Loading...
1/*
2 * Generic waiting primitives.
3 *
4 * (C) 2004 Nadia Yvette Chambers, Oracle
5 */
6#include <linux/init.h>
7#include <linux/export.h>
8#include <linux/sched.h>
9#include <linux/mm.h>
10#include <linux/wait.h>
11#include <linux/hash.h>
12#include <linux/kthread.h>
13
14void __init_waitqueue_head(wait_queue_head_t *q, const char *name, struct lock_class_key *key)
15{
16 spin_lock_init(&q->lock);
17 lockdep_set_class_and_name(&q->lock, key, name);
18 INIT_LIST_HEAD(&q->task_list);
19}
20
21EXPORT_SYMBOL(__init_waitqueue_head);
22
23void add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
24{
25 unsigned long flags;
26
27 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
28 spin_lock_irqsave(&q->lock, flags);
29 __add_wait_queue(q, wait);
30 spin_unlock_irqrestore(&q->lock, flags);
31}
32EXPORT_SYMBOL(add_wait_queue);
33
34void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait)
35{
36 unsigned long flags;
37
38 wait->flags |= WQ_FLAG_EXCLUSIVE;
39 spin_lock_irqsave(&q->lock, flags);
40 __add_wait_queue_tail(q, wait);
41 spin_unlock_irqrestore(&q->lock, flags);
42}
43EXPORT_SYMBOL(add_wait_queue_exclusive);
44
45void remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
46{
47 unsigned long flags;
48
49 spin_lock_irqsave(&q->lock, flags);
50 __remove_wait_queue(q, wait);
51 spin_unlock_irqrestore(&q->lock, flags);
52}
53EXPORT_SYMBOL(remove_wait_queue);
54
55
56/*
57 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
58 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
59 * number) then we wake all the non-exclusive tasks and one exclusive task.
60 *
61 * There are circumstances in which we can try to wake a task which has already
62 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
63 * zero in this (rare) case, and we handle it by continuing to scan the queue.
64 */
65static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
66 int nr_exclusive, int wake_flags, void *key)
67{
68 wait_queue_t *curr, *next;
69
70 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
71 unsigned flags = curr->flags;
72
73 if (curr->func(curr, mode, wake_flags, key) &&
74 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
75 break;
76 }
77}
78
79/**
80 * __wake_up - wake up threads blocked on a waitqueue.
81 * @q: the waitqueue
82 * @mode: which threads
83 * @nr_exclusive: how many wake-one or wake-many threads to wake up
84 * @key: is directly passed to the wakeup function
85 *
86 * It may be assumed that this function implies a write memory barrier before
87 * changing the task state if and only if any tasks are woken up.
88 */
89void __wake_up(wait_queue_head_t *q, unsigned int mode,
90 int nr_exclusive, void *key)
91{
92 unsigned long flags;
93
94 spin_lock_irqsave(&q->lock, flags);
95 __wake_up_common(q, mode, nr_exclusive, 0, key);
96 spin_unlock_irqrestore(&q->lock, flags);
97}
98EXPORT_SYMBOL(__wake_up);
99
100/*
101 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
102 */
103void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr)
104{
105 __wake_up_common(q, mode, nr, 0, NULL);
106}
107EXPORT_SYMBOL_GPL(__wake_up_locked);
108
109void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
110{
111 __wake_up_common(q, mode, 1, 0, key);
112}
113EXPORT_SYMBOL_GPL(__wake_up_locked_key);
114
115/**
116 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
117 * @q: the waitqueue
118 * @mode: which threads
119 * @nr_exclusive: how many wake-one or wake-many threads to wake up
120 * @key: opaque value to be passed to wakeup targets
121 *
122 * The sync wakeup differs that the waker knows that it will schedule
123 * away soon, so while the target thread will be woken up, it will not
124 * be migrated to another CPU - ie. the two threads are 'synchronized'
125 * with each other. This can prevent needless bouncing between CPUs.
126 *
127 * On UP it can prevent extra preemption.
128 *
129 * It may be assumed that this function implies a write memory barrier before
130 * changing the task state if and only if any tasks are woken up.
131 */
132void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
133 int nr_exclusive, void *key)
134{
135 unsigned long flags;
136 int wake_flags = 1; /* XXX WF_SYNC */
137
138 if (unlikely(!q))
139 return;
140
141 if (unlikely(nr_exclusive != 1))
142 wake_flags = 0;
143
144 spin_lock_irqsave(&q->lock, flags);
145 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
146 spin_unlock_irqrestore(&q->lock, flags);
147}
148EXPORT_SYMBOL_GPL(__wake_up_sync_key);
149
150/*
151 * __wake_up_sync - see __wake_up_sync_key()
152 */
153void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
154{
155 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
156}
157EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
158
159/*
160 * Note: we use "set_current_state()" _after_ the wait-queue add,
161 * because we need a memory barrier there on SMP, so that any
162 * wake-function that tests for the wait-queue being active
163 * will be guaranteed to see waitqueue addition _or_ subsequent
164 * tests in this thread will see the wakeup having taken place.
