]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Generic waiting primitives. | |
3 | * | |
6d49e352 | 4 | * (C) 2004 Nadia Yvette Chambers, Oracle |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include <linux/init.h> |
9984de1a | 7 | #include <linux/export.h> |
174cd4b1 | 8 | #include <linux/sched/signal.h> |
b17b0153 | 9 | #include <linux/sched/debug.h> |
1da177e4 LT |
10 | #include <linux/mm.h> |
11 | #include <linux/wait.h> | |
12 | #include <linux/hash.h> | |
cb6538e7 | 13 | #include <linux/kthread.h> |
1da177e4 | 14 | |
f07fdec5 | 15 | void __init_waitqueue_head(wait_queue_head_t *q, const char *name, struct lock_class_key *key) |
21d71f51 IM |
16 | { |
17 | spin_lock_init(&q->lock); | |
f07fdec5 | 18 | lockdep_set_class_and_name(&q->lock, key, name); |
21d71f51 IM |
19 | INIT_LIST_HEAD(&q->task_list); |
20 | } | |
eb4542b9 | 21 | |
2fc39111 | 22 | EXPORT_SYMBOL(__init_waitqueue_head); |
eb4542b9 | 23 | |
7ad5b3a5 | 24 | void add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait) |
1da177e4 LT |
25 | { |
26 | unsigned long flags; | |
27 | ||
28 | wait->flags &= ~WQ_FLAG_EXCLUSIVE; | |
29 | spin_lock_irqsave(&q->lock, flags); | |
30 | __add_wait_queue(q, wait); | |
31 | spin_unlock_irqrestore(&q->lock, flags); | |
32 | } | |
33 | EXPORT_SYMBOL(add_wait_queue); | |
34 | ||
7ad5b3a5 | 35 | void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait) |
1da177e4 LT |
36 | { |
37 | unsigned long flags; | |
38 | ||
39 | wait->flags |= WQ_FLAG_EXCLUSIVE; | |
40 | spin_lock_irqsave(&q->lock, flags); | |
41 | __add_wait_queue_tail(q, wait); | |
42 | spin_unlock_irqrestore(&q->lock, flags); | |
43 | } | |
44 | EXPORT_SYMBOL(add_wait_queue_exclusive); | |
45 | ||
7ad5b3a5 | 46 | void remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait) |
1da177e4 LT |
47 | { |
48 | unsigned long flags; | |
49 | ||
50 | spin_lock_irqsave(&q->lock, flags); | |
51 | __remove_wait_queue(q, wait); | |
52 | spin_unlock_irqrestore(&q->lock, flags); | |
53 | } | |
54 | EXPORT_SYMBOL(remove_wait_queue); | |
55 | ||
56 | ||
b4145872 PZ |
57 | /* |
58 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just | |
59 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
60 | * number) then we wake all the non-exclusive tasks and one exclusive task. | |
61 | * | |
62 | * There are circumstances in which we can try to wake a task which has already | |
63 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns | |
64 | * zero in this (rare) case, and we handle it by continuing to scan the queue. | |
65 | */ | |
66 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
67 | int nr_exclusive, int wake_flags, void *key) | |
68 | { | |
69 | wait_queue_t *curr, *next; | |
70 | ||
71 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { | |
72 | unsigned flags = curr->flags; | |
73 | ||
74 | if (curr->func(curr, mode, wake_flags, key) && | |
75 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) | |
76 | break; | |
77 | } | |
78 | } | |
79 | ||
80 | /** | |
81 | * __wake_up - wake up threads blocked on a waitqueue. | |
82 | * @q: the waitqueue | |
83 | * @mode: which threads | |
84 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
85 | * @key: is directly passed to the wakeup function | |
86 | * | |
87 | * It may be assumed that this function implies a write memory barrier before | |
88 | * changing the task state if and only if any tasks are woken up. | |
89 | */ | |
90 | void __wake_up(wait_queue_head_t *q, unsigned int mode, | |
91 | int nr_exclusive, void *key) | |
92 | { | |
93 | unsigned long flags; | |
94 | ||
95 | spin_lock_irqsave(&q->lock, flags); | |
96 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
97 | spin_unlock_irqrestore(&q->lock, flags); | |
98 | } | |
