f928f2a87b9b365a8e1d8684902a3ad6f917d6b0
[linux-2.6.git] / kernel / workqueue.c
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 <dwmw2@infradead.org>
12  *   Andrew Morton
13  *   Kai Petzke <wpp@marie.physik.tu-berlin.de>
14  *   Theodore Ts'o <tytso@mit.edu>
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
37 /*
38  * The per-CPU workqueue (if single thread, we always use the first
39  * possible cpu).
40  */
41 struct cpu_workqueue_struct {
42
43         spinlock_t lock;
44
45         struct list_head worklist;
46         wait_queue_head_t more_work;
47         struct work_struct *current_work;
48
49         struct workqueue_struct *wq;
50         struct task_struct *thread;
51
52         int run_depth;          /* Detect run_workqueue() recursion depth */
53 } ____cacheline_aligned;
54
55 /*
56  * The externally visible workqueue abstraction is an array of
57  * per-CPU workqueues:
58  */
59 struct 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 /* Serializes the accesses to the list of workqueues. */
72 static DEFINE_SPINLOCK(workqueue_lock);
73 static LIST_HEAD(workqueues);
74
75 static int singlethread_cpu __read_mostly;
76 static cpumask_t cpu_singlethread_map __read_mostly;
77 /*
78  * _cpu_down() first removes CPU from cpu_online_map, then CPU_DEAD
79  * flushes cwq->worklist. This means that flush_workqueue/wait_on_work
80  * which comes in between can't use for_each_online_cpu(). We could
81  * use cpu_possible_map, the cpumask below is more a documentation
82  * than optimization.
83  */
84 static cpumask_t cpu_populated_map __read_mostly;
85
86 /* If it's single threaded, it isn't in the list of workqueues. */
87 static inline int is_single_threaded(struct workqueue_struct *wq)
88 {
89         return wq->singlethread;
90 }
91
92 static const cpumask_t *wq_cpu_map(struct workqueue_struct *wq)
93 {
94         return is_single_threaded(wq)
95                 ? &cpu_singlethread_map : &cpu_populated_map;
96 }
97
98 static
99 struct cpu_workqueue_struct *wq_per_cpu(struct workqueue_struct *wq, int cpu)
100 {
101         if (unlikely(is_single_threaded(wq)))
102                 cpu = singlethread_cpu;
103         return per_cpu_ptr(wq->cpu_wq, cpu);
104 }
105
106 /*
107  * Set the workqueue on which a work item is to be run
108  * - Must *only* be called if the pending flag is set
109  */
110 static inline void set_wq_data(struct work_struct *work,
111                                 struct cpu_workqueue_struct *cwq)
112 {
113         unsigned long new;
114
115         BUG_ON(!work_pending(work));
116
117         new = (unsigned long) cwq | (1UL << WORK_STRUCT_PENDING);
118         new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work);
119         atomic_long_set(&work->data, new);
120 }
121
122 static inline
123 struct cpu_workqueue_struct *get_wq_data(struct work_struct *work)
124 {
125         return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
126 }
127
128 static void insert_work(struct cpu_workqueue_struct *cwq,
129                         struct work_struct *work, struct list_head *head)
130 {
131         set_wq_data(work, cwq);
132         /*
133          * Ensure that we get the right work->data if we see the
134          * result of list_add() below, see try_to_grab_pending().
135          */
136         smp_wmb();
137         list_add_tail(&work->entry, head);
138         wake_up(&cwq->more_work);
139 }
140
141 static void __queue_work(struct cpu_workqueue_struct *cwq,
142                          struct work_struct *work)
143 {
144         unsigned long flags;
145
146         spin_lock_irqsave(&cwq->lock, flags);
147         insert_work(cwq, work, &cwq->worklist);
148         spin_unlock_irqrestore(&cwq->lock, flags);
149 }
150
151 /**
152  * queue_work - queue work on a workqueue
153  * @wq: workqueue to use
154  * @work: work to queue
155  *
156  * Returns 0 if @work was already on a queue, non-zero otherwise.
157  *
158  * We queue the work to the CPU on which it was submitted, but if the CPU dies
159  * it can be processed by another CPU.
