2 * Generic waiting primitives.
4 * (C) 2004 Nadia Yvette Chambers, Oracle
8 void __init_waitqueue_head(struct wait_queue_head *wq_head, const char *name, struct lock_class_key *key)
10 spin_lock_init(&wq_head->lock);
11 lockdep_set_class_and_name(&wq_head->lock, key, name);
12 INIT_LIST_HEAD(&wq_head->head);
15 EXPORT_SYMBOL(__init_waitqueue_head);
17 void add_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
21 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
22 spin_lock_irqsave(&wq_head->lock, flags);
23 __add_wait_queue(wq_head, wq_entry);
24 spin_unlock_irqrestore(&wq_head->lock, flags);
26 EXPORT_SYMBOL(add_wait_queue);
28 void add_wait_queue_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
32 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
33 spin_lock_irqsave(&wq_head->lock, flags);
34 __add_wait_queue_entry_tail(wq_head, wq_entry);
35 spin_unlock_irqrestore(&wq_head->lock, flags);
37 EXPORT_SYMBOL(add_wait_queue_exclusive);
39 void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
43 spin_lock_irqsave(&wq_head->lock, flags);
44 __remove_wait_queue(wq_head, wq_entry);
45 spin_unlock_irqrestore(&wq_head->lock, flags);
47 EXPORT_SYMBOL(remove_wait_queue);
50 * Scan threshold to break wait queue walk.
51 * This allows a waker to take a break from holding the
52 * wait queue lock during the wait queue walk.
54 #define WAITQUEUE_WALK_BREAK_CNT 64
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.
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.
65 static int __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
66 int nr_exclusive, int wake_flags, void *key,
67 wait_queue_entry_t *bookmark)
69 wait_queue_entry_t *curr, *next;
72 lockdep_assert_held(&wq_head->lock);
74 if (bookmark && (bookmark->flags & WQ_FLAG_BOOKMARK)) {
75 curr = list_next_entry(bookmark, entry);
77 list_del(&bookmark->entry);
80 curr = list_first_entry(&wq_head->head, wait_queue_entry_t, entry);
82 if (&curr->entry == &wq_head->head)
85 list_for_each_entry_safe_from(curr, next, &wq_head->head, entry) {
86 unsigned flags = curr->flags;
89 if (flags & WQ_FLAG_BOOKMARK)
92 ret = curr->func(curr, mode, wake_flags, key);
95 if (ret && (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
98 if (bookmark && (++cnt > WAITQUEUE_WALK_BREAK_CNT) &&
99 (&next->entry != &wq_head->head)) {
100 bookmark->flags = WQ_FLAG_BOOKMARK;
101 list_add_tail(&bookmark->entry, &next->entry);
109 static void __wake_up_common_lock(struct wait_queue_head *wq_head, unsigned int mode,
110 int nr_exclusive, int wake_flags, void *key)
113 wait_queue_entry_t bookmark;
116 bookmark.private = NULL;
117 bookmark.func = NULL;
118 INIT_LIST_HEAD(&bookmark.entry);
120 spin_lock_irqsave(&wq_head->lock, flags);
121 nr_exclusive = __wake_up_common(wq_head, mode, nr_exclusive, wake_flags, key, &bookmark);
122 spin_unlock_irqrestore(&wq_head->lock, flags);
124 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
125 spin_lock_irqsave(&wq_head->lock, flags);
126 nr_exclusive = __wake_up_common(wq_head, mode, nr_exclusive,
127 wake_flags, key, &bookmark);
128 spin_unlock_irqrestore(&wq_head->lock, flags);
133 * __wake_up - wake up threads blocked on a waitqueue.
134 * @wq_head: the waitqueue
135 * @mode: which threads
136 * @nr_exclusive: how many wake-one or wake-many threads to wake up
137 * @key: is directly passed to the wakeup function
139 * If this function wakes up a task, it executes a full memory barrier before
140 * accessing the task state.