99 | EXPORT_SYMBOL(__wake_up); | |
100 | ||
101 | /* | |
102 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
103 | */ | |
104 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode, int nr) | |
105 | { | |
106 | __wake_up_common(q, mode, nr, 0, NULL); | |
107 | } | |
108 | EXPORT_SYMBOL_GPL(__wake_up_locked); | |
109 | ||
ac5be6b4 | 110 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
b4145872 | 111 | { |
ac5be6b4 | 112 | __wake_up_common(q, mode, 1, 0, key); |
b4145872 PZ |
113 | } |
114 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); | |
115 | ||
116 | /** | |
117 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. | |
118 | * @q: the waitqueue | |
119 | * @mode: which threads | |
120 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
121 | * @key: opaque value to be passed to wakeup targets | |
122 | * | |
123 | * The sync wakeup differs that the waker knows that it will schedule | |
124 | * away soon, so while the target thread will be woken up, it will not | |
125 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
126 | * with each other. This can prevent needless bouncing between CPUs. | |
127 | * | |
128 | * On UP it can prevent extra preemption. | |
129 | * | |
130 | * It may be assumed that this function implies a write memory barrier before | |
131 | * changing the task state if and only if any tasks are woken up. | |
132 | */ | |
133 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, | |
134 | int nr_exclusive, void *key) | |
135 | { | |
136 | unsigned long flags; | |
137 | int wake_flags = 1; /* XXX WF_SYNC */ | |
138 | ||
139 | if (unlikely(!q)) | |
140 | return; | |
141 | ||
142 | if (unlikely(nr_exclusive != 1)) | |
143 | wake_flags = 0; | |
144 | ||
145 | spin_lock_irqsave(&q->lock, flags); | |
146 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); | |
147 | spin_unlock_irqrestore(&q->lock, flags); | |
148 | } | |
149 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); | |
150 | ||
151 | /* | |
152 | * __wake_up_sync - see __wake_up_sync_key() | |
153 | */ | |
154 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
155 | { | |
156 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
157 | } | |
158 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
159 | ||
1da177e4 LT |
160 | /* |
161 | * Note: we use "set_current_state()" _after_ the wait-queue add, | |
162 | * because we need a memory barrier there on SMP, so that any | |
163 | * wake-function that tests for the wait-queue being active | |
164 | * will be guaranteed to see waitqueue addition _or_ subsequent | |
165 | * tests in this thread will see the wakeup having taken place. | |
166 | * | |
167 | * The spin_unlock() itself is semi-permeable and only protects | |
168 | * one way (it only protects stuff inside the critical region and | |
169 | * stops them from bleeding out - it would still allow subsequent | |
59c51591 | 170 | * loads to move into the critical region). |
1da177e4 | 171 | */ |
7ad5b3a5 | 172 | void |
1da177e4 LT |
173 | prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state) |
174 | { | |
175 | unsigned long flags; | |
176 | ||
177 | wait->flags &= ~WQ_FLAG_EXCLUSIVE; | |
178 | spin_lock_irqsave(&q->lock, flags); | |
179 | if (list_empty(&wait->task_list)) | |
180 | __add_wait_queue(q, wait); | |
a25d644f | 181 | set_current_state(state); |
1da177e4 LT |
182 | spin_unlock_irqrestore(&q->lock, flags); |
183 | } | |
184 | EXPORT_SYMBOL(prepare_to_wait); | |
185 | ||
7ad5b3a5 | 186 | void |
1da177e4 LT |
187 | prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state) |
188 | { | |
189 | unsigned long flags; | |
190 | ||
191 | wait->flags |= WQ_FLAG_EXCLUSIVE; | |
192 | spin_lock_irqsave(&q->lock, flags); | |
193 | if (list_empty(&wait->task_list)) | |