165 *
166 * The spin_unlock() itself is semi-permeable and only protects
167 * one way (it only protects stuff inside the critical region and
168 * stops them from bleeding out - it would still allow subsequent
169 * loads to move into the critical region).
170 */
171void
172prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
173{
174 unsigned long flags;
175
176 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
177 spin_lock_irqsave(&q->lock, flags);
178 if (list_empty(&wait->task_list))
179 __add_wait_queue(q, wait);
180 set_current_state(state);
181 spin_unlock_irqrestore(&q->lock, flags);
182}
183EXPORT_SYMBOL(prepare_to_wait);
184
185void
186prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
187{
188 unsigned long flags;
189
190 wait->flags |= WQ_FLAG_EXCLUSIVE;
191 spin_lock_irqsave(&q->lock, flags);
192 if (list_empty(&wait->task_list))
193 __add_wait_queue_tail(q, wait);
194 set_current_state(state);
195 spin_unlock_irqrestore(&q->lock, flags);
196}
197EXPORT_SYMBOL(prepare_to_wait_exclusive);
198
199void init_wait_entry(wait_queue_t *wait, int flags)
200{
201 wait->flags = flags;
202 wait->private = current;
203 wait->func = autoremove_wake_function;
204 INIT_LIST_HEAD(&wait->task_list);
205}
206EXPORT_SYMBOL(init_wait_entry);
207
208long prepare_to_wait_event(wait_queue_head_t *q, wait_queue_t *wait, int state)
209{
210 unsigned long flags;
211 long ret = 0;
212
213 spin_lock_irqsave(&q->lock, flags);
214 if (unlikely(signal_pending_state(state, current))) {
215 /*
216 * Exclusive waiter must not fail if it was selected by wakeup,
217 * it should "consume" the condition we were waiting for.
218 *
219 * The caller will recheck the condition and return success if
220 * we were already woken up, we can not miss the event because
221 * wakeup locks/unlocks the same q->lock.
222 *
223 * But we need to ensure that set-condition + wakeup after that
224 * can't see us, it should wake up another exclusive waiter if
225 * we fail.
226 */
227 list_del_init(&wait->task_list);
228 ret = -ERESTARTSYS;
229 } else {
230 if (list_empty(&wait->task_list)) {
231 if (wait->flags & WQ_FLAG_EXCLUSIVE)
232 __add_wait_queue_tail(q, wait);
233 else
234 __add_wait_queue(q, wait);
235 }
236 set_current_state(state);
237 }
238 spin_unlock_irqrestore(&q->lock, flags);
239
240 return ret;
241}
242EXPORT_SYMBOL(prepare_to_wait_event);
243
244/**
245 * finish_wait - clean up after waiting in a queue
246 * @q: waitqueue waited on
247 * @wait: wait descriptor
248 *
249 * Sets current thread back to running state and removes
250 * the wait descriptor from the given waitqueue if still
251 * queued.
252 */
253void finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
254{
255 unsigned long flags;
256
257 __set_current_state(TASK_RUNNING);
258 /*
259 * We can check for list emptiness outside the lock
260 * IFF:
261 * - we use the "careful" check that verifies both
262 * the next and prev pointers, so that there cannot
263 * be any half-pending updates in progress on other
264 * CPU's that we haven't seen yet (and that might
265 * still change the stack area.
266 * and
267 * - all other users take the lock (ie we can only
268 * have _one_ other CPU that looks at or modifies
269 * the list).
