xref: /linux/kernel/workqueue.c (revision b0148a98ec5151fec82064d95f11eb9efbc628ea)
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 
36 /*
37  * The per-CPU workqueue (if single thread, we always use the first
38  * possible cpu).
39  *
40  * The sequence counters are for flush_scheduled_work().  It wants to wait
41  * until all currently-scheduled works are completed, but it doesn't
42  * want to be livelocked by new, incoming ones.  So it waits until
43  * remove_sequence is >= the insert_sequence which pertained when
44  * flush_scheduled_work() was called.
45  */
46 struct cpu_workqueue_struct {
47 
48 	spinlock_t lock;
49 
50 	long remove_sequence;	/* Least-recently added (next to run) */
51 	long insert_sequence;	/* Next to add */
52 
53 	struct list_head worklist;
54 	wait_queue_head_t more_work;
55 	wait_queue_head_t work_done;
56 
57 	struct workqueue_struct *wq;
58 	struct task_struct *thread;
59 
60 	int run_depth;		/* Detect run_workqueue() recursion depth */
61 
62 	int freezeable;		/* Freeze the thread during suspend */
63 } ____cacheline_aligned;
64 
65 /*
66  * The externally visible workqueue abstraction is an array of
67  * per-CPU workqueues:
68  */
69 struct workqueue_struct {
70 	struct cpu_workqueue_struct *cpu_wq;
71 	const char *name;
72 	struct list_head list; 	/* Empty if single thread */
73 };
74 
75 /* All the per-cpu workqueues on the system, for hotplug cpu to add/remove
76    threads to each one as cpus come/go. */
77 static DEFINE_MUTEX(workqueue_mutex);
78 static LIST_HEAD(workqueues);
79 
80 static int singlethread_cpu;
81 
82 /* If it's single threaded, it isn't in the list of workqueues. */
83 static inline int is_single_threaded(struct workqueue_struct *wq)
84 {
85 	return list_empty(&wq->list);
86 }
87 
88 /*
89  * Set the workqueue on which a work item is to be run
90  * - Must *only* be called if the pending flag is set
91  */
92 static inline void set_wq_data(struct work_struct *work, void *wq)
93 {
94 	unsigned long new;
95 
96 	BUG_ON(!work_pending(work));
97 
98 	new = (unsigned long) wq | (1UL << WORK_STRUCT_PENDING);
99 	new |= WORK_STRUCT_FLAG_MASK & *work_data_bits(work);
100 	atomic_long_set(&work->data, new);
101 }
102 
103 static inline void *get_wq_data(struct work_struct *work)
104 {
105 	return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
106 }
107 
108 static int __run_work(struct cpu_workqueue_struct *cwq, struct work_struct *work)
109 {
110 	int ret = 0;
111 	unsigned long flags;
112 
113 	spin_lock_irqsave(&cwq->lock, flags);
114 	/*
115 	 * We need to re-validate the work info after we've gotten
116 	 * the cpu_workqueue lock. We can run the work now iff:
117 	 *
118 	 *  - the wq_data still matches the cpu_workqueue_struct
119 	 *  - AND the work is still marked pending
120 	 *  - AND the work is still on a list (which will be this
121 	 *    workqueue_struct list)
122 	 *
123 	 * All these conditions are important, because we
124 	 * need to protect against the work being run right
125 	 * now on another CPU (all but the last one might be
126 	 * true if it's currently running and has not been
127 	 * released yet, for example).
128 	 */
129 	if (get_wq_data(work) == cwq
130 	    && work_pending(work)
131 	    && !list_empty(&work->entry)) {
132 		work_func_t f = work->func;
133 		list_del_init(&work->entry);
134 		spin_unlock_irqrestore(&cwq->lock, flags);
135 
136 		if (!test_bit(WORK_STRUCT_NOAUTOREL, work_data_bits(work)))
137 			work_release(work);
138 		f(work);
139 
140 		spin_lock_irqsave(&cwq->lock, flags);
141 		cwq->remove_sequence++;
142 		wake_up(&cwq->work_done);
143 		ret = 1;
144 	}
145 	spin_unlock_irqrestore(&cwq->lock, flags);
146 	return ret;
147 }
148 
149 /**
150  * run_scheduled_work - run scheduled work synchronously
151  * @work: work to run
152  *
153  * This checks if the work was pending, and runs it
154  * synchronously if so. It returns a boolean to indicate
155  * whether it had any scheduled work to run or not.
