2 * Generic waiting primitives.
4 * (C) 2004 Nadia Yvette Chambers, Oracle
6 #include <linux/init.h>
7 #include <linux/export.h>
8 #include <linux/sched/signal.h>
9 #include <linux/sched/debug.h>
11 #include <linux/wait.h>
12 #include <linux/hash.h>
13 #include <linux/kthread.h>
15 void __init_waitqueue_head(wait_queue_head_t
*q
, const char *name
, struct lock_class_key
*key
)
17 spin_lock_init(&q
->lock
);
18 lockdep_set_class_and_name(&q
->lock
, key
, name
);
19 INIT_LIST_HEAD(&q
->task_list
);
22 EXPORT_SYMBOL(__init_waitqueue_head
);
24 void add_wait_queue(wait_queue_head_t
*q
, struct wait_queue_entry
*wq_entry
)
28 wq_entry
->flags
&= ~WQ_FLAG_EXCLUSIVE
;
29 spin_lock_irqsave(&q
->lock
, flags
);
30 __add_wait_queue_entry_tail(q
, wq_entry
);
31 spin_unlock_irqrestore(&q
->lock
, flags
);
33 EXPORT_SYMBOL(add_wait_queue
);
35 void add_wait_queue_exclusive(wait_queue_head_t
*q
, struct wait_queue_entry
*wq_entry
)
39 wq_entry
->flags
|= WQ_FLAG_EXCLUSIVE
;
40 spin_lock_irqsave(&q
->lock
, flags
);
41 __add_wait_queue_entry_tail(q
, wq_entry
);
42 spin_unlock_irqrestore(&q
->lock
, flags
);
44 EXPORT_SYMBOL(add_wait_queue_exclusive
);
46 void remove_wait_queue(wait_queue_head_t
*q
, struct wait_queue_entry
*wq_entry
)
50 spin_lock_irqsave(&q
->lock
, flags
);
51 __remove_wait_queue(q
, wq_entry
);
52 spin_unlock_irqrestore(&q
->lock
, flags
);
54 EXPORT_SYMBOL(remove_wait_queue
);
58 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
59 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
60 * number) then we wake all the non-exclusive tasks and one exclusive task.
62 * There are circumstances in which we can try to wake a task which has already
63 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
64 * zero in this (rare) case, and we handle it by continuing to scan the queue.
66 static void __wake_up_common(wait_queue_head_t
*q
, unsigned int mode
,
67 int nr_exclusive
, int wake_flags
, void *key
)
69 wait_queue_entry_t
*curr
, *next
;
71 list_for_each_entry_safe(curr
, next
, &q
->task_list
, task_list
) {
72 unsigned flags
= curr
->flags
;
74 if (curr
->func(curr
, mode
, wake_flags
, key
) &&
75 (flags
& WQ_FLAG_EXCLUSIVE
) && !--nr_exclusive
)
81 * __wake_up - wake up threads blocked on a waitqueue.
83 * @mode: which threads
84 * @nr_exclusive: how many wake-one or wake-many threads to wake up
85 * @key: is directly passed to the wakeup function
87 * It may be assumed that this function implies a write memory barrier before
88 * changing the task state if and only if any tasks are woken up.
90 void __wake_up(wait_queue_head_t
*q
, unsigned int mode
,
91 int nr_exclusive
, void *key
)
95 spin_lock_irqsave(&q
->lock
, flags
);
96 __wake_up_common(q
, mode
, nr_exclusive
, 0, key
);
97 spin_unlock_irqrestore(&q
->lock
, flags
);
99 EXPORT_SYMBOL(__wake_up
);
102 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
104 void __wake_up_locked(wait_queue_head_t
*q
, unsigned int mode
, int nr
)
106 __wake_up_common(q
, mode
, nr
, 0, NULL
);
108 EXPORT_SYMBOL_GPL(__wake_up_locked
);
110 void __wake_up_locked_key(wait_queue_head_t
*q
, unsigned int mode
, void *key
)
112 __wake_up_common(q
, mode
, 1, 0, key
);
114 EXPORT_SYMBOL_GPL(__wake_up_locked_key
);
117 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
119 * @mode: which threads
120 * @nr_exclusive: how many wake-one or wake-many threads to wake up
121 * @key: opaque value to be passed to wakeup targets
123 * The sync wakeup differs that the waker knows that it will schedule
124 * away soon, so while the target thread will be woken up, it will not
125 * be migrated to another CPU - ie. the two threads are 'synchronized'
126 * with each other. This can prevent needless bouncing between CPUs.