194 | __add_wait_queue_tail(q, wait); | |
a25d644f | 195 | set_current_state(state); |
1da177e4 LT |
196 | spin_unlock_irqrestore(&q->lock, flags); |
197 | } | |
198 | EXPORT_SYMBOL(prepare_to_wait_exclusive); | |
199 | ||
0176beaf ON |
200 | void init_wait_entry(wait_queue_t *wait, int flags) |
201 | { | |
202 | wait->flags = flags; | |
203 | wait->private = current; | |
204 | wait->func = autoremove_wake_function; | |
205 | INIT_LIST_HEAD(&wait->task_list); | |
206 | } | |
207 | EXPORT_SYMBOL(init_wait_entry); | |
208 | ||
c2d81644 ON |
209 | long prepare_to_wait_event(wait_queue_head_t *q, wait_queue_t *wait, int state) |
210 | { | |
211 | unsigned long flags; | |
b1ea06a9 | 212 | long ret = 0; |
c2d81644 | 213 | |
c2d81644 | 214 | spin_lock_irqsave(&q->lock, flags); |
b1ea06a9 ON |
215 | if (unlikely(signal_pending_state(state, current))) { |
216 | /* | |
217 | * Exclusive waiter must not fail if it was selected by wakeup, | |
218 | * it should "consume" the condition we were waiting for. | |
219 | * | |
220 | * The caller will recheck the condition and return success if | |
221 | * we were already woken up, we can not miss the event because | |
222 | * wakeup locks/unlocks the same q->lock. | |
223 | * | |
224 | * But we need to ensure that set-condition + wakeup after that | |
225 | * can't see us, it should wake up another exclusive waiter if | |
226 | * we fail. | |
227 | */ | |
228 | list_del_init(&wait->task_list); | |
229 | ret = -ERESTARTSYS; | |
230 | } else { | |
231 | if (list_empty(&wait->task_list)) { | |
232 | if (wait->flags & WQ_FLAG_EXCLUSIVE) | |
233 | __add_wait_queue_tail(q, wait); | |
234 | else | |
235 | __add_wait_queue(q, wait); | |
236 | } | |
237 | set_current_state(state); | |
c2d81644 | 238 | } |
c2d81644 ON |
239 | spin_unlock_irqrestore(&q->lock, flags); |
240 | ||
b1ea06a9 | 241 | return ret; |
c2d81644 ON |
242 | } |
243 | EXPORT_SYMBOL(prepare_to_wait_event); | |
244 | ||
bd0f9b35 LT |
245 | /* |
246 | * Note! These two wait functions are entered with the | |
247 | * wait-queue lock held (and interrupts off in the _irq | |
248 | * case), so there is no race with testing the wakeup | |
249 | * condition in the caller before they add the wait | |
250 | * entry to the wake queue. | |
251 | */ | |
252 | int do_wait_intr(wait_queue_head_t *wq, wait_queue_t *wait) | |
253 | { | |
254 | if (likely(list_empty(&wait->task_list))) | |
255 | __add_wait_queue_tail(wq, wait); | |
256 | ||
257 | set_current_state(TASK_INTERRUPTIBLE); | |
258 | if (signal_pending(current)) | |
259 | return -ERESTARTSYS; | |
260 | ||
261 | spin_unlock(&wq->lock); | |
262 | schedule(); | |
263 | spin_lock(&wq->lock); | |
264 | return 0; | |
265 | } | |
266 | EXPORT_SYMBOL(do_wait_intr); | |
267 | ||
268 | int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_t *wait) | |
269 | { | |
270 | if (likely(list_empty(&wait->task_list))) | |
271 | __add_wait_queue_tail(wq, wait); | |
272 | ||
273 | set_current_state(TASK_INTERRUPTIBLE); | |
274 | if (signal_pending(current)) | |
275 | return -ERESTARTSYS; | |
276 | ||
277 | spin_unlock_irq(&wq->lock); | |
278 | schedule(); | |
279 | spin_lock_irq(&wq->lock); | |
280 | return 0; | |
281 | } | |
282 | EXPORT_SYMBOL(do_wait_intr_irq); | |
283 | ||
ee2f154a | 284 | /** |
777c6c5f JW |
285 | * finish_wait - clean up after waiting in a queue |
286 | * @q: waitqueue waited on | |
287 | * @wait: wait descriptor | |
288 | * | |
289 | * Sets current thread back to running state and removes | |
290 | * the wait descriptor from the given waitqueue if still | |
291 | * queued. | |
292 | */ | |
7ad5b3a5 | 293 | void finish_wait(wait_queue_head_t *q, wait_queue_t *wait) |
1da177e4 LT |
294 | { |
295 | unsigned long flags; | |
296 | ||
