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
__wake_up_on_current_cpu(struct wait_queue_head * wq_head,unsigned int mode,void * key)165 void __wake_up_on_current_cpu(struct wait_queue_head *wq_head, unsigned int mode, void *key)
166 {
167 __wake_up_common_lock(wq_head, mode, 1, WF_CURRENT_CPU, key);
168 }
169
170 /*
171 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
172 */
__wake_up_locked(struct wait_queue_head * wq_head,unsigned int mode,int nr)173 void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
174 {
175 __wake_up_common(wq_head, mode, nr, 0, NULL, NULL);
176 }
177 EXPORT_SYMBOL_GPL(__wake_up_locked);
178
__wake_up_locked_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)179 void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
180 {
181 __wake_up_common(wq_head, mode, 1, 0, key, NULL);
182 }
183 EXPORT_SYMBOL_GPL(__wake_up_locked_key);
184
__wake_up_locked_key_bookmark(struct wait_queue_head * wq_head,unsigned int mode,void * key,wait_queue_entry_t * bookmark)185 void __wake_up_locked_key_bookmark(struct wait_queue_head *wq_head,
186 unsigned int mode, void *key, wait_queue_entry_t *bookmark)
187 {
188 __wake_up_common(wq_head, mode, 1, 0, key, bookmark);
189 }
190 EXPORT_SYMBOL_GPL(__wake_up_locked_key_bookmark);
191
192 /**
193 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
194 * @wq_head: the waitqueue
195 * @mode: which threads
196 * @key: opaque value to be passed to wakeup targets
197 *
198 * The sync wakeup differs that the waker knows that it will schedule
199 * away soon, so while the target thread will be woken up, it will not
200 * be migrated to another CPU - ie. the two threads are 'synchronized'
201 * with each other. This can prevent needless bouncing between CPUs.
202 *
203 * On UP it can prevent extra preemption.
204 *
205 * If this function wakes up a task, it executes a full memory barrier before
206 * accessing the task state.
207 */
__wake_up_sync_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)208 void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
209 void *key)
210 {
211 int wake_flags = WF_SYNC;
212
213 if (unlikely(!wq_head))
214 return;
215
216 trace_android_vh_set_wake_flags(&wake_flags, &mode);
217 __wake_up_common_lock(wq_head, mode, 1, wake_flags, key);
218 }
219 EXPORT_SYMBOL_GPL(__wake_up_sync_key);
220
221 /**
222 * __wake_up_locked_sync_key - wake up a thread blocked on a locked waitqueue.
223 * @wq_head: the waitqueue
224 * @mode: which threads
225 * @key: opaque value to be passed to wakeup targets
226 *
227 * The sync wakeup differs in that the waker knows that it will schedule
228 * away soon, so while the target thread will be woken up, it will not
229 * be migrated to another CPU - ie. the two threads are 'synchronized'
230 * with each other. This can prevent needless bouncing between CPUs.
231 *
232 * On UP it can prevent extra preemption.
233 *
234 * If this function wakes up a task, it executes a full memory barrier before
235 * accessing the task state.
236 */
__wake_up_locked_sync_key(struct wait_queue_head * wq_head,unsigned int mode,void * key)237 void __wake_up_locked_sync_key(struct wait_queue_head *wq_head,
238 unsigned int mode, void *key)
239 {
240 __wake_up_common(wq_head, mode, 1, WF_SYNC, key, NULL);
241 }
242 EXPORT_SYMBOL_GPL(__wake_up_locked_sync_key);
243
244 /*
245 * __wake_up_sync - see __wake_up_sync_key()
246 */
__wake_up_sync(struct wait_queue_head * wq_head,unsigned int mode)247 void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode)
248 {
249 __wake_up_sync_key(wq_head, mode, NULL);
250 }
251 EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
252
__wake_up_pollfree(struct wait_queue_head * wq_head)253 void __wake_up_pollfree(struct wait_queue_head *wq_head)
254 {
255 __wake_up(wq_head, TASK_NORMAL, 0, poll_to_key(EPOLLHUP | POLLFREE));
256 /* POLLFREE must have cleared the queue. */
257 WARN_ON_ONCE(waitqueue_active(wq_head));
258 }
259
260 /*
261 * Note: we use "set_current_state()" _after_ the wait-queue add,
262 * because we need a memory barrier there on SMP, so that any
263 * wake-function that tests for the wait-queue being active
264 * will be guaranteed to see waitqueue addition _or_ subsequent
265 * tests in this thread will see the wakeup having taken place.
