projects/linux·blob v2.6.33-rc3 · refs

1179 lines · 28.8 KiB

1/*
2 * linux/kernel/workqueue.c
3 *
4 * Generic mechanism for defining kernel helper threads for running
5 * arbitrary tasks in process context.
6 *
7 * Started by Ingo Molnar, Copyright (C) 2002
8 *
9 * Derived from the taskqueue/keventd code by:
10 *
11 * David Woodhouse <[email protected]>
12 * Andrew Morton
13 * Kai Petzke <[email protected]>
14 * Theodore Ts'o <[email protected]>
15 *
16 * Made to use alloc_percpu by Christoph Lameter.
17 */
18
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>
38
39/*
40 * The per-CPU workqueue (if single thread, we always use the first
41 * possible cpu).
42 */
43struct cpu_workqueue_struct {
44
45 spinlock_t lock;
46
47 struct list_head worklist;
48 wait_queue_head_t more_work;
49 struct work_struct *current_work;
50
51 struct workqueue_struct *wq;
52 struct task_struct *thread;
53} ____cacheline_aligned;
54
55/*
56 * The externally visible workqueue abstraction is an array of
57 * per-CPU workqueues:
58 */
59struct workqueue_struct {
60 struct cpu_workqueue_struct *cpu_wq;
61 struct list_head list;
62 const char *name;
63 int singlethread;
64 int freezeable; /* Freeze threads during suspend */
65 int rt;
66#ifdef CONFIG_LOCKDEP
67 struct lockdep_map lockdep_map;
68#endif
69};
70
71#ifdef CONFIG_DEBUG_OBJECTS_WORK
72
73static struct debug_obj_descr work_debug_descr;
74
75/*
76 * fixup_init is called when:
77 * - an active object is initialized
78 */
79static int work_fixup_init(void *addr, enum debug_obj_state state)
80{
81 struct work_struct *work = addr;
82
83 switch (state) {
84 case ODEBUG_STATE_ACTIVE:
85 cancel_work_sync(work);
86 debug_object_init(work, &work_debug_descr);
87 return 1;
88 default:
89 return 0;
90 }
91}
92
93/*
94 * fixup_activate is called when:
95 * - an active object is activated
96 * - an unknown object is activated (might be a statically initialized object)
97 */
98static int work_fixup_activate(void *addr, enum debug_obj_state state)
99{
100 struct work_struct *work = addr;
101
102 switch (state) {
103
104 case ODEBUG_STATE_NOTAVAILABLE:
105 /*
106 * This is not really a fixup. The work struct was
107 * statically initialized. We just make sure that it
108 * is tracked in the object tracker.
109 */
110 if (test_bit(WORK_STRUCT_STATIC, work_data_bits(work))) {
111 debug_object_init(work, &work_debug_descr);
112 debug_object_activate(work, &work_debug_descr);
113 return 0;
114 }
115 WARN_ON_ONCE(1);
116 return 0;
117
118 case ODEBUG_STATE_ACTIVE:
119 WARN_ON(1);
120
121 default:
122 return 0;
123 }
124}
125
126/*
127 * fixup_free is called when:
128 * - an active object is freed
129 */
130static int work_fixup_free(void *addr, enum debug_obj_state state)
131{
132 struct work_struct *work = addr;
133
134 switch (state) {
135 case ODEBUG_STATE_ACTIVE:
136 cancel_work_sync(work);
137 debug_object_free(work, &work_debug_descr);
138 return 1;
139 default:
140 return 0;
141 }
142}
143
144static struct debug_obj_descr work_debug_descr = {
145 .name = "work_struct",
146 .fixup_init = work_fixup_init,
147 .fixup_activate = work_fixup_activate,
148 .fixup_free = work_fixup_free,
149};
150
151static inline void debug_work_activate(struct work_struct *work)
152{
153 debug_object_activate(work, &work_debug_descr);
154}
155
156static inline void debug_work_deactivate(struct work_struct *work)
157{
158 debug_object_deactivate(work, &work_debug_descr);
159}
160
161void __init_work(struct work_struct *work, int onstack)
162{
163 if (onstack)
164 debug_object_init_on_stack(work, &work_debug_descr);
165 else
166 debug_object_init(work, &work_debug_descr);
167}
168EXPORT_SYMBOL_GPL(__init_work);
169
170void destroy_work_on_stack(struct work_struct *work)
171{
172 debug_object_free(work, &work_debug_descr);
173}
174EXPORT_SYMBOL_GPL(destroy_work_on_stack);
175
176#else
177static inline void debug_work_activate(struct work_struct *work) { }
178static inline void debug_work_deactivate(struct work_struct *work) { }
179#endif
180
181/* Serializes the accesses to the list of workqueues. */
182static DEFINE_SPINLOCK(workqueue_lock);
183static LIST_HEAD(workqueues);
184
185static int singlethread_cpu __read_mostly;
186static const struct cpumask *cpu_singlethread_map __read_mostly;
187/*
188 * _cpu_down() first removes CPU from cpu_online_map, then CPU_DEAD
189 * flushes cwq->worklist. This means that flush_workqueue/wait_on_work
190 * which comes in between can't use for_each_online_cpu(). We could
191 * use cpu_possible_map, the cpumask below is more a documentation
192 * than optimization.
