ocfs2: allocation hints
[linux-2.6] / 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 <andrewm@uow.edu.au>
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 <clameter@sgi.com>.
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
32 /*
33  * The per-CPU workqueue (if single thread, we always use the first
34  * possible cpu).
35  *
36  * The sequence counters are for flush_scheduled_work().  It wants to wait
37  * until until all currently-scheduled works are completed, but it doesn't
38  * want to be livelocked by new, incoming ones.  So it waits until
39  * remove_sequence is >= the insert_sequence which pertained when
40  * flush_scheduled_work() was called.
41  */
42 struct cpu_workqueue_struct {
43
44         spinlock_t lock;
45
46         long remove_sequence;   /* Least-recently added (next to run) */
47         long insert_sequence;   /* Next to add */
48
49         struct list_head worklist;
50         wait_queue_head_t more_work;
51         wait_queue_head_t work_done;
52
53         struct workqueue_struct *wq;
54         struct task_struct *thread;
55
56         int run_depth;          /* Detect run_workqueue() recursion depth */
57 } ____cacheline_aligned;
58
59 /*
60  * The externally visible workqueue abstraction is an array of
61  * per-CPU workqueues:
62  */
63 struct workqueue_struct {
64         struct cpu_workqueue_struct *cpu_wq;
65         const char *name;
66         struct list_head list;  /* Empty if single thread */
67 };
68
69 /* All the per-cpu workqueues on the system, for hotplug cpu to add/remove
70    threads to each one as cpus come/go. */
71 static DEFINE_SPINLOCK(workqueue_lock);
72 static LIST_HEAD(workqueues);
73
74 static int singlethread_cpu;
75
76 /* If it's single threaded, it isn't in the list of workqueues. */
77 static inline int is_single_threaded(struct workqueue_struct *wq)
78 {
79         return list_empty(&wq->list);
80 }
81
82 /* Preempt must be disabled. */
83 static void __queue_work(struct cpu_workqueue_struct *cwq,
84                          struct work_struct *work)
85 {
86         unsigned long flags;
87
88         spin_lock_irqsave(&cwq->lock, flags);
89         work->wq_data = cwq;
90         list_add_tail(&work->entry, &cwq->worklist);
91         cwq->insert_sequence++;
92         wake_up(&cwq->more_work);
93         spin_unlock_irqrestore(&cwq->lock, flags);
94 }
95
96 /**
97  * queue_work - queue work on a workqueue
98  * @wq: workqueue to use
99  * @work: work to queue
100  *
101  * Returns non-zero if it was successfully added.
102  *
103  * We queue the work to the CPU it was submitted, but there is no
104  * guarantee that it will be processed by that CPU.
105  */
106 int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
107 {
108         int ret = 0, cpu = get_cpu();
109
110         if (!test_and_set_bit(0, &work->pending)) {
111                 if (unlikely(is_single_threaded(wq)))
112                         cpu = singlethread_cpu;
113                 BUG_ON(!list_empty(&work->entry));
114                 __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
115                 ret = 1;
116         }
117         put_cpu();
118         return ret;
119 }
120 EXPORT_SYMBOL_GPL(queue_work);
121
122 static void delayed_work_timer_fn(unsigned long __data)
123 {
124         struct work_struct *work = (struct work_struct *)__data;
125         struct workqueue_struct *wq = work->wq_data;
126         int cpu = smp_processor_id();
127
128         if (unlikely(is_single_threaded(wq)))
129                 cpu = singlethread_cpu;
130
131         __queue_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
132 }
133
134 /**
135  * queue_delayed_work - queue work on a workqueue after delay
136  * @wq: workqueue to use
137  * @work: work to queue
138  * @delay: number of jiffies to wait before queueing
139  *
140  * Returns non-zero if it was successfully added.
141  */
142 int fastcall queue_delayed_work(struct workqueue_struct *wq,
143                         struct work_struct *work, unsigned long delay)
144 {
145         int ret = 0;
146         struct timer_list *timer = &work->timer;
147
148         if (!test_and_set_bit(0, &work->pending)) {
149                 BUG_ON(timer_pending(timer));
150                 BUG_ON(!list_empty(&work->entry));
151
152                 /* This stores wq for the moment, for the timer_fn */
153                 work->wq_data = wq;
154                 timer->expires = jiffies + delay;
155                 timer->data = (unsigned long)work;
156                 timer->function = delayed_work_timer_fn;
157                 add_timer(timer);
158                 ret = 1;
159         }
160         return ret;
161 }
162 EXPORT_SYMBOL_GPL(queue_delayed_work);
163
164 /**
165  * queue_delayed_work_on - queue work on specific CPU after delay
166  * @cpu: CPU number to execute work on
167  * @wq: workqueue to use
168  * @work: work to queue
169  * @delay: number of jiffies to wait before queueing
170  *
171  * Returns non-zero if it was successfully added.
