• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 const struct cpumask *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_var_t cpu_populated_map __read_mostly;
85 
86 /* If it's single threaded, it isn't in the list of workqueues. */
is_wq_single_threaded(struct workqueue_struct * wq)87 static inline int is_wq_single_threaded(struct workqueue_struct *wq)
88 {
89 	return wq->singlethread;
90 }
91 
wq_cpu_map(struct workqueue_struct * wq)92 static const struct cpumask *wq_cpu_map(struct workqueue_struct *wq)
93 {
94 	return is_wq_single_threaded(wq)
95 		? cpu_singlethread_map : cpu_populated_map;
96 }
97 
98 static
wq_per_cpu(struct workqueue_struct * wq,int cpu)99 struct cpu_workqueue_struct *wq_per_cpu(struct workqueue_struct *wq, int cpu)
100 {
101 	if (unlikely(is_wq_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  */
set_wq_data(struct work_struct * work,struct cpu_workqueue_struct * cwq)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
get_wq_data(struct work_struct * work)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 
insert_work(struct cpu_workqueue_struct * cwq,struct work_struct * work,struct list_head * head)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 
__queue_work(struct cpu_workqueue_struct * cwq,struct work_struct * work)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  */
queue_work(struct workqueue_struct * wq,struct work_struct * work)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
queue_work_on(int cpu,struct workqueue_struct * wq,struct work_struct * work)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 
delayed_work_timer_fn(unsigned long __data)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  */
queue_delayed_work(struct workqueue_struct * wq,struct delayed_work * dwork,unsigned long delay)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  */
queue_delayed_work_on(int cpu,struct workqueue_struct * wq,struct delayed_work * dwork,unsigned long delay)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 
run_workqueue(struct cpu_workqueue_struct * cwq)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 
worker_thread(void * __cwq)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 
wq_barrier_func(struct work_struct * work)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 
insert_wq_barrier(struct cpu_workqueue_struct * cwq,struct wq_barrier * barr,struct list_head * head)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 
flush_cpu_workqueue(struct cpu_workqueue_struct * cwq)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  */
flush_workqueue(struct workqueue_struct * wq)411 void flush_workqueue(struct workqueue_struct *wq)
412 {
413 	const struct cpumask *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  */
flush_work(struct work_struct * work)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  */
try_to_grab_pending(struct work_struct * work)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 
wait_on_cpu_work(struct cpu_workqueue_struct * cwq,struct work_struct * work)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 
wait_on_work(struct work_struct * work)531 static void wait_on_work(struct work_struct *work)
532 {
533 	struct cpu_workqueue_struct *cwq;
534 	struct workqueue_struct *wq;
535 	const struct cpumask *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 
__cancel_work_timer(struct work_struct * work,struct timer_list * timer)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  */
cancel_work_sync(struct work_struct * work)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  */
cancel_delayed_work_sync(struct delayed_work * dwork)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  */
schedule_work(struct work_struct * work)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  */
schedule_work_on(int cpu,struct work_struct * work)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  */
schedule_delayed_work(struct delayed_work * dwork,unsigned long delay)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  */
schedule_delayed_work_on(int cpu,struct delayed_work * dwork,unsigned long delay)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  */
schedule_on_each_cpu(work_func_t func)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 
flush_scheduled_work(void)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  */
execute_in_process_context(work_func_t fn,struct execute_work * ew)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 
keventd_up(void)734 int keventd_up(void)
735 {
736 	return keventd_wq != NULL;
737 }
738 
current_is_keventd(void)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 *
init_cpu_workqueue(struct workqueue_struct * wq,int cpu)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 
create_workqueue_thread(struct cpu_workqueue_struct * cwq,int cpu)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_wq_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 
start_workqueue_thread(struct cpu_workqueue_struct * cwq,int cpu)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 
__create_workqueue_key(const char * name,int singlethread,int freezeable,int rt,struct lock_class_key * key,const char * lock_name)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 
cleanup_workqueue_thread(struct cpu_workqueue_struct * cwq)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  */
destroy_workqueue(struct workqueue_struct * wq)904 void destroy_workqueue(struct workqueue_struct *wq)
905 {
906 	const struct cpumask *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 
workqueue_cpu_callback(struct notifier_block * nfb,unsigned long action,void * hcpu)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 		cpumask_set_cpu(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 		cpumask_clear_cpu(cpu, cpu_populated_map);
968 	}
969 
970 	return ret;
971 }
972 
973 #ifdef CONFIG_SMP
974 static struct workqueue_struct *work_on_cpu_wq __read_mostly;
975 
976 struct work_for_cpu {
977 	struct work_struct work;
978 	long (*fn)(void *);
979 	void *arg;
980 	long ret;
981 };
982 
do_work_for_cpu(struct work_struct * w)983 static void do_work_for_cpu(struct work_struct *w)
984 {
985 	struct work_for_cpu *wfc = container_of(w, struct work_for_cpu, work);
986 
987 	wfc->ret = wfc->fn(wfc->arg);
988 }
989 
990 /**
991  * work_on_cpu - run a function in user context on a particular cpu
992  * @cpu: the cpu to run on
993  * @fn: the function to run
994  * @arg: the function arg
995  *
996  * This will return the value @fn returns.
997  * It is up to the caller to ensure that the cpu doesn't go offline.
998  */
work_on_cpu(unsigned int cpu,long (* fn)(void *),void * arg)999 long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg)
1000 {
1001 	struct work_for_cpu wfc;
1002 
1003 	INIT_WORK(&wfc.work, do_work_for_cpu);
1004 	wfc.fn = fn;
1005 	wfc.arg = arg;
1006 	queue_work_on(cpu, work_on_cpu_wq, &wfc.work);
1007 	flush_work(&wfc.work);
1008 
1009 	return wfc.ret;
1010 }
1011 EXPORT_SYMBOL_GPL(work_on_cpu);
1012 #endif /* CONFIG_SMP */
1013 
init_workqueues(void)1014 void __init init_workqueues(void)
1015 {
1016 	alloc_cpumask_var(&cpu_populated_map, GFP_KERNEL);
1017 
1018 	cpumask_copy(cpu_populated_map, cpu_online_mask);
1019 	singlethread_cpu = cpumask_first(cpu_possible_mask);
1020 	cpu_singlethread_map = cpumask_of(singlethread_cpu);
1021 	hotcpu_notifier(workqueue_cpu_callback, 0);
1022 	keventd_wq = create_workqueue("events");
1023 	BUG_ON(!keventd_wq);
1024 #ifdef CONFIG_SMP
1025 	work_on_cpu_wq = create_workqueue("work_on_cpu");
1026 	BUG_ON(!work_on_cpu_wq);
1027 #endif
1028 }
1029