156  *
157  * NOTE! This _only_ works for normal work_structs. You
158  * CANNOT use this for delayed work, because the wq data
159  * for delayed work will not point properly to the per-
160  * CPU workqueue struct, but will change!
161  */
162 int fastcall run_scheduled_work(struct work_struct *work)
163 {
164 	for (;;) {
165 		struct cpu_workqueue_struct *cwq;
166 
167 		if (!work_pending(work))
168 			return 0;
169 		if (list_empty(&work->entry))
170 			return 0;
171 		/* NOTE! This depends intimately on __queue_work! */
172 		cwq = get_wq_data(work);
173 		if (!cwq)
174 			return 0;
175 		if (__run_work(cwq, work))
176 			return 1;
177 	}
178 }
179 EXPORT_SYMBOL(run_scheduled_work);
180 
181 /* Preempt must be disabled. */
182 static void __queue_work(struct cpu_workqueue_struct *cwq,
183 			 struct work_struct *work)
184 {
185 	unsigned long flags;
186 
187 	spin_lock_irqsave(&cwq->lock, flags);
188 	set_wq_data(work, cwq);
189 	list_add_tail(&work->entry, &cwq->worklist);
190 	cwq->insert_sequence++;
191 	wake_up(&cwq->more_work);
192 	spin_unlock_irqrestore(&cwq->lock, flags);
193 }
194 
195 /**
196  * queue_work - queue work on a workqueue
197  * @wq: workqueue to use
198  * @work: work to queue
199  *
200  * Returns 0 if @work was already on a queue, non-zero otherwise.
201  *
202  * We queue the work to the CPU it was submitted, but there is no
203  * guarantee that it will be processed by that CPU.
204  */
205 int fastcall queue_work(struct workqueue_struct *wq, struct work_struct *work)
206 {
207 	int ret = 0, cpu = get_cpu();
208 
209 	if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
210 		if (unlikely(is_single_threaded(wq)))
211 			cpu = singlethread_cpu;
212 		BUG_ON(!list_empty(&work->entry));
213 		__queue_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
214 		ret = 1;
215 	}
216 	put_cpu();
217 	return ret;
218 }
219 EXPORT_SYMBOL_GPL(queue_work);
220 
221 static void delayed_work_timer_fn(unsigned long __data)
222 {
223 	struct delayed_work *dwork = (struct delayed_work *)__data;
224 	struct workqueue_struct *wq = get_wq_data(&dwork->work);
225 	int cpu = smp_processor_id();
226 
227 	if (unlikely(is_single_threaded(wq)))
228 		cpu = singlethread_cpu;
229 
230 	__queue_work(per_cpu_ptr(wq->cpu_wq, cpu), &dwork->work);
231 }
232 
233 /**
234  * queue_delayed_work - queue work on a workqueue after delay
235  * @wq: workqueue to use
236  * @dwork: delayable work to queue
237  * @delay: number of jiffies to wait before queueing
238  *
239  * Returns 0 if @work was already on a queue, non-zero otherwise.
240  */
241 int fastcall queue_delayed_work(struct workqueue_struct *wq,
242 			struct delayed_work *dwork, unsigned long delay)
243 {
244 	int ret = 0;
245 	struct timer_list *timer = &dwork->timer;
246 	struct work_struct *work = &dwork->work;
247 
248 	if (delay == 0)
249 		return queue_work(wq, work);
250 
251 	if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
252 		BUG_ON(timer_pending(timer));
253 		BUG_ON(!list_empty(&work->entry));
254 
255 		/* This stores wq for the moment, for the timer_fn */
256 		set_wq_data(work, wq);
257 		timer->expires = jiffies + delay;
258 		timer->data = (unsigned long)dwork;
259 		timer->function = delayed_work_timer_fn;
260 		add_timer(timer);
261 		ret = 1;
262 	}
263 	return ret;
264 }
265 EXPORT_SYMBOL_GPL(queue_delayed_work);
266 
267 /**
268  * queue_delayed_work_on - queue work on specific CPU after delay
269  * @cpu: CPU number to execute work on
270  * @wq: workqueue to use
271  * @dwork: work to queue
272  * @delay: number of jiffies to wait before queueing
273  *
274  * Returns 0 if @work was already on a queue, non-zero otherwise.