172  */
173 int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
174                         struct work_struct *work, unsigned long delay)
175 {
176         int ret = 0;
177         struct timer_list *timer = &work->timer;
178
179         if (!test_and_set_bit(0, &work->pending)) {
180                 BUG_ON(timer_pending(timer));
181                 BUG_ON(!list_empty(&work->entry));
182
183                 /* This stores wq for the moment, for the timer_fn */
184                 work->wq_data = wq;
185                 timer->expires = jiffies + delay;
186                 timer->data = (unsigned long)work;
187                 timer->function = delayed_work_timer_fn;
188                 add_timer_on(timer, cpu);
189                 ret = 1;
190         }
191         return ret;
192 }
193 EXPORT_SYMBOL_GPL(queue_delayed_work_on);
194
195 static void run_workqueue(struct cpu_workqueue_struct *cwq)
196 {
197         unsigned long flags;
198
199         /*
200          * Keep taking off work from the queue until
201          * done.
202          */
203         spin_lock_irqsave(&cwq->lock, flags);
204         cwq->run_depth++;
205         if (cwq->run_depth > 3) {
206                 /* morton gets to eat his hat */
207                 printk("%s: recursion depth exceeded: %d\n",
208                         __FUNCTION__, cwq->run_depth);
209                 dump_stack();
210         }
211         while (!list_empty(&cwq->worklist)) {
212                 struct work_struct *work = list_entry(cwq->worklist.next,
213                                                 struct work_struct, entry);
214                 void (*f) (void *) = work->func;
215                 void *data = work->data;
216
217                 list_del_init(cwq->worklist.next);
218                 spin_unlock_irqrestore(&cwq->lock, flags);
219
220                 BUG_ON(work->wq_data != cwq);
221                 clear_bit(0, &work->pending);
222                 f(data);
223
224                 spin_lock_irqsave(&cwq->lock, flags);
225                 cwq->remove_sequence++;
226                 wake_up(&cwq->work_done);
227         }
228         cwq->run_depth--;
229         spin_unlock_irqrestore(&cwq->lock, flags);
230 }
231
232 static int worker_thread(void *__cwq)
233 {
234         struct cpu_workqueue_struct *cwq = __cwq;
235         DECLARE_WAITQUEUE(wait, current);
236         struct k_sigaction sa;
237         sigset_t blocked;
238
239         current->flags |= PF_NOFREEZE;
240
241         set_user_nice(current, -5);
242
243         /* Block and flush all signals */
244         sigfillset(&blocked);
245         sigprocmask(SIG_BLOCK, &blocked, NULL);
246         flush_signals(current);
247
248         /* SIG_IGN makes children autoreap: see do_notify_parent(). */
249         sa.sa.sa_handler = SIG_IGN;
250         sa.sa.sa_flags = 0;
251         siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
252         do_sigaction(SIGCHLD, &sa, (struct k_sigaction *)0);
253
254         set_current_state(TASK_INTERRUPTIBLE);
255         while (!kthread_should_stop()) {
256                 add_wait_queue(&cwq->more_work, &wait);
257                 if (list_empty(&cwq->worklist))
258                         schedule();
259                 else
260                         __set_current_state(TASK_RUNNING);
261                 remove_wait_queue(&cwq->more_work, &wait);
262
263                 if (!list_empty(&cwq->worklist))
264                         run_workqueue(cwq);
265                 set_current_state(TASK_INTERRUPTIBLE);
266         }
267         __set_current_state(TASK_RUNNING);
268         return 0;
269 }
270
271 static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
272 {
273         if (cwq->thread == current) {
274                 /*
275                  * Probably keventd trying to flush its own queue. So simply run
276                  * it by hand rather than deadlocking.
