]>
Commit | Line | Data |
---|---|---|
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 | #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 | #ifdef CONFIG_LOCKDEP | |
66 | struct lockdep_map lockdep_map; | |
67 | #endif | |
68 | }; | |
69 | ||
70 | /* Serializes the accesses to the list of workqueues. */ | |
71 | static DEFINE_SPINLOCK(workqueue_lock); | |
72 | static LIST_HEAD(workqueues); | |
73 | ||
74 | static int singlethread_cpu __read_mostly; | |
75 | static cpumask_t cpu_singlethread_map __read_mostly; | |
76 | /* | |
77 | * _cpu_down() first removes CPU from cpu_online_map, then CPU_DEAD | |
78 | * flushes cwq->worklist. This means that flush_workqueue/wait_on_work | |
79 | * which comes in between can't use for_each_online_cpu(). We could | |
80 | * use cpu_possible_map, the cpumask below is more a documentation | |
81 | * than optimization. | |
82 | */ | |
83 | static cpumask_t cpu_populated_map __read_mostly; | |
84 | ||
85 | /* If it's single threaded, it isn't in the list of workqueues. */ | |
86 | static inline int is_single_threaded(struct workqueue_struct *wq) | |
87 | { | |
88 | return wq->singlethread; | |
89 | } | |
90 | ||
91 | static const cpumask_t *wq_cpu_map(struct workqueue_struct *wq) | |
92 | { | |
93 | return is_single_threaded(wq) | |
94 | ? &cpu_singlethread_map : &cpu_populated_map; | |
95 | } | |
96 | ||
97 | static | |
98 | struct cpu_workqueue_struct *wq_per_cpu(struct workqueue_struct *wq, int cpu) | |
99 | { | |
100 | if (unlikely(is_single_threaded(wq))) | |
101 | cpu = singlethread_cpu; | |
102 | return per_cpu_ptr(wq->cpu_wq, cpu); | |
103 | } | |
104 | ||
105 | /* | |
106 | * Set the workqueue on which a work item is to be run | |
107 | * - Must *only* be called if the pending flag is set | |
108 | */ | |
109 | static inline void set_wq_data(struct work_struct *work, | |
110 | struct cpu_workqueue_struct *cwq) | |
111 | { | |
112 | unsigned long new; | |
113 | ||
114 | BUG_ON(!work_pending(work)); | |
115 | ||
116 | new = (unsigned long) cwq | (1UL << WORK_STRUCT_PENDING); | |
117 | new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work); | |
118 | atomic_long_set(&work->data, new); | |
119 | } | |
120 | ||
121 | static inline | |
122 | struct cpu_workqueue_struct *get_wq_data(struct work_struct *work) | |
123 | { | |
124 | return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK); | |
125 | } | |
126 | ||
127 | static void insert_work(struct cpu_workqueue_struct *cwq, | |
128 | struct work_struct *work, int tail) | |
129 | { | |
130 | set_wq_data(work, cwq); | |
131 | /* | |
132 | * Ensure that we get the right work->data if we see the | |
133 | * result of list_add() below, see try_to_grab_pending(). | |
134 | */ | |
135 | smp_wmb(); | |
136 | if (tail) | |
137 | list_add_tail(&work->entry, &cwq->worklist); | |
138 | else | |
139 | list_add(&work->entry, &cwq->worklist); | |
140 | wake_up(&cwq->more_work); | |
141 | } | |
142 | ||
143 | /* Preempt must be disabled. */ | |
144 | static void __queue_work(struct cpu_workqueue_struct *cwq, | |
145 | struct work_struct *work) | |
146 | { | |
147 | unsigned long flags; | |
148 | ||
149 | spin_lock_irqsave(&cwq->lock, flags); | |
150 | insert_work(cwq, work, 1); | |
151 | spin_unlock_irqrestore(&cwq->lock, flags); | |
152 | } | |
153 | ||
154 | /** | |
155 | * queue_work - queue work on a workqueue | |
156 | * @wq: workqueue to use | |
157 | * @work: work to queue | |
158 | * | |
