1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Generic waiting primitives.
4  *
5  * (C) 2004 Nadia Yvette Chambers, Oracle
6  */
7 
__init_waitqueue_head(struct wait_queue_head * wq_head,const char * name,struct lock_class_key * key)8 void __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(&wq_head->head);
13 }
14 
15 EXPORT_SYMBOL(__init_waitqueue_head);
16 
add_wait_queue(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)17 void 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(&wq_head->lock, flags);
25 }
26 EXPORT_SYMBOL(add_wait_queue);
27 
add_wait_queue_exclusive(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)28 void 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(&wq_head->lock, flags);
36 }
37 EXPORT_SYMBOL(add_wait_queue_exclusive);
38 
add_wait_queue_priority(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)39 void 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(&wq_head->lock, flags);
47 }
48 EXPORT_SYMBOL_GPL(add_wait_queue_priority);
49 
remove_wait_queue(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)50 void 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(&wq_head->lock, flags);
57 }
58 EXPORT_SYMBOL(remove_wait_queue);
59 
60 /*
61  * Scan threshold to break wait queue walk.
62  * This allows a waker to take a break from holding the
63  * wait queue lock during the wait queue walk.
64  */
65 #define WAITQUEUE_WALK_BREAK_CNT 64
66 
67 /*
68  * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
69  * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
70  * number) then we wake that number of exclusive tasks, and potentially all
71  * the non-exclusive tasks. Normally, exclusive tasks will be at the end of
72  * the list and any non-exclusive tasks will be woken first. A priority task
73  * may be at the head of the list, and can consume the event without any other
74  * tasks being woken.
75  *
76  * There are circumstances in which we can try to wake a task which has already
77  * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
78  * zero in this (rare) case, and we handle it by continuing to scan the queue.
79  */
__wake_up_common(struct wait_queue_head * wq_head,unsigned int mode,int nr_exclusive,int wake_flags,void * key,wait_queue_entry_t * bookmark)80 static int __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
81 			int nr_exclusive, int wake_flags, void *key,
82 			wait_queue_entry_t *bookmark)
83 {
84 	wait_queue_entry_t *curr, *next;
85 	int cnt = 0;
86 
87 	lockdep_assert_held(&wq_head->lock);
88 
89 	if (bookmark && (bookmark->flags & WQ_FLAG_BOOKMARK)) {
90 		curr = list_next_entry(bookmark, entry);
91 
92 		list_del(&bookmark->entry);
93 		bookmark->flags = 0;
94 	} else
95 		curr = list_first_entry(&wq_head->head, wait_queue_entry_t, entry);
96 
97 	if (&curr->entry == &wq_head->head)
98 		return nr_exclusive;
99 
100 	list_for_each_entry_safe_from(curr, next, &wq_head->head, entry) {
101 		unsigned flags = curr->flags;
102 		int ret;
103 
104 		if (flags & WQ_FLAG_BOOKMARK)
105 			continue;
106 
107 		ret = curr->func(curr, mode, wake_flags, key);
108 		if (ret < 0)
109 			break;
110 		if (ret && (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
111 			break;
112 
113 		if (bookmark && (++cnt > WAITQUEUE_WALK_BREAK_CNT) &&
114 				(&next->entry != &wq_head->head)) {
115 			bookmark->flags = WQ_FLAG_BOOKMARK;
116 			list_add_tail(&bookmark->entry, &next->entry);
117 			break;
118 		}
119 	}
120 
121 	return nr_exclusive;
122 }
123 
__wake_up_common_lock(struct wait_queue_head * wq_head,unsigned int mode,int nr_exclusive,int wake_flags,void * key)124 static int __wake_up_common_lock(struct wait_queue_head *wq_head, unsigned int mode,
125 			int nr_exclusive, int wake_flags, void *key)
126 {
127 	unsigned long flags;
128 	wait_queue_entry_t bookmark;
129 	int remaining = nr_exclusive;
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 		remaining = __wake_up_common(wq_head, mode, remaining,
139 						wake_flags, key, &bookmark);
140 		spin_unlock_irqrestore(&wq_head->lock, flags);
141 	} while (bookmark.flags & WQ_FLAG_BOOKMARK);
142 
143 	return nr_exclusive - remaining;
144 }
145 
146 /**
147  * __wake_up - wake up threads blocked on a waitqueue.
148  * @wq_head: the waitqueue
149  * @mode: which threads
150  * @nr_exclusive: how many wake-one or wake-many threads to wake up
151  * @key: is directly passed to the wakeup function
152  *
153  * If this function wakes up a task, it executes a full memory barrier
154  * before accessing the task state.  Returns the number of exclusive
155  * tasks that were awaken.
