Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Generic waiting primitives.
4 *
5 * (C) 2004 Nadia Yvette Chambers, Oracle
6 */
7#include "sched.h"
8
9void __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
16EXPORT_SYMBOL(__init_waitqueue_head);
17
18void 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}
27EXPORT_SYMBOL(add_wait_queue);
28
29void 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}
38EXPORT_SYMBOL(add_wait_queue_exclusive);
39
40void 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}
49EXPORT_SYMBOL_GPL(add_wait_queue_priority);
50
51void 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}
59EXPORT_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 */
81static 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
125static void __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
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 nr_exclusive = __wake_up_common(wq_head, mode, nr_exclusive,
139 wake_flags, key, &bookmark);
140 spin_unlock_irqrestore(&wq_head->lock, flags);
141 } while (bookmark.flags & WQ_FLAG_BOOKMARK);
142}
143
144/**
145 * __wake_up - wake up threads blocked on a waitqueue.
146 * @wq_head: the waitqueue
147 * @mode: which threads
148 * @nr_exclusive: how many wake-one or wake-many threads to wake up
149 * @key: is directly passed to the wakeup function
150 *
151 * If this function wakes up a task, it executes a full memory barrier before
152 * accessing the task state.
153 */
154void __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
155 int nr_exclusive, void *key)
156{
157 __wake_up_common_lock(wq_head, mode, nr_exclusive, 0, key);
158}
159EXPORT_SYMBOL(__wake_up);
160
161/*
162 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
163 */
164void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
165{
166 __wake_up_common(wq_head, mode, nr, 0, NULL, NULL);
167}
168EXPORT_SYMBOL_GPL(__wake_up_locked);
169
170void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
171{
172 __wake_up_common(wq_head, mode, 1, 0, key, NULL);
173}
174EXPORT_SYMBOL_GPL(__wake_up_locked_key);
175
176void __wake_up_locked_key_bookmark(struct wait_queue_head *wq_head,
177 unsigned int mode, void *key, wait_queue_entry_t *bookmark)
178{
179 __wake_up_common(wq_head, mode, 1, 0, key, bookmark);
180}
181EXPORT_SYMBOL_GPL(__wake_up_locked_key_bookmark);
182
183/**
184 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
185 * @wq_head: the waitqueue
186 * @mode: which threads
187 * @key: opaque value to be passed to wakeup targets
188 *
189 * The sync wakeup differs that the waker knows that it will schedule
190 * away soon, so while the target thread will be woken up, it will not
191 * be migrated to another CPU - ie. the two threads are 'synchronized'
192 * with each other. This can prevent needless bouncing between CPUs.
193 *
194 * On UP it can prevent extra preemption.
195 *
196 * If this function wakes up a task, it executes a full memory barrier before
197 * accessing the task state.
198 */
199void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
200 void *key)
201{
202 if (unlikely(!wq_head))
203 return;
204
205 __wake_up_common_lock(wq_head, mode, 1, WF_SYNC, key);
206}
207EXPORT_SYMBOL_GPL(__wake_up_sync_key);
208
209/**
210 * __wake_up_locked_sync_key - wake up a thread blocked on a locked waitqueue.
211 * @wq_head: the waitqueue
212 * @mode: which threads
213 * @key: opaque value to be passed to wakeup targets
214 *
215 * The sync wakeup differs in that the waker knows that it will schedule
216 * away soon, so while the target thread will be woken up, it will not
217 * be migrated to another CPU - ie. the two threads are 'synchronized'
218 * with each other. This can prevent needless bouncing between CPUs.
219 *
220 * On UP it can prevent extra preemption.
221 *
222 * If this function wakes up a task, it executes a full memory barrier before
223 * accessing the task state.
224 */
225void __wake_up_locked_sync_key(struct wait_queue_head *wq_head,
226 unsigned int mode, void *key)
227{
228 __wake_up_common(wq_head, mode, 1, WF_SYNC, key, NULL);
229}
230EXPORT_SYMBOL_GPL(__wake_up_locked_sync_key);
231
232/*
233 * __wake_up_sync - see __wake_up_sync_key()
234 */
235void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode)
236{
237 __wake_up_sync_key(wq_head, mode, NULL);
238}
239EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
240
241/*
242 * Note: we use "set_current_state()" _after_ the wait-queue add,
243 * because we need a memory barrier there on SMP, so that any
244 * wake-function that tests for the wait-queue being active
245 * will be guaranteed to see waitqueue addition _or_ subsequent
246 * tests in this thread will see the wakeup having taken place.
