• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Generic waiting primitives.
4  *
5  * (C) 2004 Nadia Yvette Chambers, Oracle
6  */
7 #include <trace/hooks/sched.h>
8 
__init_waitqueue_head(struct wait_queue_head * wq_head,const char * name,struct lock_class_key * key)9 void __init_waitqueue_head(struct wait_queue_head *wq_head, const char *name, struct lock_class_key *key)
10 {
11 	spin_lock_init(&wq_head->lock);
12 	lockdep_set_class_and_name(&wq_head->lock, key, name);
13 	INIT_LIST_HEAD(&wq_head->head);
14 }
15 
16 EXPORT_SYMBOL(__init_waitqueue_head);
17 
add_wait_queue(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)18 void add_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
19 {
20 	unsigned long flags;
21 
22 	wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
23 	spin_lock_irqsave(&wq_head->lock, flags);
24 	__add_wait_queue(wq_head, wq_entry);
25 	spin_unlock_irqrestore(&wq_head->lock, flags);
26 }
27 EXPORT_SYMBOL(add_wait_queue);
28 
add_wait_queue_exclusive(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)29 void add_wait_queue_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
30 {
31 	unsigned long flags;
32 
33 	wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
34 	spin_lock_irqsave(&wq_head->lock, flags);
35 	__add_wait_queue_entry_tail(wq_head, wq_entry);
36 	spin_unlock_irqrestore(&wq_head->lock, flags);
37 }
38 EXPORT_SYMBOL(add_wait_queue_exclusive);
39 
add_wait_queue_priority(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)40 void add_wait_queue_priority(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
41 {
42 	unsigned long flags;
43 
44 	wq_entry->flags |= WQ_FLAG_EXCLUSIVE | WQ_FLAG_PRIORITY;
45 	spin_lock_irqsave(&wq_head->lock, flags);
46 	__add_wait_queue(wq_head, wq_entry);
47 	spin_unlock_irqrestore(&wq_head->lock, flags);
48 }
49 EXPORT_SYMBOL_GPL(add_wait_queue_priority);
50 
remove_wait_queue(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)51 void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
52 {
53 	unsigned long flags;
54 
55 	spin_lock_irqsave(&wq_head->lock, flags);
56 	__remove_wait_queue(wq_head, wq_entry);
57 	spin_unlock_irqrestore(&wq_head->lock, flags);
58 }
59 EXPORT_SYMBOL(remove_wait_queue);
60 
61 /*
62  * Scan threshold to break wait queue walk.
63  * This allows a waker to take a break from holding the
64  * wait queue lock during the wait queue walk.
65  */
66 #define WAITQUEUE_WALK_BREAK_CNT 64
67 
68 /*
69  * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
70  * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
71  * number) then we wake that number of exclusive tasks, and potentially all
72  * the non-exclusive tasks. Normally, exclusive tasks will be at the end of
73  * the list and any non-exclusive tasks will be woken first. A priority task
74  * may be at the head of the list, and can consume the event without any other
75  * tasks being woken.
76  *
77  * There are circumstances in which we can try to wake a task which has already
78  * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
79  * zero in this (rare) case, and we handle it by continuing to scan the queue.