193 */
194static cpumask_var_t cpu_populated_map __read_mostly;
195
196/* If it's single threaded, it isn't in the list of workqueues. */
197static inline int is_wq_single_threaded(struct workqueue_struct *wq)
198{
199 return wq->singlethread;
200}
201
202static const struct cpumask *wq_cpu_map(struct workqueue_struct *wq)
203{
204 return is_wq_single_threaded(wq)
205 ? cpu_singlethread_map : cpu_populated_map;
206}
207
208static
209struct cpu_workqueue_struct *wq_per_cpu(struct workqueue_struct *wq, int cpu)
210{
211 if (unlikely(is_wq_single_threaded(wq)))
212 cpu = singlethread_cpu;
213 return per_cpu_ptr(wq->cpu_wq, cpu);
214}
215
216/*
217 * Set the workqueue on which a work item is to be run
218 * - Must *only* be called if the pending flag is set
219 */
220static inline void set_wq_data(struct work_struct *work,
221 struct cpu_workqueue_struct *cwq)
222{
223 unsigned long new;
224
225 BUG_ON(!work_pending(work));
226
227 new = (unsigned long) cwq | (1UL << WORK_STRUCT_PENDING);
228 new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work);
229 atomic_long_set(&work->data, new);
230}
231
232static inline
233struct cpu_workqueue_struct *get_wq_data(struct work_struct *work)
234{
235 return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
236}
237
238static void insert_work(struct cpu_workqueue_struct *cwq,
239 struct work_struct *work, struct list_head *head)
240{
241 trace_workqueue_insertion(cwq->thread, work);
242
243 set_wq_data(work, cwq);
244 /*
245 * Ensure that we get the right work->data if we see the
246 * result of list_add() below, see try_to_grab_pending().
247 */
248 smp_wmb();
249 list_add_tail(&work->entry, head);
250 wake_up(&cwq->more_work);
251}
252
253static void __queue_work(struct cpu_workqueue_struct *cwq,
254 struct work_struct *work)
255{
256 unsigned long flags;
257
258 debug_work_activate(work);
259 spin_lock_irqsave(&cwq->lock, flags);
260 insert_work(cwq, work, &cwq->worklist);
261 spin_unlock_irqrestore(&cwq->lock, flags);
262}
263
264/**
265 * queue_work - queue work on a workqueue
266 * @wq: workqueue to use
267 * @work: work to queue
268 *
269 * Returns 0 if @work was already on a queue, non-zero otherwise.
270 *
271 * We queue the work to the CPU on which it was submitted, but if the CPU dies
272 * it can be processed by another CPU.
273 */
274int queue_work(struct workqueue_struct *wq, struct work_struct *work)
275{
276 int ret;
277
278 ret = queue_work_on(get_cpu(), wq, work);
279 put_cpu();
280
281 return ret;
282}
283EXPORT_SYMBOL_GPL(queue_work);
284
285/**
286 * queue_work_on - queue work on specific cpu
287 * @cpu: CPU number to execute work on
288 * @wq: workqueue to use
289 * @work: work to queue
290 *
291 * Returns 0 if @work was already on a queue, non-zero otherwise.
292 *
293 * We queue the work to a specific CPU, the caller must ensure it
294 * can't go away.
295 */
296int
297queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work)
298{
299 int ret = 0;
300
301 if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
302 BUG_ON(!list_empty(&work->entry));
303 __queue_work(wq_per_cpu(wq, cpu), work);
304 ret = 1;
305 }
306 return ret;
307}
308EXPORT_SYMBOL_GPL(queue_work_on);
309
310static void delayed_work_timer_fn(unsigned long __data)
311{
312 struct delayed_work *dwork = (struct delayed_work *)__data;
313 struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work);
314 struct workqueue_struct *wq = cwq->wq;
315
316 __queue_work(wq_per_cpu(wq, smp_processor_id()), &dwork->work);
317}
318
319/**
320 * queue_delayed_work - queue work on a workqueue after delay
321 * @wq: workqueue to use
322 * @dwork: delayable work to queue
323 * @delay: number of jiffies to wait before queueing
324 *
325 * Returns 0 if @work was already on a queue, non-zero otherwise.