142 void __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
143 int nr_exclusive, void *key)
145 __wake_up_common_lock(wq_head, mode, nr_exclusive, 0, key);
147 EXPORT_SYMBOL(__wake_up);
150 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
152 void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
154 __wake_up_common(wq_head, mode, nr, 0, NULL, NULL);
156 EXPORT_SYMBOL_GPL(__wake_up_locked);
158 void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
160 __wake_up_common(wq_head, mode, 1, 0, key, NULL);
162 EXPORT_SYMBOL_GPL(__wake_up_locked_key);
164 void __wake_up_locked_key_bookmark(struct wait_queue_head *wq_head,
165 unsigned int mode, void *key, wait_queue_entry_t *bookmark)
167 __wake_up_common(wq_head, mode, 1, 0, key, bookmark);
169 EXPORT_SYMBOL_GPL(__wake_up_locked_key_bookmark);
172 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
173 * @wq_head: the waitqueue
174 * @mode: which threads
175 * @nr_exclusive: how many wake-one or wake-many threads to wake up
176 * @key: opaque value to be passed to wakeup targets
178 * The sync wakeup differs that the waker knows that it will schedule
179 * away soon, so while the target thread will be woken up, it will not
180 * be migrated to another CPU - ie. the two threads are 'synchronized'
181 * with each other. This can prevent needless bouncing between CPUs.
183 * On UP it can prevent extra preemption.
185 * If this function wakes up a task, it executes a full memory barrier before
186 * accessing the task state.
188 void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
189 int nr_exclusive, void *key)
191 int wake_flags = 1; /* XXX WF_SYNC */
193 if (unlikely(!wq_head))
196 if (unlikely(nr_exclusive != 1))
199 __wake_up_common_lock(wq_head, mode, nr_exclusive, wake_flags, key);
201 EXPORT_SYMBOL_GPL(__wake_up_sync_key);
204 * __wake_up_sync - see __wake_up_sync_key()
206 void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode, int nr_exclusive)
208 __wake_up_sync_key(wq_head, mode, nr_exclusive, NULL);
210 EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
212 void __wake_up_pollfree(struct wait_queue_head *wq_head)
214 __wake_up(wq_head, TASK_NORMAL, 0, poll_to_key(EPOLLHUP | POLLFREE));
215 /* POLLFREE must have cleared the queue. */
216 WARN_ON_ONCE(waitqueue_active(wq_head));
220 * Note: we use "set_current_state()" _after_ the wait-queue add,
221 * because we need a memory barrier there on SMP, so that any
222 * wake-function that tests for the wait-queue being active
223 * will be guaranteed to see waitqueue addition _or_ subsequent
224 * tests in this thread will see the wakeup having taken place.
226 * The spin_unlock() itself is semi-permeable and only protects
227 * one way (it only protects stuff inside the critical region and
228 * stops them from bleeding out - it would still allow subsequent
229 * loads to move into the critical region).
232 prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
236 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
237 spin_lock_irqsave(&wq_head->lock, flags);
238 if (list_empty(&wq_entry->entry))
239 __add_wait_queue(wq_head, wq_entry);
240 set_current_state(state);
241 spin_unlock_irqrestore(&wq_head->lock, flags);
243 EXPORT_SYMBOL(prepare_to_wait);
246 prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
250 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
251 spin_lock_irqsave(&wq_head->lock, flags);
252 if (list_empty(&wq_entry->entry))
253 __add_wait_queue_entry_tail(wq_head, wq_entry);
254 set_current_state(state);
255 spin_unlock_irqrestore(&wq_head->lock, flags);
257 EXPORT_SYMBOL(prepare_to_wait_exclusive);
259 void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
261 wq_entry->flags = flags;
262 wq_entry->private = current;
263 wq_entry->func = autoremove_wake_function;
264 INIT_LIST_HEAD(&wq_entry->entry);
266 EXPORT_SYMBOL(init_wait_entry);
268 long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
273 spin_lock_irqsave(&wq_head->lock, flags);
274 if (unlikely(signal_pending_state(state, current))) {
276 * Exclusive waiter must not fail if it was selected by wakeup,
277 * it should "consume" the condition we were waiting for.
279 * The caller will recheck the condition and return success if
280 * we were already woken up, we can not miss the event because
281 * wakeup locks/unlocks the same wq_head->lock.