80  */
__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)81 static int __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
82 			int nr_exclusive, int wake_flags, void *key,
83 			wait_queue_entry_t *bookmark)
84 {
85 	wait_queue_entry_t *curr, *next;
86 	int cnt = 0;
87 
88 	lockdep_assert_held(&wq_head->lock);
89 
90 	if (bookmark && (bookmark->flags & WQ_FLAG_BOOKMARK)) {
91 		curr = list_next_entry(bookmark, entry);
92 
93 		list_del(&bookmark->entry);
94 		bookmark->flags = 0;
95 	} else
96 		curr = list_first_entry(&wq_head->head, wait_queue_entry_t, entry);
97 
98 	if (&curr->entry == &wq_head->head)
99 		return nr_exclusive;
100 
101 	list_for_each_entry_safe_from(curr, next, &wq_head->head, entry) {
102 		unsigned flags = curr->flags;
103 		int ret;
104 
105 		if (flags & WQ_FLAG_BOOKMARK)
106 			continue;
107 
108 		ret = curr->func(curr, mode, wake_flags, key);
109 		if (ret < 0)
110 			break;
111 		if (ret && (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
112 			break;
113 
114 		if (bookmark && (++cnt > WAITQUEUE_WALK_BREAK_CNT) &&
115 				(&next->entry != &wq_head->head)) {
116 			bookmark->flags = WQ_FLAG_BOOKMARK;
117 			list_add_tail(&bookmark->entry, &next->entry);
118 			break;
119 		}
120 	}
121 
122 	return nr_exclusive;
123 }
124 
__wake_up_common_lock(struct wait_queue_head * wq_head,unsigned int mode,int nr_exclusive,int wake_flags,void * key)125 static int __wake_up_common_lock(struct wait_queue_head *wq_head, unsigned int mode,
126 			int nr_exclusive, int wake_flags, void *key)
127 {
128 	unsigned long flags;
129 	wait_queue_entry_t bookmark;
130 	int remaining = nr_exclusive;
131 
132 	bookmark.flags = 0;
133 	bookmark.private = NULL;
134 	bookmark.func = NULL;
135 	INIT_LIST_HEAD(&bookmark.entry);
136 
137 	do {
138 		spin_lock_irqsave(&wq_head->lock, flags);
139 		remaining = __wake_up_common(wq_head, mode, remaining,
140 						wake_flags, key, &bookmark);
141 		spin_unlock_irqrestore(&wq_head->lock, flags);
142 	} while (bookmark.flags & WQ_FLAG_BOOKMARK);
143 
144 	return nr_exclusive - remaining;
145 }
146 
147 /**
148  * __wake_up - wake up threads blocked on a waitqueue.
149  * @wq_head: the waitqueue
150  * @mode: which threads
151  * @nr_exclusive: how many wake-one or wake-many threads to wake up
152  * @key: is directly passed to the wakeup function
153  *
154  * If this function wakes up a task, it executes a full memory barrier
155  * before accessing the task state.  Returns the number of exclusive
156  * tasks that were awaken.
157  */
__wake_up(struct wait_queue_head * wq_head,unsigned int mode,int nr_exclusive,void * key)158 int __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
159 	      int nr_exclusive, void *key)
160 {
161 	return __wake_up_common_lock(wq_head, mode, nr_exclusive, 0, key);
162 }
163 EXPORT_SYMBOL(__wake_up);
164 
165 /*
166  * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
167  */
__wake_up_locked(struct wait_queue_head * wq_head,unsigned int mode,int nr)168 void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
169 {
170 	__wake_up_common(wq_head, mode, nr, 0, NULL, NULL);
171 }
172 EXPORT_SYMBOL_GPL(__wake_up_locked);
173 
__wake_up_locked_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)174 void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
175 {
176 	__wake_up_common(wq_head, mode, 1, 0, key, NULL);
177 }
178 EXPORT_SYMBOL_GPL(__wake_up_locked_key);
179 
__wake_up_locked_key_bookmark(struct wait_queue_head * wq_head,unsigned int mode,void * key,wait_queue_entry_t * bookmark)180 void __wake_up_locked_key_bookmark(struct wait_queue_head *wq_head,
181 		unsigned int mode, void *key, wait_queue_entry_t *bookmark)
182 {
183 	__wake_up_common(wq_head, mode, 1, 0, key, bookmark);
184 }
185 EXPORT_SYMBOL_GPL(__wake_up_locked_key_bookmark);
186 
187 /**
188  * __wake_up_sync_key - wake up threads blocked on a waitqueue.