128 * On UP it can prevent extra preemption.
130 * It may be assumed that this function implies a write memory barrier before
131 * changing the task state if and only if any tasks are woken up.
133 void __wake_up_sync_key(wait_queue_head_t
*q
, unsigned int mode
,
134 int nr_exclusive
, void *key
)
137 int wake_flags
= 1; /* XXX WF_SYNC */
142 if (unlikely(nr_exclusive
!= 1))
145 spin_lock_irqsave(&q
->lock
, flags
);
146 __wake_up_common(q
, mode
, nr_exclusive
, wake_flags
, key
);
147 spin_unlock_irqrestore(&q
->lock
, flags
);
149 EXPORT_SYMBOL_GPL(__wake_up_sync_key
);
152 * __wake_up_sync - see __wake_up_sync_key()
154 void __wake_up_sync(wait_queue_head_t
*q
, unsigned int mode
, int nr_exclusive
)
156 __wake_up_sync_key(q
, mode
, nr_exclusive
, NULL
);
158 EXPORT_SYMBOL_GPL(__wake_up_sync
); /* For internal use only */
161 * Note: we use "set_current_state()" _after_ the wait-queue add,
162 * because we need a memory barrier there on SMP, so that any
163 * wake-function that tests for the wait-queue being active
164 * will be guaranteed to see waitqueue addition _or_ subsequent
165 * tests in this thread will see the wakeup having taken place.
167 * The spin_unlock() itself is semi-permeable and only protects
168 * one way (it only protects stuff inside the critical region and
169 * stops them from bleeding out - it would still allow subsequent
170 * loads to move into the critical region).
173 prepare_to_wait(wait_queue_head_t
*q
, struct wait_queue_entry
*wq_entry
, int state
)
177 wq_entry
->flags
&= ~WQ_FLAG_EXCLUSIVE
;
178 spin_lock_irqsave(&q
->lock
, flags
);
179 if (list_empty(&wq_entry
->task_list
))
180 __add_wait_queue(q
, wq_entry
);
181 set_current_state(state
);
182 spin_unlock_irqrestore(&q
->lock
, flags
);
184 EXPORT_SYMBOL(prepare_to_wait
);
187 prepare_to_wait_exclusive(wait_queue_head_t
*q
, struct wait_queue_entry
*wq_entry
, int state
)
191 wq_entry
->flags
|= WQ_FLAG_EXCLUSIVE
;
192 spin_lock_irqsave(&q
->lock
, flags
);
193 if (list_empty(&wq_entry
->task_list
))
194 __add_wait_queue_entry_tail(q
, wq_entry
);
195 set_current_state(state
);
196 spin_unlock_irqrestore(&q
->lock
, flags
);
198 EXPORT_SYMBOL(prepare_to_wait_exclusive
);
200 void init_wait_entry(struct wait_queue_entry
*wq_entry
, int flags
)
202 wq_entry
->flags
= flags
;
203 wq_entry
->private = current
;
204 wq_entry
->func
= autoremove_wake_function
;
205 INIT_LIST_HEAD(&wq_entry
->task_list
);
207 EXPORT_SYMBOL(init_wait_entry
);
209 long prepare_to_wait_event(wait_queue_head_t
*q
, struct wait_queue_entry
*wq_entry
, int state
)
214 spin_lock_irqsave(&q
->lock
, flags
);
215 if (unlikely(signal_pending_state(state
, current
))) {
217 * Exclusive waiter must not fail if it was selected by wakeup,
218 * it should "consume" the condition we were waiting for.