156  */
__wake_up(struct wait_queue_head * wq_head,unsigned int mode,int nr_exclusive,void * key)157 int __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
158 	      int nr_exclusive, void *key)
159 {
160 	return __wake_up_common_lock(wq_head, mode, nr_exclusive, 0, key);
161 }
162 EXPORT_SYMBOL(__wake_up);
163 
__wake_up_on_current_cpu(struct wait_queue_head * wq_head,unsigned int mode,void * key)164 void __wake_up_on_current_cpu(struct wait_queue_head *wq_head, unsigned int mode, void *key)
165 {
166 	__wake_up_common_lock(wq_head, mode, 1, WF_CURRENT_CPU, key);
167 }
168 
169 /*
170  * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
171  */
__wake_up_locked(struct wait_queue_head * wq_head,unsigned int mode,int nr)172 void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
173 {
174 	__wake_up_common(wq_head, mode, nr, 0, NULL, NULL);
175 }
176 EXPORT_SYMBOL_GPL(__wake_up_locked);
177 
__wake_up_locked_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)178 void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
179 {
180 	__wake_up_common(wq_head, mode, 1, 0, key, NULL);
181 }
182 EXPORT_SYMBOL_GPL(__wake_up_locked_key);
183 
__wake_up_locked_key_bookmark(struct wait_queue_head * wq_head,unsigned int mode,void * key,wait_queue_entry_t * bookmark)184 void __wake_up_locked_key_bookmark(struct wait_queue_head *wq_head,
185 		unsigned int mode, void *key, wait_queue_entry_t *bookmark)
186 {
187 	__wake_up_common(wq_head, mode, 1, 0, key, bookmark);
188 }
189 EXPORT_SYMBOL_GPL(__wake_up_locked_key_bookmark);
190 
191 /**
192  * __wake_up_sync_key - wake up threads blocked on a waitqueue.
193  * @wq_head: the waitqueue
194  * @mode: which threads
195  * @key: opaque value to be passed to wakeup targets
196  *
197  * The sync wakeup differs that the waker knows that it will schedule
198  * away soon, so while the target thread will be woken up, it will not
199  * be migrated to another CPU - ie. the two threads are 'synchronized'
200  * with each other. This can prevent needless bouncing between CPUs.
201  *
202  * On UP it can prevent extra preemption.
203  *
204  * If this function wakes up a task, it executes a full memory barrier before
205  * accessing the task state.
206  */
__wake_up_sync_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)207 void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
208 			void *key)
209 {
210 	if (unlikely(!wq_head))
211 		return;
212 
213 	__wake_up_common_lock(wq_head, mode, 1, WF_SYNC, key);
214 }
215 EXPORT_SYMBOL_GPL(__wake_up_sync_key);
216 
217 /**
218  * __wake_up_locked_sync_key - wake up a thread blocked on a locked waitqueue.
219  * @wq_head: the waitqueue
220  * @mode: which threads
221  * @key: opaque value to be passed to wakeup targets
222  *
223  * The sync wakeup differs in that the waker knows that it will schedule
224  * away soon, so while the target thread will be woken up, it will not
225  * be migrated to another CPU - ie. the two threads are 'synchronized'
226  * with each other. This can prevent needless bouncing between CPUs.
227  *
228  * On UP it can prevent extra preemption.
229  *
230  * If this function wakes up a task, it executes a full memory barrier before
231  * accessing the task state.
232  */
__wake_up_locked_sync_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)233 void __wake_up_locked_sync_key(struct wait_queue_head *wq_head,
234 			       unsigned int mode, void *key)
235 {
236         __wake_up_common(wq_head, mode, 1, WF_SYNC, key, NULL);
237 }
238 EXPORT_SYMBOL_GPL(__wake_up_locked_sync_key);
239 
240 /*
241  * __wake_up_sync - see __wake_up_sync_key()
242  */
__wake_up_sync(struct wait_queue_head * wq_head,unsigned int mode)243 void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode)
244 {
245 	__wake_up_sync_key(wq_head, mode, NULL);
246 }
247 EXPORT_SYMBOL_GPL(__wake_up_sync);	/* For internal use only */
248 
__wake_up_pollfree(struct wait_queue_head * wq_head)249 void __wake_up_pollfree(struct wait_queue_head *wq_head)
250 {
251 	__wake_up(wq_head, TASK_NORMAL, 0, poll_to_key(EPOLLHUP | POLLFREE));
252 	/* POLLFREE must have cleared the queue. */
253 	WARN_ON_ONCE(waitqueue_active(wq_head));
254 }
255 
256 /*
257  * Note: we use "set_current_state()" _after_ the wait-queue add,
258  * because we need a memory barrier there on SMP, so that any
259  * wake-function that tests for the wait-queue being active
260  * will be guaranteed to see waitqueue addition _or_ subsequent
261  * tests in this thread will see the wakeup having taken place.