326 */
327int queue_delayed_work(struct workqueue_struct *wq,
328 struct delayed_work *dwork, unsigned long delay)
329{
330 if (delay == 0)
331 return queue_work(wq, &dwork->work);
332
333 return queue_delayed_work_on(-1, wq, dwork, delay);
334}
335EXPORT_SYMBOL_GPL(queue_delayed_work);
336
337/**
338 * queue_delayed_work_on - queue work on specific CPU after delay
339 * @cpu: CPU number to execute work on
340 * @wq: workqueue to use
341 * @dwork: work to queue
342 * @delay: number of jiffies to wait before queueing
343 *
344 * Returns 0 if @work was already on a queue, non-zero otherwise.
345 */
346int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
347 struct delayed_work *dwork, unsigned long delay)
348{
349 int ret = 0;
350 struct timer_list *timer = &dwork->timer;
351 struct work_struct *work = &dwork->work;
352
353 if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
354 BUG_ON(timer_pending(timer));
355 BUG_ON(!list_empty(&work->entry));
356
357 timer_stats_timer_set_start_info(&dwork->timer);
358
359 /* This stores cwq for the moment, for the timer_fn */
360 set_wq_data(work, wq_per_cpu(wq, raw_smp_processor_id()));
361 timer->expires = jiffies + delay;
362 timer->data = (unsigned long)dwork;
363 timer->function = delayed_work_timer_fn;
364
365 if (unlikely(cpu >= 0))
366 add_timer_on(timer, cpu);
367 else
368 add_timer(timer);
369 ret = 1;
370 }
371 return ret;
372}
373EXPORT_SYMBOL_GPL(queue_delayed_work_on);
374
375static void run_workqueue(struct cpu_workqueue_struct *cwq)
376{
377 spin_lock_irq(&cwq->lock);
378 while (!list_empty(&cwq->worklist)) {
379 struct work_struct *work = list_entry(cwq->worklist.next,
380 struct work_struct, entry);
381 work_func_t f = work->func;
382#ifdef CONFIG_LOCKDEP
383 /*
384 * It is permissible to free the struct work_struct
385 * from inside the function that is called from it,
386 * this we need to take into account for lockdep too.
387 * To avoid bogus "held lock freed" warnings as well
388 * as problems when looking into work->lockdep_map,
389 * make a copy and use that here.
390 */
391 struct lockdep_map lockdep_map = work->lockdep_map;
392#endif
393 trace_workqueue_execution(cwq->thread, work);
394 debug_work_deactivate(work);
395 cwq->current_work = work;
396 list_del_init(cwq->worklist.next);
397 spin_unlock_irq(&cwq->lock);
398
399 BUG_ON(get_wq_data(work) != cwq);
400 work_clear_pending(work);
401 lock_map_acquire(&cwq->wq->lockdep_map);
402 lock_map_acquire(&lockdep_map);
403 f(work);
404 lock_map_release(&lockdep_map);
405 lock_map_release(&cwq->wq->lockdep_map);
406
407 if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
408 printk(KERN_ERR "BUG: workqueue leaked lock or atomic: "
409 "%s/0x%08x/%d\n",
410 current->comm, preempt_count(),
411 task_pid_nr(current));
412 printk(KERN_ERR " last function: ");
413 print_symbol("%s\n", (unsigned long)f);
414 debug_show_held_locks(current);
415 dump_stack();
416 }
417
418 spin_lock_irq(&cwq->lock);
419 cwq->current_work = NULL;
420 }
421 spin_unlock_irq(&cwq->lock);
422}
423
424static int worker_thread(void *__cwq)
425{
426 struct cpu_workqueue_struct *cwq = __cwq;
427 DEFINE_WAIT(wait);
428
429 if (cwq->wq->freezeable)
430 set_freezable();
431
432 for (;;) {
433 prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE);
434 if (!freezing(current) &&
435 !kthread_should_stop() &&
436 list_empty(&cwq->worklist))
437 schedule();
438 finish_wait(&cwq->more_work, &wait);
439
440 try_to_freeze();
441
442 if (kthread_should_stop())
443 break;
444
445 run_workqueue(cwq);
446 }
447
448 return 0;
449}
450
451struct wq_barrier {
452 struct work_struct work;
453 struct completion done;
454};
455
456static void wq_barrier_func(struct work_struct *work)
457{
458 struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
459 complete(&barr->done);
460}
461
462static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
463 struct wq_barrier *barr, struct list_head *head)
464{
465 /*
466 * debugobject calls are safe here even with cwq->lock locked
467 * as we know for sure that this will not trigger any of the
468 * checks and call back into the fixup functions where we
469 * might deadlock.