266 *
267 * The spin_unlock() itself is semi-permeable and only protects
268 * one way (it only protects stuff inside the critical region and
269 * stops them from bleeding out - it would still allow subsequent
270 * loads to move into the critical region).
271 */
272 void
prepare_to_wait(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)273 prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
274 {
275 unsigned long flags;
276
277 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
278 spin_lock_irqsave(&wq_head->lock, flags);
279 if (list_empty(&wq_entry->entry))
280 __add_wait_queue(wq_head, wq_entry);
281 set_current_state(state);
282 spin_unlock_irqrestore(&wq_head->lock, flags);
283 }
284 EXPORT_SYMBOL(prepare_to_wait);
285
286 /* Returns true if we are the first waiter in the queue, false otherwise. */
287 bool
prepare_to_wait_exclusive(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)288 prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
289 {
290 unsigned long flags;
291 bool was_empty = false;
292
293 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
294 spin_lock_irqsave(&wq_head->lock, flags);
295 if (list_empty(&wq_entry->entry)) {
296 was_empty = list_empty(&wq_head->head);
297 __add_wait_queue_entry_tail(wq_head, wq_entry);
298 }
299 set_current_state(state);
300 spin_unlock_irqrestore(&wq_head->lock, flags);
301 return was_empty;
302 }
303 EXPORT_SYMBOL(prepare_to_wait_exclusive);
304
init_wait_entry(struct wait_queue_entry * wq_entry,int flags)305 void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
306 {
307 wq_entry->flags = flags;
308 wq_entry->private = current;
309 wq_entry->func = autoremove_wake_function;
310 INIT_LIST_HEAD(&wq_entry->entry);
311 }
312 EXPORT_SYMBOL(init_wait_entry);
313
prepare_to_wait_event(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry,int state)314 long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
315 {
316 unsigned long flags;
317 long ret = 0;
318
319 spin_lock_irqsave(&wq_head->lock, flags);
320 if (signal_pending_state(state, current)) {
321 /*
322 * Exclusive waiter must not fail if it was selected by wakeup,
323 * it should "consume" the condition we were waiting for.
324 *
325 * The caller will recheck the condition and return success if
326 * we were already woken up, we can not miss the event because
327 * wakeup locks/unlocks the same wq_head->lock.
328 *
329 * But we need to ensure that set-condition + wakeup after that
330 * can't see us, it should wake up another exclusive waiter if
331 * we fail.
332 */
333 list_del_init(&wq_entry->entry);
334 ret = -ERESTARTSYS;
335 } else {
336 if (list_empty(&wq_entry->entry)) {
337 if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
338 __add_wait_queue_entry_tail(wq_head, wq_entry);
339 else
340 __add_wait_queue(wq_head, wq_entry);
341 }
342 set_current_state(state);
343 }
344 spin_unlock_irqrestore(&wq_head->lock, flags);
345
346 return ret;
347 }
348 EXPORT_SYMBOL(prepare_to_wait_event);
349
350 /*
351 * Note! These two wait functions are entered with the
352 * wait-queue lock held (and interrupts off in the _irq
353 * case), so there is no race with testing the wakeup
354 * condition in the caller before they add the wait
355 * entry to the wake queue.
356 */
do_wait_intr(wait_queue_head_t * wq,wait_queue_entry_t * wait)357 int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
358 {
359 if (likely(list_empty(&wait->entry)))
360 __add_wait_queue_entry_tail(wq, wait);
361
362 set_current_state(TASK_INTERRUPTIBLE);
363 if (signal_pending(current))
364 return -ERESTARTSYS;
365
366 spin_unlock(&wq->lock);
367 schedule();
368 spin_lock(&wq->lock);
369
370 return 0;
371 }
372 EXPORT_SYMBOL(do_wait_intr);
373
do_wait_intr_irq(wait_queue_head_t * wq,wait_queue_entry_t * wait)374 int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
375 {
376 if (likely(list_empty(&wait->entry)))
377 __add_wait_queue_entry_tail(wq, wait);
378
379 set_current_state(TASK_INTERRUPTIBLE);
380 if (signal_pending(current))
381 return -ERESTARTSYS;
382
383 spin_unlock_irq(&wq->lock);
384 schedule();
385 spin_lock_irq(&wq->lock);
386
387 return 0;
388 }
389 EXPORT_SYMBOL(do_wait_intr_irq);
390
391 /**
392 * finish_wait - clean up after waiting in a queue
393 * @wq_head: waitqueue waited on
394 * @wq_entry: wait descriptor
395 *
396 * Sets current thread back to running state and removes
397 * the wait descriptor from the given waitqueue if still
398 * queued.