247 *
248 * The spin_unlock() itself is semi-permeable and only protects
249 * one way (it only protects stuff inside the critical region and
250 * stops them from bleeding out - it would still allow subsequent
251 * loads to move into the critical region).
252 */
253void
254prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
255{
256 unsigned long flags;
257
258 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
259 spin_lock_irqsave(&wq_head->lock, flags);
260 if (list_empty(&wq_entry->entry))
261 __add_wait_queue(wq_head, wq_entry);
262 set_current_state(state);
263 spin_unlock_irqrestore(&wq_head->lock, flags);
264}
265EXPORT_SYMBOL(prepare_to_wait);
266
267/* Returns true if we are the first waiter in the queue, false otherwise. */
268bool
269prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
270{
271 unsigned long flags;
272 bool was_empty = false;
273
274 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
275 spin_lock_irqsave(&wq_head->lock, flags);
276 if (list_empty(&wq_entry->entry)) {
277 was_empty = list_empty(&wq_head->head);
278 __add_wait_queue_entry_tail(wq_head, wq_entry);
279 }
280 set_current_state(state);
281 spin_unlock_irqrestore(&wq_head->lock, flags);
282 return was_empty;
283}
284EXPORT_SYMBOL(prepare_to_wait_exclusive);
285
286void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
287{
288 wq_entry->flags = flags;
289 wq_entry->private = current;
290 wq_entry->func = autoremove_wake_function;
291 INIT_LIST_HEAD(&wq_entry->entry);
292}
293EXPORT_SYMBOL(init_wait_entry);
294
295long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
296{
297 unsigned long flags;
298 long ret = 0;
299
300 spin_lock_irqsave(&wq_head->lock, flags);
301 if (signal_pending_state(state, current)) {
302 /*
303 * Exclusive waiter must not fail if it was selected by wakeup,
304 * it should "consume" the condition we were waiting for.
305 *
306 * The caller will recheck the condition and return success if
307 * we were already woken up, we can not miss the event because
308 * wakeup locks/unlocks the same wq_head->lock.
309 *
310 * But we need to ensure that set-condition + wakeup after that
311 * can't see us, it should wake up another exclusive waiter if
312 * we fail.
313 */
314 list_del_init(&wq_entry->entry);
315 ret = -ERESTARTSYS;
316 } else {
317 if (list_empty(&wq_entry->entry)) {
318 if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
319 __add_wait_queue_entry_tail(wq_head, wq_entry);
320 else
321 __add_wait_queue(wq_head, wq_entry);
322 }
323 set_current_state(state);
324 }
325 spin_unlock_irqrestore(&wq_head->lock, flags);
326
327 return ret;
328}
329EXPORT_SYMBOL(prepare_to_wait_event);
330
331/*
332 * Note! These two wait functions are entered with the
333 * wait-queue lock held (and interrupts off in the _irq
334 * case), so there is no race with testing the wakeup
335 * condition in the caller before they add the wait
336 * entry to the wake queue.
337 */
338int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
339{
340 if (likely(list_empty(&wait->entry)))
341 __add_wait_queue_entry_tail(wq, wait);
342
343 set_current_state(TASK_INTERRUPTIBLE);
344 if (signal_pending(current))
345 return -ERESTARTSYS;
346
347 spin_unlock(&wq->lock);
348 schedule();
349 spin_lock(&wq->lock);
350
351 return 0;
352}
353EXPORT_SYMBOL(do_wait_intr);
354
355int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
356{
357 if (likely(list_empty(&wait->entry)))
358 __add_wait_queue_entry_tail(wq, wait);
359
360 set_current_state(TASK_INTERRUPTIBLE);
361 if (signal_pending(current))
362 return -ERESTARTSYS;
363
364 spin_unlock_irq(&wq->lock);
365 schedule();
366 spin_lock_irq(&wq->lock);
367
368 return 0;
369}
370EXPORT_SYMBOL(do_wait_intr_irq);
371
372/**
373 * finish_wait - clean up after waiting in a queue
374 * @wq_head: waitqueue waited on
375 * @wq_entry: wait descriptor
376 *
377 * Sets current thread back to running state and removes
378 * the wait descriptor from the given waitqueue if still
379 * queued.