277                  */
278                 run_workqueue(cwq);
279         } else {
280                 DEFINE_WAIT(wait);
281                 long sequence_needed;
282
283                 spin_lock_irq(&cwq->lock);
284                 sequence_needed = cwq->insert_sequence;
285
286                 while (sequence_needed - cwq->remove_sequence > 0) {
287                         prepare_to_wait(&cwq->work_done, &wait,
288                                         TASK_UNINTERRUPTIBLE);
289                         spin_unlock_irq(&cwq->lock);
290                         schedule();
291                         spin_lock_irq(&cwq->lock);
292                 }
293                 finish_wait(&cwq->work_done, &wait);
294                 spin_unlock_irq(&cwq->lock);
295         }
296 }
297
298 /**
299  * flush_workqueue - ensure that any scheduled work has run to completion.
300  * @wq: workqueue to flush
301  *
302  * Forces execution of the workqueue and blocks until its completion.
303  * This is typically used in driver shutdown handlers.
304  *
305  * This function will sample each workqueue's current insert_sequence number and
306  * will sleep until the head sequence is greater than or equal to that.  This
307  * means that we sleep until all works which were queued on entry have been
308  * handled, but we are not livelocked by new incoming ones.
309  *
310  * This function used to run the workqueues itself.  Now we just wait for the
311  * helper threads to do it.
312  */
313 void fastcall flush_workqueue(struct workqueue_struct *wq)
314 {
315         might_sleep();
316
317         if (is_single_threaded(wq)) {
318                 /* Always use first cpu's area. */
319                 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, singlethread_cpu));
320         } else {
321                 int cpu;
322
323                 lock_cpu_hotplug();
324                 for_each_online_cpu(cpu)
325                         flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
326                 unlock_cpu_hotplug();
327         }
328 }
329 EXPORT_SYMBOL_GPL(flush_workqueue);
330
331 static struct task_struct *create_workqueue_thread(struct workqueue_struct *wq,
332                                                    int cpu)
333 {
334         struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
335         struct task_struct *p;
336
337         spin_lock_init(&cwq->lock);
338         cwq->wq = wq;
339         cwq->thread = NULL;
340         cwq->insert_sequence = 0;
341         cwq->remove_sequence = 0;
342         INIT_LIST_HEAD(&cwq->worklist);
343         init_waitqueue_head(&cwq->more_work);
344         init_waitqueue_head(&cwq->work_done);
345
346         if (is_single_threaded(wq))
347                 p = kthread_create(worker_thread, cwq, "%s", wq->name);
348         else
349                 p = kthread_create(worker_thread, cwq, "%s/%d", wq->name, cpu);
350         if (IS_ERR(p))
351                 return NULL;
352         cwq->thread = p;
353         return p;
354 }
355
356 struct workqueue_struct *__create_workqueue(const char *name,
357                                             int singlethread)
358 {
359         int cpu, destroy = 0;
360         struct workqueue_struct *wq;
361         struct task_struct *p;
362
363         wq = kzalloc(sizeof(*wq), GFP_KERNEL);
364         if (!wq)
365                 return NULL;
366
367         wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
368         if (!wq->cpu_wq) {
369                 kfree(wq);
370                 return NULL;
371         }
372
373         wq->name = name;
374         /* We don't need the distraction of CPUs appearing and vanishing. */
375         lock_cpu_hotplug();
376         if (singlethread) {
377                 INIT_LIST_HEAD(&wq->list);
378                 p = create_workqueue_thread(wq, singlethread_cpu);
379                 if (!p)
380                         destroy = 1;
381                 else
382                         wake_up_process(p);
383         } else {
384                 spin_lock(&workqueue_lock);
385                 list_add(&wq->list, &workqueues);
386                 spin_unlock(&workqueue_lock);
387                 for_each_online_cpu(cpu) {
388                         p = create_workqueue_thread(wq, cpu);
389                         if (p) {
390                                 kthread_bind(p, cpu);
391                                 wake_up_process(p);
392                         } else
393                                 destroy = 1;
394                 }
395         }
396         unlock_cpu_hotplug();
397
398         /*
399          * Was there any error during startup? If yes then clean up:
400          */
401         if (destroy) {
402                 destroy_workqueue(wq);
403                 wq = NULL;
404         }
405         return wq;
406 }
407 EXPORT_SYMBOL_GPL(__create_workqueue);
408
409 static void cleanup_workqueue_thread(struct workqueue_struct *wq, int cpu)
410 {
411         struct cpu_workqueue_struct *cwq;
412         unsigned long flags;
413         struct task_struct *p;
414
415         cwq = per_cpu_ptr(wq->cpu_wq, cpu);
416         spin_lock_irqsave(&cwq->lock, flags);
417         p = cwq->thread;
418         cwq->thread = NULL;
419         spin_unlock_irqrestore(&cwq->lock, flags);
420         if (p)
421                 kthread_stop(p);
422 }
423
424 /**
425  * destroy_workqueue - safely terminate a workqueue
426  * @wq: target workqueue
427  *
428  * Safely destroy a workqueue. All work currently pending will be done first.