220 * The caller will recheck the condition and return success if
221 * we were already woken up, we can not miss the event because
222 * wakeup locks/unlocks the same q->lock.
224 * But we need to ensure that set-condition + wakeup after that
225 * can't see us, it should wake up another exclusive waiter if
228 list_del_init(&wq_entry
->task_list
);
231 if (list_empty(&wq_entry
->task_list
)) {
232 if (wq_entry
->flags
& WQ_FLAG_EXCLUSIVE
)
233 __add_wait_queue_entry_tail(q
, wq_entry
);
235 __add_wait_queue(q
, wq_entry
);
237 set_current_state(state
);
239 spin_unlock_irqrestore(&q
->lock
, flags
);
243 EXPORT_SYMBOL(prepare_to_wait_event
);
246 * Note! These two wait functions are entered with the
247 * wait-queue lock held (and interrupts off in the _irq
248 * case), so there is no race with testing the wakeup
249 * condition in the caller before they add the wait
250 * entry to the wake queue.
252 int do_wait_intr(wait_queue_head_t
*wq
, wait_queue_entry_t
*wait
)
254 if (likely(list_empty(&wait
->task_list
)))
255 __add_wait_queue_entry_tail(wq
, wait
);
257 set_current_state(TASK_INTERRUPTIBLE
);
258 if (signal_pending(current
))
261 spin_unlock(&wq
->lock
);
263 spin_lock(&wq
->lock
);
266 EXPORT_SYMBOL(do_wait_intr
);
268 int do_wait_intr_irq(wait_queue_head_t
*wq
, wait_queue_entry_t
*wait
)
270 if (likely(list_empty(&wait
->task_list
)))
271 __add_wait_queue_entry_tail(wq
, wait
);
273 set_current_state(TASK_INTERRUPTIBLE
);
274 if (signal_pending(current
))
277 spin_unlock_irq(&wq
->lock
);
279 spin_lock_irq(&wq
->lock
);
282 EXPORT_SYMBOL(do_wait_intr_irq
);
285 * finish_wait - clean up after waiting in a queue
286 * @q: waitqueue waited on
287 * @wq_entry: wait descriptor
289 * Sets current thread back to running state and removes
290 * the wait descriptor from the given waitqueue if still
293 void finish_wait(wait_queue_head_t
*q
, struct wait_queue_entry
*wq_entry
)
297 __set_current_state(TASK_RUNNING
);
299 * We can check for list emptiness outside the lock
301 * - we use the "careful" check that verifies both
302 * the next and prev pointers, so that there cannot
303 * be any half-pending updates in progress on other
304 * CPU's that we haven't seen yet (and that might
305 * still change the stack area.
307 * - all other users take the lock (ie we can only
308 * have _one_ other CPU that looks at or modifies
311 if (!list_empty_careful(&wq_entry
->task_list
)) {
312 spin_lock_irqsave(&q
->lock
, flags
);
313 list_del_init(&wq_entry
->task_list
);
314 spin_unlock_irqrestore(&q
->lock
, flags
);
317 EXPORT_SYMBOL(finish_wait
);
319 int autoremove_wake_function(struct wait_queue_entry
*wq_entry
, unsigned mode
, int sync
, void *key
)
321 int ret
= default_wake_function(wq_entry
, mode
, sync
, key
);
324 list_del_init(&wq_entry
->task_list
);
327 EXPORT_SYMBOL(autoremove_wake_function
);
329 static inline bool is_kthread_should_stop(void)
331 return (current
->flags
& PF_KTHREAD
) && kthread_should_stop();
335 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
337 * add_wait_queue(&wq, &wait);
342 * p->state = mode; condition = true;
343 * smp_mb(); // A smp_wmb(); // C
344 * if (!wq_entry->flags & WQ_FLAG_WOKEN) wq_entry->flags |= WQ_FLAG_WOKEN;
345 * schedule() try_to_wake_up();
346 * p->state = TASK_RUNNING; ~~~~~~~~~~~~~~~~~~
347 * wq_entry->flags &= ~WQ_FLAG_WOKEN; condition = true;
348 * smp_mb() // B smp_wmb(); // C
349 * wq_entry->flags |= WQ_FLAG_WOKEN;
351 * remove_wait_queue(&wq, &wait);
354 long wait_woken(struct wait_queue_entry
*wq_entry
, unsigned mode
, long timeout
)
356 set_current_state(mode
); /* A */
358 * The above implies an smp_mb(), which matches with the smp_wmb() from
359 * woken_wake_function() such that if we observe WQ_FLAG_WOKEN we must
360 * also observe all state before the wakeup.