275  */
276 int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
277 			struct delayed_work *dwork, unsigned long delay)
278 {
279 	int ret = 0;
280 	struct timer_list *timer = &dwork->timer;
281 	struct work_struct *work = &dwork->work;
282 
283 	if (!test_and_set_bit(WORK_STRUCT_PENDING, work_data_bits(work))) {
284 		BUG_ON(timer_pending(timer));
285 		BUG_ON(!list_empty(&work->entry));
286 
287 		/* This stores wq for the moment, for the timer_fn */
288 		set_wq_data(work, wq);
289 		timer->expires = jiffies + delay;
290 		timer->data = (unsigned long)dwork;
291 		timer->function = delayed_work_timer_fn;
292 		add_timer_on(timer, cpu);
293 		ret = 1;
294 	}
295 	return ret;
296 }
297 EXPORT_SYMBOL_GPL(queue_delayed_work_on);
298 
299 static void run_workqueue(struct cpu_workqueue_struct *cwq)
300 {
301 	unsigned long flags;
302 
303 	/*
304 	 * Keep taking off work from the queue until
305 	 * done.
306 	 */
307 	spin_lock_irqsave(&cwq->lock, flags);
308 	cwq->run_depth++;
309 	if (cwq->run_depth > 3) {
310 		/* morton gets to eat his hat */
311 		printk("%s: recursion depth exceeded: %d\n",
312 			__FUNCTION__, cwq->run_depth);
313 		dump_stack();
314 	}
315 	while (!list_empty(&cwq->worklist)) {
316 		struct work_struct *work = list_entry(cwq->worklist.next,
317 						struct work_struct, entry);
318 		work_func_t f = work->func;
319 
320 		list_del_init(cwq->worklist.next);
321 		spin_unlock_irqrestore(&cwq->lock, flags);
322 
323 		BUG_ON(get_wq_data(work) != cwq);
324 		if (!test_bit(WORK_STRUCT_NOAUTOREL, work_data_bits(work)))
325 			work_release(work);
326 		f(work);
327 
328 		if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
329 			printk(KERN_ERR "BUG: workqueue leaked lock or atomic: "
330 					"%s/0x%08x/%d\n",
331 					current->comm, preempt_count(),
332 				       	current->pid);
333 			printk(KERN_ERR "    last function: ");
334 			print_symbol("%s\n", (unsigned long)f);
335 			debug_show_held_locks(current);
336 			dump_stack();
337 		}
338 
339 		spin_lock_irqsave(&cwq->lock, flags);
340 		cwq->remove_sequence++;
341 		wake_up(&cwq->work_done);
342 	}
343 	cwq->run_depth--;
344 	spin_unlock_irqrestore(&cwq->lock, flags);
345 }
346 
347 static int worker_thread(void *__cwq)
348 {
349 	struct cpu_workqueue_struct *cwq = __cwq;
350 	DECLARE_WAITQUEUE(wait, current);
351 	struct k_sigaction sa;
352 	sigset_t blocked;
353 
354 	if (!cwq->freezeable)
355 		current->flags |= PF_NOFREEZE;
356 
357 	set_user_nice(current, -5);
358 
359 	/* Block and flush all signals */
360 	sigfillset(&blocked);
361 	sigprocmask(SIG_BLOCK, &blocked, NULL);
362 	flush_signals(current);
363 
364 	/*
365 	 * We inherited MPOL_INTERLEAVE from the booting kernel.
366 	 * Set MPOL_DEFAULT to insure node local allocations.