189  * @wq_head: the waitqueue
190  * @mode: which threads
191  * @key: opaque value to be passed to wakeup targets
192  *
193  * The sync wakeup differs that the waker knows that it will schedule
194  * away soon, so while the target thread will be woken up, it will not
195  * be migrated to another CPU - ie. the two threads are 'synchronized'
196  * with each other. This can prevent needless bouncing between CPUs.
197  *
198  * On UP it can prevent extra preemption.
199  *
200  * If this function wakes up a task, it executes a full memory barrier before
201  * accessing the task state.
202  */
__wake_up_sync_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)203 void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
204 			void *key)
205 {
206 	int wake_flags = WF_SYNC;
207 
208 	if (unlikely(!wq_head))
209 		return;
210 
211 	trace_android_vh_set_wake_flags(&wake_flags, &mode);
212 	__wake_up_common_lock(wq_head, mode, 1, wake_flags, key);
213 }
214 EXPORT_SYMBOL_GPL(__wake_up_sync_key);
215 
216 /**
217  * __wake_up_locked_sync_key - wake up a thread blocked on a locked waitqueue.
218  * @wq_head: the waitqueue
219  * @mode: which threads
220  * @key: opaque value to be passed to wakeup targets
221  *
222  * The sync wakeup differs in that the waker knows that it will schedule
223  * away soon, so while the target thread will be woken up, it will not
224  * be migrated to another CPU - ie. the two threads are 'synchronized'
225  * with each other. This can prevent needless bouncing between CPUs.
226  *
227  * On UP it can prevent extra preemption.
228  *
229  * If this function wakes up a task, it executes a full memory barrier before
230  * accessing the task state.
231  */
__wake_up_locked_sync_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)232 void __wake_up_locked_sync_key(struct wait_queue_head *wq_head,
233 			       unsigned int mode, void *key)
234 {
235         __wake_up_common(wq_head, mode, 1, WF_SYNC, key, NULL);
236 }
237 EXPORT_SYMBOL_GPL(__wake_up_locked_sync_key);
238 
239 /*
240  * __wake_up_sync - see __wake_up_sync_key()
241  */
__wake_up_sync(struct wait_queue_head * wq_head,unsigned int mode)242 void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode)
243 {
244 	__wake_up_sync_key(wq_head, mode, NULL);
245 }
246 EXPORT_SYMBOL_GPL(__wake_up_sync);	/* For internal use only */
247 
__wake_up_pollfree(struct wait_queue_head * wq_head)248 void __wake_up_pollfree(struct wait_queue_head *wq_head)
249 {
250 	__wake_up(wq_head, TASK_NORMAL, 0, poll_to_key(EPOLLHUP | POLLFREE));
251 	/* POLLFREE must have cleared the queue. */
252 	WARN_ON_ONCE(waitqueue_active(wq_head));
253 }
254 
255 /*
256  * Note: we use "set_current_state()" _after_ the wait-queue add,
257  * because we need a memory barrier there on SMP, so that any
258  * wake-function that tests for the wait-queue being active
259  * will be guaranteed to see waitqueue addition _or_ subsequent
260  * tests in this thread will see the wakeup having taken place.
261  *
262  * The spin_unlock() itself is semi-permeable and only protects
263  * one way (it only protects stuff inside the critical region and
264  * stops them from bleeding out - it would still allow subsequent
265  * loads to move into the critical region).