159 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
160 | * | |
161 | * We queue the work to the CPU on which it was submitted, but if the CPU dies | |
162 | * it can be processed by another CPU. | |
163 | */ | |
164 | int queue_work(struct workqueue_struct *wq, struct work_struct *work) | |
165 | { | |
166 | int ret = 0; | |
167 | ||
168 | if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) { | |
169 | BUG_ON(!list_empty(&work->entry)); | |
170 | __queue_work(wq_per_cpu(wq, get_cpu()), work); | |
171 | put_cpu(); | |
172 | ret = 1; | |
173 | } | |
174 | return ret; | |
175 | } | |
176 | EXPORT_SYMBOL_GPL(queue_work); | |
177 | ||
178 | static void delayed_work_timer_fn(unsigned long __data) | |
179 | { | |
180 | struct delayed_work *dwork = (struct delayed_work *)__data; | |
181 | struct cpu_workqueue_struct *cwq = get_wq_data(&dwork->work); | |
182 | struct workqueue_struct *wq = cwq->wq; | |
183 | ||
184 | __queue_work(wq_per_cpu(wq, smp_processor_id()), &dwork->work); | |
185 | } | |
186 | ||
187 | /** | |
188 | * queue_delayed_work - queue work on a workqueue after delay | |
189 | * @wq: workqueue to use | |
190 | * @dwork: delayable work to queue | |
191 | * @delay: number of jiffies to wait before queueing | |
192 | * | |
193 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
194 | */ | |
195 | int queue_delayed_work(struct workqueue_struct *wq, | |
196 | struct delayed_work *dwork, unsigned long delay) | |
197 | { | |
198 | timer_stats_timer_set_start_info(&dwork->timer); | |
199 | if (delay == 0) | |
200 | return queue_work(wq, &dwork->work); | |
201 | ||
202 | return queue_delayed_work_on(-1, wq, dwork, delay); | |
203 | } | |
204 | EXPORT_SYMBOL_GPL(queue_delayed_work); | |
205 | ||
206 | /** | |
207 | * queue_delayed_work_on - queue work on specific CPU after delay | |
208 | * @cpu: CPU number to execute work on | |
209 | * @wq: workqueue to use | |
210 | * @dwork: work to queue | |
211 | * @delay: number of jiffies to wait before queueing | |
212 | * | |
213 | * Returns 0 if @work was already on a queue, non-zero otherwise. | |
214 | */ | |
215 | int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, | |
216 | struct delayed_work *dwork, unsigned long delay) | |
217 | { | |
218 | int ret = 0; | |
219 | struct timer_list *timer = &dwork->timer; | |
220 | struct work_struct *work = &dwork->work; | |
221 | ||
222 | timer_stats_timer_set_start_info(&dwork->timer); | |
223 | if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) { | |
224 | BUG_ON(timer_pending(timer)); | |
225 | BUG_ON(!list_empty(&work->entry)); | |
226 | ||
227 | /* This stores cwq for the moment, for the timer_fn */ | |
228 | set_wq_data(work, wq_per_cpu(wq, raw_smp_processor_id())); | |
229 | timer->expires = jiffies + delay; | |
230 | timer->data = (unsigned long)dwork; | |
231 | timer->function = delayed_work_timer_fn; | |
232 | ||
233 | if (unlikely(cpu >= 0)) | |
234 | add_timer_on(timer, cpu); | |
235 | else | |
236 | add_timer(timer); | |
237 | ret = 1; | |
238 | } | |
239 | return ret; | |
240 | } | |
241 | EXPORT_SYMBOL_GPL(queue_delayed_work_on); | |
242 | ||
243 | static void run_workqueue(struct cpu_workqueue_struct *cwq) | |
244 | { | |
245 | spin_lock_irq(&cwq->lock); | |
246 | cwq->run_depth++; | |
247 | if (cwq->run_depth > 3) { | |
248 | /* morton gets to eat his hat */ | |
249 | printk("%s: recursion depth exceeded: %d\n", | |
250 | __func__, cwq->run_depth); | |
251 | dump_stack(); | |
252 | } | |