367 	 */
368 	numa_default_policy();
369 
370 	/* SIG_IGN makes children autoreap: see do_notify_parent(). */
371 	sa.sa.sa_handler = SIG_IGN;
372 	sa.sa.sa_flags = 0;
373 	siginitset(&sa.sa.sa_mask, sigmask(SIGCHLD));
374 	do_sigaction(SIGCHLD, &sa, (struct k_sigaction *)0);
375 
376 	set_current_state(TASK_INTERRUPTIBLE);
377 	while (!kthread_should_stop()) {
378 		if (cwq->freezeable)
379 			try_to_freeze();
380 
381 		add_wait_queue(&cwq->more_work, &wait);
382 		if (list_empty(&cwq->worklist))
383 			schedule();
384 		else
385 			__set_current_state(TASK_RUNNING);
386 		remove_wait_queue(&cwq->more_work, &wait);
387 
388 		if (!list_empty(&cwq->worklist))
389 			run_workqueue(cwq);
390 		set_current_state(TASK_INTERRUPTIBLE);
391 	}
392 	__set_current_state(TASK_RUNNING);
393 	return 0;
394 }
395 
396 static void flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
397 {
398 	if (cwq->thread == current) {
399 		/*
400 		 * Probably keventd trying to flush its own queue. So simply run
401 		 * it by hand rather than deadlocking.
402 		 */
403 		run_workqueue(cwq);
404 	} else {
405 		DEFINE_WAIT(wait);
406 		long sequence_needed;
407 
408 		spin_lock_irq(&cwq->lock);
409 		sequence_needed = cwq->insert_sequence;
410 
411 		while (sequence_needed - cwq->remove_sequence > 0) {
412 			prepare_to_wait(&cwq->work_done, &wait,
413 					TASK_UNINTERRUPTIBLE);
414 			spin_unlock_irq(&cwq->lock);
415 			schedule();
416 			spin_lock_irq(&cwq->lock);
417 		}
418 		finish_wait(&cwq->work_done, &wait);
419 		spin_unlock_irq(&cwq->lock);
420 	}
421 }
422 
423 /**
424  * flush_workqueue - ensure that any scheduled work has run to completion.
425  * @wq: workqueue to flush
426  *
427  * Forces execution of the workqueue and blocks until its completion.
428  * This is typically used in driver shutdown handlers.
429  *
430  * This function will sample each workqueue's current insert_sequence number and
431  * will sleep until the head sequence is greater than or equal to that.  This
432  * means that we sleep until all works which were queued on entry have been
433  * handled, but we are not livelocked by new incoming ones.
434  *
435  * This function used to run the workqueues itself.  Now we just wait for the
436  * helper threads to do it.
437  */
438 void fastcall flush_workqueue(struct workqueue_struct *wq)
439 {
440 	might_sleep();
441 
442 	if (is_single_threaded(wq)) {
443 		/* Always use first cpu's area. */
444 		flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, singlethread_cpu));
445 	} else {
446 		int cpu;
447 
448 		mutex_lock(&workqueue_mutex);
449 		for_each_online_cpu(cpu)
450 			flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
451 		mutex_unlock(&workqueue_mutex);
452 	}
453 }
454 EXPORT_SYMBOL_GPL(flush_workqueue);
455 
456 static struct task_struct *create_workqueue_thread(struct workqueue_struct *wq,
457 						   int cpu, int freezeable)
458 {
459 	struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
460 	struct task_struct *p;
461 
462 	spin_lock_init(&cwq->lock);
463 	cwq->wq = wq;
464 	cwq->thread = NULL;
465 	cwq->insert_sequence = 0;
466 	cwq->remove_sequence = 0;
467 	cwq->freezeable = freezeable;
468 	INIT_LIST_HEAD(&cwq->worklist);
469 	init_waitqueue_head(&cwq->more_work);
470 	init_waitqueue_head(&cwq->work_done);