160  */
161 int queue_work(struct workqueue_struct *wq, struct work_struct *work)
162 {
163         int ret;
164
165         ret = queue_work_on(get_cpu(), wq, work);
166         put_cpu();
167
168         return ret;
169 }
170 EXPORT_SYMBOL_GPL(queue_work);
171
172 /**
173  * queue_work_on - queue work on specific cpu
174  * @cpu: CPU number to execute work on
175  * @wq: workqueue to use
176  * @work: work to queue
177  *
178  * Returns 0 if @work was already on a queue, non-zero otherwise.
179  *
180  * We queue the work to a specific CPU, the caller must ensure it
181  * can't go away.
182  */
183 int
184 queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work)
185 {
186         int ret = 0;
187
188         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
189                 BUG_ON(!list_empty(&work->entry));
190                 __queue_work(wq_per_cpu(wq, cpu), work);
191                 ret = 1;
192         }
193         return ret;
194 }
195 EXPORT_SYMBOL_GPL(queue_work_on);
196
197 static void delayed_work_timer_fn(unsigned long __data)
198 {
199         struct delayed_work *dwork = (struct delayed_work *)__data;
200         struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work);
201         struct workqueue_struct *wq = cwq->wq;
202
203         __queue_work(wq_per_cpu(wq, smp_processor_id()), &dwork->work);
204 }
205
206 /**
207  * queue_delayed_work - queue work on a workqueue after delay
208  * @wq: workqueue to use
209  * @dwork: delayable work to queue
210  * @delay: number of jiffies to wait before queueing
211  *
212  * Returns 0 if @work was already on a queue, non-zero otherwise.
213  */
214 int queue_delayed_work(struct workqueue_struct *wq,
215                         struct delayed_work *dwork, unsigned long delay)
216 {
217         if (delay == 0)
218                 return queue_work(wq, &dwork->work);
219
220         return queue_delayed_work_on(-1, wq, dwork, delay);
221 }
222 EXPORT_SYMBOL_GPL(queue_delayed_work);
223
224 /**
225  * queue_delayed_work_on - queue work on specific CPU after delay
226  * @cpu: CPU number to execute work on
227  * @wq: workqueue to use
228  * @dwork: work to queue
229  * @delay: number of jiffies to wait before queueing
230  *
231  * Returns 0 if @work was already on a queue, non-zero otherwise.
232  */
233 int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
234                         struct delayed_work *dwork, unsigned long delay)
235 {
236         int ret = 0;
237         struct timer_list *timer = &dwork->timer;
238         struct work_struct *work = &dwork->work;
239
240         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
241                 BUG_ON(timer_pending(timer));
242                 BUG_ON(!list_empty(&work->entry));
243
244                 timer_stats_timer_set_start_info(&dwork->timer);
245
246                 /* This stores cwq for the moment, for the timer_fn */
247                 set_wq_data(work, wq_per_cpu(wq, raw_smp_processor_id()));
248                 timer->expires = jiffies + delay;
249                 timer->data = (unsigned long)dwork;
250                 timer->function = delayed_work_timer_fn;
251
252                 if (unlikely(cpu >= 0))
253                         add_timer_on(timer, cpu);
254                 else
255                         add_timer(timer);
256                 ret = 1;
257         }
258         return ret;
259 }
260 EXPORT_SYMBOL_GPL(queue_delayed_work_on);
261
262 static void run_workqueue(struct cpu_workqueue_struct *cwq)
263 {
264         spin_lock_irq(&cwq->lock);
265         cwq->run_depth++;
266         if (cwq->run_depth > 3) {
267                 /* morton gets to eat his hat */
268                 printk("%s: recursion depth exceeded: %d\n",
269                         __func__, cwq->run_depth);
270                 dump_stack();
271         }
272         while (!list_empty(&cwq->worklist)) {
273                 struct work_struct *work = list_entry(cwq->worklist.next,
274                                                 struct work_struct, entry);
275                 work_func_t f = work->func;
276 #ifdef CONFIG_LOCKDEP
277                 /*
278                  * It is permissible to free the struct work_struct
279                  * from inside the function that is called from it,
280                  * this we need to take into account for lockdep too.
281                  * To avoid bogus "held lock freed" warnings as well
282                  * as problems when looking into work->lockdep_map,
283                  * make a copy and use that here.
