cpu_workqueue_struct *cwq;
445 cwq = get_wq_data(work);
446 /* Was it ever queued ? */
451 * This work can't be re-queued, no need to re-check that
452 * get_wq_data() is still the same when we take cwq->lock.
454 spin_lock_irq(&cwq->lock);
455 list_del_init(&work->entry);
457 spin_unlock_irq(&cwq->lock);
459 for_each_cpu_mask(cpu, *cpu_map)
460 wait_on_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
462 EXPORT_SYMBOL_GPL(flush_work);
465 static struct workqueue_struct *keventd_wq;
468 * schedule_work - put work task in global workqueue
469 * @work: job to be done
471 * This puts a job in the kernel-global workqueue.
473 int fastcall schedule_work(struct work_struct *work)
475 return queue_work(keventd_wq, work);
477 EXPORT_SYMBOL(schedule_work);
480 * schedule_delayed_work - put work task in global workqueue after delay
481 * @dwork: job to be done
482 * @delay: number of jiffies to wait or 0 for immediate execution
484 * After waiting for a given time this puts a job in the kernel-global
487 int fastcall schedule_delayed_work(struct delayed_work *dwork,
490 timer_stats_timer_set_start_info(&dwork->timer);
491 return queue_delayed_work(keventd_wq, dwork, delay);
493 EXPORT_SYMBOL(schedule_delayed_work);
496 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
498 * @dwork: job to be done
499 * @delay: number of jiffies to wait
501 * After waiting for a given time this puts a job in the kernel-global
502 * workqueue on the specified CPU.
504 int schedule_delayed_work_on(int cpu,
505 struct delayed_work *dwork, unsigned long delay)
507 return queue_delayed_work_on(cpu, keventd_wq, dwork, delay);
509 EXPORT_SYMBOL(schedule_delayed_work_on);
512 * schedule_on_each_cpu - call a function on each online CPU from keventd
513 * @func: the function to call
515 * Returns zero on success.
516 * Returns -ve errno on failure.
518 * Appears to be racy against CPU hotplug.
520 * schedule_on_each_cpu() is very slow.
522 int schedule_on_each_cpu(work_func_t func)
525 struct work_struct *works;
527 works = alloc_percpu(struct work_struct);
531 preempt_disable(); /* CPU hotplug */
532 for_each_online_cpu(cpu) {
533 struct work_struct *work = per_cpu_ptr(works, cpu);
535 INIT_WORK(work, func);
536 set_bit(WORK_STRUCT_PENDING, work_data_bits(work));
537 __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu), work);
540 flush_workqueue(keventd_wq);
545 void flush_scheduled_work(void)
547 flush_workqueue(keventd_wq);
549 EXPORT_SYMBOL(flush_scheduled_work);
551 void flush_work_keventd(struct work_struct *work)
553 flush_work(keventd_wq, work);
555 EXPORT_SYMBOL(flush_work_keventd);
558 * cancel_rearming_delayed_workqueue - kill off a delayed work whose handler rearms the delayed work.
559 * @wq: the controlling workqueue structure
560 * @dwork: the delayed work struct
562 * Note that the work callback function may still be running on return from
563 * cancel_delayed_work(). Run flush_workqueue() or flush_work() to wait on it.
565 void cancel_rearming_delayed_workqueue(struct workqueue_struct *wq,
566 struct delayed_work *dwork)
568 /* Was it ever queued ? */
569 if (!get_wq_data(&dwork->work))
572 while (!cancel_delayed_work(dwork))
575 EXPORT_SYMBOL(cancel_rearming_delayed_workqueue);
578 * cancel_rearming_delayed_work - kill off a delayed keventd work whose handler rearms the delayed work.
579 * @dwork: the delayed work struct
581 void cancel_rearming_delayed_work(struct delayed_work *dwork)
583 cancel_rearming_delayed_workqueue(keventd_wq, dwork);
585 EXPORT_SYMBOL(cancel_rearming_delayed_work);
588 * execute_in_process_context - reliably execute the routine with user context
589 * @fn: the function to execute
590 * @ew: guaranteed storage for the execute work structure (must
591 * be available when the work executes)
593 * Executes the function immediately if process context is available,
594 * otherwise schedules the function for delayed execution.