270 */
271 if (!list_empty_careful(&wait->task_list)) {
272 spin_lock_irqsave(&q->lock, flags);
273 list_del_init(&wait->task_list);
274 spin_unlock_irqrestore(&q->lock, flags);
275 }
276}
277EXPORT_SYMBOL(finish_wait);
278
279int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
280{
281 int ret = default_wake_function(wait, mode, sync, key);
282
283 if (ret)
284 list_del_init(&wait->task_list);
285 return ret;
286}
287EXPORT_SYMBOL(autoremove_wake_function);
288
289static inline bool is_kthread_should_stop(void)
290{
291 return (current->flags & PF_KTHREAD) && kthread_should_stop();
292}
293
294/*
295 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
296 *
297 * add_wait_queue(&wq, &wait);
298 * for (;;) {
299 * if (condition)
300 * break;
301 *
302 * p->state = mode; condition = true;
303 * smp_mb(); // A smp_wmb(); // C
304 * if (!wait->flags & WQ_FLAG_WOKEN) wait->flags |= WQ_FLAG_WOKEN;
305 * schedule() try_to_wake_up();
306 * p->state = TASK_RUNNING; ~~~~~~~~~~~~~~~~~~
307 * wait->flags &= ~WQ_FLAG_WOKEN; condition = true;
308 * smp_mb() // B smp_wmb(); // C
309 * wait->flags |= WQ_FLAG_WOKEN;
310 * }
311 * remove_wait_queue(&wq, &wait);
312 *
313 */
314long wait_woken(wait_queue_t *wait, unsigned mode, long timeout)
315{
316 set_current_state(mode); /* A */
317 /*
318 * The above implies an smp_mb(), which matches with the smp_wmb() from
319 * woken_wake_function() such that if we observe WQ_FLAG_WOKEN we must
320 * also observe all state before the wakeup.
321 */
322 if (!(wait->flags & WQ_FLAG_WOKEN) && !is_kthread_should_stop())
323 timeout = schedule_timeout(timeout);
324 __set_current_state(TASK_RUNNING);
325
326 /*
327 * The below implies an smp_mb(), it too pairs with the smp_wmb() from
328 * woken_wake_function() such that we must either observe the wait
329 * condition being true _OR_ WQ_FLAG_WOKEN such that we will not miss
330 * an event.
331 */
332 smp_store_mb(wait->flags, wait->flags & ~WQ_FLAG_WOKEN); /* B */
333
334 return timeout;
335}
336EXPORT_SYMBOL(wait_woken);
337
338int woken_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
339{
340 /*
341 * Although this function is called under waitqueue lock, LOCK
342 * doesn't imply write barrier and the users expects write
343 * barrier semantics on wakeup functions. The following
344 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
345 * and is paired with smp_store_mb() in wait_woken().
346 */
347 smp_wmb(); /* C */
348 wait->flags |= WQ_FLAG_WOKEN;
349
350 return default_wake_function(wait, mode, sync, key);
351}
352EXPORT_SYMBOL(woken_wake_function);
353
354int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
355{
356 struct wait_bit_key *key = arg;
357 struct wait_bit_queue *wait_bit
358 = container_of(wait, struct wait_bit_queue, wait);
359
360 if (wait_bit->key.flags != key->flags ||
361 wait_bit->key.bit_nr != key->bit_nr ||
362 test_bit(key->bit_nr, key->flags))
363 return 0;
364 else
365 return autoremove_wake_function(wait, mode, sync, key);
366}
367EXPORT_SYMBOL(wake_bit_function);
368
369/*
370 * To allow interruptible waiting and asynchronous (i.e. nonblocking)
371 * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are
372 * permitted return codes. Nonzero return codes halt waiting and return.
373 */
374int __sched
375__wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q,
376 wait_bit_action_f *action, unsigned mode)
377{
378 int ret = 0;
379
380 do {
381 prepare_to_wait(wq, &q->wait, mode);
382 if (test_bit(q->key.bit_nr, q->key.flags))
383 ret = (*action)(&q->key, mode);
384 } while (test_bit(q->key.bit_nr, q->key.flags) && !ret);
385 finish_wait(wq, &q->wait);
386 return ret;
387}
388EXPORT_SYMBOL(__wait_on_bit);
389
390int __sched out_of_line_wait_on_bit(void *word, int bit,
391 wait_bit_action_f *action, unsigned mode)
392{
393 wait_queue_head_t *wq = bit_waitqueue(word, bit);
394 DEFINE_WAIT_BIT(wait, word, bit);
395
396 return __wait_on_bit(wq, &wait, action, mode);
397}
398EXPORT_SYMBOL(out_of_line_wait_on_bit);
399
400int __sched out_of_line_wait_on_bit_timeout(
401 void *word, int bit, wait_bit_action_f *action,
402 unsigned mode, unsigned long timeout)
403{
404 wait_queue_head_t *wq = bit_waitqueue(word, bit);
405 DEFINE_WAIT_BIT(wait, word, bit);
406
407 wait.key.timeout = jiffies + timeout;
408 return __wait_on_bit(wq, &wait, action, mode);
409}
410EXPORT_SYMBOL_GPL(out_of_line_wait_on_bit_timeout);
411
412int __sched
413__wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q,
414 wait_bit_action_f *action, unsigned mode)
415{
416 int ret = 0;
417
418 for (;;) {
419 prepare_to_wait_exclusive(wq, &q->wait, mode);
420 if (test_bit(q->key.bit_nr, q->key.flags)) {
421 ret = action(&q->key, mode);
422 /*
423 * See the comment in prepare_to_wait_event().