262  *
263  * The spin_unlock() itself is semi-permeable and only protects
264  * one way (it only protects stuff inside the critical region and
265  * stops them from bleeding out - it would still allow subsequent
266  * loads to move into the critical region).
267  */
268 void
prepare_to_wait(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)269 prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
270 {
271 	unsigned long flags;
272 
273 	wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
274 	spin_lock_irqsave(&wq_head->lock, flags);
275 	if (list_empty(&wq_entry->entry))
276 		__add_wait_queue(wq_head, wq_entry);
277 	set_current_state(state);
278 	spin_unlock_irqrestore(&wq_head->lock, flags);
279 }
280 EXPORT_SYMBOL(prepare_to_wait);
281 
282 /* Returns true if we are the first waiter in the queue, false otherwise. */
283 bool
prepare_to_wait_exclusive(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)284 prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
285 {
286 	unsigned long flags;
287 	bool was_empty = false;
288 
289 	wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
290 	spin_lock_irqsave(&wq_head->lock, flags);
291 	if (list_empty(&wq_entry->entry)) {
292 		was_empty = list_empty(&wq_head->head);
293 		__add_wait_queue_entry_tail(wq_head, wq_entry);
294 	}
295 	set_current_state(state);
296 	spin_unlock_irqrestore(&wq_head->lock, flags);
297 	return was_empty;
298 }
299 EXPORT_SYMBOL(prepare_to_wait_exclusive);
300 
init_wait_entry(struct wait_queue_entry * wq_entry,int flags)301 void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
302 {
303 	wq_entry->flags = flags;
304 	wq_entry->private = current;
305 	wq_entry->func = autoremove_wake_function;
306 	INIT_LIST_HEAD(&wq_entry->entry);
307 }
308 EXPORT_SYMBOL(init_wait_entry);
309 
prepare_to_wait_event(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)310 long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
311 {
312 	unsigned long flags;
313 	long ret = 0;
314 
315 	spin_lock_irqsave(&wq_head->lock, flags);
316 	if (signal_pending_state(state, current)) {
317 		/*
318 		 * Exclusive waiter must not fail if it was selected by wakeup,
319 		 * it should "consume" the condition we were waiting for.
320 		 *
321 		 * The caller will recheck the condition and return success if
322 		 * we were already woken up, we can not miss the event because
323 		 * wakeup locks/unlocks the same wq_head->lock.
324 		 *
325 		 * But we need to ensure that set-condition + wakeup after that
326 		 * can't see us, it should wake up another exclusive waiter if
327 		 * we fail.
328 		 */
329 		list_del_init(&wq_entry->entry);
330 		ret = -ERESTARTSYS;
331 	} else {
332 		if (list_empty(&wq_entry->entry)) {
333 			if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
334 				__add_wait_queue_entry_tail(wq_head, wq_entry);
335 			else
336 				__add_wait_queue(wq_head, wq_entry);
337 		}
338 		set_current_state(state);
339 	}
340 	spin_unlock_irqrestore(&wq_head->lock, flags);
341 
342 	return ret;
343 }
344 EXPORT_SYMBOL(prepare_to_wait_event);
345 
346 /*
347  * Note! These two wait functions are entered with the
348  * wait-queue lock held (and interrupts off in the _irq
349  * case), so there is no race with testing the wakeup
350  * condition in the caller before they add the wait
351  * entry to the wake queue.
352  */
do_wait_intr(wait_queue_head_t * wq,wait_queue_entry_t * wait)353 int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
354 {
355 	if (likely(list_empty(&wait->entry)))
356 		__add_wait_queue_entry_tail(wq, wait);
357 
358 	set_current_state(TASK_INTERRUPTIBLE);
359 	if (signal_pending(current))
360 		return -ERESTARTSYS;
361 
362 	spin_unlock(&wq->lock);
363 	schedule();
364 	spin_lock(&wq->lock);
365 
366 	return 0;
367 }
368 EXPORT_SYMBOL(do_wait_intr);
369 
do_wait_intr_irq(wait_queue_head_t * wq,wait_queue_entry_t * wait)370 int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
371 {
372 	if (likely(list_empty(&wait->entry)))
373 		__add_wait_queue_entry_tail(wq, wait);
374 
375 	set_current_state(TASK_INTERRUPTIBLE);
376 	if (signal_pending(current))
377 		return -ERESTARTSYS;
378 
379 	spin_unlock_irq(&wq->lock);
380 	schedule();
381 	spin_lock_irq(&wq->lock);
382 
383 	return 0;
384 }
385 EXPORT_SYMBOL(do_wait_intr_irq);
386 
387 /**
388  * finish_wait - clean up after waiting in a queue
389  * @wq_head: waitqueue waited on
390  * @wq_entry: wait descriptor
391  *
392  * Sets current thread back to running state and removes
393  * the wait descriptor from the given waitqueue if still
394  * queued.