471 
472 	if (is_single_threaded(wq))
473 		p = kthread_create(worker_thread, cwq, "%s", wq->name);
474 	else
475 		p = kthread_create(worker_thread, cwq, "%s/%d", wq->name, cpu);
476 	if (IS_ERR(p))
477 		return NULL;
478 	cwq->thread = p;
479 	return p;
480 }
481 
482 struct workqueue_struct *__create_workqueue(const char *name,
483 					    int singlethread, int freezeable)
484 {
485 	int cpu, destroy = 0;
486 	struct workqueue_struct *wq;
487 	struct task_struct *p;
488 
489 	wq = kzalloc(sizeof(*wq), GFP_KERNEL);
490 	if (!wq)
491 		return NULL;
492 
493 	wq->cpu_wq = alloc_percpu(struct cpu_workqueue_struct);
494 	if (!wq->cpu_wq) {
495 		kfree(wq);
496 		return NULL;
497 	}
498 
499 	wq->name = name;
500 	mutex_lock(&workqueue_mutex);
501 	if (singlethread) {
502 		INIT_LIST_HEAD(&wq->list);
503 		p = create_workqueue_thread(wq, singlethread_cpu, freezeable);
504 		if (!p)
505 			destroy = 1;
506 		else
507 			wake_up_process(p);
508 	} else {
509 		list_add(&wq->list, &workqueues);
510 		for_each_online_cpu(cpu) {
511 			p = create_workqueue_thread(wq, cpu, freezeable);
512 			if (p) {
513 				kthread_bind(p, cpu);
514 				wake_up_process(p);
515 			} else
516 				destroy = 1;
517 		}
518 	}
519 	mutex_unlock(&workqueue_mutex);
520 
521 	/*
522 	 * Was there any error during startup? If yes then clean up:
523 	 */
524 	if (destroy) {
525 		destroy_workqueue(wq);
526 		wq = NULL;
527 	}
528 	return wq;
529 }
530 EXPORT_SYMBOL_GPL(__create_workqueue);
531 
532 static void cleanup_workqueue_thread(struct workqueue_struct *wq, int cpu)
533 {
534 	struct cpu_workqueue_struct *cwq;
535 	unsigned long flags;
536 	struct task_struct *p;
537 
538 	cwq = per_cpu_ptr(wq->cpu_wq, cpu);
539 	spin_lock_irqsave(&cwq->lock, flags);
540 	p = cwq->thread;
541 	cwq->thread = NULL;
542 	spin_unlock_irqrestore(&cwq->lock, flags);
543 	if (p)
544 		kthread_stop(p);
545 }
546 
547 /**
548  * destroy_workqueue - safely terminate a workqueue
549  * @wq: target workqueue
550  *
551  * Safely destroy a workqueue. All work currently pending will be done first.
552  */
553 void destroy_workqueue(struct workqueue_struct *wq)
554 {
555 	int cpu;
556 
557 	flush_workqueue(wq);
558 
559 	/* We don't need the distraction of CPUs appearing and vanishing. */
560 	mutex_lock(&workqueue_mutex);
561 	if (is_single_threaded(wq))
562 		cleanup_workqueue_thread(wq, singlethread_cpu);
563 	else {
564 		for_each_online_cpu(cpu)
565 			cleanup_workqueue_thread(wq, cpu);
566 		list_del(&wq->list);
567 	}
568 	mutex_unlock(&workqueue_mutex);
569 	free_percpu(wq->cpu_wq);
570 	kfree(wq);
571 }
572 EXPORT_SYMBOL_GPL(destroy_workqueue);
573 
574 static struct workqueue_struct *keventd_wq;
575 
576 /**
577  * schedule_work - put work task in global workqueue
578  * @work: job to be done
579  *
580  * This puts a job in the kernel-global workqueue.
581  */
582 int fastcall schedule_work(struct work_struct *work)
583 {
584 	return queue_work(keventd_wq, work);
585 }
586 EXPORT_SYMBOL(schedule_work);
587 
588 /**
589  * schedule_delayed_work - put work task in global workqueue after delay
590  * @dwork: job to be done
591  * @delay: number of jiffies to wait or 0 for immediate execution
592  *
593  * After waiting for a given time this puts a job in the kernel-global
594  * workqueue.