470 */
471 INIT_WORK_ON_STACK(&barr->work, wq_barrier_func);
472 __set_bit(WORK_STRUCT_PENDING, work_data_bits(&barr->work));
473
474 init_completion(&barr->done);
475
476 debug_work_activate(&barr->work);
477 insert_work(cwq, &barr->work, head);
478}
479
480static int flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
481{
482 int active = 0;
483 struct wq_barrier barr;
484
485 WARN_ON(cwq->thread == current);
486
487 spin_lock_irq(&cwq->lock);
488 if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) {
489 insert_wq_barrier(cwq, &barr, &cwq->worklist);
490 active = 1;
491 }
492 spin_unlock_irq(&cwq->lock);
493
494 if (active) {
495 wait_for_completion(&barr.done);
496 destroy_work_on_stack(&barr.work);
497 }
498
499 return active;
500}
501
502/**
503 * flush_workqueue - ensure that any scheduled work has run to completion.
504 * @wq: workqueue to flush
505 *
506 * Forces execution of the workqueue and blocks until its completion.
507 * This is typically used in driver shutdown handlers.
508 *
509 * We sleep until all works which were queued on entry have been handled,
510 * but we are not livelocked by new incoming ones.
511 *
512 * This function used to run the workqueues itself. Now we just wait for the
513 * helper threads to do it.
514 */
515void flush_workqueue(struct workqueue_struct *wq)
516{
517 const struct cpumask *cpu_map = wq_cpu_map(wq);
518 int cpu;
519
520 might_sleep();
521 lock_map_acquire(&wq->lockdep_map);
522 lock_map_release(&wq->lockdep_map);
523 for_each_cpu(cpu, cpu_map)
524 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
525}
526EXPORT_SYMBOL_GPL(flush_workqueue);
527
528/**
529 * flush_work - block until a work_struct's callback has terminated
530 * @work: the work which is to be flushed
531 *
532 * Returns false if @work has already terminated.
533 *
534 * It is expected that, prior to calling flush_work(), the caller has
535 * arranged for the work to not be requeued, otherwise it doesn't make
536 * sense to use this function.
537 */
538int flush_work(struct work_struct *work)
539{
540 struct cpu_workqueue_struct *cwq;
541 struct list_head *prev;
542 struct wq_barrier barr;
543
544 might_sleep();
545 cwq = get_wq_data(work);
546 if (!cwq)
547 return 0;
548
549 lock_map_acquire(&cwq->wq->lockdep_map);
550 lock_map_release(&cwq->wq->lockdep_map);
551
552 prev = NULL;
553 spin_lock_irq(&cwq->lock);
554 if (!list_empty(&work->entry)) {
555 /*
556 * See the comment near try_to_grab_pending()->smp_rmb().
557 * If it was re-queued under us we are not going to wait.
558 */
559 smp_rmb();
560 if (unlikely(cwq != get_wq_data(work)))
561 goto out;
562 prev = &work->entry;
563 } else {
564 if (cwq->current_work != work)
565 goto out;
566 prev = &cwq->worklist;
567 }
568 insert_wq_barrier(cwq, &barr, prev->next);
569out:
570 spin_unlock_irq(&cwq->lock);
571 if (!prev)
572 return 0;
573
574 wait_for_completion(&barr.done);
575 destroy_work_on_stack(&barr.work);
576 return 1;
577}
578EXPORT_SYMBOL_GPL(flush_work);
579
580/*
581 * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit,
582 * so this work can't be re-armed in any way.
583 */
584static int try_to_grab_pending(struct work_struct *work)
585{
586 struct cpu_workqueue_struct *cwq;
587 int ret = -1;
588
589 if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work)))
590 return 0;
591
592 /*
593 * The queueing is in progress, or it is already queued. Try to
594 * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
595 */
596
597 cwq = get_wq_data(work);
598 if (!cwq)
599 return ret;
600
601 spin_lock_irq(&cwq->lock);
602 if (!list_empty(&work->entry)) {
603 /*
604 * This work is queued, but perhaps we locked the wrong cwq.
605 * In that case we must see the new value after rmb(), see
606 * insert_work()->wmb().