297 | __set_current_state(TASK_RUNNING); | |
298 | /* | |
299 | * We can check for list emptiness outside the lock | |
300 | * IFF: | |
301 | * - we use the "careful" check that verifies both | |
302 | * the next and prev pointers, so that there cannot | |
303 | * be any half-pending updates in progress on other | |
304 | * CPU's that we haven't seen yet (and that might | |
305 | * still change the stack area. | |
306 | * and | |
307 | * - all other users take the lock (ie we can only | |
308 | * have _one_ other CPU that looks at or modifies | |
309 | * the list). | |
310 | */ | |
311 | if (!list_empty_careful(&wait->task_list)) { | |
312 | spin_lock_irqsave(&q->lock, flags); | |
313 | list_del_init(&wait->task_list); | |
314 | spin_unlock_irqrestore(&q->lock, flags); | |
315 | } | |
316 | } | |
317 | EXPORT_SYMBOL(finish_wait); | |
318 | ||
319 | int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
320 | { | |
321 | int ret = default_wake_function(wait, mode, sync, key); | |
322 | ||
323 | if (ret) | |
324 | list_del_init(&wait->task_list); | |
325 | return ret; | |
326 | } | |
327 | EXPORT_SYMBOL(autoremove_wake_function); | |
328 | ||
cb6538e7 PZ |
329 | static inline bool is_kthread_should_stop(void) |
330 | { | |
331 | return (current->flags & PF_KTHREAD) && kthread_should_stop(); | |
332 | } | |
61ada528 PZ |
333 | |
334 | /* | |
335 | * DEFINE_WAIT_FUNC(wait, woken_wake_func); | |
336 | * | |
337 | * add_wait_queue(&wq, &wait); | |
338 | * for (;;) { | |
339 | * if (condition) | |
340 | * break; | |
341 | * | |
342 | * p->state = mode; condition = true; | |
343 | * smp_mb(); // A smp_wmb(); // C | |
344 | * if (!wait->flags & WQ_FLAG_WOKEN) wait->flags |= WQ_FLAG_WOKEN; | |
345 | * schedule() try_to_wake_up(); | |
346 | * p->state = TASK_RUNNING; ~~~~~~~~~~~~~~~~~~ | |
347 | * wait->flags &= ~WQ_FLAG_WOKEN; condition = true; | |
348 | * smp_mb() // B smp_wmb(); // C | |
349 | * wait->flags |= WQ_FLAG_WOKEN; | |
350 | * } | |
351 | * remove_wait_queue(&wq, &wait); | |
352 | * | |
353 | */ | |
354 | long wait_woken(wait_queue_t *wait, unsigned mode, long timeout) | |
355 | { | |
356 | set_current_state(mode); /* A */ | |
357 | /* | |
358 | * The above implies an smp_mb(), which matches with the smp_wmb() from | |
359 | * woken_wake_function() such that if we observe WQ_FLAG_WOKEN we must | |
360 | * also observe all state before the wakeup. | |
361 | */ | |
cb6538e7 | 362 | if (!(wait->flags & WQ_FLAG_WOKEN) && !is_kthread_should_stop()) |
61ada528 PZ |
363 | timeout = schedule_timeout(timeout); |
364 | __set_current_state(TASK_RUNNING); | |
365 | ||
366 | /* | |
367 | * The below implies an smp_mb(), it too pairs with the smp_wmb() from | |
368 | * woken_wake_function() such that we must either observe the wait | |
369 | * condition being true _OR_ WQ_FLAG_WOKEN such that we will not miss | |
370 | * an event. | |
371 | */ | |
b92b8b35 | 372 | smp_store_mb(wait->flags, wait->flags & ~WQ_FLAG_WOKEN); /* B */ |
61ada528 PZ |
373 | |
374 | return timeout; | |
375 | } | |
376 | EXPORT_SYMBOL(wait_woken); | |
377 | ||
378 | int woken_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key) | |
379 | { | |
380 | /* | |
381 | * Although this function is called under waitqueue lock, LOCK | |
382 | * doesn't imply write barrier and the users expects write | |
383 | * barrier semantics on wakeup functions. The following | |
384 | * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() | |
b92b8b35 | 385 | * and is paired with smp_store_mb() in wait_woken(). |
61ada528 PZ |
386 | */ |
387 | smp_wmb(); /* C */ | |
388 | wait->flags |= WQ_FLAG_WOKEN; | |
389 | ||
390 | return default_wake_function(wait, mode, sync, key); | |
391 | } | |
392 | EXPORT_SYMBOL(woken_wake_function); | |
393 | ||
1da177e4 LT |