595  */
596 int fastcall schedule_delayed_work(struct delayed_work *dwork, unsigned long delay)
597 {
598 	return queue_delayed_work(keventd_wq, dwork, delay);
599 }
600 EXPORT_SYMBOL(schedule_delayed_work);
601 
602 /**
603  * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
604  * @cpu: cpu to use
605  * @dwork: job to be done
606  * @delay: number of jiffies to wait
607  *
608  * After waiting for a given time this puts a job in the kernel-global
609  * workqueue on the specified CPU.
610  */
611 int schedule_delayed_work_on(int cpu,
612 			struct delayed_work *dwork, unsigned long delay)
613 {
614 	return queue_delayed_work_on(cpu, keventd_wq, dwork, delay);
615 }
616 EXPORT_SYMBOL(schedule_delayed_work_on);
617 
618 /**
619  * schedule_on_each_cpu - call a function on each online CPU from keventd
620  * @func: the function to call
621  *
622  * Returns zero on success.
623  * Returns -ve errno on failure.
624  *
625  * Appears to be racy against CPU hotplug.
626  *
627  * schedule_on_each_cpu() is very slow.
628  */
629 int schedule_on_each_cpu(work_func_t func)
630 {
631 	int cpu;
632 	struct work_struct *works;
633 
634 	works = alloc_percpu(struct work_struct);
635 	if (!works)
636 		return -ENOMEM;
637 
638 	mutex_lock(&workqueue_mutex);
639 	for_each_online_cpu(cpu) {
640 		struct work_struct *work = per_cpu_ptr(works, cpu);
641 
642 		INIT_WORK(work, func);
643 		set_bit(WORK_STRUCT_PENDING, work_data_bits(work));
644 		__queue_work(per_cpu_ptr(keventd_wq->cpu_wq, cpu), work);
645 	}
646 	mutex_unlock(&workqueue_mutex);
647 	flush_workqueue(keventd_wq);
648 	free_percpu(works);
649 	return 0;
650 }
651 
652 void flush_scheduled_work(void)
653 {
654 	flush_workqueue(keventd_wq);
655 }
656 EXPORT_SYMBOL(flush_scheduled_work);
657 
658 /**
659  * cancel_rearming_delayed_workqueue - reliably kill off a delayed
660  *			work whose handler rearms the delayed work.
661  * @wq:   the controlling workqueue structure
662  * @dwork: the delayed work struct
663  */
664 void cancel_rearming_delayed_workqueue(struct workqueue_struct *wq,
665 				       struct delayed_work *dwork)
666 {
667 	while (!cancel_delayed_work(dwork))
668 		flush_workqueue(wq);
669 }
670 EXPORT_SYMBOL(cancel_rearming_delayed_workqueue);
671 
672 /**
673  * cancel_rearming_delayed_work - reliably kill off a delayed keventd
674  *			work whose handler rearms the delayed work.
675  * @dwork: the delayed work struct
676  */
677 void cancel_rearming_delayed_work(struct delayed_work *dwork)
678 {
679 	cancel_rearming_delayed_workqueue(keventd_wq, dwork);
680 }
681 EXPORT_SYMBOL(cancel_rearming_delayed_work);
682 
683 /**
684  * execute_in_process_context - reliably execute the routine with user context
685  * @fn:		the function to execute
686  * @ew:		guaranteed storage for the execute work structure (must
687  *		be available when the work executes)
688  *
689  * Executes the function immediately if process context is available,
690  * otherwise schedules the function for delayed execution.