380 */
381void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
382{
383 unsigned long flags;
384
385 __set_current_state(TASK_RUNNING);
386 /*
387 * We can check for list emptiness outside the lock
388 * IFF:
389 * - we use the "careful" check that verifies both
390 * the next and prev pointers, so that there cannot
391 * be any half-pending updates in progress on other
392 * CPU's that we haven't seen yet (and that might
393 * still change the stack area.
394 * and
395 * - all other users take the lock (ie we can only
396 * have _one_ other CPU that looks at or modifies
397 * the list).
398 */
399 if (!list_empty_careful(&wq_entry->entry)) {
400 spin_lock_irqsave(&wq_head->lock, flags);
401 list_del_init(&wq_entry->entry);
402 spin_unlock_irqrestore(&wq_head->lock, flags);
403 }
404}
405EXPORT_SYMBOL(finish_wait);
406
407int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
408{
409 int ret = default_wake_function(wq_entry, mode, sync, key);
410
411 if (ret)
412 list_del_init_careful(&wq_entry->entry);
413
414 return ret;
415}
416EXPORT_SYMBOL(autoremove_wake_function);
417
418static inline bool is_kthread_should_stop(void)
419{
420 return (current->flags & PF_KTHREAD) && kthread_should_stop();
421}
422
423/*
424 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
425 *
426 * add_wait_queue(&wq_head, &wait);
427 * for (;;) {
428 * if (condition)
429 * break;
430 *
431 * // in wait_woken() // in woken_wake_function()
432 *
433 * p->state = mode; wq_entry->flags |= WQ_FLAG_WOKEN;
434 * smp_mb(); // A try_to_wake_up():
435 * if (!(wq_entry->flags & WQ_FLAG_WOKEN)) <full barrier>
436 * schedule() if (p->state & mode)
437 * p->state = TASK_RUNNING; p->state = TASK_RUNNING;
438 * wq_entry->flags &= ~WQ_FLAG_WOKEN; ~~~~~~~~~~~~~~~~~~
439 * smp_mb(); // B condition = true;
440 * } smp_mb(); // C
441 * remove_wait_queue(&wq_head, &wait); wq_entry->flags |= WQ_FLAG_WOKEN;
442 */
443long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
444{
445 /*
446 * The below executes an smp_mb(), which matches with the full barrier
447 * executed by the try_to_wake_up() in woken_wake_function() such that
448 * either we see the store to wq_entry->flags in woken_wake_function()
449 * or woken_wake_function() sees our store to current->state.
450 */
451 set_current_state(mode); /* A */
452 if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !is_kthread_should_stop())
453 timeout = schedule_timeout(timeout);
454 __set_current_state(TASK_RUNNING);
455
456 /*
457 * The below executes an smp_mb(), which matches with the smp_mb() (C)
458 * in woken_wake_function() such that either we see the wait condition
459 * being true or the store to wq_entry->flags in woken_wake_function()
460 * follows ours in the coherence order.
461 */
462 smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
463
464 return timeout;
465}
466EXPORT_SYMBOL(wait_woken);
467
468int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
469{
470 /* Pairs with the smp_store_mb() in wait_woken(). */
471 smp_mb(); /* C */
472 wq_entry->flags |= WQ_FLAG_WOKEN;
473
474 return default_wake_function(wq_entry, mode, sync, key);
475}
476EXPORT_SYMBOL(woken_wake_function);
1/*
2 * Generic waiting primitives.