429  */
430 void destroy_workqueue(struct workqueue_struct *wq)
431 {
432         int cpu;
433
434         flush_workqueue(wq);
435
436         /* We don't need the distraction of CPUs appearing and vanishing. */
437         lock_cpu_hotplug();
438         if (is_single_threaded(wq))
439                 cleanup_workqueue_thread(wq, singlethread_cpu);
440         else {
441                 for_each_online_cpu(cpu)
442                         cleanup_workqueue_thread(wq, cpu);
443                 spin_lock(&workqueue_lock);
444                 list_del(&wq->list);
445                 spin_unlock(&workqueue_lock);
446         }
447         unlock_cpu_hotplug();
448         free_percpu(wq->cpu_wq);
449         kfree(wq);
450 }
451 EXPORT_SYMBOL_GPL(destroy_workqueue);
452
453 static struct workqueue_struct *keventd_wq;
454
455 /**
456  * schedule_work - put work task in global workqueue
457  * @work: job to be done
458  *
459  * This puts a job in the kernel-global workqueue.
460  */
461 int fastcall schedule_work(struct work_struct *work)
462 {
463         return queue_work(keventd_wq, work);
464 }
465 EXPORT_SYMBOL(schedule_work);
466
467 /**
468  * schedule_delayed_work - put work task in global workqueue after delay
469  * @work: job to be done
470  * @delay: number of jiffies to wait
471  *
472  * After waiting for a given time this puts a job in the kernel-global
473  * workqueue.
474  */
475 int fastcall schedule_delayed_work(struct work_struct *work, unsigned long delay)
476 {
477         return queue_delayed_work(keventd_wq, work, delay);
478 }
479 EXPORT_SYMBOL(schedule_delayed_work);
480
481 /**
482  * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
483  * @cpu: cpu to use
484  * @work: job to be done
485  * @delay: number of jiffies to wait
486  *
487  * After waiting for a given time this puts a job in the kernel-global
488  * workqueue on the specified CPU.
489  */
490 int schedule_delayed_work_on(int cpu,
491                         struct work_struct *work, unsigned long delay)
492 {
493         return queue_delayed_work_on(cpu, keventd_wq, work, delay);
494 }
495 EXPORT_SYMBOL(schedule_delayed_work_on);
496
497 /**
498  * schedule_on_each_cpu - call a function on each online CPU from keventd
499  * @func: the function to call
500  * @info: a pointer to pass to func()
501  *
502  * Returns zero on success.
503  * Returns -ve errno on failure.
504  *
505  * Appears to be racy against CPU hotplug.
506  *
507  * schedule_on_each_cpu() is very slow.
508  */
509 int schedule_on_each_cpu(void (*func)(void *info), void *info)
510 {
511         int cpu;
512         struct work_struct *works;
513
514         works = alloc_percpu(struct work_struct);
515         if (!works)
516                 return -ENOMEM;
517
518         for_each_online_cpu(cpu) {
519                 INIT_WORK(per_cpu_ptr(works, cpu), func, info);
520                 __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu),
521                                 per_cpu_ptr(works, cpu));
522         }
523         flush_workqueue(keventd_wq);
524         free_percpu(works);
525         return 0;
526 }
527
528 void flush_scheduled_work(void)
529 {
530         flush_workqueue(keventd_wq);
531 }
532 EXPORT_SYMBOL(flush_scheduled_work);
533
534 /**
535  * cancel_rearming_delayed_workqueue - reliably kill off a delayed
536  *                      work whose handler rearms the delayed work.
537  * @wq:   the controlling workqueue structure
538  * @work: the delayed work struct
539  */
540 void cancel_rearming_delayed_workqueue(struct workqueue_struct *wq,
541                                        struct work_struct *work)
542 {
543         while (!cancel_delayed_work(work))
544                 flush_workqueue(wq);
545 }
546 EXPORT_SYMBOL(cancel_rearming_delayed_workqueue);
547
548 /**
549  * cancel_rearming_delayed_work - reliably kill off a delayed keventd
550  *                      work whose handler rearms the delayed work.