395  */
finish_wait(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)396 void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
397 {
398 	unsigned long flags;
399 
400 	__set_current_state(TASK_RUNNING);
401 	/*
402 	 * We can check for list emptiness outside the lock
403 	 * IFF:
404 	 *  - we use the "careful" check that verifies both
405 	 *    the next and prev pointers, so that there cannot
406 	 *    be any half-pending updates in progress on other
407 	 *    CPU's that we haven't seen yet (and that might
408 	 *    still change the stack area.
409 	 * and
410 	 *  - all other users take the lock (ie we can only
411 	 *    have _one_ other CPU that looks at or modifies
412 	 *    the list).
413 	 */
414 	if (!list_empty_careful(&wq_entry->entry)) {
415 		spin_lock_irqsave(&wq_head->lock, flags);
416 		list_del_init(&wq_entry->entry);
417 		spin_unlock_irqrestore(&wq_head->lock, flags);
418 	}
419 }
420 EXPORT_SYMBOL(finish_wait);
421 
autoremove_wake_function(struct wait_queue_entry * wq_entry,unsigned mode,int sync,void * key)422 int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
423 {
424 	int ret = default_wake_function(wq_entry, mode, sync, key);
425 
426 	if (ret)
427 		list_del_init_careful(&wq_entry->entry);
428 
429 	return ret;
430 }
431 EXPORT_SYMBOL(autoremove_wake_function);
432 
433 /*
434  * DEFINE_WAIT_FUNC(wait, woken_wake_func);
435  *
436  * add_wait_queue(&wq_head, &wait);
437  * for (;;) {
438  *     if (condition)
439  *         break;
440  *
441  *     // in wait_woken()			// in woken_wake_function()
442  *
443  *     p->state = mode;				wq_entry->flags |= WQ_FLAG_WOKEN;
444  *     smp_mb(); // A				try_to_wake_up():
445  *     if (!(wq_entry->flags & WQ_FLAG_WOKEN))	   <full barrier>
446  *         schedule()				   if (p->state & mode)
447  *     p->state = TASK_RUNNING;			      p->state = TASK_RUNNING;
448  *     wq_entry->flags &= ~WQ_FLAG_WOKEN;	~~~~~~~~~~~~~~~~~~
449  *     smp_mb(); // B				condition = true;
450  * }						smp_mb(); // C
451  * remove_wait_queue(&wq_head, &wait);		wq_entry->flags |= WQ_FLAG_WOKEN;
452  */
wait_woken(struct wait_queue_entry * wq_entry,unsigned mode,long timeout)453 long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
454 {
455 	/*
456 	 * The below executes an smp_mb(), which matches with the full barrier
457 	 * executed by the try_to_wake_up() in woken_wake_function() such that
458 	 * either we see the store to wq_entry->flags in woken_wake_function()
459 	 * or woken_wake_function() sees our store to current->state.
460 	 */
461 	set_current_state(mode); /* A */
462 	if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !kthread_should_stop_or_park())
463 		timeout = schedule_timeout(timeout);
464 	__set_current_state(TASK_RUNNING);
465 
466 	/*
467 	 * The below executes an smp_mb(), which matches with the smp_mb() (C)
468 	 * in woken_wake_function() such that either we see the wait condition
469 	 * being true or the store to wq_entry->flags in woken_wake_function()
470 	 * follows ours in the coherence order.
471 	 */
472 	smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
473 
474 	return timeout;
475 }
476 EXPORT_SYMBOL(wait_woken);
477 
woken_wake_function(struct wait_queue_entry * wq_entry,unsigned mode,int sync,void * key)478 int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
479 {
480 	/* Pairs with the smp_store_mb() in wait_woken(). */
481 	smp_mb(); /* C */
482 	wq_entry->flags |= WQ_FLAG_WOKEN;
483 
484 	return default_wake_function(wq_entry, mode, sync, key);
485 }
486 EXPORT_SYMBOL(woken_wake_function);
487