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(struct wait_queue_head *wq_head, const char *name, struct lock_class_key *key)
17 spin_lock_init(&wq_head->lock);
18 lockdep_set_class_and_name(&wq_head->lock, key, name);
19 INIT_LIST_HEAD(&wq_head->head);
22 EXPORT_SYMBOL(__init_waitqueue_head);
24 void add_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
28 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
29 spin_lock_irqsave(&wq_head->lock, flags);
30 __add_wait_queue(wq_head, wq_entry);
31 spin_unlock_irqrestore(&wq_head->lock, flags);
33 EXPORT_SYMBOL(add_wait_queue);
35 void add_wait_queue_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
39 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
40 spin_lock_irqsave(&wq_head->lock, flags);
41 __add_wait_queue_entry_tail(wq_head, wq_entry);
42 spin_unlock_irqrestore(&wq_head->lock, flags);
44 EXPORT_SYMBOL(add_wait_queue_exclusive);
46 void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
50 spin_lock_irqsave(&wq_head->lock, flags);
51 __remove_wait_queue(wq_head, wq_entry);
52 spin_unlock_irqrestore(&wq_head->lock, flags);
54 EXPORT_SYMBOL(remove_wait_queue);
57 * Scan threshold to break wait queue walk.
58 * This allows a waker to take a break from holding the
59 * wait queue lock during the wait queue walk.
61 #define WAITQUEUE_WALK_BREAK_CNT 64
64 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
65 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
66 * number) then we wake all the non-exclusive tasks and one exclusive task.
68 * There are circumstances in which we can try to wake a task which has already
69 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
70 * zero in this (rare) case, and we handle it by continuing to scan the queue.
72 static int __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
73 int nr_exclusive, int wake_flags, void *key,
74 wait_queue_entry_t *bookmark)
76 wait_queue_entry_t *curr, *next;
79 if (bookmark && (bookmark->flags & WQ_FLAG_BOOKMARK)) {
80 curr = list_next_entry(bookmark, entry);
82 list_del(&bookmark->entry);
85 curr = list_first_entry(&wq_head->head, wait_queue_entry_t, entry);
87 if (&curr->entry == &wq_head->head)
90 list_for_each_entry_safe_from(curr, next, &wq_head->head, entry) {
91 unsigned flags = curr->flags;
94 if (flags & WQ_FLAG_BOOKMARK)
97 ret = curr->func(curr, mode, wake_flags, key);
100 if (ret && (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
103 if (bookmark && (++cnt > WAITQUEUE_WALK_BREAK_CNT) &&
104 (&next->entry != &wq_head->head)) {
105 bookmark->flags = WQ_FLAG_BOOKMARK;
106 list_add_tail(&bookmark->entry, &next->entry);
113 static void __wake_up_common_lock(struct wait_queue_head *wq_head, unsigned int mode,
114 int nr_exclusive, int wake_flags, void *key)
117 wait_queue_entry_t bookmark;
120 bookmark.private = NULL;
121 bookmark.func = NULL;
122 INIT_LIST_HEAD(&bookmark.entry);
124 spin_lock_irqsave(&wq_head->lock, flags);
125 nr_exclusive = __wake_up_common(wq_head, mode, nr_exclusive, wake_flags, key, &bookmark);
126 spin_unlock_irqrestore(&wq_head->lock, flags);
128 while (bookmark.flags & WQ_FLAG_BOOKMARK) {
129 spin_lock_irqsave(&wq_head->lock, flags);
130 nr_exclusive = __wake_up_common(wq_head, mode, nr_exclusive,
131 wake_flags, key, &bookmark);
132 spin_unlock_irqrestore(&wq_head->lock, flags);
137 * __wake_up - wake up threads blocked on a waitqueue.
138 * @wq_head: the waitqueue
139 * @mode: which threads
140 * @nr_exclusive: how many wake-one or wake-many threads to wake up
141 * @key: is directly passed to the wakeup function
143 * It may be assumed that this function implies a write memory barrier before
144 * changing the task state if and only if any tasks are woken up.