551  * @work: the delayed work struct
552  */
553 void cancel_rearming_delayed_work(struct work_struct *work)
554 {
555         cancel_rearming_delayed_workqueue(keventd_wq, work);
556 }
557 EXPORT_SYMBOL(cancel_rearming_delayed_work);
558
559 /**
560  * execute_in_process_context - reliably execute the routine with user context
561  * @fn:         the function to execute
562  * @data:       data to pass to the function
563  * @ew:         guaranteed storage for the execute work structure (must
564  *              be available when the work executes)
565  *
566  * Executes the function immediately if process context is available,
567  * otherwise schedules the function for delayed execution.
568  *
569  * Returns:     0 - function was executed
570  *              1 - function was scheduled for execution
571  */
572 int execute_in_process_context(void (*fn)(void *data), void *data,
573                                struct execute_work *ew)
574 {
575         if (!in_interrupt()) {
576                 fn(data);
577                 return 0;
578         }
579
580         INIT_WORK(&ew->work, fn, data);
581         schedule_work(&ew->work);
582
583         return 1;
584 }
585 EXPORT_SYMBOL_GPL(execute_in_process_context);
586
587 int keventd_up(void)
588 {
589         return keventd_wq != NULL;
590 }
591
592 int current_is_keventd(void)
593 {
594         struct cpu_workqueue_struct *cwq;
595         int cpu = smp_processor_id();   /* preempt-safe: keventd is per-cpu */
596         int ret = 0;
597
598         BUG_ON(!keventd_wq);
599
600         cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
601         if (current == cwq->thread)
602                 ret = 1;
603
604         return ret;
605
606 }
607
608 #ifdef CONFIG_HOTPLUG_CPU
609 /* Take the work from this (downed) CPU. */
610 static void take_over_work(struct workqueue_struct *wq, unsigned int cpu)
611 {
612         struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
613         struct list_head list;
614         struct work_struct *work;
615
616         spin_lock_irq(&cwq->lock);
617         list_replace_init(&cwq->worklist, &list);
618
619         while (!list_empty(&list)) {
620                 printk("Taking work for %s\n", wq->name);
621                 work = list_entry(list.next,struct work_struct,entry);
622                 list_del(&work->entry);
623                 __queue_work(per_cpu_ptr(wq->cpu_wq, smp_processor_id()), work);
624         }
625         spin_unlock_irq(&cwq->lock);
626 }
627
628 /* We're holding the cpucontrol mutex here */
629 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
630                                   unsigned long action,
631                                   void *hcpu)
632 {
633         unsigned int hotcpu = (unsigned long)hcpu;
634         struct workqueue_struct *wq;
635
636         switch (action) {
637         case CPU_UP_PREPARE:
638                 /* Create a new workqueue thread for it. */
639                 list_for_each_entry(wq, &workqueues, list) {
640                         if (!create_workqueue_thread(wq, hotcpu)) {
641                                 printk("workqueue for %i failed\n", hotcpu);
642                                 return NOTIFY_BAD;
643                         }
644                 }
645                 break;
646
647         case CPU_ONLINE:
648                 /* Kick off worker threads. */
649                 list_for_each_entry(wq, &workqueues, list) {
650                         struct cpu_workqueue_struct *cwq;
651
652                         cwq = per_cpu_ptr(wq->cpu_wq, hotcpu);
653                         kthread_bind(cwq->thread, hotcpu);
654                         wake_up_process(cwq->thread);
655                 }
656                 break;
657
658         case CPU_UP_CANCELED:
659                 list_for_each_entry(wq, &workqueues, list) {
660                         if (!per_cpu_ptr(wq->cpu_wq, hotcpu)->thread)
661                                 continue;
662                         /* Unbind so it can run. */
663                         kthread_bind(per_cpu_ptr(wq->cpu_wq, hotcpu)->thread,
664                                      any_online_cpu(cpu_online_map));
665                         cleanup_workqueue_thread(wq, hotcpu);
666                 }
667                 break;
668
669         case CPU_DEAD:
670                 list_for_each_entry(wq, &workqueues, list)
671                         cleanup_workqueue_thread(wq, hotcpu);
672                 list_for_each_entry(wq, &workqueues, list)
673                         take_over_work(wq, hotcpu);
674                 break;
675         }
676
677         return NOTIFY_OK;
678 }
679 #endif
680
681 void init_workqueues(void)
682 {
683         singlethread_cpu = first_cpu(cpu_possible_map);
684         hotcpu_notifier(workqueue_cpu_callback, 0);
685         keventd_wq = create_workqueue("events");
686         BUG_ON(!keventd_wq);
687 }
688