607 */
608 smp_rmb();
609 if (cwq == get_wq_data(work)) {
610 debug_work_deactivate(work);
611 list_del_init(&work->entry);
612 ret = 1;
613 }
614 }
615 spin_unlock_irq(&cwq->lock);
616
617 return ret;
618}
619
620static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq,
621 struct work_struct *work)
622{
623 struct wq_barrier barr;
624 int running = 0;
625
626 spin_lock_irq(&cwq->lock);
627 if (unlikely(cwq->current_work == work)) {
628 insert_wq_barrier(cwq, &barr, cwq->worklist.next);
629 running = 1;
630 }
631 spin_unlock_irq(&cwq->lock);
632
633 if (unlikely(running)) {
634 wait_for_completion(&barr.done);
635 destroy_work_on_stack(&barr.work);
636 }
637}
638
639static void wait_on_work(struct work_struct *work)
640{
641 struct cpu_workqueue_struct *cwq;
642 struct workqueue_struct *wq;
643 const struct cpumask *cpu_map;
644 int cpu;
645
646 might_sleep();
647
648 lock_map_acquire(&work->lockdep_map);
649 lock_map_release(&work->lockdep_map);
650
651 cwq = get_wq_data(work);
652 if (!cwq)
653 return;
654
655 wq = cwq->wq;
656 cpu_map = wq_cpu_map(wq);
657
658 for_each_cpu(cpu, cpu_map)
659 wait_on_cpu_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
660}
661
662static int __cancel_work_timer(struct work_struct *work,
663 struct timer_list* timer)
664{
665 int ret;
666
667 do {
668 ret = (timer && likely(del_timer(timer)));
669 if (!ret)
670 ret = try_to_grab_pending(work);
671 wait_on_work(work);
672 } while (unlikely(ret < 0));
673
674 work_clear_pending(work);
675 return ret;
676}
677
678/**
679 * cancel_work_sync - block until a work_struct's callback has terminated
680 * @work: the work which is to be flushed
681 *
682 * Returns true if @work was pending.
683 *
684 * cancel_work_sync() will cancel the work if it is queued. If the work's
685 * callback appears to be running, cancel_work_sync() will block until it
686 * has completed.
687 *
688 * It is possible to use this function if the work re-queues itself. It can
689 * cancel the work even if it migrates to another workqueue, however in that
690 * case it only guarantees that work->func() has completed on the last queued
691 * workqueue.
692 *
693 * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not
694 * pending, otherwise it goes into a busy-wait loop until the timer expires.
695 *
696 * The caller must ensure that workqueue_struct on which this work was last
697 * queued can't be destroyed before this function returns.
698 */
699int cancel_work_sync(struct work_struct *work)
700{
701 return __cancel_work_timer(work, NULL);
702}
703EXPORT_SYMBOL_GPL(cancel_work_sync);
704
705/**
706 * cancel_delayed_work_sync - reliably kill off a delayed work.
707 * @dwork: the delayed work struct
708 *
709 * Returns true if @dwork was pending.
710 *
711 * It is possible to use this function if @dwork rearms itself via queue_work()
712 * or queue_delayed_work(). See also the comment for cancel_work_sync().
713 */
714int cancel_delayed_work_sync(struct delayed_work *dwork)
715{
716 return __cancel_work_timer(&dwork->work, &dwork->timer);
717}
718EXPORT_SYMBOL(cancel_delayed_work_sync);
719
720static struct workqueue_struct *keventd_wq __read_mostly;
721
722/**
723 * schedule_work - put work task in global workqueue
724 * @work: job to be done
725 *
726 * Returns zero if @work was already on the kernel-global workqueue and
727 * non-zero otherwise.
728 *
729 * This puts a job in the kernel-global workqueue if it was not already
730 * queued and leaves it in the same position on the kernel-global
731 * workqueue otherwise.
732 */
733int schedule_work(struct work_struct *work)
734{
735 return queue_work(keventd_wq, work);
736}
737EXPORT_SYMBOL(schedule_work);
738
739/*
740 * schedule_work_on - put work task on a specific cpu
741 * @cpu: cpu to put the work task on
742 * @work: job to be done
743 *
744 * This puts a job on a specific cpu
745 */
746int schedule_work_on(int cpu, struct work_struct *work)
747{
748 return queue_work_on(cpu, keventd_wq, work);
749}
750EXPORT_SYMBOL(schedule_work_on);
751
752/**
753 * schedule_delayed_work - put work task in global workqueue after delay
754 * @dwork: job to be done
755 * @delay: number of jiffies to wait or 0 for immediate execution
756 *
757 * After waiting for a given time this puts a job in the kernel-global
758 * workqueue.