424 * finish_wait() does not necessarily takes wq->lock,
425 * but test_and_set_bit() implies mb() which pairs with
426 * smp_mb__after_atomic() before wake_up_page().
427 */
428 if (ret)
429 finish_wait(wq, &q->wait);
430 }
431 if (!test_and_set_bit(q->key.bit_nr, q->key.flags)) {
432 if (!ret)
433 finish_wait(wq, &q->wait);
434 return 0;
435 } else if (ret) {
436 return ret;
437 }
438 }
439}
440EXPORT_SYMBOL(__wait_on_bit_lock);
441
442int __sched out_of_line_wait_on_bit_lock(void *word, int bit,
443 wait_bit_action_f *action, unsigned mode)
444{
445 wait_queue_head_t *wq = bit_waitqueue(word, bit);
446 DEFINE_WAIT_BIT(wait, word, bit);
447
448 return __wait_on_bit_lock(wq, &wait, action, mode);
449}
450EXPORT_SYMBOL(out_of_line_wait_on_bit_lock);
451
452void __wake_up_bit(wait_queue_head_t *wq, void *word, int bit)
453{
454 struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit);
455 if (waitqueue_active(wq))
456 __wake_up(wq, TASK_NORMAL, 1, &key);
457}
458EXPORT_SYMBOL(__wake_up_bit);
459
460/**
461 * wake_up_bit - wake up a waiter on a bit
462 * @word: the word being waited on, a kernel virtual address
463 * @bit: the bit of the word being waited on
464 *
465 * There is a standard hashed waitqueue table for generic use. This
466 * is the part of the hashtable's accessor API that wakes up waiters
467 * on a bit. For instance, if one were to have waiters on a bitflag,
468 * one would call wake_up_bit() after clearing the bit.
469 *
470 * In order for this to function properly, as it uses waitqueue_active()
471 * internally, some kind of memory barrier must be done prior to calling
472 * this. Typically, this will be smp_mb__after_atomic(), but in some
473 * cases where bitflags are manipulated non-atomically under a lock, one
474 * may need to use a less regular barrier, such fs/inode.c's smp_mb(),
475 * because spin_unlock() does not guarantee a memory barrier.
476 */
477void wake_up_bit(void *word, int bit)
478{
479 __wake_up_bit(bit_waitqueue(word, bit), word, bit);
480}
481EXPORT_SYMBOL(wake_up_bit);
482
483/*
484 * Manipulate the atomic_t address to produce a better bit waitqueue table hash
485 * index (we're keying off bit -1, but that would produce a horrible hash
486 * value).
487 */
488static inline wait_queue_head_t *atomic_t_waitqueue(atomic_t *p)
489{
490 if (BITS_PER_LONG == 64) {
491 unsigned long q = (unsigned long)p;
492 return bit_waitqueue((void *)(q & ~1), q & 1);
493 }
494 return bit_waitqueue(p, 0);
495}
496
497static int wake_atomic_t_function(wait_queue_t *wait, unsigned mode, int sync,
498 void *arg)
499{
500 struct wait_bit_key *key = arg;
501 struct wait_bit_queue *wait_bit
502 = container_of(wait, struct wait_bit_queue, wait);
503 atomic_t *val = key->flags;
504
505 if (wait_bit->key.flags != key->flags ||
506 wait_bit->key.bit_nr != key->bit_nr ||
507 atomic_read(val) != 0)
508 return 0;
509 return autoremove_wake_function(wait, mode, sync, key);
510}
511
512/*
513 * To allow interruptible waiting and asynchronous (i.e. nonblocking) waiting,
514 * the actions of __wait_on_atomic_t() are permitted return codes. Nonzero
515 * return codes halt waiting and return.
516 */
517static __sched
518int __wait_on_atomic_t(wait_queue_head_t *wq, struct wait_bit_queue *q,
519 int (*action)(atomic_t *), unsigned mode)
520{
521 atomic_t *val;
522 int ret = 0;
523
524 do {
525 prepare_to_wait(wq, &q->wait, mode);
526 val = q->key.flags;
527 if (atomic_read(val) == 0)
528 break;
529 ret = (*action)(val);
530 } while (!ret && atomic_read(val) != 0);
531 finish_wait(wq, &q->wait);
532 return ret;
533}
534
535#define DEFINE_WAIT_ATOMIC_T(name, p) \
536 struct wait_bit_queue name = { \
537 .key = __WAIT_ATOMIC_T_KEY_INITIALIZER(p), \
538 .wait = { \
539 .private = current, \
540 .func = wake_atomic_t_function, \
541 .task_list = \
542 LIST_HEAD_INIT((name).wait.task_list), \
543 }, \
544 }
545
546__sched int out_of_line_wait_on_atomic_t(atomic_t *p, int (*action)(atomic_t *),
547 unsigned mode)
548{
549 wait_queue_head_t *wq = atomic_t_waitqueue(p);
550 DEFINE_WAIT_ATOMIC_T(wait, p);
551
552 return __wait_on_atomic_t(wq, &wait, action, mode);
553}
554EXPORT_SYMBOL(out_of_line_wait_on_atomic_t);
555
556/**
557 * wake_up_atomic_t - Wake up a waiter on a atomic_t
558 * @p: The atomic_t being waited on, a kernel virtual address
559 *
560 * Wake up anyone waiting for the atomic_t to go to zero.