253 | while (!list_empty(&cwq->worklist)) { | |
254 | struct work_struct *work = list_entry(cwq->worklist.next, | |
255 | struct work_struct, entry); | |
256 | work_func_t f = work->func; | |
257 | #ifdef CONFIG_LOCKDEP | |
258 | /* | |
259 | * It is permissible to free the struct work_struct | |
260 | * from inside the function that is called from it, | |
261 | * this we need to take into account for lockdep too. | |
262 | * To avoid bogus "held lock freed" warnings as well | |
263 | * as problems when looking into work->lockdep_map, | |
264 | * make a copy and use that here. | |
265 | */ | |
266 | struct lockdep_map lockdep_map = work->lockdep_map; | |
267 | #endif | |
268 | ||
269 | cwq->current_work = work; | |
270 | list_del_init(cwq->worklist.next); | |
271 | spin_unlock_irq(&cwq->lock); | |
272 | ||
273 | BUG_ON(get_wq_data(work) != cwq); | |
274 | work_clear_pending(work); | |
275 | lock_acquire(&cwq->wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_); | |
276 | lock_acquire(&lockdep_map, 0, 0, 0, 2, _THIS_IP_); | |
277 | f(work); | |
278 | lock_release(&lockdep_map, 1, _THIS_IP_); | |
279 | lock_release(&cwq->wq->lockdep_map, 1, _THIS_IP_); | |
280 | ||
281 | if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { | |
282 | printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " | |
283 | "%s/0x%08x/%d\n", | |
284 | current->comm, preempt_count(), | |
285 | task_pid_nr(current)); | |
286 | printk(KERN_ERR " last function: "); | |
287 | print_symbol("%s\n", (unsigned long)f); | |
288 | debug_show_held_locks(current); | |
289 | dump_stack(); | |
290 | } | |
291 | ||
292 | spin_lock_irq(&cwq->lock); | |
293 | cwq->current_work = NULL; | |
294 | } | |
295 | cwq->run_depth--; | |
296 | spin_unlock_irq(&cwq->lock); | |
297 | } | |
298 | ||
299 | static int worker_thread(void *__cwq) | |
300 | { | |
301 | struct cpu_workqueue_struct *cwq = __cwq; | |
302 | DEFINE_WAIT(wait); | |
303 | ||
304 | if (cwq->wq->freezeable) | |
305 | set_freezable(); | |
306 | ||
307 | set_user_nice(current, -5); | |
308 | ||
309 | for (;;) { | |
310 | prepare_to_wait(&cwq->more_work, &wait, TASK_INTERRUPTIBLE); | |
311 | if (!freezing(current) && | |
312 | !kthread_should_stop() && | |
313 | list_empty(&cwq->worklist)) | |
314 | schedule(); | |
315 | finish_wait(&cwq->more_work, &wait); | |
316 | ||
317 | try_to_freeze(); | |
318 | ||
319 | if (kthread_should_stop()) | |
320 | break; | |
321 | ||
322 | run_workqueue(cwq); | |
323 | } | |
324 | ||
325 | return 0; | |
326 | } | |
327 | ||
328 | struct wq_barrier { | |
329 | struct work_struct work; | |
330 | struct completion done; | |
331 | }; | |
332 | ||
333 | static void wq_barrier_func(struct work_struct *work) | |
334 | { | |
335 | struct wq_barrier *barr = container_of(work, struct wq_barrier, work); | |
336 | complete(&barr->done); | |
337 | } | |
338 | ||
339 | static void insert_wq_barrier(struct cpu_workqueue_struct *cwq, | |
340 | struct wq_barrier *barr, int tail) | |
341 | { | |
342 | INIT_WORK(&barr->work, wq_barrier_func); | |
343 | __set_bit(WORK_STRUCT_PENDING, work_data_bits(&barr->work)); | |
344 | ||
345 | init_completion(&barr->done); | |
346 | ||
347 | insert_work(cwq, &barr->work, tail); | |
348 | } | |
349 | ||
350 | static int flush_cpu_workqueue(struct cpu_workqueue_struct *cwq) | |
351 | { | |
352 | int active; | |
353 | ||
354 | if (cwq->thread == current) { | |
355 | /* | |
356 | * Probably keventd trying to flush its own queue. So simply run | |