691  *
692  * Returns:	0 - function was executed
693  *		1 - function was scheduled for execution
694  */
695 int execute_in_process_context(work_func_t fn, struct execute_work *ew)
696 {
697 	if (!in_interrupt()) {
698 		fn(&ew->work);
699 		return 0;
700 	}
701 
702 	INIT_WORK(&ew->work, fn);
703 	schedule_work(&ew->work);
704 
705 	return 1;
706 }
707 EXPORT_SYMBOL_GPL(execute_in_process_context);
708 
709 int keventd_up(void)
710 {
711 	return keventd_wq != NULL;
712 }
713 
714 int current_is_keventd(void)
715 {
716 	struct cpu_workqueue_struct *cwq;
717 	int cpu = smp_processor_id();	/* preempt-safe: keventd is per-cpu */
718 	int ret = 0;
719 
720 	BUG_ON(!keventd_wq);
721 
722 	cwq = per_cpu_ptr(keventd_wq->cpu_wq, cpu);
723 	if (current == cwq->thread)
724 		ret = 1;
725 
726 	return ret;
727 
728 }
729 
730 /* Take the work from this (downed) CPU. */
731 static void take_over_work(struct workqueue_struct *wq, unsigned int cpu)
732 {
733 	struct cpu_workqueue_struct *cwq = per_cpu_ptr(wq->cpu_wq, cpu);
734 	struct list_head list;
735 	struct work_struct *work;
736 
737 	spin_lock_irq(&cwq->lock);
738 	list_replace_init(&cwq->worklist, &list);
739 
740 	while (!list_empty(&list)) {
741 		printk("Taking work for %s\n", wq->name);
742 		work = list_entry(list.next,struct work_struct,entry);
743 		list_del(&work->entry);
744 		__queue_work(per_cpu_ptr(wq->cpu_wq, smp_processor_id()), work);
745 	}
746 	spin_unlock_irq(&cwq->lock);
747 }
748 
749 /* We're holding the cpucontrol mutex here */
750 static int __devinit workqueue_cpu_callback(struct notifier_block *nfb,
751 				  unsigned long action,
752 				  void *hcpu)
753 {
754 	unsigned int hotcpu = (unsigned long)hcpu;
755 	struct workqueue_struct *wq;
756 
757 	switch (action) {
758 	case CPU_UP_PREPARE:
759 		mutex_lock(&workqueue_mutex);
760 		/* Create a new workqueue thread for it. */
761 		list_for_each_entry(wq, &workqueues, list) {
762 			if (!create_workqueue_thread(wq, hotcpu, 0)) {
763 				printk("workqueue for %i failed\n", hotcpu);
764 				return NOTIFY_BAD;
765 			}
766 		}
767 		break;
768 
769 	case CPU_ONLINE:
770 		/* Kick off worker threads. */
771 		list_for_each_entry(wq, &workqueues, list) {
772 			struct cpu_workqueue_struct *cwq;
773 
774 			cwq = per_cpu_ptr(wq->cpu_wq, hotcpu);
775 			kthread_bind(cwq->thread, hotcpu);
776 			wake_up_process(cwq->thread);
777 		}
778 		mutex_unlock(&workqueue_mutex);
779 		break;
780 
781 	case CPU_UP_CANCELED:
782 		list_for_each_entry(wq, &workqueues, list) {
783 			if (!per_cpu_ptr(wq->cpu_wq, hotcpu)->thread)
784 				continue;
785 			/* Unbind so it can run. */
786 			kthread_bind(per_cpu_ptr(wq->cpu_wq, hotcpu)->thread,
787 				     any_online_cpu(cpu_online_map));
788 			cleanup_workqueue_thread(wq, hotcpu);
789 		}
790 		mutex_unlock(&workqueue_mutex);
791 		break;
792 
793 	case CPU_DOWN_PREPARE:
794 		mutex_lock(&workqueue_mutex);
795 		break;
796 
797 	case CPU_DOWN_FAILED:
798 		mutex_unlock(&workqueue_mutex);
799 		break;
800 
801 	case CPU_DEAD:
802 		list_for_each_entry(wq, &workqueues, list)
803 			cleanup_workqueue_thread(wq, hotcpu);
804 		list_for_each_entry(wq, &workqueues, list)
805 			take_over_work(wq, hotcpu);
806 		mutex_unlock(&workqueue_mutex);
807 		break;
808 	}
809 
810 	return NOTIFY_OK;
811 }
812 
813 void init_workqueues(void)
814 {
815 	singlethread_cpu = first_cpu(cpu_possible_map);
816 	hotcpu_notifier(workqueue_cpu_callback, 0);
817 	keventd_wq = create_workqueue("events");
818 	BUG_ON(!keventd_wq);
819 }
820 
821