146 void __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
147 int nr_exclusive, void *key)
149 __wake_up_common_lock(wq_head, mode, nr_exclusive, 0, key);
151 EXPORT_SYMBOL(__wake_up);
154 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
156 void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
158 __wake_up_common(wq_head, mode, nr, 0, NULL, NULL);
160 EXPORT_SYMBOL_GPL(__wake_up_locked);
162 void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
164 __wake_up_common(wq_head, mode, 1, 0, key, NULL);
166 EXPORT_SYMBOL_GPL(__wake_up_locked_key);
168 void __wake_up_locked_key_bookmark(struct wait_queue_head *wq_head,
169 unsigned int mode, void *key, wait_queue_entry_t *bookmark)
171 __wake_up_common(wq_head, mode, 1, 0, key, bookmark);
173 EXPORT_SYMBOL_GPL(__wake_up_locked_key_bookmark);
176 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
177 * @wq_head: the waitqueue
178 * @mode: which threads
179 * @nr_exclusive: how many wake-one or wake-many threads to wake up
180 * @key: opaque value to be passed to wakeup targets
182 * The sync wakeup differs that the waker knows that it will schedule
183 * away soon, so while the target thread will be woken up, it will not
184 * be migrated to another CPU - ie. the two threads are 'synchronized'
185 * with each other. This can prevent needless bouncing between CPUs.
187 * On UP it can prevent extra preemption.
189 * It may be assumed that this function implies a write memory barrier before
190 * changing the task state if and only if any tasks are woken up.
192 void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
193 int nr_exclusive, void *key)
195 int wake_flags = 1; /* XXX WF_SYNC */
197 if (unlikely(!wq_head))
200 if (unlikely(nr_exclusive != 1))
203 __wake_up_common_lock(wq_head, mode, nr_exclusive, wake_flags, key);
205 EXPORT_SYMBOL_GPL(__wake_up_sync_key);
208 * __wake_up_sync - see __wake_up_sync_key()
210 void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode, int nr_exclusive)
212 __wake_up_sync_key(wq_head, mode, nr_exclusive, NULL);
214 EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
217 * Note: we use "set_current_state()" _after_ the wait-queue add,
218 * because we need a memory barrier there on SMP, so that any
219 * wake-function that tests for the wait-queue being active
220 * will be guaranteed to see waitqueue addition _or_ subsequent
221 * tests in this thread will see the wakeup having taken place.
223 * The spin_unlock() itself is semi-permeable and only protects
224 * one way (it only protects stuff inside the critical region and
225 * stops them from bleeding out - it would still allow subsequent
226 * loads to move into the critical region).
229 prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
233 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
234 spin_lock_irqsave(&wq_head->lock, flags);
235 if (list_empty(&wq_entry->entry))
236 __add_wait_queue(wq_head, wq_entry);
237 set_current_state(state);
238 spin_unlock_irqrestore(&wq_head->lock, flags);
240 EXPORT_SYMBOL(prepare_to_wait);
243 prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
247 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
248 spin_lock_irqsave(&wq_head->lock, flags);
249 if (list_empty(&wq_entry->entry))
250 __add_wait_queue_entry_tail(wq_head, wq_entry);
251 set_current_state(state);
252 spin_unlock_irqrestore(&wq_head->lock, flags);
254 EXPORT_SYMBOL(prepare_to_wait_exclusive);
256 void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
258 wq_entry->flags = flags;
259 wq_entry->private = current;
260 wq_entry->func = autoremove_wake_function;
261 INIT_LIST_HEAD(&wq_entry->entry);
263 EXPORT_SYMBOL(init_wait_entry);
265 long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
270 spin_lock_irqsave(&wq_head->lock, flags);
271 if (unlikely(signal_pending_state(state, current))) {
273 * Exclusive waiter must not fail if it was selected by wakeup,
274 * it should "consume" the condition we were waiting for.
276 * The caller will recheck the condition and return success if
277 * we were already woken up, we can not miss the event because
278 * wakeup locks/unlocks the same wq_head->lock.
280 * But we need to ensure that set-condition + wakeup after that
281 * can't see us, it should wake up another exclusive waiter if
284 list_del_init(&wq_entry->entry);
287 if (list_empty(&wq_entry->entry)) {
288 if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
289 __add_wait_queue_entry_tail(wq_head, wq_entry);
291 __add_wait_queue(wq_head, wq_entry);
293 set_current_state(state);
295 spin_unlock_irqrestore(&wq_head->lock, flags);
299 EXPORT_SYMBOL(prepare_to_wait_event);
302 * Note! These two wait functions are entered with the
303 * wait-queue lock held (and interrupts off in the _irq
304 * case), so there is no race with testing the wakeup
305 * condition in the caller before they add the wait
306 * entry to the wake queue.