357 | * it by hand rather than deadlocking. | |
358 | */ | |
359 | run_workqueue(cwq); | |
360 | active = 1; | |
361 | } else { | |
362 | struct wq_barrier barr; | |
363 | ||
364 | active = 0; | |
365 | spin_lock_irq(&cwq->lock); | |
366 | if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) { | |
367 | insert_wq_barrier(cwq, &barr, 1); | |
368 | active = 1; | |
369 | } | |
370 | spin_unlock_irq(&cwq->lock); | |
371 | ||
372 | if (active) | |
373 | wait_for_completion(&barr.done); | |
374 | } | |
375 | ||
376 | return active; | |
377 | } | |
378 | ||
379 | /** | |
380 | * flush_workqueue - ensure that any scheduled work has run to completion. | |
381 | * @wq: workqueue to flush | |
382 | * | |
383 | * Forces execution of the workqueue and blocks until its completion. | |
384 | * This is typically used in driver shutdown handlers. | |
385 | * | |
386 | * We sleep until all works which were queued on entry have been handled, | |
387 | * but we are not livelocked by new incoming ones. | |
388 | * | |
389 | * This function used to run the workqueues itself. Now we just wait for the | |
390 | * helper threads to do it. | |
391 | */ | |
392 | void flush_workqueue(struct workqueue_struct *wq) | |
393 | { | |
394 | const cpumask_t *cpu_map = wq_cpu_map(wq); | |
395 | int cpu; | |
396 | ||
397 | might_sleep(); | |
398 | lock_acquire(&wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_); | |
399 | lock_release(&wq->lockdep_map, 1, _THIS_IP_); | |
400 | for_each_cpu_mask(cpu, *cpu_map) | |
401 | flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu)); | |
402 | } | |
403 | EXPORT_SYMBOL_GPL(flush_workqueue); | |
404 | ||
405 | /* | |
406 | * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, | |
407 | * so this work can't be re-armed in any way. | |
408 | */ | |
409 | static int try_to_grab_pending(struct work_struct *work) | |
410 | { | |
411 | struct cpu_workqueue_struct *cwq; | |
412 | int ret = -1; | |
413 | ||
414 | if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) | |
415 | return 0; | |
416 | ||
417 | /* | |
418 | * The queueing is in progress, or it is already queued. Try to | |
419 | * steal it from ->worklist without clearing WORK_STRUCT_PENDING. | |
420 | */ | |
421 | ||
422 | cwq = get_wq_data(work); | |
423 | if (!cwq) | |
424 | return ret; | |
425 | ||
426 | spin_lock_irq(&cwq->lock); | |
427 | if (!list_empty(&work->entry)) { | |
428 | /* | |
429 | * This work is queued, but perhaps we locked the wrong cwq. | |
430 | * In that case we must see the new value after rmb(), see | |
431 | * insert_work()->wmb(). | |
432 | */ | |
433 | smp_rmb(); | |
434 | if (cwq == get_wq_data(work)) { | |
435 | list_del_init(&work->entry); | |
436 | ret = 1; | |
437 | } | |
438 | } | |
439 | spin_unlock_irq(&cwq->lock); | |
440 | ||
441 | return ret; | |
442 | } | |
443 | ||
444 | static void wait_on_cpu_work(struct cpu_workqueue_struct *cwq, | |
445 | struct work_struct *work) | |
446 | { | |
447 | struct wq_barrier barr; | |
448 | int running = 0; | |
449 | ||
450 | spin_lock_irq(&cwq->lock); | |
451 | if (unlikely(cwq->current_work == work)) { | |
452 | insert_wq_barrier(cwq, &barr, 0); | |
453 | running = 1; | |
454 | } | |
455 | spin_unlock_irq(&cwq->lock); | |
456 | ||
457 | if (unlikely(running)) | |
458 | wait_for_completion(&barr.done); | |
459 | } | |
460 | ||
461 | static void wait_on_work(struct work_struct *work) | |
462 | { | |
463 | struct cpu_workqueue_struct *cwq; | |
464 | struct workqueue_struct *wq; | |