3 *
4 * (C) 2004 Nadia Yvette Chambers, Oracle
5 */
6#include <linux/init.h>
7#include <linux/export.h>
8#include <linux/sched.h>
9#include <linux/mm.h>
10#include <linux/wait.h>
11#include <linux/hash.h>
12
13void __init_waitqueue_head(wait_queue_head_t *q, const char *name, struct lock_class_key *key)
14{
15 spin_lock_init(&q->lock);
16 lockdep_set_class_and_name(&q->lock, key, name);
17 INIT_LIST_HEAD(&q->task_list);
18}
19
20EXPORT_SYMBOL(__init_waitqueue_head);
21
22void add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
23{
24 unsigned long flags;
25
26 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
27 spin_lock_irqsave(&q->lock, flags);
28 __add_wait_queue(q, wait);
29 spin_unlock_irqrestore(&q->lock, flags);
30}
31EXPORT_SYMBOL(add_wait_queue);
32
33void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait)
34{
35 unsigned long flags;
36
37 wait->flags |= WQ_FLAG_EXCLUSIVE;
38 spin_lock_irqsave(&q->lock, flags);
39 __add_wait_queue_tail(q, wait);
40 spin_unlock_irqrestore(&q->lock, flags);
41}
42EXPORT_SYMBOL(add_wait_queue_exclusive);
43
44void remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
45{
46 unsigned long flags;
47
48 spin_lock_irqsave(&q->lock, flags);
49 __remove_wait_queue(q, wait);
50 spin_unlock_irqrestore(&q->lock, flags);
51}
52EXPORT_SYMBOL(remove_wait_queue);
53
54
55/*
56 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
57 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
58 * number) then we wake all the non-exclusive tasks and one exclusive task.
59 *
60 * There are circumstances in which we can try to wake a task which has already
61 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
62 * zero in this (rare) case, and we handle it by continuing to scan the queue.
63 */
64static void __wake_up_common(wait_queue_head_t *q, unsigned int mode,
65 int nr_exclusive, int wake_flags, void *key)
66{
67 wait_queue_t *curr, *next;
68
69 list_for_each_entry_safe(curr, next, &q->task_list, task_list) {
70 unsigned flags = curr->flags;
71
72 if (curr->func(curr, mode, wake_flags, key) &&
73 (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
74 break;
75 }
76}
77
78/**
79 * __wake_up - wake up threads blocked on a waitqueue.
80 * @q: the waitqueue
81 * @mode: which threads
82 * @nr_exclusive: how many wake-one or wake-many threads to wake up
83 * @key: is directly passed to the wakeup function
84 *
85 * It may be assumed that this function implies a write memory barrier before
86 * changing the task state if and only if any tasks are woken up.
87 */
88void __wake_up(wait_queue_head_t *q, unsigned int mode,
89 int nr_exclusive, void *key)
90{
91 unsigned long flags;
92
93 spin_lock_irqsave(&q->lock, flags);
94 __wake_up_common(q, mode, nr_exclusive, 0, key);
95 spin_unlock_irqrestore(&q->lock, flags);
96}
97EXPORT_SYMBOL(__wake_up);
98
99/*
100 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
101 */
102void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr)
103{
104 __wake_up_common(q, mode, nr, 0, NULL);
105}
106EXPORT_SYMBOL_GPL(__wake_up_locked);
107
108void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
109{
110 __wake_up_common(q, mode, 1, 0, key);
111}
112EXPORT_SYMBOL_GPL(__wake_up_locked_key);
113
114/**
115 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
116 * @q: the waitqueue
117 * @mode: which threads
118 * @nr_exclusive: how many wake-one or wake-many threads to wake up
119 * @key: opaque value to be passed to wakeup targets
120 *
121 * The sync wakeup differs that the waker knows that it will schedule
122 * away soon, so while the target thread will be woken up, it will not
123 * be migrated to another CPU - ie. the two threads are 'synchronized'
124 * with each other. This can prevent needless bouncing between CPUs.
125 *
126 * On UP it can prevent extra preemption.
127 *
128 * It may be assumed that this function implies a write memory barrier before
129 * changing the task state if and only if any tasks are woken up.
130 */
131void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
132 int nr_exclusive, void *key)
133{
134 unsigned long flags;
135 int wake_flags = 1; /* XXX WF_SYNC */
136
137 if (unlikely(!q))
138 return;
139
140 if (unlikely(nr_exclusive != 1))
141 wake_flags = 0;
142
143 spin_lock_irqsave(&q->lock, flags);
144 __wake_up_common(q, mode, nr_exclusive, wake_flags, key);
145 spin_unlock_irqrestore(&q->lock, flags);
146}
147EXPORT_SYMBOL_GPL(__wake_up_sync_key);
148
149/*
150 * __wake_up_sync - see __wake_up_sync_key()
151 */
152void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
153{
154 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
155}
156EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
157
158/*
159 * Note: we use "set_current_state()" _after_ the wait-queue add,
160 * because we need a memory barrier there on SMP, so that any
161 * wake-function that tests for the wait-queue being active
162 * will be guaranteed to see waitqueue addition _or_ subsequent
163 * tests in this thread will see the wakeup having taken place.