284                  */
285                 struct lockdep_map lockdep_map = work->lockdep_map;
286 #endif
287
288                 cwq->current_work = work;
289                 list_del_init(cwq->worklist.next);
290                 spin_unlock_irq(&cwq->lock);
291
292                 BUG_ON(get_wq_data(work) != cwq);
293                 work_clear_pending(work);
294                 lock_map_acquire(&cwq->wq->lockdep_map);
295                 lock_map_acquire(&lockdep_map);
296                 f(work);
297                 lock_map_release(&lockdep_map);
298                 lock_map_release(&cwq->wq->lockdep_map);
299
300                 if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
301                         printk(KERN_ERR "BUG: workqueue leaked lock or atomic: "
302                                         "%s/0x%08x/%d\n",
303                                         current->comm, preempt_count(),
304                                         task_pid_nr(current));
305                         printk(KERN_ERR "    last function: ");
306                         print_symbol("%s\n", (unsigned long)f);
307                         debug_show_held_locks(current);
308                         dump_stack();
309                 }
310
311                 spin_lock_irq(&cwq->lock);
312                 cwq->current_work = NULL;
313         }
314         cwq->run_depth--;
315         spin_unlock_irq(&cwq->lock);
316 }
317
318 static int worker_thread(void *__cwq)
319 {
320         struct cpu_workqueue_struct *cwq = __cwq;
321         DEFINE_WAIT(wait);
322
323         if (cwq->wq->freezeable)
324                 set_freezable();
325
326         set_user_nice(current, -5);
327
328         for (;;) {
329                 prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE);
330                 if (!freezing(current) &&
331                     !kthread_should_stop() &&
332                     list_empty(&cwq->worklist))
333                         schedule();
334                 finish_wait(&cwq->more_work, &wait);
335
336                 try_to_freeze();
337
338                 if (kthread_should_stop())
339                         break;
340
341                 run_workqueue(cwq);
342         }
343
344         return 0;
345 }
346
347 struct wq_barrier {
348         struct work_struct      work;
349         struct completion       done;
350 };
351
352 static void wq_barrier_func(struct work_struct *work)
353 {
354         struct wq_barrier *barr = container_of(work, struct wq_barrier, work);
355         complete(&barr->done);
356 }
357
358 static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
359                         struct wq_barrier *barr, struct list_head *head)
360 {
361         INIT_WORK(&barr->work, wq_barrier_func);
362         __set_bit(WORK_STRUCT_PENDING, work_data_bits(&barr->work));
363
364         init_completion(&barr->done);
365
366         insert_work(cwq, &barr->work, head);
367 }
368
369 static int flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
370 {
371         int active;
372
373         if (cwq->thread == current) {
374                 /*
375                  * Probably keventd trying to flush its own queue. So simply run
376                  * it by hand rather than deadlocking.
377                  */
378                 run_workqueue(cwq);
379                 active = 1;
380         } else {
381                 struct wq_barrier barr;
382
383                 active = 0;
384                 spin_lock_irq(&cwq->lock);
385                 if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) {
386                         insert_wq_barrier(cwq, &barr, &cwq->worklist);
387                         active = 1;
388                 }
389                 spin_unlock_irq(&cwq->lock);
390
391                 if (active)
392                         wait_for_completion(&barr.done);
393         }
394
395         return active;
396 }
397
398 /**
399  * flush_workqueue - ensure that any scheduled work has run to completion.
400  * @wq: workqueue to flush
401  *
402  * Forces execution of the workqueue and blocks until its completion.
403  * This is typically used in driver shutdown handlers.
404  *
405  * We sleep until all works which were queued on entry have been handled,
406  * but we are not livelocked by new incoming ones.
407  *
408  * This function used to run the workqueues itself.  Now we just wait for the
409  * helper threads to do it.
410  */
411 void flush_workqueue(struct workqueue_struct *wq)
412 {
413         const cpumask_t *cpu_map = wq_cpu_map(wq);
414         int cpu;
415
416         might_sleep();
417         lock_map_acquire(&wq->lockdep_map);
418         lock_map_release(&wq->lockdep_map);
419         for_each_cpu_mask_nr(cpu, *cpu_map)
420                 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
421 }
422 EXPORT_SYMBOL_GPL(flush_workqueue);
423
424 /**
425  * flush_work - block until a work_struct's callback has terminated
426  * @work: the work which is to be flushed
427  *
428  * Returns false if @work has already terminated.
429  *
430  * It is expected that, prior to calling flush_work(), the caller has
431  * arranged for the work to not be requeued, otherwise it doesn't make
432  * sense to use this function.