266  */
267 void
prepare_to_wait(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)268 prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
269 {
270 	unsigned long flags;
271 
272 	wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
273 	spin_lock_irqsave(&wq_head->lock, flags);
274 	if (list_empty(&wq_entry->entry))
275 		__add_wait_queue(wq_head, wq_entry);
276 	set_current_state(state);
277 	spin_unlock_irqrestore(&wq_head->lock, flags);
278 }
279 EXPORT_SYMBOL(prepare_to_wait);
280 
281 /* Returns true if we are the first waiter in the queue, false otherwise. */
282 bool
prepare_to_wait_exclusive(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)283 prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
284 {
285 	unsigned long flags;
286 	bool was_empty = false;
287 
288 	wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
289 	spin_lock_irqsave(&wq_head->lock, flags);
290 	if (list_empty(&wq_entry->entry)) {
291 		was_empty = list_empty(&wq_head->head);
292 		__add_wait_queue_entry_tail(wq_head, wq_entry);
293 	}
294 	set_current_state(state);
295 	spin_unlock_irqrestore(&wq_head->lock, flags);
296 	return was_empty;
297 }
298 EXPORT_SYMBOL(prepare_to_wait_exclusive);
299 
init_wait_entry(struct wait_queue_entry * wq_entry,int flags)300 void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
301 {
302 	wq_entry->flags = flags;
303 	wq_entry->private = current;
304 	wq_entry->func = autoremove_wake_function;
305 	INIT_LIST_HEAD(&wq_entry->entry);
306 }
307 EXPORT_SYMBOL(init_wait_entry);
308 
prepare_to_wait_event(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)309 long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
310 {
311 	unsigned long flags;
312 	long ret = 0;
313 
314 	spin_lock_irqsave(&wq_head->lock, flags);
315 	if (signal_pending_state(state, current)) {
316 		/*
317 		 * Exclusive waiter must not fail if it was selected by wakeup,
318 		 * it should "consume" the condition we were waiting for.
319 		 *
320 		 * The caller will recheck the condition and return success if
321 		 * we were already woken up, we can not miss the event because
322 		 * wakeup locks/unlocks the same wq_head->lock.
323 		 *
324 		 * But we need to ensure that set-condition + wakeup after that
325 		 * can't see us, it should wake up another exclusive waiter if
326 		 * we fail.
327 		 */
328 		list_del_init(&wq_entry->entry);
329 		ret = -ERESTARTSYS;
330 	} else {
331 		if (list_empty(&wq_entry->entry)) {
332 			if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
333 				__add_wait_queue_entry_tail(wq_head, wq_entry);
334 			else
335 				__add_wait_queue(wq_head, wq_entry);
336 		}
337 		set_current_state(state);
338 	}
339 	spin_unlock_irqrestore(&wq_head->lock, flags);
340 
341 	return ret;
342 }
343 EXPORT_SYMBOL(prepare_to_wait_event);
344 
345 /*
346  * Note! These two wait functions are entered with the
347  * wait-queue lock held (and interrupts off in the _irq
348  * case), so there is no race with testing the wakeup
349  * condition in the caller before they add the wait
350  * entry to the wake queue.
351  */
do_wait_intr(wait_queue_head_t * wq,wait_queue_entry_t * wait)352 int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
353 {
354 	if (likely(list_empty(&wait->entry)))
355 		__add_wait_queue_entry_tail(wq, wait);
356 
357 	set_current_state(TASK_INTERRUPTIBLE);
358 	if (signal_pending(current))
359 		return -ERESTARTSYS;
360 
361 	spin_unlock(&wq->lock);
362 	schedule();
363 	spin_lock(&wq->lock);
364 
365 	return 0;
366 }
367 EXPORT_SYMBOL(do_wait_intr);
368 
do_wait_intr_irq(wait_queue_head_t * wq,wait_queue_entry_t * wait)369 int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
370 {
371 	if (likely(list_empty(&wait->entry)))
372 		__add_wait_queue_entry_tail(wq, wait);
373 
374 	set_current_state(TASK_INTERRUPTIBLE);
375 	if (signal_pending(current))
376 		return -ERESTARTSYS;
377 
378 	spin_unlock_irq(&wq->lock);
379 	schedule();
380 	spin_lock_irq(&wq->lock);
381 
382 	return 0;
383 }
384 EXPORT_SYMBOL(do_wait_intr_irq);
385 
386 /**
387  * finish_wait - clean up after waiting in a queue
388  * @wq_head: waitqueue waited on
389  * @wq_entry: wait descriptor
390  *
391  * Sets current thread back to running state and removes
392  * the wait descriptor from the given waitqueue if still
393  * queued.