283 * But we need to ensure that set-condition + wakeup after that
284 * can't see us, it should wake up another exclusive waiter if
287 list_del_init(&wq_entry->entry);
290 if (list_empty(&wq_entry->entry)) {
291 if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
292 __add_wait_queue_entry_tail(wq_head, wq_entry);
294 __add_wait_queue(wq_head, wq_entry);
296 set_current_state(state);
298 spin_unlock_irqrestore(&wq_head->lock, flags);
302 EXPORT_SYMBOL(prepare_to_wait_event);
305 * Note! These two wait functions are entered with the
306 * wait-queue lock held (and interrupts off in the _irq
307 * case), so there is no race with testing the wakeup
308 * condition in the caller before they add the wait
309 * entry to the wake queue.
311 int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
313 if (likely(list_empty(&wait->entry)))
314 __add_wait_queue_entry_tail(wq, wait);
316 set_current_state(TASK_INTERRUPTIBLE);
317 if (signal_pending(current))
320 spin_unlock(&wq->lock);
322 spin_lock(&wq->lock);
326 EXPORT_SYMBOL(do_wait_intr);
328 int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
330 if (likely(list_empty(&wait->entry)))
331 __add_wait_queue_entry_tail(wq, wait);
333 set_current_state(TASK_INTERRUPTIBLE);
334 if (signal_pending(current))
337 spin_unlock_irq(&wq->lock);
339 spin_lock_irq(&wq->lock);
343 EXPORT_SYMBOL(do_wait_intr_irq);
346 * finish_wait - clean up after waiting in a queue
347 * @wq_head: waitqueue waited on
348 * @wq_entry: wait descriptor
350 * Sets current thread back to running state and removes
351 * the wait descriptor from the given waitqueue if still
354 void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
358 __set_current_state(TASK_RUNNING);
360 * We can check for list emptiness outside the lock
362 * - we use the "careful" check that verifies both
363 * the next and prev pointers, so that there cannot
364 * be any half-pending updates in progress on other
365 * CPU's that we haven't seen yet (and that might
366 * still change the stack area.
368 * - all other users take the lock (ie we can only
369 * have _one_ other CPU that looks at or modifies
372 if (!list_empty_careful(&wq_entry->entry)) {
373 spin_lock_irqsave(&wq_head->lock, flags);
374 list_del_init(&wq_entry->entry);
375 spin_unlock_irqrestore(&wq_head->lock, flags);
378 EXPORT_SYMBOL(finish_wait);
380 int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
382 int ret = default_wake_function(wq_entry, mode, sync, key);
385 list_del_init(&wq_entry->entry);
389 EXPORT_SYMBOL(autoremove_wake_function);
391 static inline bool is_kthread_should_stop(void)
393 return (current->flags & PF_KTHREAD) && kthread_should_stop();
397 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
399 * add_wait_queue(&wq_head, &wait);
404 * // in wait_woken() // in woken_wake_function()
406 * p->state = mode; wq_entry->flags |= WQ_FLAG_WOKEN;
407 * smp_mb(); // A try_to_wake_up():
408 * if (!(wq_entry->flags & WQ_FLAG_WOKEN)) <full barrier>
409 * schedule() if (p->state & mode)
410 * p->state = TASK_RUNNING; p->state = TASK_RUNNING;
411 * wq_entry->flags &= ~WQ_FLAG_WOKEN; ~~~~~~~~~~~~~~~~~~
412 * smp_mb(); // B condition = true;
414 * remove_wait_queue(&wq_head, &wait); wq_entry->flags |= WQ_FLAG_WOKEN;
416 long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
419 * The below executes an smp_mb(), which matches with the full barrier
420 * executed by the try_to_wake_up() in woken_wake_function() such that
421 * either we see the store to wq_entry->flags in woken_wake_function()
422 * or woken_wake_function() sees our store to current->state.
424 set_current_state(mode); /* A */
425 if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !is_kthread_should_stop())
426 timeout = schedule_timeout(timeout);
427 __set_current_state(TASK_RUNNING);
430 * The below executes an smp_mb(), which matches with the smp_mb() (C)
431 * in woken_wake_function() such that either we see the wait condition
432 * being true or the store to wq_entry->flags in woken_wake_function()
433 * follows ours in the coherence order.
435 smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
439 EXPORT_SYMBOL(wait_woken);
441 int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
443 /* Pairs with the smp_store_mb() in wait_woken(). */
445 wq_entry->flags |= WQ_FLAG_WOKEN;
447 return default_wake_function(wq_entry, mode, sync, key);
449 EXPORT_SYMBOL(woken_wake_function);