433  */
434 int flush_work(struct work_struct *work)
435 {
436         struct cpu_workqueue_struct *cwq;
437         struct list_head *prev;
438         struct wq_barrier barr;
439
440         might_sleep();
441         cwq = get_wq_data(work);
442         if (!cwq)
443                 return 0;
444
445         lock_map_acquire(&cwq->wq->lockdep_map);
446         lock_map_release(&cwq->wq->lockdep_map);
447
448         prev = NULL;
449         spin_lock_irq(&cwq->lock);
450         if (!list_empty(&work->entry)) {
451                 /*
452                  * See the comment near try_to_grab_pending()->smp_rmb().
453                  * If it was re-queued under us we are not going to wait.
454                  */
455                 smp_rmb();
456                 if (unlikely(cwq != get_wq_data(work)))
457                         goto out;
458                 prev = &work->entry;
459         } else {
460                 if (cwq->current_work != work)
461                         goto out;
462                 prev = &cwq->worklist;
463         }
464         insert_wq_barrier(cwq, &barr, prev->next);
465 out:
466         spin_unlock_irq(&cwq->lock);
467         if (!prev)
468                 return 0;
469
470         wait_for_completion(&barr.done);
471         return 1;
472 }
473 EXPORT_SYMBOL_GPL(flush_work);
474
475 /*
476  * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit,
477  * so this work can't be re-armed in any way.
478  */
479 static int try_to_grab_pending(struct work_struct *work)
480 {
481         struct cpu_workqueue_struct *cwq;
482         int ret = -1;
483
484         if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work)))
485                 return 0;
486
487         /*
488          * The queueing is in progress, or it is already queued. Try to
489          * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
490          */
491
492         cwq = get_wq_data(work);
493         if (!cwq)
494                 return ret;
495
496         spin_lock_irq(&cwq->lock);
497         if (!list_empty(&work->entry)) {
498                 /*
499                  * This work is queued, but perhaps we locked the wrong cwq.
500                  * In that case we must see the new value after rmb(), see
501                  * insert_work()->wmb().
502                  */
503                 smp_rmb();
504                 if (cwq == get_wq_data(work)) {
505                         list_del_init(&work->entry);
506                         ret = 1;
507                 }
508         }
509         spin_unlock_irq(&cwq->lock);
510
511         return ret;
512 }
513
514 static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq,
515                                 struct work_struct *work)
516 {
517         struct wq_barrier barr;
518         int running = 0;
519
520         spin_lock_irq(&cwq->lock);
521         if (unlikely(cwq->current_work == work)) {
522                 insert_wq_barrier(cwq, &barr, cwq->worklist.next);
523                 running = 1;
524         }
525         spin_unlock_irq(&cwq->lock);
526
527         if (unlikely(running))
528                 wait_for_completion(&barr.done);
529 }
530
531 static void wait_on_work(struct work_struct *work)
532 {
533         struct cpu_workqueue_struct *cwq;
534         struct workqueue_struct *wq;
535         const cpumask_t *cpu_map;
536         int cpu;
537
538         might_sleep();
539
540         lock_map_acquire(&work->lockdep_map);
541         lock_map_release(&work->lockdep_map);
542
543         cwq = get_wq_data(work);
544         if (!cwq)
545                 return;
546
547         wq = cwq->wq;
548         cpu_map = wq_cpu_map(wq);
549
550         for_each_cpu_mask_nr(cpu, *cpu_map)
551                 wait_on_cpu_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
552 }
553
554 static int __cancel_work_timer(struct work_struct *work,
555                                 struct timer_list* timer)
556 {
557         int ret;
558
559         do {
560                 ret = (timer && likely(del_timer(timer)));
561                 if (!ret)
562                         ret = try_to_grab_pending(work);
563                 wait_on_work(work);
564         } while (unlikely(ret < 0));
565
566         work_clear_pending(work);
567         return ret;
568 }
569
570 /**
571  * cancel_work_sync - block until a work_struct's callback has terminated
572  * @work: the work which is to be flushed
573  *
574  * Returns true if @work was pending.
575  *
576  * cancel_work_sync() will cancel the work if it is queued. If the work's
577  * callback appears to be running, cancel_work_sync() will block until it
578  * has completed.
579  *
580  * It is possible to use this function if the work re-queues itself. It can
581  * cancel the work even if it migrates to another workqueue, however in that
582  * case it only guarantees that work->func() has completed on the last queued
583  * workqueue.