465 | const cpumask_t *cpu_map; | |
466 | int cpu; | |
467 | ||
468 | might_sleep(); | |
469 | ||
470 | lock_acquire(&work->lockdep_map, 0, 0, 0, 2, _THIS_IP_); | |
471 | lock_release(&work->lockdep_map, 1, _THIS_IP_); | |
472 | ||
473 | cwq = get_wq_data(work); | |
474 | if (!cwq) | |
475 | return; | |
476 | ||
477 | wq = cwq->wq; | |
478 | cpu_map = wq_cpu_map(wq); | |
479 | ||
480 | for_each_cpu_mask(cpu, *cpu_map) | |
481 | wait_on_cpu_work(per_cpu_ptr(wq->cpu_wq, cpu), work); | |
482 | } | |
483 | ||
484 | static int __cancel_work_timer(struct work_struct *work, | |
485 | struct timer_list* timer) | |
486 | { | |
487 | int ret; | |
488 | ||
489 | do { | |
490 | ret = (timer && likely(del_timer(timer))); | |
491 | if (!ret) | |
492 | ret = try_to_grab_pending(work); | |
493 | wait_on_work(work); | |
494 | } while (unlikely(ret < 0)); | |
495 | ||
496 | work_clear_pending(work); | |
497 | return ret; | |
498 | } | |
499 | ||
500 | /** | |
501 | * cancel_work_sync - block until a work_struct's callback has terminated | |
502 | * @work: the work which is to be flushed | |
503 | * | |
504 | * Returns true if @work was pending. | |
505 | * | |
506 | * cancel_work_sync() will cancel the work if it is queued. If the work's | |
507 | * callback appears to be running, cancel_work_sync() will block until it | |
508 | * has completed. | |
509 | * | |
510 | * It is possible to use this function if the work re-queues itself. It can | |
511 | * cancel the work even if it migrates to another workqueue, however in that | |
512 | * case it only guarantees that work->func() has completed on the last queued | |
513 | * workqueue. | |
514 | * | |
515 | * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not | |
516 | * pending, otherwise it goes into a busy-wait loop until the timer expires. | |
517 | * | |
518 | * The caller must ensure that workqueue_struct on which this work was last | |
519 | * queued can't be destroyed before this function returns. | |
520 | */ | |
521 | int cancel_work_sync(struct work_struct *work) | |
522 | { | |
523 | return __cancel_work_timer(work, NULL); | |
524 | } | |
525 | EXPORT_SYMBOL_GPL(cancel_work_sync); | |
526 | ||
527 | /** | |
528 | * cancel_delayed_work_sync - reliably kill off a delayed work. | |
529 | * @dwork: the delayed work struct | |
530 | * | |
531 | * Returns true if @dwork was pending. | |
532 | * | |
533 | * It is possible to use this function if @dwork rearms itself via queue_work() | |
534 | * or queue_delayed_work(). See also the comment for cancel_work_sync(). | |
535 | */ | |
536 | int cancel_delayed_work_sync(struct delayed_work *dwork) | |
537 | { | |
538 | return __cancel_work_timer(&dwork->work, &dwork->timer); | |
539 | } | |
540 | EXPORT_SYMBOL(cancel_delayed_work_sync); | |
541 | ||
542 | static struct workqueue_struct *keventd_wq __read_mostly; | |
543 | ||
544 | /** | |
545 | * schedule_work - put work task in global workqueue | |
546 | * @work: job to be done | |
547 | * | |
548 | * This puts a job in the kernel-global workqueue. | |
549 | */ | |
550 | int schedule_work(struct work_struct *work) | |
551 | { | |
552 | return queue_work(keventd_wq, work); | |
553 | } | |
554 | EXPORT_SYMBOL(schedule_work); | |
555 | ||
556 | /** | |
557 | * schedule_delayed_work - put work task in global workqueue after delay | |
558 | * @dwork: job to be done | |
559 | * @delay: number of jiffies to wait or 0 for immediate execution | |
560 | * | |