164 *
165 * The spin_unlock() itself is semi-permeable and only protects
166 * one way (it only protects stuff inside the critical region and
167 * stops them from bleeding out - it would still allow subsequent
168 * loads to move into the critical region).
169 */
170void
171prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
172{
173 unsigned long flags;
174
175 wait->flags &= ~WQ_FLAG_EXCLUSIVE;
176 spin_lock_irqsave(&q->lock, flags);
177 if (list_empty(&wait->task_list))
178 __add_wait_queue(q, wait);
179 set_current_state(state);
180 spin_unlock_irqrestore(&q->lock, flags);
181}
182EXPORT_SYMBOL(prepare_to_wait);
183
184void
185prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
186{
187 unsigned long flags;
188
189 wait->flags |= WQ_FLAG_EXCLUSIVE;
190 spin_lock_irqsave(&q->lock, flags);
191 if (list_empty(&wait->task_list))
192 __add_wait_queue_tail(q, wait);
193 set_current_state(state);
194 spin_unlock_irqrestore(&q->lock, flags);
195}
196EXPORT_SYMBOL(prepare_to_wait_exclusive);
197
198long prepare_to_wait_event(wait_queue_head_t *q, wait_queue_t *wait, int state)
199{
200 unsigned long flags;
201
202 if (signal_pending_state(state, current))
203 return -ERESTARTSYS;
204
205 wait->private = current;
206 wait->func = autoremove_wake_function;
207
208 spin_lock_irqsave(&q->lock, flags);
209 if (list_empty(&wait->task_list)) {
210 if (wait->flags & WQ_FLAG_EXCLUSIVE)
211 __add_wait_queue_tail(q, wait);
212 else
213 __add_wait_queue(q, wait);
214 }
215 set_current_state(state);
216 spin_unlock_irqrestore(&q->lock, flags);
217
218 return 0;
219}
220EXPORT_SYMBOL(prepare_to_wait_event);
221
222/**
223 * finish_wait - clean up after waiting in a queue
224 * @q: waitqueue waited on
225 * @wait: wait descriptor
226 *
227 * Sets current thread back to running state and removes
228 * the wait descriptor from the given waitqueue if still
229 * queued.
230 */
231void finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
232{
233 unsigned long flags;
234
235 __set_current_state(TASK_RUNNING);
236 /*
237 * We can check for list emptiness outside the lock
238 * IFF:
239 * - we use the "careful" check that verifies both
240 * the next and prev pointers, so that there cannot
241 * be any half-pending updates in progress on other
242 * CPU's that we haven't seen yet (and that might
243 * still change the stack area.
244 * and
245 * - all other users take the lock (ie we can only
246 * have _one_ other CPU that looks at or modifies
247 * the list).
248 */
249 if (!list_empty_careful(&wait->task_list)) {
250 spin_lock_irqsave(&q->lock, flags);
251 list_del_init(&wait->task_list);
252 spin_unlock_irqrestore(&q->lock, flags);
253 }
254}
255EXPORT_SYMBOL(finish_wait);
256
257/**
258 * abort_exclusive_wait - abort exclusive waiting in a queue
259 * @q: waitqueue waited on
260 * @wait: wait descriptor
261 * @mode: runstate of the waiter to be woken
262 * @key: key to identify a wait bit queue or %NULL
263 *
264 * Sets current thread back to running state and removes
265 * the wait descriptor from the given waitqueue if still
266 * queued.
267 *
268 * Wakes up the next waiter if the caller is concurrently
269 * woken up through the queue.
270 *
271 * This prevents waiter starvation where an exclusive waiter
272 * aborts and is woken up concurrently and no one wakes up
273 * the next waiter.