584  *
585  * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not
586  * pending, otherwise it goes into a busy-wait loop until the timer expires.
587  *
588  * The caller must ensure that workqueue_struct on which this work was last
589  * queued can't be destroyed before this function returns.
590  */
591 int cancel_work_sync(struct work_struct *work)
592 {
593         return __cancel_work_timer(work, NULL);
594 }
595 EXPORT_SYMBOL_GPL(cancel_work_sync);
596
597 /**
598  * cancel_delayed_work_sync - reliably kill off a delayed work.
599  * @dwork: the delayed work struct
600  *
601  * Returns true if @dwork was pending.
602  *
603  * It is possible to use this function if @dwork rearms itself via queue_work()
604  * or queue_delayed_work(). See also the comment for cancel_work_sync().
605  */
606 int cancel_delayed_work_sync(struct delayed_work *dwork)
607 {
608         return __cancel_work_timer(&dwork->work, &dwork->timer);
609 }
610 EXPORT_SYMBOL(cancel_delayed_work_sync);
611
612 static struct workqueue_struct *keventd_wq __read_mostly;
613
614 /**
615  * schedule_work - put work task in global workqueue
616  * @work: job to be done
617  *
618  * This puts a job in the kernel-global workqueue.
619  */
620 int schedule_work(struct work_struct *work)
621 {
622         return queue_work(keventd_wq, work);
623 }
624 EXPORT_SYMBOL(schedule_work);
625
626 /*
627  * schedule_work_on - put work task on a specific cpu
628  * @cpu: cpu to put the work task on
629  * @work: job to be done
630  *
631  * This puts a job on a specific cpu
632  */
633 int schedule_work_on(int cpu, struct work_struct *work)
634 {
635         return queue_work_on(cpu, keventd_wq, work);
636 }
637 EXPORT_SYMBOL(schedule_work_on);
638
639 /**
640  * schedule_delayed_work - put work task in global workqueue after delay
641  * @dwork: job to be done
642  * @delay: number of jiffies to wait or 0 for immediate execution
643  *
644  * After waiting for a given time this puts a job in the kernel-global
645  * workqueue.
646  */
647 int schedule_delayed_work(struct delayed_work *dwork,
648                                         unsigned long delay)
649 {
650         return queue_delayed_work(keventd_wq, dwork, delay);
651 }
652 EXPORT_SYMBOL(schedule_delayed_work);
653
654 /**
655  * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
656  * @cpu: cpu to use
657  * @dwork: job to be done
658  * @delay: number of jiffies to wait
659  *
660  * After waiting for a given time this puts a job in the kernel-global
661  * workqueue on the specified CPU.
662  */
663 int schedule_delayed_work_on(int cpu,
664                         struct delayed_work *dwork, unsigned long delay)
665 {
666         return queue_delayed_work_on(cpu, keventd_wq, dwork, delay);
667 }
668 EXPORT_SYMBOL(schedule_delayed_work_on);
669
670 /**
671  * schedule_on_each_cpu - call a function on each online CPU from keventd
672  * @func: the function to call
673  *
674  * Returns zero on success.
675  * Returns -ve errno on failure.
676  *
677  * schedule_on_each_cpu() is very slow.
678  */
679 int schedule_on_each_cpu(work_func_t func)
680 {
681         int cpu;
682         struct work_struct *works;
683
684         works = alloc_percpu(struct work_struct);
685         if (!works)
686                 return -ENOMEM;
687
688         get_online_cpus();
689         for_each_online_cpu(cpu) {
690                 struct work_struct *work = per_cpu_ptr(works, cpu);
691
692                 INIT_WORK(work, func);
693                 schedule_work_on(cpu, work);
694         }
695         for_each_online_cpu(cpu)
696                 flush_work(per_cpu_ptr(works, cpu));
697         put_online_cpus();
698         free_percpu(works);
699         return 0;
700 }
701
702 void flush_scheduled_work(void)
703 {
704         flush_workqueue(keventd_wq);
705 }
706 EXPORT_SYMBOL(flush_scheduled_work);
707
708 /**
709  * execute_in_process_context - reliably execute the routine with user context
710  * @fn:         the function to execute
711  * @ew:         guaranteed storage for the execute work structure (must
712  *              be available when the work executes)
713  *
714  * Executes the function immediately if process context is available,
715  * otherwise schedules the function for delayed execution.