561 | * After waiting for a given time this puts a job in the kernel-global | |
562 | * workqueue. | |
563 | */ | |
564 | int schedule_delayed_work(struct delayed_work *dwork, | |
565 | unsigned long delay) | |
566 | { | |
567 | timer_stats_timer_set_start_info(&dwork->timer); | |
568 | return queue_delayed_work(keventd_wq, dwork, delay); | |
569 | } | |
570 | EXPORT_SYMBOL(schedule_delayed_work); | |
571 | ||
572 | /** | |
573 | * schedule_delayed_work_on - queue work in global workqueue on CPU after delay | |
574 | * @cpu: cpu to use | |
575 | * @dwork: job to be done | |
576 | * @delay: number of jiffies to wait | |
577 | * | |
578 | * After waiting for a given time this puts a job in the kernel-global | |
579 | * workqueue on the specified CPU. | |
580 | */ | |
581 | int schedule_delayed_work_on(int cpu, | |
582 | struct delayed_work *dwork, unsigned long delay) | |
583 | { | |
584 | timer_stats_timer_set_start_info(&dwork->timer); | |
585 | return queue_delayed_work_on(cpu, keventd_wq, dwork, delay); | |
586 | } | |
587 | EXPORT_SYMBOL(schedule_delayed_work_on); | |
588 | ||
589 | /** | |
590 | * schedule_on_each_cpu - call a function on each online CPU from keventd | |
591 | * @func: the function to call | |
592 | * | |
593 | * Returns zero on success. | |
594 | * Returns -ve errno on failure. | |
595 | * | |
596 | * schedule_on_each_cpu() is very slow. | |
597 | */ | |
598 | int schedule_on_each_cpu(work_func_t func) | |
599 | { | |
600 | int cpu; | |
601 | struct work_struct *works; | |
602 | ||
603 | works = alloc_percpu(struct work_struct); | |
604 | if (!works) | |
605 | return -ENOMEM; | |
606 | ||
607 | get_online_cpus(); | |
608 | for_each_online_cpu(cpu) { | |
609 | struct work_struct *work = per_cpu_ptr(works, cpu); | |
610 | ||
611 | INIT_WORK(work, func); | |
612 | set_bit(WORK_STRUCT_PENDING, work_data_bits(work)); | |
613 | __queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu), work); | |
614 | } | |
615 | flush_workqueue(keventd_wq); | |
616 | put_online_cpus(); | |
617 | free_percpu(works); | |
618 | return 0; | |
619 | } | |
620 | ||
621 | void flush_scheduled_work(void) | |
622 | { | |
623 | flush_workqueue(keventd_wq); | |
624 | } | |
625 | EXPORT_SYMBOL(flush_scheduled_work); | |
626 | ||
627 | /** | |
628 | * execute_in_process_context - reliably execute the routine with user context | |
629 | * @fn: the function to execute | |
630 | * @ew: guaranteed storage for the execute work structure (must | |
631 | * be available when the work executes) | |
632 | * | |
633 | * Executes the function immediately if process context is available, | |
634 | * otherwise schedules the function for delayed execution. | |
635 | * | |
636 | * Returns: 0 - function was executed | |
637 | * 1 - function was scheduled for execution | |
638 | */ | |
639 | int execute_in_process_context(work_func_t fn, struct execute_work *ew) | |
640 | { | |
641 | if (!in_interrupt()) { | |
642 | fn(&ew->work); | |
643 | return 0; | |
644 | } | |
645 | ||
646 | INIT_WORK(&ew->work, fn); | |
647 | schedule_work(&ew->work); | |
648 | ||
649 | return 1; | |
650 | } | |
651 | EXPORT_SYMBOL_GPL(execute_in_process_context); | |
652 | ||
653 | int keventd_up(void) | |
654 | { | |
655 | return keventd_wq != NULL; | |
656 | } | |
657 | ||
658 | int current_is_keventd(void) | |
659 | { | |
660 | struct cpu_workqueue_struct *cwq; | |
661 | int cpu = raw_smp_processor_id(); /* preempt-safe: keventd is per-cpu */ | |
662 | int ret = 0; | |