274 */
275void abort_exclusive_wait(wait_queue_head_t *q, wait_queue_t *wait,
276 unsigned int mode, void *key)
277{
278 unsigned long flags;
279
280 __set_current_state(TASK_RUNNING);
281 spin_lock_irqsave(&q->lock, flags);
282 if (!list_empty(&wait->task_list))
283 list_del_init(&wait->task_list);
284 else if (waitqueue_active(q))
285 __wake_up_locked_key(q, mode, key);
286 spin_unlock_irqrestore(&q->lock, flags);
287}
288EXPORT_SYMBOL(abort_exclusive_wait);
289
290int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
291{
292 int ret = default_wake_function(wait, mode, sync, key);
293
294 if (ret)
295 list_del_init(&wait->task_list);
296 return ret;
297}
298EXPORT_SYMBOL(autoremove_wake_function);
299
300int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
301{
302 struct wait_bit_key *key = arg;
303 struct wait_bit_queue *wait_bit
304 = container_of(wait, struct wait_bit_queue, wait);
305
306 if (wait_bit->key.flags != key->flags ||
307 wait_bit->key.bit_nr != key->bit_nr ||
308 test_bit(key->bit_nr, key->flags))
309 return 0;
310 else
311 return autoremove_wake_function(wait, mode, sync, key);
312}
313EXPORT_SYMBOL(wake_bit_function);
314
315/*
316 * To allow interruptible waiting and asynchronous (i.e. nonblocking)
317 * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are
318 * permitted return codes. Nonzero return codes halt waiting and return.
319 */
320int __sched
321__wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q,
322 int (*action)(void *), unsigned mode)
323{
324 int ret = 0;
325
326 do {
327 prepare_to_wait(wq, &q->wait, mode);
328 if (test_bit(q->key.bit_nr, q->key.flags))
329 ret = (*action)(q->key.flags);
330 } while (test_bit(q->key.bit_nr, q->key.flags) && !ret);
331 finish_wait(wq, &q->wait);
332 return ret;
333}
334EXPORT_SYMBOL(__wait_on_bit);
335
336int __sched out_of_line_wait_on_bit(void *word, int bit,
337 int (*action)(void *), unsigned mode)
338{
339 wait_queue_head_t *wq = bit_waitqueue(word, bit);
340 DEFINE_WAIT_BIT(wait, word, bit);
341
342 return __wait_on_bit(wq, &wait, action, mode);
343}
344EXPORT_SYMBOL(out_of_line_wait_on_bit);
345
346int __sched
347__wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q,
348 int (*action)(void *), unsigned mode)
349{
350 do {
351 int ret;
352
353 prepare_to_wait_exclusive(wq, &q->wait, mode);
354 if (!test_bit(q->key.bit_nr, q->key.flags))
355 continue;
356 ret = action(q->key.flags);
357 if (!ret)
358 continue;
359 abort_exclusive_wait(wq, &q->wait, mode, &q->key);
360 return ret;
361 } while (test_and_set_bit(q->key.bit_nr, q->key.flags));
362 finish_wait(wq, &q->wait);
363 return 0;
364}
365EXPORT_SYMBOL(__wait_on_bit_lock);
366
367int __sched out_of_line_wait_on_bit_lock(void *word, int bit,
368 int (*action)(void *), unsigned mode)
369{
370 wait_queue_head_t *wq = bit_waitqueue(word, bit);
371 DEFINE_WAIT_BIT(wait, word, bit);
372
373 return __wait_on_bit_lock(wq, &wait, action, mode);
374}
375EXPORT_SYMBOL(out_of_line_wait_on_bit_lock);
376
377void __wake_up_bit(wait_queue_head_t *wq, void *word, int bit)
378{
379 struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit);
380 if (waitqueue_active(wq))
381 __wake_up(wq, TASK_NORMAL, 1, &key);
382}
383EXPORT_SYMBOL(__wake_up_bit);
384
385/**
386 * wake_up_bit - wake up a waiter on a bit
387 * @word: the word being waited on, a kernel virtual address
388 * @bit: the bit of the word being waited on
389 *
390 * There is a standard hashed waitqueue table for generic use. This
391 * is the part of the hashtable's accessor API that wakes up waiters
392 * on a bit. For instance, if one were to have waiters on a bitflag,
393 * one would call wake_up_bit() after clearing the bit.
394 *
395 * In order for this to function properly, as it uses waitqueue_active()
396 * internally, some kind of memory barrier must be done prior to calling
397 * this. Typically, this will be smp_mb__after_clear_bit(), but in some
398 * cases where bitflags are manipulated non-atomically under a lock, one
399 * may need to use a less regular barrier, such fs/inode.c's smp_mb(),
400 * because spin_unlock() does not guarantee a memory barrier.