759 */
760int schedule_delayed_work(struct delayed_work *dwork,
761 unsigned long delay)
762{
763 return queue_delayed_work(keventd_wq, dwork, delay);
764}
765EXPORT_SYMBOL(schedule_delayed_work);
766
767/**
768 * flush_delayed_work - block until a dwork_struct's callback has terminated
769 * @dwork: the delayed work which is to be flushed
770 *
771 * Any timeout is cancelled, and any pending work is run immediately.
772 */
773void flush_delayed_work(struct delayed_work *dwork)
774{
775 if (del_timer_sync(&dwork->timer)) {
776 struct cpu_workqueue_struct *cwq;
777 cwq = wq_per_cpu(keventd_wq, get_cpu());
778 __queue_work(cwq, &dwork->work);
779 put_cpu();
780 }
781 flush_work(&dwork->work);
782}
783EXPORT_SYMBOL(flush_delayed_work);
784
785/**
786 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
787 * @cpu: cpu to use
788 * @dwork: job to be done
789 * @delay: number of jiffies to wait
790 *
791 * After waiting for a given time this puts a job in the kernel-global
792 * workqueue on the specified CPU.
793 */
794int schedule_delayed_work_on(int cpu,
795 struct delayed_work *dwork, unsigned long delay)
796{
797 return queue_delayed_work_on(cpu, keventd_wq, dwork, delay);
798}
799EXPORT_SYMBOL(schedule_delayed_work_on);
800
801/**
802 * schedule_on_each_cpu - call a function on each online CPU from keventd
803 * @func: the function to call
804 *
805 * Returns zero on success.
806 * Returns -ve errno on failure.
807 *
808 * schedule_on_each_cpu() is very slow.
809 */
810int schedule_on_each_cpu(work_func_t func)
811{
812 int cpu;
813 int orig = -1;
814 struct work_struct *works;
815
816 works = alloc_percpu(struct work_struct);
817 if (!works)
818 return -ENOMEM;
819
820 get_online_cpus();
821
822 /*
823 * When running in keventd don't schedule a work item on
824 * itself. Can just call directly because the work queue is
825 * already bound. This also is faster.
826 */
827 if (current_is_keventd())
828 orig = raw_smp_processor_id();
829
830 for_each_online_cpu(cpu) {
831 struct work_struct *work = per_cpu_ptr(works, cpu);
832
833 INIT_WORK(work, func);
834 if (cpu != orig)
835 schedule_work_on(cpu, work);
836 }
837 if (orig >= 0)
838 func(per_cpu_ptr(works, orig));
839
840 for_each_online_cpu(cpu)
841 flush_work(per_cpu_ptr(works, cpu));
842
843 put_online_cpus();
844 free_percpu(works);
845 return 0;
846}
847
848void flush_scheduled_work(void)
849{
850 flush_workqueue(keventd_wq);
851}
852EXPORT_SYMBOL(flush_scheduled_work);
853
854/**
855 * execute_in_process_context - reliably execute the routine with user context
856 * @fn: the function to execute
857 * @ew: guaranteed storage for the execute work structure (must
858 * be available when the work executes)
859 *
860 * Executes the function immediately if process context is available,
861 * otherwise schedules the function for delayed execution.
862 *
863 * Returns: 0 - function was executed
864 * 1 - function was scheduled for execution
865 */
866int execute_in_process_context(work_func_t fn, struct execute_work *ew)
867{
868 if (!in_interrupt()) {
869 fn(&ew->work);
870 return 0;
871 }
872
873 INIT_WORK(&ew->work, fn);
874 schedule_work(&ew->work);
875
876 return 1;
877}
878EXPORT_SYMBOL_GPL(execute_in_process_context);
879
880int keventd_up(void)
881{
882 return keventd_wq != NULL;
883}
884
885int current_is_keventd(void)
886{
887 struct cpu_workqueue_struct *cwq;
888 int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */
889 int ret = 0;
890
891 BUG_ON(!keventd_wq);
892
893 cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
894 if (current == cwq->thread)
895 ret = 1;
896
897 return ret;
898
899}
900
901static struct cpu_workqueue_struct *
902init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
903{
904 struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
905
906 cwq->wq = wq;
907 spin_lock_init(&cwq->lock);
908 INIT_LIST_HEAD(&cwq->worklist);
909 init_waitqueue_head(&cwq->more_work);
910
911 return cwq;
912}
913
914static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
915{
916 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
917 struct workqueue_struct *wq = cwq->wq;
918 const char *fmt = is_wq_single_threaded(wq) ? "%s" : "%s/%d";
919 struct task_struct *p;
920
921 p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu);
922 /*
923 * Nobody can add the work_struct to this cwq,
924 * if (caller is __create_workqueue)
925 * nobody should see this wq
926 * else // caller is CPU_UP_PREPARE
927 * cpu is not on cpu_online_map
928 * so we can abort safely.
