1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Generic waiting primitives.
4 *
5 * (C) 2004 Nadia Yvette Chambers, Oracle
6 */
7
8void __init_waitqueue_head(struct wait_queue_head *wq_head, const char *name, struct lock_class_key *key)
9{
10 spin_lock_init(&wq_head->lock);
11 lockdep_set_class_and_name(&wq_head->lock, key, name);
12 INIT_LIST_HEAD(list: &wq_head->head);
13}
14
15EXPORT_SYMBOL(__init_waitqueue_head);
16
17void add_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
18{
19 unsigned long flags;
20
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(lock: &wq_head->lock, flags);
25}
26EXPORT_SYMBOL(add_wait_queue);
27
28void add_wait_queue_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
29{
30 unsigned long flags;
31
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(lock: &wq_head->lock, flags);
36}
37EXPORT_SYMBOL(add_wait_queue_exclusive);
38
39void add_wait_queue_priority(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
40{
41 unsigned long flags;
42
43 wq_entry->flags |= WQ_FLAG_EXCLUSIVE | WQ_FLAG_PRIORITY;
44 spin_lock_irqsave(&wq_head->lock, flags);
45 __add_wait_queue(wq_head, wq_entry);
46 spin_unlock_irqrestore(lock: &wq_head->lock, flags);
47}
48EXPORT_SYMBOL_GPL(add_wait_queue_priority);
49
50void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
51{
52 unsigned long flags;
53
54 spin_lock_irqsave(&wq_head->lock, flags);
55 __remove_wait_queue(wq_head, wq_entry);
56 spin_unlock_irqrestore(lock: &wq_head->lock, flags);
57}
58EXPORT_SYMBOL(remove_wait_queue);
59
60/*
61 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
62 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
63 * number) then we wake that number of exclusive tasks, and potentially all
64 * the non-exclusive tasks. Normally, exclusive tasks will be at the end of
65 * the list and any non-exclusive tasks will be woken first. A priority task
66 * may be at the head of the list, and can consume the event without any other
67 * tasks being woken.
68 *
69 * There are circumstances in which we can try to wake a task which has already
70 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
71 * zero in this (rare) case, and we handle it by continuing to scan the queue.
72 */
73static int __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
74 int nr_exclusive, int wake_flags, void *key)
75{
76 wait_queue_entry_t *curr, *next;
77
78 lockdep_assert_held(&wq_head->lock);
79
80 curr = list_first_entry(&wq_head->head, wait_queue_entry_t, entry);
81
82 if (&curr->entry == &wq_head->head)
83 return nr_exclusive;
84
85 list_for_each_entry_safe_from(curr, next, &wq_head->head, entry) {
86 unsigned flags = curr->flags;
87 int ret;
88
89 ret = curr->func(curr, mode, wake_flags, key);
90 if (ret < 0)
91 break;
92 if (ret && (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
93 break;
94 }
95
96 return nr_exclusive;
97}
98
99static int __wake_up_common_lock(struct wait_queue_head *wq_head, unsigned int mode,
100 int nr_exclusive, int wake_flags, void *key)
101{
102 unsigned long flags;
103 int remaining;
104
105 spin_lock_irqsave(&wq_head->lock, flags);
106 remaining = __wake_up_common(wq_head, mode, nr_exclusive, wake_flags,
107 key);
108 spin_unlock_irqrestore(lock: &wq_head->lock, flags);
109
110 return nr_exclusive - remaining;
111}
112
113/**
114 * __wake_up - wake up threads blocked on a waitqueue.
115 * @wq_head: the waitqueue
116 * @mode: which threads
117 * @nr_exclusive: how many wake-one or wake-many threads to wake up
118 * @key: is directly passed to the wakeup function
119 *
120 * If this function wakes up a task, it executes a full memory barrier
121 * before accessing the task state. Returns the number of exclusive
122 * tasks that were awaken.
123 */
124int __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
125 int nr_exclusive, void *key)
126{
127 return __wake_up_common_lock(wq_head, mode, nr_exclusive, wake_flags: 0, key);
128}
129EXPORT_SYMBOL(__wake_up);
130
131void __wake_up_on_current_cpu(struct wait_queue_head *wq_head, unsigned int mode, void *key)
132{
133 __wake_up_common_lock(wq_head, mode, nr_exclusive: 1, WF_CURRENT_CPU, key);
134}
135
136/*
137 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
138 */
139void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
140{
141 __wake_up_common(wq_head, mode, nr_exclusive: nr, wake_flags: 0, NULL);
142}
143EXPORT_SYMBOL_GPL(__wake_up_locked);
144
145void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
146{
147 __wake_up_common(wq_head, mode, nr_exclusive: 1, wake_flags: 0, key);
148}
149EXPORT_SYMBOL_GPL(__wake_up_locked_key);
150
151/**
152 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
153 * @wq_head: the waitqueue
154 * @mode: which threads
155 * @key: opaque value to be passed to wakeup targets
156 *
157 * The sync wakeup differs that the waker knows that it will schedule
158 * away soon, so while the target thread will be woken up, it will not
159 * be migrated to another CPU - ie. the two threads are 'synchronized'
160 * with each other. This can prevent needless bouncing between CPUs.