596 * Returns: 0 - function was executed
597 * 1 - function was scheduled for execution
599 int execute_in_process_context(work_func_t fn, struct execute_work *ew)
601 if (!in_interrupt()) {
606 INIT_WORK(&ew->work, fn);
607 schedule_work(&ew->work);
611 EXPORT_SYMBOL_GPL(execute_in_process_context);
615 return keventd_wq != NULL;
618 int current_is_keventd(void)
620 struct cpu_workqueue_struct *cwq;
621 int cpu = smp_processor_id(); /* preempt-safe: keventd is per-cpu */
626 cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
627 if (current == cwq->thread)
634 static struct cpu_workqueue_struct *
635 init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
637 struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
640 spin_lock_init(&cwq->lock);
641 INIT_LIST_HEAD(&cwq->worklist);
642 init_waitqueue_head(&cwq->more_work);
647 static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
649 struct workqueue_struct *wq = cwq->wq;
650 const char *fmt = is_single_threaded(wq) ? "%s" : "%s/%d";
651 struct task_struct *p;
653 p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu);
655 * Nobody can add the work_struct to this cwq,
656 * if (caller is __create_workqueue)
657 * nobody should see this wq
658 * else // caller is CPU_UP_PREPARE
659 * cpu is not on cpu_online_map
660 * so we can abort safely.
666 cwq->should_stop = 0;
671 static void start_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
673 struct task_struct *p = cwq->thread;
677 kthread_bind(p, cpu);
682 struct workqueue_struct *__create_workqueue(const char *name,
683 int singlethread, int freezeable)
685 struct workqueue_struct *wq;
686 struct cpu_workqueue_struct *cwq;
689 wq = kzalloc(sizeof(*wq), GFP_KERNEL);
693 wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
700 wq->singlethread = singlethread;
701 wq->freezeable = freezeable;
702 INIT_LIST_HEAD(&wq->list);
705 cwq = init_cpu_workqueue(wq, singlethread_cpu);
706 err = create_workqueue_thread(cwq, singlethread_cpu);
707 start_workqueue_thread(cwq, -1);
709 mutex_lock(&workqueue_mutex);
710 list_add(&wq->list, &workqueues);
712 for_each_possible_cpu(cpu) {
713 cwq = init_cpu_workqueue(wq, cpu);
714 if (err || !cpu_online(cpu))
716 err = create_workqueue_thread(cwq, cpu);
717 start_workqueue_thread(cwq, cpu);
719 mutex_unlock(&workqueue_mutex);
723 destroy_workqueue(wq);
728 EXPORT_SYMBOL_GPL(__create_workqueue);
730 static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
732 struct wq_barrier barr;
735 spin_lock_irq(&cwq->lock);
736 if (cwq->thread != NULL) {
737 insert_wq_barrier(cwq, &barr, 1);
738 cwq->should_stop = 1;
741 spin_unlock_irq(&cwq->lock);
744 wait_for_completion(&barr.done);
746 while (unlikely(cwq->thread != NULL))
749 * Wait until cwq->thread unlocks cwq->lock,
750 * it won't touch *cwq after that.
753 spin_unlock_wait(&cwq->lock);
758 * destroy_workqueue - safely terminate a workqueue
759 * @wq: target workqueue
761 * Safely destroy a workqueue. All work currently pending will be done first.
763 void destroy_workqueue(struct workqueue_struct *wq)
765 const cpumask_t *cpu_map = wq_cpu_map(wq);
766 struct cpu_workqueue_struct *cwq;
769 mutex_lock(&workqueue_mutex);
771 mutex_unlock(&workqueue_mutex);
773 for_each_cpu_mask(cpu, *cpu_map) {
774 cwq = per_cpu_ptr(wq->cpu_wq, cpu);
775 cleanup_workqueue_thread(cwq, cpu);
778 free_percpu(wq->cpu_wq);
781 EXPORT_SYMBOL_GPL(destroy_workqueue);
783 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
784 unsigned long action,
787 unsigned int cpu = (unsigned long)hcpu;
788 struct cpu_workqueue_struct *cwq;
789 struct workqueue_struct *wq;
792 case CPU_LOCK_ACQUIRE:
793 mutex_lock(&workqueue_mutex);
796 case CPU_LOCK_RELEASE:
797 mutex_unlock(&workqueue_mutex);
801 cpu_set(cpu, cpu_populated_map);
804 list_for_each_entry(wq, &workqueues, list) {
805 cwq = per_cpu_ptr(wq->cpu_wq, cpu);
809 if (!create_workqueue_thread(cwq, cpu))
811 printk(KERN_ERR "workqueue for %i failed\n", cpu);
815 start_workqueue_thread(cwq, cpu);
818 case CPU_UP_CANCELED:
819 start_workqueue_thread(cwq, -1);
821 cleanup_workqueue_thread(cwq, cpu);
829 void __init init_workqueues(void)
831 cpu_populated_map = cpu_online_map;
832 singlethread_cpu = first_cpu(cpu_possible_map);
833 cpu_singlethread_map = cpumask_of_cpu(singlethread_cpu);
834 hotcpu_notifier(workqueue_cpu_callback, 0);
835 keventd_wq = create_workqueue("events");