561 *
562 * Abuse the bit-waker function and its waitqueue hash table set (the atomic_t
563 * check is done by the waiter's wake function, not the by the waker itself).
564 */
565void wake_up_atomic_t(atomic_t *p)
566{
567 __wake_up_bit(atomic_t_waitqueue(p), p, WAIT_ATOMIC_T_BIT_NR);
568}
569EXPORT_SYMBOL(wake_up_atomic_t);
570
571__sched int bit_wait(struct wait_bit_key *word, int mode)
572{
573 schedule();
574 if (signal_pending_state(mode, current))
575 return -EINTR;
576 return 0;
577}
578EXPORT_SYMBOL(bit_wait);
579
580__sched int bit_wait_io(struct wait_bit_key *word, int mode)
581{
582 io_schedule();
583 if (signal_pending_state(mode, current))
584 return -EINTR;
585 return 0;
586}
587EXPORT_SYMBOL(bit_wait_io);
588
589__sched int bit_wait_timeout(struct wait_bit_key *word, int mode)
590{
591 unsigned long now = READ_ONCE(jiffies);
592 if (time_after_eq(now, word->timeout))
593 return -EAGAIN;
594 schedule_timeout(word->timeout - now);
595 if (signal_pending_state(mode, current))
596 return -EINTR;
597 return 0;
598}
599EXPORT_SYMBOL_GPL(bit_wait_timeout);
600
601__sched int bit_wait_io_timeout(struct wait_bit_key *word, int mode)
602{
603 unsigned long now = READ_ONCE(jiffies);
604 if (time_after_eq(now, word->timeout))
605 return -EAGAIN;
606 io_schedule_timeout(word->timeout - now);
607 if (signal_pending_state(mode, current))
608 return -EINTR;
609 return 0;
610}
611EXPORT_SYMBOL_GPL(bit_wait_io_timeout);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Generic waiting primitives.
4 *
5 * (C) 2004 Nadia Yvette Chambers, Oracle
6 */
7#include "sched.h"
8
9void __init_waitqueue_head(struct wait_queue_head *wq_head, const char *name, struct lock_class_key *key)
10{
11 spin_lock_init(&wq_head->lock);
12 lockdep_set_class_and_name(&wq_head->lock, key, name);
13 INIT_LIST_HEAD(&wq_head->head);
14}
15
16EXPORT_SYMBOL(__init_waitqueue_head);
17
18void add_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
19{
20 unsigned long flags;
21
22 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
23 spin_lock_irqsave(&wq_head->lock, flags);
24 __add_wait_queue(wq_head, wq_entry);
25 spin_unlock_irqrestore(&wq_head->lock, flags);
26}
27EXPORT_SYMBOL(add_wait_queue);
28
29void add_wait_queue_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
30{
31 unsigned long flags;
32
33 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
34 spin_lock_irqsave(&wq_head->lock, flags);
35 __add_wait_queue_entry_tail(wq_head, wq_entry);
36 spin_unlock_irqrestore(&wq_head->lock, flags);
37}
38EXPORT_SYMBOL(add_wait_queue_exclusive);
39
40void add_wait_queue_priority(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
41{
42 unsigned long flags;
43
44 wq_entry->flags |= WQ_FLAG_EXCLUSIVE | WQ_FLAG_PRIORITY;
45 spin_lock_irqsave(&wq_head->lock, flags);
46 __add_wait_queue(wq_head, wq_entry);
47 spin_unlock_irqrestore(&wq_head->lock, flags);
48}
49EXPORT_SYMBOL_GPL(add_wait_queue_priority);
50
51void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
52{
53 unsigned long flags;
54
55 spin_lock_irqsave(&wq_head->lock, flags);
56 __remove_wait_queue(wq_head, wq_entry);
57 spin_unlock_irqrestore(&wq_head->lock, flags);
58}
59EXPORT_SYMBOL(remove_wait_queue);
60
61/*
62 * Scan threshold to break wait queue walk.