663 | ||
664 | BUG_ON(!keventd_wq); | |
665 | ||
666 | cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu); | |
667 | if (current == cwq->thread) | |
668 | ret = 1; | |
669 | ||
670 | return ret; | |
671 | ||
672 | } | |
673 | ||
674 | static struct cpu_workqueue_struct * | |
675 | init_cpu_workqueue(struct workqueue_struct *wq, int cpu) | |
676 | { | |
677 | struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu); | |
678 | ||
679 | cwq->wq = wq; | |
680 | spin_lock_init(&cwq->lock); | |
681 | INIT_LIST_HEAD(&cwq->worklist); | |
682 | init_waitqueue_head(&cwq->more_work); | |
683 | ||
684 | return cwq; | |
685 | } | |
686 | ||
687 | static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu) | |
688 | { | |
689 | struct workqueue_struct *wq = cwq->wq; | |
690 | const char *fmt = is_single_threaded(wq) ? "%s" : "%s/%d"; | |
691 | struct task_struct *p; | |
692 | ||
693 | p = kthread_create(worker_thread, cwq, fmt, wq->name, cpu); | |
694 | /* | |
695 | * Nobody can add the work_struct to this cwq, | |
696 | * if (caller is __create_workqueue) | |
697 | * nobody should see this wq | |
698 | * else // caller is CPU_UP_PREPARE | |
699 | * cpu is not on cpu_online_map | |
700 | * so we can abort safely. | |
701 | */ | |
702 | if (IS_ERR(p)) | |
703 | return PTR_ERR(p); | |
704 | ||
705 | cwq->thread = p; | |
706 | ||
707 | return 0; | |
708 | } | |
709 | ||
710 | static void start_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu) | |
711 | { | |
712 | struct task_struct *p = cwq->thread; | |
713 | ||
714 | if (p != NULL) { | |
715 | if (cpu >= 0) | |
716 | kthread_bind(p, cpu); | |
717 | wake_up_process(p); | |
718 | } | |
719 | } | |
720 | ||
721 | struct workqueue_struct *__create_workqueue_key(const char *name, | |
722 | int singlethread, | |
723 | int freezeable, | |
724 | struct lock_class_key *key, | |
725 | const char *lock_name) | |
726 | { | |
727 | struct workqueue_struct *wq; | |
728 | struct cpu_workqueue_struct *cwq; | |
729 | int err = 0, cpu; | |
730 | ||
731 | wq = kzalloc(sizeof(*wq), GFP_KERNEL); | |
732 | if (!wq) | |
733 | return NULL; | |
734 | ||
735 | wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct); | |
736 | if (!wq->cpu_wq) { | |
737 | kfree(wq); | |
738 | return NULL; | |
739 | } | |
740 | ||
741 | wq->name = name; | |
742 | lockdep_init_map(&wq->lockdep_map, lock_name, key, 0); | |
743 | wq->singlethread = singlethread; | |
744 | wq->freezeable = freezeable; | |
745 | INIT_LIST_HEAD(&wq->list); | |
746 | ||
747 | if (singlethread) { | |
748 | cwq = init_cpu_workqueue(wq, singlethread_cpu); | |
749 | err = create_workqueue_thread(cwq, singlethread_cpu); | |
750 | start_workqueue_thread(cwq, -1); | |
751 | } else { | |
752 | get_online_cpus(); | |
753 | spin_lock(&workqueue_lock); | |
754 | list_add(&wq->list, &workqueues); | |
755 | spin_unlock(&workqueue_lock); | |
756 | ||
757 | for_each_possible_cpu(cpu) { | |
758 | cwq = init_cpu_workqueue(wq, cpu); | |
759 | if (err || !cpu_online(cpu)) | |
760 | continue; | |
761 | err = create_workqueue_thread(cwq, cpu); | |
762 | start_workqueue_thread(cwq, cpu); | |
763 | } | |
764 | put_online_cpus(); | |
765 | } | |
766 | ||
767 | if (err) { | |
768 | destroy_workqueue(wq); | |
769 | wq = NULL; | |
770 | } | |
771 | return wq; | |
772 | } | |
773 | EXPORT_SYMBOL_GPL(__create_workqueue_key); | |
774 | ||
775 | static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq) | |
776 | { | |
777 | /* | |