161 *
162 * On UP it can prevent extra preemption.
163 *
164 * If this function wakes up a task, it executes a full memory barrier before
165 * accessing the task state.
166 */
167void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
168 void *key)
169{
170 if (unlikely(!wq_head))
171 return;
172
173 __wake_up_common_lock(wq_head, mode, nr_exclusive: 1, WF_SYNC, key);
174}
175EXPORT_SYMBOL_GPL(__wake_up_sync_key);
176
177/**
178 * __wake_up_locked_sync_key - wake up a thread blocked on a locked waitqueue.
179 * @wq_head: the waitqueue
180 * @mode: which threads
181 * @key: opaque value to be passed to wakeup targets
182 *
183 * The sync wakeup differs in that the waker knows that it will schedule
184 * away soon, so while the target thread will be woken up, it will not
185 * be migrated to another CPU - ie. the two threads are 'synchronized'
186 * with each other. This can prevent needless bouncing between CPUs.
187 *
188 * On UP it can prevent extra preemption.
189 *
190 * If this function wakes up a task, it executes a full memory barrier before
191 * accessing the task state.
192 */
193void __wake_up_locked_sync_key(struct wait_queue_head *wq_head,
194 unsigned int mode, void *key)
195{
196 __wake_up_common(wq_head, mode, nr_exclusive: 1, WF_SYNC, key);
197}
198EXPORT_SYMBOL_GPL(__wake_up_locked_sync_key);
199
200/*
201 * __wake_up_sync - see __wake_up_sync_key()
202 */
203void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode)
204{
205 __wake_up_sync_key(wq_head, mode, NULL);
206}
207EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
208
209void __wake_up_pollfree(struct wait_queue_head *wq_head)
210{
211 __wake_up(wq_head, TASK_NORMAL, 0, poll_to_key(EPOLLHUP | POLLFREE));
212 /* POLLFREE must have cleared the queue. */
213 WARN_ON_ONCE(waitqueue_active(wq_head));
214}
215
216/*
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.
222 *
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).
227 */
228void
229prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
230{
231 unsigned long flags;
232
233 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
234 spin_lock_irqsave(&wq_head->lock, flags);
235 if (list_empty(head: &wq_entry->entry))
236 __add_wait_queue(wq_head, wq_entry);
237 set_current_state(state);
238 spin_unlock_irqrestore(lock: &wq_head->lock, flags);
239}
240EXPORT_SYMBOL(prepare_to_wait);
241
242/* Returns true if we are the first waiter in the queue, false otherwise. */
243bool
244prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
245{
246 unsigned long flags;
247 bool was_empty = false;
248
249 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
250 spin_lock_irqsave(&wq_head->lock, flags);
251 if (list_empty(head: &wq_entry->entry)) {
252 was_empty = list_empty(head: &wq_head->head);
253 __add_wait_queue_entry_tail(wq_head, wq_entry);
254 }
255 set_current_state(state);
256 spin_unlock_irqrestore(lock: &wq_head->lock, flags);
257 return was_empty;
258}
259EXPORT_SYMBOL(prepare_to_wait_exclusive);
260
261void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
262{
263 wq_entry->flags = flags;
264 wq_entry->private = current;
265 wq_entry->func = autoremove_wake_function;
266 INIT_LIST_HEAD(list: &wq_entry->entry);
267}
268EXPORT_SYMBOL(init_wait_entry);
269
270long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
271{
272 unsigned long flags;
273 long ret = 0;
274
275 spin_lock_irqsave(&wq_head->lock, flags);
276 if (signal_pending_state(state, current)) {
277 /*
278 * Exclusive waiter must not fail if it was selected by wakeup,
279 * it should "consume" the condition we were waiting for.
280 *
281 * The caller will recheck the condition and return success if
282 * we were already woken up, we can not miss the event because
283 * wakeup locks/unlocks the same wq_head->lock.
284 *
285 * But we need to ensure that set-condition + wakeup after that
286 * can't see us, it should wake up another exclusive waiter if
287 * we fail.
288 */
289 list_del_init(entry: &wq_entry->entry);
290 ret = -ERESTARTSYS;
291 } else {
292 if (list_empty(head: &wq_entry->entry)) {
293 if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
294 __add_wait_queue_entry_tail(wq_head, wq_entry);
295 else
296 __add_wait_queue(wq_head, wq_entry);
297 }
298 set_current_state(state);
299 }
300 spin_unlock_irqrestore(lock: &wq_head->lock, flags);
301
302 return ret;
303}
304EXPORT_SYMBOL(prepare_to_wait_event);
305
306/*
307 * Note! These two wait functions are entered with the
308 * wait-queue lock held (and interrupts off in the _irq
309 * case), so there is no race with testing the wakeup
310 * condition in the caller before they add the wait
311 * entry to the wake queue.