362 if (!(wq_entry
->flags
& WQ_FLAG_WOKEN
) && !is_kthread_should_stop())
363 timeout
= schedule_timeout(timeout
);
364 __set_current_state(TASK_RUNNING
);
367 * The below implies an smp_mb(), it too pairs with the smp_wmb() from
368 * woken_wake_function() such that we must either observe the wait
369 * condition being true _OR_ WQ_FLAG_WOKEN such that we will not miss
372 smp_store_mb(wq_entry
->flags
, wq_entry
->flags
& ~WQ_FLAG_WOKEN
); /* B */
376 EXPORT_SYMBOL(wait_woken
);
378 int woken_wake_function(struct wait_queue_entry
*wq_entry
, unsigned mode
, int sync
, void *key
)
381 * Although this function is called under waitqueue lock, LOCK
382 * doesn't imply write barrier and the users expects write
383 * barrier semantics on wakeup functions. The following
384 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
385 * and is paired with smp_store_mb() in wait_woken().
388 wq_entry
->flags
|= WQ_FLAG_WOKEN
;
390 return default_wake_function(wq_entry
, mode
, sync
, key
);
392 EXPORT_SYMBOL(woken_wake_function
);
394 int wake_bit_function(struct wait_queue_entry
*wq_entry
, unsigned mode
, int sync
, void *arg
)
396 struct wait_bit_key
*key
= arg
;
397 struct wait_bit_queue
*wait_bit
398 = container_of(wq_entry
, struct wait_bit_queue
, wait
);
400 if (wait_bit
->key
.flags
!= key
->flags
||
401 wait_bit
->key
.bit_nr
!= key
->bit_nr
||
402 test_bit(key
->bit_nr
, key
->flags
))
405 return autoremove_wake_function(wq_entry
, mode
, sync
, key
);
407 EXPORT_SYMBOL(wake_bit_function
);
410 * To allow interruptible waiting and asynchronous (i.e. nonblocking)
411 * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are
412 * permitted return codes. Nonzero return codes halt waiting and return.