394  */
finish_wait(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)395 void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
396 {
397 	unsigned long flags;
398 
399 	__set_current_state(TASK_RUNNING);
400 	/*
401 	 * We can check for list emptiness outside the lock
402 	 * IFF:
403 	 *  - we use the "careful" check that verifies both
404 	 *    the next and prev pointers, so that there cannot
405 	 *    be any half-pending updates in progress on other
406 	 *    CPU's that we haven't seen yet (and that might
407 	 *    still change the stack area.
408 	 * and
409 	 *  - all other users take the lock (ie we can only
410 	 *    have _one_ other CPU that looks at or modifies
411 	 *    the list).
412 	 */
413 	if (!list_empty_careful(&wq_entry->entry)) {
414 		spin_lock_irqsave(&wq_head->lock, flags);
415 		list_del_init(&wq_entry->entry);
416 		spin_unlock_irqrestore(&wq_head->lock, flags);
417 	}
418 }
419 EXPORT_SYMBOL(finish_wait);
420 
autoremove_wake_function(struct wait_queue_entry * wq_entry,unsigned mode,int sync,void * key)421 int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
422 {
423 	int ret = default_wake_function(wq_entry, mode, sync, key);
424 
425 	if (ret)
426 		list_del_init_careful(&wq_entry->entry);
427 
428 	return ret;
429 }
430 EXPORT_SYMBOL(autoremove_wake_function);
431 
432 /*
433  * DEFINE_WAIT_FUNC(wait, woken_wake_func);
434  *
435  * add_wait_queue(&wq_head, &wait);
436  * for (;;) {
437  *     if (condition)
438  *         break;
439  *
440  *     // in wait_woken()			// in woken_wake_function()
441  *
442  *     p->state = mode;				wq_entry->flags |= WQ_FLAG_WOKEN;
443  *     smp_mb(); // A				try_to_wake_up():
444  *     if (!(wq_entry->flags & WQ_FLAG_WOKEN))	   <full barrier>
445  *         schedule()				   if (p->state & mode)
446  *     p->state = TASK_RUNNING;			      p->state = TASK_RUNNING;
447  *     wq_entry->flags &= ~WQ_FLAG_WOKEN;	~~~~~~~~~~~~~~~~~~
448  *     smp_mb(); // B				condition = true;
449  * }						smp_mb(); // C
450  * remove_wait_queue(&wq_head, &wait);		wq_entry->flags |= WQ_FLAG_WOKEN;
451  */
wait_woken(struct wait_queue_entry * wq_entry,unsigned mode,long timeout)452 long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
453 {
454 	/*
455 	 * The below executes an smp_mb(), which matches with the full barrier
456 	 * executed by the try_to_wake_up() in woken_wake_function() such that
457 	 * either we see the store to wq_entry->flags in woken_wake_function()
458 	 * or woken_wake_function() sees our store to current->state.
459 	 */
460 	set_current_state(mode); /* A */
461 	if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !kthread_should_stop_or_park())
462 		timeout = schedule_timeout(timeout);
463 	__set_current_state(TASK_RUNNING);
464 
465 	/*
466 	 * The below executes an smp_mb(), which matches with the smp_mb() (C)
467 	 * in woken_wake_function() such that either we see the wait condition
468 	 * being true or the store to wq_entry->flags in woken_wake_function()
469 	 * follows ours in the coherence order.
470 	 */
471 	smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
472 
473 	return timeout;
474 }
475 EXPORT_SYMBOL(wait_woken);
476 
woken_wake_function(struct wait_queue_entry * wq_entry,unsigned mode,int sync,void * key)477 int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
478 {
479 	/* Pairs with the smp_store_mb() in wait_woken(). */
480 	smp_mb(); /* C */
481 	wq_entry->flags |= WQ_FLAG_WOKEN;
482 
483 	return default_wake_function(wq_entry, mode, sync, key);
484 }
485 EXPORT_SYMBOL(woken_wake_function);
486