308 int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
310 if (likely(list_empty(&wait->entry)))
311 __add_wait_queue_entry_tail(wq, wait);
313 set_current_state(TASK_INTERRUPTIBLE);
314 if (signal_pending(current))
317 spin_unlock(&wq->lock);
319 spin_lock(&wq->lock);
322 EXPORT_SYMBOL(do_wait_intr);
324 int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
326 if (likely(list_empty(&wait->entry)))
327 __add_wait_queue_entry_tail(wq, wait);
329 set_current_state(TASK_INTERRUPTIBLE);
330 if (signal_pending(current))
333 spin_unlock_irq(&wq->lock);
335 spin_lock_irq(&wq->lock);
338 EXPORT_SYMBOL(do_wait_intr_irq);
341 * finish_wait - clean up after waiting in a queue
342 * @wq_head: waitqueue waited on
343 * @wq_entry: wait descriptor
345 * Sets current thread back to running state and removes
346 * the wait descriptor from the given waitqueue if still
349 void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
353 __set_current_state(TASK_RUNNING);
355 * We can check for list emptiness outside the lock
357 * - we use the "careful" check that verifies both
358 * the next and prev pointers, so that there cannot
359 * be any half-pending updates in progress on other
360 * CPU's that we haven't seen yet (and that might
361 * still change the stack area.
363 * - all other users take the lock (ie we can only
364 * have _one_ other CPU that looks at or modifies
367 if (!list_empty_careful(&wq_entry->entry)) {
368 spin_lock_irqsave(&wq_head->lock, flags);
369 list_del_init(&wq_entry->entry);
370 spin_unlock_irqrestore(&wq_head->lock, flags);
373 EXPORT_SYMBOL(finish_wait);
375 int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
377 int ret = default_wake_function(wq_entry, mode, sync, key);
380 list_del_init(&wq_entry->entry);
383 EXPORT_SYMBOL(autoremove_wake_function);
385 static inline bool is_kthread_should_stop(void)
387 return (current->flags & PF_KTHREAD) && kthread_should_stop();
391 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
393 * add_wait_queue(&wq_head, &wait);
398 * // in wait_woken() // in woken_wake_function()
400 * p->state = mode; wq_entry->flags |= WQ_FLAG_WOKEN;
401 * smp_mb(); // A try_to_wake_up():
402 * if (!(wq_entry->flags & WQ_FLAG_WOKEN)) <full barrier>
403 * schedule() if (p->state & mode)
404 * p->state = TASK_RUNNING; p->state = TASK_RUNNING;
405 * wq_entry->flags &= ~WQ_FLAG_WOKEN; ~~~~~~~~~~~~~~~~~~
406 * smp_mb(); // B condition = true;
408 * remove_wait_queue(&wq_head, &wait); wq_entry->flags |= WQ_FLAG_WOKEN;
410 long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
413 * The below executes an smp_mb(), which matches with the full barrier
414 * executed by the try_to_wake_up() in woken_wake_function() such that
415 * either we see the store to wq_entry->flags in woken_wake_function()
416 * or woken_wake_function() sees our store to current->state.
418 set_current_state(mode); /* A */
419 if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !is_kthread_should_stop())
420 timeout = schedule_timeout(timeout);
421 __set_current_state(TASK_RUNNING);
424 * The below executes an smp_mb(), which matches with the smp_mb() (C)
425 * in woken_wake_function() such that either we see the wait condition
426 * being true or the store to wq_entry->flags in woken_wake_function()
427 * follows ours in the coherence order.
429 smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
433 EXPORT_SYMBOL(wait_woken);
435 int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
437 /* Pairs with the smp_store_mb() in wait_woken(). */
439 wq_entry->flags |= WQ_FLAG_WOKEN;
441 return default_wake_function(wq_entry, mode, sync, key);
443 EXPORT_SYMBOL(woken_wake_function);