929 */
930 if (IS_ERR(p))
931 return PTR_ERR(p);
932 if (cwq->wq->rt)
933 sched_setscheduler_nocheck(p, SCHED_FIFO, &param);
934 cwq->thread = p;
935
936 trace_workqueue_creation(cwq->thread, cpu);
937
938 return 0;
939}
940
941static void start_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
942{
943 struct task_struct *p = cwq->thread;
944
945 if (p != NULL) {
946 if (cpu >= 0)
947 kthread_bind(p, cpu);
948 wake_up_process(p);
949 }
950}
951
952struct workqueue_struct *__create_workqueue_key(const char *name,
953 int singlethread,
954 int freezeable,
955 int rt,
956 struct lock_class_key *key,
957 const char *lock_name)
958{
959 struct workqueue_struct *wq;
960 struct cpu_workqueue_struct *cwq;
961 int err = 0, cpu;
962
963 wq = kzalloc(sizeof(*wq), GFP_KERNEL);
964 if (!wq)
965 return NULL;
966
967 wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
968 if (!wq->cpu_wq) {
969 kfree(wq);
970 return NULL;
971 }
972
973 wq->name = name;
974 lockdep_init_map(&wq->lockdep_map, lock_name, key, 0);
975 wq->singlethread = singlethread;
976 wq->freezeable = freezeable;
977 wq->rt = rt;
978 INIT_LIST_HEAD(&wq->list);
979
980 if (singlethread) {
981 cwq = init_cpu_workqueue(wq, singlethread_cpu);
982 err = create_workqueue_thread(cwq, singlethread_cpu);
983 start_workqueue_thread(cwq, -1);
984 } else {
985 cpu_maps_update_begin();
986 /*
987 * We must place this wq on list even if the code below fails.
988 * cpu_down(cpu) can remove cpu from cpu_populated_map before
989 * destroy_workqueue() takes the lock, in that case we leak
990 * cwq[cpu]->thread.
991 */
992 spin_lock(&workqueue_lock);
993 list_add(&wq->list, &workqueues);
994 spin_unlock(&workqueue_lock);
995 /*
996 * We must initialize cwqs for each possible cpu even if we
997 * are going to call destroy_workqueue() finally. Otherwise
998 * cpu_up() can hit the uninitialized cwq once we drop the
999 * lock.
1000 */
1001 for_each_possible_cpu(cpu) {
1002 cwq = init_cpu_workqueue(wq, cpu);
1003 if (err || !cpu_online(cpu))
1004 continue;
1005 err = create_workqueue_thread(cwq, cpu);
1006 start_workqueue_thread(cwq, cpu);
1007 }
1008 cpu_maps_update_done();
1009 }
1010
1011 if (err) {
1012 destroy_workqueue(wq);
1013 wq = NULL;
1014 }
1015 return wq;
1016}
1017EXPORT_SYMBOL_GPL(__create_workqueue_key);
1018
1019static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq)
1020{
1021 /*
1022 * Our caller is either destroy_workqueue() or CPU_POST_DEAD,
1023 * cpu_add_remove_lock protects cwq->thread.
1024 */
1025 if (cwq->thread == NULL)
1026 return;
1027
1028 lock_map_acquire(&cwq->wq->lockdep_map);
1029 lock_map_release(&cwq->wq->lockdep_map);
1030
1031 flush_cpu_workqueue(cwq);
1032 /*
1033 * If the caller is CPU_POST_DEAD and cwq->worklist was not empty,
1034 * a concurrent flush_workqueue() can insert a barrier after us.
1035 * However, in that case run_workqueue() won't return and check
1036 * kthread_should_stop() until it flushes all work_struct's.
1037 * When ->worklist becomes empty it is safe to exit because no
1038 * more work_structs can be queued on this cwq: flush_workqueue
1039 * checks list_empty(), and a "normal" queue_work() can't use
1040 * a dead CPU.
1041 */
1042 trace_workqueue_destruction(cwq->thread);
1043 kthread_stop(cwq->thread);
1044 cwq->thread = NULL;
1045}
1046
1047/**
1048 * destroy_workqueue - safely terminate a workqueue
1049 * @wq: target workqueue
1050 *
1051 * Safely destroy a workqueue. All work currently pending will be done first.