399 */
finish_wait(struct wait_queue_head * wq_head,struct wait_queue_entry * wq_entry)400 void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
401 {
402 unsigned long flags;
403
404 __set_current_state(TASK_RUNNING);
405 /*
406 * We can check for list emptiness outside the lock
407 * IFF:
408 * - we use the "careful" check that verifies both
409 * the next and prev pointers, so that there cannot
410 * be any half-pending updates in progress on other
411 * CPU's that we haven't seen yet (and that might
412 * still change the stack area.
413 * and
414 * - all other users take the lock (ie we can only
415 * have _one_ other CPU that looks at or modifies
416 * the list).
417 */
418 if (!list_empty_careful(&wq_entry->entry)) {
419 spin_lock_irqsave(&wq_head->lock, flags);
420 list_del_init(&wq_entry->entry);
421 spin_unlock_irqrestore(&wq_head->lock, flags);
422 }
423 }
424 EXPORT_SYMBOL(finish_wait);
425
autoremove_wake_function(struct wait_queue_entry * wq_entry,unsigned mode,int sync,void * key)426 int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
427 {
428 int ret = default_wake_function(wq_entry, mode, sync, key);
429
430 if (ret)
431 list_del_init_careful(&wq_entry->entry);
432
433 return ret;
434 }
435 EXPORT_SYMBOL(autoremove_wake_function);
436
437 /*
438 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
439 *
440 * add_wait_queue(&wq_head, &wait);
441 * for (;;) {
442 * if (condition)
443 * break;
444 *
445 * // in wait_woken() // in woken_wake_function()
446 *
447 * p->state = mode; wq_entry->flags |= WQ_FLAG_WOKEN;
448 * smp_mb(); // A try_to_wake_up():
449 * if (!(wq_entry->flags & WQ_FLAG_WOKEN)) <full barrier>
450 * schedule() if (p->state & mode)
451 * p->state = TASK_RUNNING; p->state = TASK_RUNNING;
452 * wq_entry->flags &= ~WQ_FLAG_WOKEN; ~~~~~~~~~~~~~~~~~~
453 * smp_mb(); // B condition = true;
454 * } smp_mb(); // C
455 * remove_wait_queue(&wq_head, &wait); wq_entry->flags |= WQ_FLAG_WOKEN;
456 */
wait_woken(struct wait_queue_entry * wq_entry,unsigned mode,long timeout)457 long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
458 {
459 /*
460 * The below executes an smp_mb(), which matches with the full barrier
461 * executed by the try_to_wake_up() in woken_wake_function() such that
462 * either we see the store to wq_entry->flags in woken_wake_function()
463 * or woken_wake_function() sees our store to current->state.
464 */
465 set_current_state(mode); /* A */
466 if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !kthread_should_stop_or_park())
467 timeout = schedule_timeout(timeout);
468 __set_current_state(TASK_RUNNING);
469
470 /*
471 * The below executes an smp_mb(), which matches with the smp_mb() (C)
472 * in woken_wake_function() such that either we see the wait condition
473 * being true or the store to wq_entry->flags in woken_wake_function()
474 * follows ours in the coherence order.
475 */
476 smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
477
478 return timeout;
479 }
480 EXPORT_SYMBOL(wait_woken);
481
woken_wake_function(struct wait_queue_entry * wq_entry,unsigned mode,int sync,void * key)482 int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
483 {
484 /* Pairs with the smp_store_mb() in wait_woken(). */
485 smp_mb(); /* C */
486 wq_entry->flags |= WQ_FLAG_WOKEN;
487
488 return default_wake_function(wq_entry, mode, sync, key);
489 }
490 EXPORT_SYMBOL(woken_wake_function);
491