401 */
402void wake_up_bit(void *word, int bit)
403{
404 __wake_up_bit(bit_waitqueue(word, bit), word, bit);
405}
406EXPORT_SYMBOL(wake_up_bit);
407
408wait_queue_head_t *bit_waitqueue(void *word, int bit)
409{
410 const int shift = BITS_PER_LONG == 32 ? 5 : 6;
411 const struct zone *zone = page_zone(virt_to_page(word));
412 unsigned long val = (unsigned long)word << shift | bit;
413
414 return &zone->wait_table[hash_long(val, zone->wait_table_bits)];
415}
416EXPORT_SYMBOL(bit_waitqueue);
417
418/*
419 * Manipulate the atomic_t address to produce a better bit waitqueue table hash
420 * index (we're keying off bit -1, but that would produce a horrible hash
421 * value).
422 */
423static inline wait_queue_head_t *atomic_t_waitqueue(atomic_t *p)
424{
425 if (BITS_PER_LONG == 64) {
426 unsigned long q = (unsigned long)p;
427 return bit_waitqueue((void *)(q & ~1), q & 1);
428 }
429 return bit_waitqueue(p, 0);
430}
431
432static int wake_atomic_t_function(wait_queue_t *wait, unsigned mode, int sync,
433 void *arg)
434{
435 struct wait_bit_key *key = arg;
436 struct wait_bit_queue *wait_bit
437 = container_of(wait, struct wait_bit_queue, wait);
438 atomic_t *val = key->flags;
439
440 if (wait_bit->key.flags != key->flags ||
441 wait_bit->key.bit_nr != key->bit_nr ||
442 atomic_read(val) != 0)
443 return 0;
444 return autoremove_wake_function(wait, mode, sync, key);
445}
446
447/*
448 * To allow interruptible waiting and asynchronous (i.e. nonblocking) waiting,
449 * the actions of __wait_on_atomic_t() are permitted return codes. Nonzero
450 * return codes halt waiting and return.
451 */
452static __sched
453int __wait_on_atomic_t(wait_queue_head_t *wq, struct wait_bit_queue *q,
454 int (*action)(atomic_t *), unsigned mode)
455{
456 atomic_t *val;
457 int ret = 0;
458
459 do {
460 prepare_to_wait(wq, &q->wait, mode);
461 val = q->key.flags;
462 if (atomic_read(val) == 0)
463 break;
464 ret = (*action)(val);
465 } while (!ret && atomic_read(val) != 0);
466 finish_wait(wq, &q->wait);
467 return ret;
468}
469
470#define DEFINE_WAIT_ATOMIC_T(name, p) \
471 struct wait_bit_queue name = { \
472 .key = __WAIT_ATOMIC_T_KEY_INITIALIZER(p), \
473 .wait = { \
474 .private = current, \
475 .func = wake_atomic_t_function, \
476 .task_list = \
477 LIST_HEAD_INIT((name).wait.task_list), \
478 }, \
479 }
480
481__sched int out_of_line_wait_on_atomic_t(atomic_t *p, int (*action)(atomic_t *),
482 unsigned mode)
483{
484 wait_queue_head_t *wq = atomic_t_waitqueue(p);
485 DEFINE_WAIT_ATOMIC_T(wait, p);
486
487 return __wait_on_atomic_t(wq, &wait, action, mode);
488}
489EXPORT_SYMBOL(out_of_line_wait_on_atomic_t);
490
491/**
492 * wake_up_atomic_t - Wake up a waiter on a atomic_t
493 * @p: The atomic_t being waited on, a kernel virtual address
494 *
495 * Wake up anyone waiting for the atomic_t to go to zero.
496 *
497 * Abuse the bit-waker function and its waitqueue hash table set (the atomic_t
498 * check is done by the waiter's wake function, not the by the waker itself).
499 */
500void wake_up_atomic_t(atomic_t *p)
501{
502 __wake_up_bit(atomic_t_waitqueue(p), p, WAIT_ATOMIC_T_BIT_NR);
503}
504EXPORT_SYMBOL(wake_up_atomic_t);