716  *
717  * Returns:     0 - function was executed
718  *              1 - function was scheduled for execution
719  */
720 int execute_in_process_context(work_func_t fn, struct execute_work *ew)
721 {
722         if (!in_interrupt()) {
723                 fn(&ew->work);
724                 return 0;
725         }
726
727         INIT_WORK(&ew->work, fn);
728         schedule_work(&ew->work);
729
730         return 1;
731 }
732 EXPORT_SYMBOL_GPL(execute_in_process_context);
733
734 int keventd_up(void)
735 {
736         return keventd_wq != NULL;
737 }
738
739 int current_is_keventd(void)
740 {
741         struct cpu_workqueue_struct *cwq;
742         int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */
743         int ret = 0;
744
745         BUG_ON(!keventd_wq);
746
747         cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
748         if (current == cwq->thread)
749                 ret = 1;
750
751         return ret;
752
753 }
754
755 static struct cpu_workqueue_struct *
756 init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
757 {
758         struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
759
760         cwq->wq = wq;
761         spin_lock_init(&cwq->lock);
762         INIT_LIST_HEAD(&cwq->worklist);
763         init_waitqueue_head(&cwq->more_work);
764
765         return cwq;
766 }
767
768 static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
769 {
770         struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
771         struct workqueue_struct *wq = cwq->wq;
772         const char *fmt = is_single_threaded(wq) ? "%s" : "%s/%d";
773         struct task_struct *p;
774
775         p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu);
776         /*
777          * Nobody can add the work_struct to this cwq,
778          *      if (caller is __create_workqueue)
779          *              nobody should see this wq
780          *      else // caller is CPU_UP_PREPARE
781          *              cpu is not on cpu_online_map
782          * so we can abort safely.
783          */
784         if (IS_ERR(p))
785                 return PTR_ERR(p);
786         if (cwq->wq->rt)
787                 sched_setscheduler_nocheck(p, SCHED_FIFO, &param);
788         cwq->thread = p;
789
790         return 0;
791 }
792
793 static void start_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
794 {
795         struct task_struct *p = cwq->thread;
796
797         if (p != NULL) {
798                 if (cpu >= 0)
799                         kthread_bind(p, cpu);
800                 wake_up_process(p);
801         }
802 }
803
804 struct workqueue_struct *__create_workqueue_key(const char *name,
805                                                 int singlethread,
806                                                 int freezeable,
807                                                 int rt,
808                                                 struct lock_class_key *key,
809                                                 const char *lock_name)
810 {
811         struct workqueue_struct *wq;
812         struct cpu_workqueue_struct *cwq;
813         int err = 0, cpu;
814
815         wq = kzalloc(sizeof(*wq), GFP_KERNEL);
816         if (!wq)
817                 return NULL;
818
819         wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
820         if (!wq->cpu_wq) {
821                 kfree(wq);
822                 return NULL;
823         }
824
825         wq->name = name;
826         lockdep_init_map(&wq->lockdep_map, lock_name, key, 0);
827         wq->singlethread = singlethread;
828         wq->freezeable = freezeable;
829         wq->rt = rt;
830         INIT_LIST_HEAD(&wq->list);
831
832         if (singlethread) {
833                 cwq = init_cpu_workqueue(wq, singlethread_cpu);
834                 err = create_workqueue_thread(cwq, singlethread_cpu);
835                 start_workqueue_thread(cwq, -1);
836         } else {
837                 cpu_maps_update_begin();
838                 /*
839                  * We must place this wq on list even if the code below fails.
840                  * cpu_down(cpu) can remove cpu from cpu_populated_map before
841                  * destroy_workqueue() takes the lock, in that case we leak
842                  * cwq[cpu]->thread.
843                  */
844                 spin_lock(&workqueue_lock);
845                 list_add(&wq->list, &workqueues);
846                 spin_unlock(&workqueue_lock);
847                 /*
848                  * We must initialize cwqs for each possible cpu even if we
849                  * are going to call destroy_workqueue() finally. Otherwise
850                  * cpu_up() can hit the uninitialized cwq once we drop the
851                  * lock.