312 */
313int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
314{
315 if (likely(list_empty(&wait->entry)))
316 __add_wait_queue_entry_tail(wq_head: wq, wq_entry: wait);
317
318 set_current_state(TASK_INTERRUPTIBLE);
319 if (signal_pending(current))
320 return -ERESTARTSYS;
321
322 spin_unlock(lock: &wq->lock);
323 schedule();
324 spin_lock(lock: &wq->lock);
325
326 return 0;
327}
328EXPORT_SYMBOL(do_wait_intr);
329
330int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
331{
332 if (likely(list_empty(&wait->entry)))
333 __add_wait_queue_entry_tail(wq_head: wq, wq_entry: wait);
334
335 set_current_state(TASK_INTERRUPTIBLE);
336 if (signal_pending(current))
337 return -ERESTARTSYS;
338
339 spin_unlock_irq(lock: &wq->lock);
340 schedule();
341 spin_lock_irq(lock: &wq->lock);
342
343 return 0;
344}
345EXPORT_SYMBOL(do_wait_intr_irq);
346
347/**
348 * finish_wait - clean up after waiting in a queue
349 * @wq_head: waitqueue waited on
350 * @wq_entry: wait descriptor
351 *
352 * Sets current thread back to running state and removes
353 * the wait descriptor from the given waitqueue if still
354 * queued.
355 */
356void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
357{
358 unsigned long flags;
359
360 __set_current_state(TASK_RUNNING);
361 /*
362 * We can check for list emptiness outside the lock
363 * IFF:
364 * - we use the "careful" check that verifies both
365 * the next and prev pointers, so that there cannot
366 * be any half-pending updates in progress on other
367 * CPU's that we haven't seen yet (and that might
368 * still change the stack area.
369 * and
370 * - all other users take the lock (ie we can only
371 * have _one_ other CPU that looks at or modifies
372 * the list).
373 */
374 if (!list_empty_careful(head: &wq_entry->entry)) {
375 spin_lock_irqsave(&wq_head->lock, flags);
376 list_del_init(entry: &wq_entry->entry);
377 spin_unlock_irqrestore(lock: &wq_head->lock, flags);
378 }
379}
380EXPORT_SYMBOL(finish_wait);
381
382int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
383{
384 int ret = default_wake_function(wq_entry, mode, flags: sync, key);
385
386 if (ret)
387 list_del_init_careful(entry: &wq_entry->entry);
388
389 return ret;
390}
391EXPORT_SYMBOL(autoremove_wake_function);
392
393/*
394 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
395 *
396 * add_wait_queue(&wq_head, &wait);
397 * for (;;) {
398 * if (condition)
399 * break;
400 *
401 * // in wait_woken() // in woken_wake_function()
402 *
403 * p->state = mode; wq_entry->flags |= WQ_FLAG_WOKEN;
404 * smp_mb(); // A try_to_wake_up():
405 * if (!(wq_entry->flags & WQ_FLAG_WOKEN)) <full barrier>
406 * schedule() if (p->state & mode)
407 * p->state = TASK_RUNNING; p->state = TASK_RUNNING;
408 * wq_entry->flags &= ~WQ_FLAG_WOKEN; ~~~~~~~~~~~~~~~~~~
409 * smp_mb(); // B condition = true;
410 * } smp_mb(); // C
411 * remove_wait_queue(&wq_head, &wait); wq_entry->flags |= WQ_FLAG_WOKEN;
412 */
413long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
414{
415 /*
416 * The below executes an smp_mb(), which matches with the full barrier
417 * executed by the try_to_wake_up() in woken_wake_function() such that
418 * either we see the store to wq_entry->flags in woken_wake_function()
419 * or woken_wake_function() sees our store to current->state.
420 */
421 set_current_state(mode); /* A */
422 if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !kthread_should_stop_or_park())
423 timeout = schedule_timeout(timeout);
424 __set_current_state(TASK_RUNNING);
425
426 /*
427 * The below executes an smp_mb(), which matches with the smp_mb() (C)
428 * in woken_wake_function() such that either we see the wait condition
429 * being true or the store to wq_entry->flags in woken_wake_function()
430 * follows ours in the coherence order.
431 */
432 smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
433
434 return timeout;
435}
436EXPORT_SYMBOL(wait_woken);
437
438int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
439{
440 /* Pairs with the smp_store_mb() in wait_woken(). */
441 smp_mb(); /* C */
442 wq_entry->flags |= WQ_FLAG_WOKEN;
443
444 return default_wake_function(wq_entry, mode, flags: sync, key);
445}
446EXPORT_SYMBOL(woken_wake_function);
447

source code of linux/kernel/sched/wait.c