778 | * Our caller is either destroy_workqueue() or CPU_DEAD, | |
779 | * get_online_cpus() protects cwq->thread. | |
780 | */ | |
781 | if (cwq->thread == NULL) | |
782 | return; | |
783 | ||
784 | lock_acquire(&cwq->wq->lockdep_map, 0, 0, 0, 2, _THIS_IP_); | |
785 | lock_release(&cwq->wq->lockdep_map, 1, _THIS_IP_); | |
786 | ||
787 | flush_cpu_workqueue(cwq); | |
788 | /* | |
789 | * If the caller is CPU_DEAD and cwq->worklist was not empty, | |
790 | * a concurrent flush_workqueue() can insert a barrier after us. | |
791 | * However, in that case run_workqueue() won't return and check | |
792 | * kthread_should_stop() until it flushes all work_struct's. | |
793 | * When ->worklist becomes empty it is safe to exit because no | |
794 | * more work_structs can be queued on this cwq: flush_workqueue | |
795 | * checks list_empty(), and a "normal" queue_work() can't use | |
796 | * a dead CPU. | |
797 | */ | |
798 | kthread_stop(cwq->thread); | |
799 | cwq->thread = NULL; | |
800 | } | |
801 | ||
802 | /** | |
803 | * destroy_workqueue - safely terminate a workqueue | |
804 | * @wq: target workqueue | |
805 | * | |
806 | * Safely destroy a workqueue. All work currently pending will be done first. | |
807 | */ | |
808 | void destroy_workqueue(struct workqueue_struct *wq) | |
809 | { | |
810 | const cpumask_t *cpu_map = wq_cpu_map(wq); | |
811 | int cpu; | |
812 | ||
813 | get_online_cpus(); | |
814 | spin_lock(&workqueue_lock); | |
815 | list_del(&wq->list); | |
816 | spin_unlock(&workqueue_lock); | |
817 | ||
818 | for_each_cpu_mask(cpu, *cpu_map) | |
819 | cleanup_workqueue_thread(per_cpu_ptr(wq->cpu_wq, cpu)); | |
820 | put_online_cpus(); | |
821 | ||
822 | free_percpu(wq->cpu_wq); | |
823 | kfree(wq); | |
824 | } | |
825 | EXPORT_SYMBOL_GPL(destroy_workqueue); | |
826 | ||
827 | static int __devinit workqueue_cpu_callback(struct notifier_block *nfb, | |
828 | unsigned long action, | |
829 | void *hcpu) | |
830 | { | |
831 | unsigned int cpu = (unsigned long)hcpu; | |
832 | struct cpu_workqueue_struct *cwq; | |
833 | struct workqueue_struct *wq; | |
834 | ||
835 | action &= ~CPU_TASKS_FROZEN; | |
836 | ||
837 | switch (action) { | |
838 | case CPU_UP_PREPARE: | |
839 | cpu_set(cpu, cpu_populated_map); | |
840 | } | |
841 | ||
842 | list_for_each_entry(wq, &workqueues, list) { | |
843 | cwq = per_cpu_ptr(wq->cpu_wq, cpu); | |
844 | ||
845 | switch (action) { | |
846 | case CPU_UP_PREPARE: | |
847 | if (!create_workqueue_thread(cwq, cpu)) | |
848 | break; | |
849 | printk(KERN_ERR "workqueue [%s] for %i failed\n", | |
850 | wq->name, cpu); | |
851 | return NOTIFY_BAD; | |
852 | ||
853 | case CPU_ONLINE: | |
854 | start_workqueue_thread(cwq, cpu); | |
855 | break; | |
856 | ||
857 | case CPU_UP_CANCELED: | |
858 | start_workqueue_thread(cwq, -1); | |
859 | case CPU_DEAD: | |
860 | cleanup_workqueue_thread(cwq); | |
861 | break; | |
862 | } | |
863 | } | |
864 | ||
865 | switch (action) { | |
866 | case CPU_UP_CANCELED: | |
867 | case CPU_DEAD: | |
868 | cpu_clear(cpu, cpu_populated_map); | |
869 | } | |
870 | ||
871 | return NOTIFY_OK; | |
872 | } | |
873 | ||
874 | void __init init_workqueues(void) | |
875 | { | |
876 | cpu_populated_map = cpu_online_map; | |
877 | singlethread_cpu = first_cpu(cpu_possible_map); | |
878 | cpu_singlethread_map = cpumask_of_cpu(singlethread_cpu); | |
879 | hotcpu_notifier(workqueue_cpu_callback, 0); | |
880 | keventd_wq = create_workqueue("events"); | |
881 | BUG_ON(!keventd_wq); | |
882 | } |