852                  */
853                 for_each_possible_cpu(cpu) {
854                         cwq = init_cpu_workqueue(wq, cpu);
855                         if (err || !cpu_online(cpu))
856                                 continue;
857                         err = create_workqueue_thread(cwq, cpu);
858                         start_workqueue_thread(cwq, cpu);
859                 }
860                 cpu_maps_update_done();
861         }
862
863         if (err) {
864                 destroy_workqueue(wq);
865                 wq = NULL;
866         }
867         return wq;
868 }
869 EXPORT_SYMBOL_GPL(__create_workqueue_key);
870
871 static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq)
872 {
873         /*
874          * Our caller is either destroy_workqueue() or CPU_POST_DEAD,
875          * cpu_add_remove_lock protects cwq->thread.
876          */
877         if (cwq->thread == NULL)
878                 return;
879
880         lock_map_acquire(&cwq->wq->lockdep_map);
881         lock_map_release(&cwq->wq->lockdep_map);
882
883         flush_cpu_workqueue(cwq);
884         /*
885          * If the caller is CPU_POST_DEAD and cwq->worklist was not empty,
886          * a concurrent flush_workqueue() can insert a barrier after us.
887          * However, in that case run_workqueue() won't return and check
888          * kthread_should_stop() until it flushes all work_struct's.
889          * When ->worklist becomes empty it is safe to exit because no
890          * more work_structs can be queued on this cwq: flush_workqueue
891          * checks list_empty(), and a "normal" queue_work() can't use
892          * a dead CPU.
893          */
894         kthread_stop(cwq->thread);
895         cwq->thread = NULL;
896 }
897
898 /**
899  * destroy_workqueue - safely terminate a workqueue
900  * @wq: target workqueue
901  *
902  * Safely destroy a workqueue. All work currently pending will be done first.
903  */
904 void destroy_workqueue(struct workqueue_struct *wq)
905 {
906         const cpumask_t *cpu_map = wq_cpu_map(wq);
907         int cpu;
908
909         cpu_maps_update_begin();
910         spin_lock(&workqueue_lock);
911         list_del(&wq->list);
912         spin_unlock(&workqueue_lock);
913
914         for_each_cpu_mask_nr(cpu, *cpu_map)
915                 cleanup_workqueue_thread(per_cpu_ptr(wq->cpu_wq, cpu));
916         cpu_maps_update_done();
917
918         free_percpu(wq->cpu_wq);
919         kfree(wq);
920 }
921 EXPORT_SYMBOL_GPL(destroy_workqueue);
922
923 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
924                                                 unsigned long action,
925                                                 void *hcpu)
926 {
927         unsigned int cpu = (unsigned long)hcpu;
928         struct cpu_workqueue_struct *cwq;
929         struct workqueue_struct *wq;
930         int ret = NOTIFY_OK;
931
932         action &= ~CPU_TASKS_FROZEN;
933
934         switch (action) {
935         case CPU_UP_PREPARE:
936                 cpu_set(cpu, cpu_populated_map);
937         }
938 undo:
939         list_for_each_entry(wq, &workqueues, list) {
940                 cwq = per_cpu_ptr(wq->cpu_wq, cpu);
941
942                 switch (action) {
943                 case CPU_UP_PREPARE:
944                         if (!create_workqueue_thread(cwq, cpu))
945                                 break;
946                         printk(KERN_ERR "workqueue [%s] for %i failed\n",
947                                 wq->name, cpu);
948                         action = CPU_UP_CANCELED;
949                         ret = NOTIFY_BAD;
950                         goto undo;
951
952                 case CPU_ONLINE:
953                         start_workqueue_thread(cwq, cpu);
954                         break;
955
956                 case CPU_UP_CANCELED:
957                         start_workqueue_thread(cwq, -1);
958                 case CPU_POST_DEAD:
959                         cleanup_workqueue_thread(cwq);
960                         break;
961                 }
962         }
963
964         switch (action) {
965         case CPU_UP_CANCELED:
966         case CPU_POST_DEAD:
967                 cpu_clear(cpu, cpu_populated_map);
968         }
969
970         return ret;
971 }
972
973 void __init init_workqueues(void)
974 {
975         cpu_populated_map = cpu_online_map;
976         singlethread_cpu = first_cpu(cpu_possible_map);
977         cpu_singlethread_map = cpumask_of_cpu(singlethread_cpu);
978         hotcpu_notifier(workqueue_cpu_callback, 0);
979         keventd_wq = create_workqueue("events");
980         BUG_ON(!keventd_wq);
981 }