63 * This allows a waker to take a break from holding the
64 * wait queue lock during the wait queue walk.
65 */
66#define WAITQUEUE_WALK_BREAK_CNT 64
67
68/*
69 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
70 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
71 * number) then we wake that number of exclusive tasks, and potentially all
72 * the non-exclusive tasks. Normally, exclusive tasks will be at the end of
73 * the list and any non-exclusive tasks will be woken first. A priority task
74 * may be at the head of the list, and can consume the event without any other
75 * tasks being woken.
76 *
77 * There are circumstances in which we can try to wake a task which has already
78 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
79 * zero in this (rare) case, and we handle it by continuing to scan the queue.
80 */
81static int __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
82 int nr_exclusive, int wake_flags, void *key,
83 wait_queue_entry_t *bookmark)
84{
85 wait_queue_entry_t *curr, *next;
86 int cnt = 0;
87
88 lockdep_assert_held(&wq_head->lock);
89
90 if (bookmark && (bookmark->flags & WQ_FLAG_BOOKMARK)) {
91 curr = list_next_entry(bookmark, entry);
92
93 list_del(&bookmark->entry);
94 bookmark->flags = 0;
95 } else
96 curr = list_first_entry(&wq_head->head, wait_queue_entry_t, entry);
97
98 if (&curr->entry == &wq_head->head)
99 return nr_exclusive;
100
101 list_for_each_entry_safe_from(curr, next, &wq_head->head, entry) {
102 unsigned flags = curr->flags;
103 int ret;
104
105 if (flags & WQ_FLAG_BOOKMARK)
106 continue;
107
108 ret = curr->func(curr, mode, wake_flags, key);
109 if (ret < 0)
110 break;
111 if (ret && (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
112 break;
113
114 if (bookmark && (++cnt > WAITQUEUE_WALK_BREAK_CNT) &&
115 (&next->entry != &wq_head->head)) {
116 bookmark->flags = WQ_FLAG_BOOKMARK;
117 list_add_tail(&bookmark->entry, &next->entry);
118 break;
119 }
120 }
121
122 return nr_exclusive;
123}
124
125static void __wake_up_common_lock(struct wait_queue_head *wq_head, unsigned int mode,
126 int nr_exclusive, int wake_flags, void *key)
127{
128 unsigned long flags;
129 wait_queue_entry_t bookmark;
130
131 bookmark.flags = 0;
132 bookmark.private = NULL;
133 bookmark.func = NULL;
134 INIT_LIST_HEAD(&bookmark.entry);
135
136 do {
137 spin_lock_irqsave(&wq_head->lock, flags);
138 nr_exclusive = __wake_up_common(wq_head, mode, nr_exclusive,
139 wake_flags, key, &bookmark);
140 spin_unlock_irqrestore(&wq_head->lock, flags);
141 } while (bookmark.flags & WQ_FLAG_BOOKMARK);
142}
143
144/**
145 * __wake_up - wake up threads blocked on a waitqueue.
146 * @wq_head: the waitqueue
147 * @mode: which threads
148 * @nr_exclusive: how many wake-one or wake-many threads to wake up
149 * @key: is directly passed to the wakeup function
150 *
151 * If this function wakes up a task, it executes a full memory barrier before
152 * accessing the task state.
153 */
154void __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
155 int nr_exclusive, void *key)
156{
157 __wake_up_common_lock(wq_head, mode, nr_exclusive, 0, key);
158}
159EXPORT_SYMBOL(__wake_up);
160
161/*
162 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
163 */
164void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
165{
166 __wake_up_common(wq_head, mode, nr, 0, NULL, NULL);
167}
168EXPORT_SYMBOL_GPL(__wake_up_locked);
169
170void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
171{
172 __wake_up_common(wq_head, mode, 1, 0, key, NULL);
173}
174EXPORT_SYMBOL_GPL(__wake_up_locked_key);
175
176void __wake_up_locked_key_bookmark(struct wait_queue_head *wq_head,
177 unsigned int mode, void *key, wait_queue_entry_t *bookmark)
178{
179 __wake_up_common(wq_head, mode, 1, 0, key, bookmark);
180}
181EXPORT_SYMBOL_GPL(__wake_up_locked_key_bookmark);
182
183/**
184 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
185 * @wq_head: the waitqueue
186 * @mode: which threads
187 * @key: opaque value to be passed to wakeup targets
188 *
189 * The sync wakeup differs that the waker knows that it will schedule
190 * away soon, so while the target thread will be woken up, it will not
191 * be migrated to another CPU - ie. the two threads are 'synchronized'
192 * with each other. This can prevent needless bouncing between CPUs.
193 *
194 * On UP it can prevent extra preemption.
195 *
196 * If this function wakes up a task, it executes a full memory barrier before
197 * accessing the task state.