415 __wait_on_bit(wait_queue_head_t
*wq
, struct wait_bit_queue
*q
,
416 wait_bit_action_f
*action
, unsigned mode
)
421 prepare_to_wait(wq
, &q
->wait
, mode
);
422 if (test_bit(q
->key
.bit_nr
, q
->key
.flags
))
423 ret
= (*action
)(&q
->key
, mode
);
424 } while (test_bit(q
->key
.bit_nr
, q
->key
.flags
) && !ret
);
425 finish_wait(wq
, &q
->wait
);
428 EXPORT_SYMBOL(__wait_on_bit
);
430 int __sched
out_of_line_wait_on_bit(void *word
, int bit
,
431 wait_bit_action_f
*action
, unsigned mode
)
433 wait_queue_head_t
*wq
= bit_waitqueue(word
, bit
);
434 DEFINE_WAIT_BIT(wait
, word
, bit
);
436 return __wait_on_bit(wq
, &wait
, action
, mode
);
438 EXPORT_SYMBOL(out_of_line_wait_on_bit
);
440 int __sched
out_of_line_wait_on_bit_timeout(
441 void *word
, int bit
, wait_bit_action_f
*action
,
442 unsigned mode
, unsigned long timeout
)
444 wait_queue_head_t
*wq
= bit_waitqueue(word
, bit
);
445 DEFINE_WAIT_BIT(wait
, word
, bit
);
447 wait
.key
.timeout
= jiffies
+ timeout
;
448 return __wait_on_bit(wq
, &wait
, action
, mode
);
450 EXPORT_SYMBOL_GPL(out_of_line_wait_on_bit_timeout
);
453 __wait_on_bit_lock(wait_queue_head_t
*wq
, struct wait_bit_queue
*q
,
454 wait_bit_action_f
*action
, unsigned mode
)
459 prepare_to_wait_exclusive(wq
, &q
->wait
, mode
);
460 if (test_bit(q
->key
.bit_nr
, q
->key
.flags
)) {
461 ret
= action(&q
->key
, mode
);
463 * See the comment in prepare_to_wait_event().
464 * finish_wait() does not necessarily takes wq->lock,
465 * but test_and_set_bit() implies mb() which pairs with
466 * smp_mb__after_atomic() before wake_up_page().
469 finish_wait(wq
, &q
->wait
);
471 if (!test_and_set_bit(q
->key
.bit_nr
, q
->key
.flags
)) {
473 finish_wait(wq
, &q
->wait
);
480 EXPORT_SYMBOL(__wait_on_bit_lock
);
482 int __sched
out_of_line_wait_on_bit_lock(void *word
, int bit
,
483 wait_bit_action_f
*action
, unsigned mode
)
485 wait_queue_head_t
*wq
= bit_waitqueue(word
, bit
);
486 DEFINE_WAIT_BIT(wait
, word
, bit
);
488 return __wait_on_bit_lock(wq
, &wait
, action
, mode
);
490 EXPORT_SYMBOL(out_of_line_wait_on_bit_lock
);
492 void __wake_up_bit(wait_queue_head_t
*wq
, void *word
, int bit
)
494 struct wait_bit_key key
= __WAIT_BIT_KEY_INITIALIZER(word
, bit
);
495 if (waitqueue_active(wq
))
496 __wake_up(wq
, TASK_NORMAL
, 1, &key
);
498 EXPORT_SYMBOL(__wake_up_bit
);
501 * wake_up_bit - wake up a waiter on a bit
502 * @word: the word being waited on, a kernel virtual address
503 * @bit: the bit of the word being waited on
505 * There is a standard hashed waitqueue table for generic use. This
506 * is the part of the hashtable's accessor API that wakes up waiters
507 * on a bit. For instance, if one were to have waiters on a bitflag,
508 * one would call wake_up_bit() after clearing the bit.
510 * In order for this to function properly, as it uses waitqueue_active()
511 * internally, some kind of memory barrier must be done prior to calling
512 * this. Typically, this will be smp_mb__after_atomic(), but in some
513 * cases where bitflags are manipulated non-atomically under a lock, one
514 * may need to use a less regular barrier, such fs/inode.c's smp_mb(),
515 * because spin_unlock() does not guarantee a memory barrier.