1052 */
1053void destroy_workqueue(struct workqueue_struct *wq)
1054{
1055 const struct cpumask *cpu_map = wq_cpu_map(wq);
1056 int cpu;
1057
1058 cpu_maps_update_begin();
1059 spin_lock(&workqueue_lock);
1060 list_del(&wq->list);
1061 spin_unlock(&workqueue_lock);
1062
1063 for_each_cpu(cpu, cpu_map)
1064 cleanup_workqueue_thread(per_cpu_ptr(wq->cpu_wq, cpu));
1065 cpu_maps_update_done();
1066
1067 free_percpu(wq->cpu_wq);
1068 kfree(wq);
1069}
1070EXPORT_SYMBOL_GPL(destroy_workqueue);
1071
1072static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
1073 unsigned long action,
1074 void *hcpu)
1075{
1076 unsigned int cpu = (unsigned long)hcpu;
1077 struct cpu_workqueue_struct *cwq;
1078 struct workqueue_struct *wq;
1079 int ret = NOTIFY_OK;
1080
1081 action &= ~CPU_TASKS_FROZEN;
1082
1083 switch (action) {
1084 case CPU_UP_PREPARE:
1085 cpumask_set_cpu(cpu, cpu_populated_map);
1086 }
1087undo:
1088 list_for_each_entry(wq, &workqueues, list) {
1089 cwq = per_cpu_ptr(wq->cpu_wq, cpu);
1090
1091 switch (action) {
1092 case CPU_UP_PREPARE:
1093 if (!create_workqueue_thread(cwq, cpu))
1094 break;
1095 printk(KERN_ERR "workqueue [%s] for %i failed\n",
1096 wq->name, cpu);
1097 action = CPU_UP_CANCELED;
1098 ret = NOTIFY_BAD;
1099 goto undo;
1100
1101 case CPU_ONLINE:
1102 start_workqueue_thread(cwq, cpu);
1103 break;
1104
1105 case CPU_UP_CANCELED:
1106 start_workqueue_thread(cwq, -1);
1107 case CPU_POST_DEAD:
1108 cleanup_workqueue_thread(cwq);
1109 break;
1110 }
1111 }
1112
1113 switch (action) {
1114 case CPU_UP_CANCELED:
1115 case CPU_POST_DEAD:
1116 cpumask_clear_cpu(cpu, cpu_populated_map);
1117 }
1118
1119 return ret;
1120}
1121
1122#ifdef CONFIG_SMP
1123
1124struct work_for_cpu {
1125 struct completion completion;
1126 long (*fn)(void *);
1127 void *arg;
1128 long ret;
1129};
1130
1131static int do_work_for_cpu(void *_wfc)
1132{
1133 struct work_for_cpu *wfc = _wfc;
1134 wfc->ret = wfc->fn(wfc->arg);
1135 complete(&wfc->completion);
1136 return 0;
1137}
1138
1139/**
1140 * work_on_cpu - run a function in user context on a particular cpu
1141 * @cpu: the cpu to run on
1142 * @fn: the function to run
1143 * @arg: the function arg
1144 *
1145 * This will return the value @fn returns.
1146 * It is up to the caller to ensure that the cpu doesn't go offline.
1147 * The caller must not hold any locks which would prevent @fn from completing.
1148 */
1149long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg)
1150{
1151 struct task_struct *sub_thread;
1152 struct work_for_cpu wfc = {
1153 .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion),
1154 .fn = fn,
1155 .arg = arg,
1156 };
1157
1158 sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu");
1159 if (IS_ERR(sub_thread))
1160 return PTR_ERR(sub_thread);
1161 kthread_bind(sub_thread, cpu);
1162 wake_up_process(sub_thread);
1163 wait_for_completion(&wfc.completion);
1164 return wfc.ret;
1165}
1166EXPORT_SYMBOL_GPL(work_on_cpu);
1167#endif /* CONFIG_SMP */
1168
1169void __init init_workqueues(void)
1170{
1171 alloc_cpumask_var(&cpu_populated_map, GFP_KERNEL);
1172
1173 cpumask_copy(cpu_populated_map, cpu_online_mask);
1174 singlethread_cpu = cpumask_first(cpu_possible_mask);
1175 cpu_singlethread_map = cpumask_of(singlethread_cpu);
1176 hotcpu_notifier(workqueue_cpu_callback, 0);
1177 keventd_wq = create_workqueue("events");
1178 BUG_ON(!keventd_wq);
1179}