198 */
199void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
200 void *key)
201{
202 if (unlikely(!wq_head))
203 return;
204
205 __wake_up_common_lock(wq_head, mode, 1, WF_SYNC, key);
206}
207EXPORT_SYMBOL_GPL(__wake_up_sync_key);
208
209/**
210 * __wake_up_locked_sync_key - wake up a thread blocked on a locked waitqueue.
211 * @wq_head: the waitqueue
212 * @mode: which threads
213 * @key: opaque value to be passed to wakeup targets
214 *
215 * The sync wakeup differs in that the waker knows that it will schedule
216 * away soon, so while the target thread will be woken up, it will not
217 * be migrated to another CPU - ie. the two threads are 'synchronized'
218 * with each other. This can prevent needless bouncing between CPUs.
219 *
220 * On UP it can prevent extra preemption.
221 *
222 * If this function wakes up a task, it executes a full memory barrier before
223 * accessing the task state.
224 */
225void __wake_up_locked_sync_key(struct wait_queue_head *wq_head,
226 unsigned int mode, void *key)
227{
228 __wake_up_common(wq_head, mode, 1, WF_SYNC, key, NULL);
229}
230EXPORT_SYMBOL_GPL(__wake_up_locked_sync_key);
231
232/*
233 * __wake_up_sync - see __wake_up_sync_key()
234 */
235void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode)
236{
237 __wake_up_sync_key(wq_head, mode, NULL);
238}
239EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
240
241/*
242 * Note: we use "set_current_state()" _after_ the wait-queue add,
243 * because we need a memory barrier there on SMP, so that any
244 * wake-function that tests for the wait-queue being active
245 * will be guaranteed to see waitqueue addition _or_ subsequent
246 * tests in this thread will see the wakeup having taken place.
247 *
248 * The spin_unlock() itself is semi-permeable and only protects
249 * one way (it only protects stuff inside the critical region and
250 * stops them from bleeding out - it would still allow subsequent
251 * loads to move into the critical region).
252 */
253void
254prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
255{
256 unsigned long flags;
257
258 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
259 spin_lock_irqsave(&wq_head->lock, flags);
260 if (list_empty(&wq_entry->entry))
261 __add_wait_queue(wq_head, wq_entry);
262 set_current_state(state);
263 spin_unlock_irqrestore(&wq_head->lock, flags);
264}
265EXPORT_SYMBOL(prepare_to_wait);
266
267/* Returns true if we are the first waiter in the queue, false otherwise. */
268bool
269prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
270{
271 unsigned long flags;
272 bool was_empty = false;
273
274 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
275 spin_lock_irqsave(&wq_head->lock, flags);
276 if (list_empty(&wq_entry->entry)) {
277 was_empty = list_empty(&wq_head->head);
278 __add_wait_queue_entry_tail(wq_head, wq_entry);
279 }
280 set_current_state(state);
281 spin_unlock_irqrestore(&wq_head->lock, flags);
282 return was_empty;
283}
284EXPORT_SYMBOL(prepare_to_wait_exclusive);
285
286void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
287{
288 wq_entry->flags = flags;
289 wq_entry->private = current;
290 wq_entry->func = autoremove_wake_function;
291 INIT_LIST_HEAD(&wq_entry->entry);
292}
293EXPORT_SYMBOL(init_wait_entry);
294
295long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
296{
297 unsigned long flags;
298 long ret = 0;
299
300 spin_lock_irqsave(&wq_head->lock, flags);
301 if (signal_pending_state(state, current)) {
302 /*
303 * Exclusive waiter must not fail if it was selected by wakeup,
304 * it should "consume" the condition we were waiting for.
305 *
306 * The caller will recheck the condition and return success if
307 * we were already woken up, we can not miss the event because
308 * wakeup locks/unlocks the same wq_head->lock.
309 *
310 * But we need to ensure that set-condition + wakeup after that
311 * can't see us, it should wake up another exclusive waiter if
312 * we fail.
313 */
314 list_del_init(&wq_entry->entry);
315 ret = -ERESTARTSYS;
316 } else {
317 if (list_empty(&wq_entry->entry)) {
318 if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
319 __add_wait_queue_entry_tail(wq_head, wq_entry);
320 else
321 __add_wait_queue(wq_head, wq_entry);
322 }
323 set_current_state(state);
324 }
325 spin_unlock_irqrestore(&wq_head->lock, flags);
326
327 return ret;
328}
329EXPORT_SYMBOL(prepare_to_wait_event);
330
331/*
332 * Note! These two wait functions are entered with the
333 * wait-queue lock held (and interrupts off in the _irq
334 * case), so there is no race with testing the wakeup
335 * condition in the caller before they add the wait
336 * entry to the wake queue.