517 void wake_up_bit(void *word
, int bit
)
519 __wake_up_bit(bit_waitqueue(word
, bit
), word
, bit
);
521 EXPORT_SYMBOL(wake_up_bit
);
524 * Manipulate the atomic_t address to produce a better bit waitqueue table hash
525 * index (we're keying off bit -1, but that would produce a horrible hash
528 static inline wait_queue_head_t
*atomic_t_waitqueue(atomic_t
*p
)
530 if (BITS_PER_LONG
== 64) {
531 unsigned long q
= (unsigned long)p
;
532 return bit_waitqueue((void *)(q
& ~1), q
& 1);
534 return bit_waitqueue(p
, 0);
537 static int wake_atomic_t_function(struct wait_queue_entry
*wq_entry
, unsigned mode
, int sync
,
540 struct wait_bit_key
*key
= arg
;
541 struct wait_bit_queue
*wait_bit
542 = container_of(wq_entry
, struct wait_bit_queue
, wait
);
543 atomic_t
*val
= key
->flags
;
545 if (wait_bit
->key
.flags
!= key
->flags
||
546 wait_bit
->key
.bit_nr
!= key
->bit_nr
||
547 atomic_read(val
) != 0)
549 return autoremove_wake_function(wq_entry
, mode
, sync
, key
);
553 * To allow interruptible waiting and asynchronous (i.e. nonblocking) waiting,
554 * the actions of __wait_on_atomic_t() are permitted return codes. Nonzero
555 * return codes halt waiting and return.
558 int __wait_on_atomic_t(wait_queue_head_t
*wq
, struct wait_bit_queue
*q
,
559 int (*action
)(atomic_t
*), unsigned mode
)
565 prepare_to_wait(wq
, &q
->wait
, mode
);
567 if (atomic_read(val
) == 0)
569 ret
= (*action
)(val
);
570 } while (!ret
&& atomic_read(val
) != 0);
571 finish_wait(wq
, &q
->wait
);
575 #define DEFINE_WAIT_ATOMIC_T(name, p) \
576 struct wait_bit_queue name = { \
577 .key = __WAIT_ATOMIC_T_KEY_INITIALIZER(p), \
579 .private = current, \
580 .func = wake_atomic_t_function, \
582 LIST_HEAD_INIT((name).wait.task_list), \
586 __sched
int out_of_line_wait_on_atomic_t(atomic_t
*p
, int (*action
)(atomic_t
*),
589 wait_queue_head_t
*wq
= atomic_t_waitqueue(p
);
590 DEFINE_WAIT_ATOMIC_T(wait
, p
);
592 return __wait_on_atomic_t(wq
, &wait
, action
, mode
);
594 EXPORT_SYMBOL(out_of_line_wait_on_atomic_t
);
597 * wake_up_atomic_t - Wake up a waiter on a atomic_t
598 * @p: The atomic_t being waited on, a kernel virtual address
600 * Wake up anyone waiting for the atomic_t to go to zero.
602 * Abuse the bit-waker function and its waitqueue hash table set (the atomic_t
603 * check is done by the waiter's wake function, not the by the waker itself).
605 void wake_up_atomic_t(atomic_t
*p
)
607 __wake_up_bit(atomic_t_waitqueue(p
), p
, WAIT_ATOMIC_T_BIT_NR
);
609 EXPORT_SYMBOL(wake_up_atomic_t
);
611 __sched
int bit_wait(struct wait_bit_key
*word
, int mode
)
614 if (signal_pending_state(mode
, current
))
618 EXPORT_SYMBOL(bit_wait
);
620 __sched
int bit_wait_io(struct wait_bit_key
*word
, int mode
)
623 if (signal_pending_state(mode
, current
))
627 EXPORT_SYMBOL(bit_wait_io
);
629 __sched
int bit_wait_timeout(struct wait_bit_key
*word
, int mode
)
631 unsigned long now
= READ_ONCE(jiffies
);
632 if (time_after_eq(now
, word
->timeout
))
634 schedule_timeout(word
->timeout
- now
);
635 if (signal_pending_state(mode
, current
))
639 EXPORT_SYMBOL_GPL(bit_wait_timeout
);
641 __sched
int bit_wait_io_timeout(struct wait_bit_key
*word
, int mode
)
643 unsigned long now
= READ_ONCE(jiffies
);
644 if (time_after_eq(now
, word
->timeout
))
646 io_schedule_timeout(word
->timeout
- now
);
647 if (signal_pending_state(mode
, current
))
651 EXPORT_SYMBOL_GPL(bit_wait_io_timeout
);