337 */
338int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
339{
340 if (likely(list_empty(&wait->entry)))
341 __add_wait_queue_entry_tail(wq, wait);
342
343 set_current_state(TASK_INTERRUPTIBLE);
344 if (signal_pending(current))
345 return -ERESTARTSYS;
346
347 spin_unlock(&wq->lock);
348 schedule();
349 spin_lock(&wq->lock);
350
351 return 0;
352}
353EXPORT_SYMBOL(do_wait_intr);
354
355int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
356{
357 if (likely(list_empty(&wait->entry)))
358 __add_wait_queue_entry_tail(wq, wait);
359
360 set_current_state(TASK_INTERRUPTIBLE);
361 if (signal_pending(current))
362 return -ERESTARTSYS;
363
364 spin_unlock_irq(&wq->lock);
365 schedule();
366 spin_lock_irq(&wq->lock);
367
368 return 0;
369}
370EXPORT_SYMBOL(do_wait_intr_irq);
371
372/**
373 * finish_wait - clean up after waiting in a queue
374 * @wq_head: waitqueue waited on
375 * @wq_entry: wait descriptor
376 *
377 * Sets current thread back to running state and removes
378 * the wait descriptor from the given waitqueue if still
379 * queued.
380 */
381void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
382{
383 unsigned long flags;
384
385 __set_current_state(TASK_RUNNING);
386 /*
387 * We can check for list emptiness outside the lock
388 * IFF:
389 * - we use the "careful" check that verifies both
390 * the next and prev pointers, so that there cannot
391 * be any half-pending updates in progress on other
392 * CPU's that we haven't seen yet (and that might
393 * still change the stack area.
394 * and
395 * - all other users take the lock (ie we can only
396 * have _one_ other CPU that looks at or modifies
397 * the list).
398 */
399 if (!list_empty_careful(&wq_entry->entry)) {
400 spin_lock_irqsave(&wq_head->lock, flags);
401 list_del_init(&wq_entry->entry);
402 spin_unlock_irqrestore(&wq_head->lock, flags);
403 }
404}
405EXPORT_SYMBOL(finish_wait);
406
407int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
408{
409 int ret = default_wake_function(wq_entry, mode, sync, key);
410
411 if (ret)
412 list_del_init_careful(&wq_entry->entry);
413
414 return ret;
415}
416EXPORT_SYMBOL(autoremove_wake_function);
417
418static inline bool is_kthread_should_stop(void)
419{
420 return (current->flags & PF_KTHREAD) && kthread_should_stop();
421}
422
423/*
424 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
425 *
426 * add_wait_queue(&wq_head, &wait);
427 * for (;;) {
428 * if (condition)
429 * break;
430 *
431 * // in wait_woken() // in woken_wake_function()
432 *
433 * p->state = mode; wq_entry->flags |= WQ_FLAG_WOKEN;
434 * smp_mb(); // A try_to_wake_up():
435 * if (!(wq_entry->flags & WQ_FLAG_WOKEN)) <full barrier>
436 * schedule() if (p->state & mode)
437 * p->state = TASK_RUNNING; p->state = TASK_RUNNING;
438 * wq_entry->flags &= ~WQ_FLAG_WOKEN; ~~~~~~~~~~~~~~~~~~
439 * smp_mb(); // B condition = true;
440 * } smp_mb(); // C
441 * remove_wait_queue(&wq_head, &wait); wq_entry->flags |= WQ_FLAG_WOKEN;
442 */
443long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
444{
445 /*
446 * The below executes an smp_mb(), which matches with the full barrier
447 * executed by the try_to_wake_up() in woken_wake_function() such that
448 * either we see the store to wq_entry->flags in woken_wake_function()
449 * or woken_wake_function() sees our store to current->state.
450 */
451 set_current_state(mode); /* A */
452 if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !is_kthread_should_stop())
453 timeout = schedule_timeout(timeout);
454 __set_current_state(TASK_RUNNING);
455
456 /*
457 * The below executes an smp_mb(), which matches with the smp_mb() (C)
458 * in woken_wake_function() such that either we see the wait condition
459 * being true or the store to wq_entry->flags in woken_wake_function()
460 * follows ours in the coherence order.
461 */
462 smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
463
464 return timeout;
465}
466EXPORT_SYMBOL(wait_woken);
467
468int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
469{
470 /* Pairs with the smp_store_mb() in wait_woken(). */
471 smp_mb(); /* C */
472 wq_entry->flags |= WQ_FLAG_WOKEN;
473
474 return default_wake_function(wq_entry, mode, sync, key);
475}
476EXPORT_SYMBOL(woken_wake_function);