394 | int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg) |
395 | { | |
396 | struct wait_bit_key *key = arg; | |
397 | struct wait_bit_queue *wait_bit | |
398 | = container_of(wait, struct wait_bit_queue, wait); | |
399 | ||
400 | if (wait_bit->key.flags != key->flags || | |
401 | wait_bit->key.bit_nr != key->bit_nr || | |
402 | test_bit(key->bit_nr, key->flags)) | |
403 | return 0; | |
404 | else | |
405 | return autoremove_wake_function(wait, mode, sync, key); | |
406 | } | |
407 | EXPORT_SYMBOL(wake_bit_function); | |
408 | ||
409 | /* | |
410 | * To allow interruptible waiting and asynchronous (i.e. nonblocking) | |
411 | * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are | |
412 | * permitted return codes. Nonzero return codes halt waiting and return. | |
413 | */ | |
7ad5b3a5 | 414 | int __sched |
1da177e4 | 415 | __wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q, |
c1221321 | 416 | wait_bit_action_f *action, unsigned mode) |
1da177e4 LT |
417 | { |
418 | int ret = 0; | |
419 | ||
420 | do { | |
421 | prepare_to_wait(wq, &q->wait, mode); | |
422 | if (test_bit(q->key.bit_nr, q->key.flags)) | |
dfd01f02 | 423 | ret = (*action)(&q->key, mode); |
1da177e4 LT |
424 | } while (test_bit(q->key.bit_nr, q->key.flags) && !ret); |
425 | finish_wait(wq, &q->wait); | |
426 | return ret; | |
427 | } | |
428 | EXPORT_SYMBOL(__wait_on_bit); | |
429 | ||
7ad5b3a5 | 430 | int __sched out_of_line_wait_on_bit(void *word, int bit, |
c1221321 | 431 | wait_bit_action_f *action, unsigned mode) |
1da177e4 LT |
432 | { |
433 | wait_queue_head_t *wq = bit_waitqueue(word, bit); | |
434 | DEFINE_WAIT_BIT(wait, word, bit); | |
435 | ||
436 | return __wait_on_bit(wq, &wait, action, mode); | |
437 | } | |
438 | EXPORT_SYMBOL(out_of_line_wait_on_bit); | |
439 | ||
cbbce822 N |
440 | int __sched out_of_line_wait_on_bit_timeout( |
441 | void *word, int bit, wait_bit_action_f *action, | |
442 | unsigned mode, unsigned long timeout) | |
443 | { | |
444 | wait_queue_head_t *wq = bit_waitqueue(word, bit); | |
445 | DEFINE_WAIT_BIT(wait, word, bit); | |
446 | ||
447 | wait.key.timeout = jiffies + timeout; | |
448 | return __wait_on_bit(wq, &wait, action, mode); | |
449 | } | |
450 | EXPORT_SYMBOL_GPL(out_of_line_wait_on_bit_timeout); | |
451 | ||
7ad5b3a5 | 452 | int __sched |
1da177e4 | 453 | __wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q, |
c1221321 | 454 | wait_bit_action_f *action, unsigned mode) |
1da177e4 | 455 | { |
eaf9ef52 | 456 | int ret = 0; |
777c6c5f | 457 | |
eaf9ef52 | 458 | for (;;) { |
1da177e4 | 459 | prepare_to_wait_exclusive(wq, &q->wait, mode); |
eaf9ef52 ON |
460 | if (test_bit(q->key.bit_nr, q->key.flags)) { |
461 | ret = action(&q->key, mode); | |
462 | /* | |
463 | * See the comment in prepare_to_wait_event(). | |
464 | * finish_wait() does not necessarily takes wq->lock, | |
465 | * but test_and_set_bit() implies mb() which pairs with | |
466 | * smp_mb__after_atomic() before wake_up_page(). | |
467 | */ | |
468 | if (ret) | |
469 | finish_wait(wq, &q->wait); | |
470 | } | |
471 | if (!test_and_set_bit(q->key.bit_nr, q->key.flags)) { | |
472 | if (!ret) | |
473 | finish_wait(wq, &q->wait); | |
474 | return 0; | |
475 | } else if (ret) { | |
476 | return ret; | |
477 | } | |
478 | } | |
1da177e4 LT |
479 | } |
480 | EXPORT_SYMBOL(__wait_on_bit_lock); | |
481 | ||
7ad5b3a5 | 482 | int __sched out_of_line_wait_on_bit_lock(void *word, int bit, |
c1221321 | 483 | wait_bit_action_f *action, unsigned mode) |
1da177e4 LT |
484 | { |
485 | wait_queue_head_t *wq = bit_waitqueue(word, bit); | |
486 | DEFINE_WAIT_BIT(wait, word, bit); | |
487 | ||
488 | return __wait_on_bit_lock(wq, &wait, action, mode); | |
489 | } | |
490 | EXPORT_SYMBOL(out_of_line_wait_on_bit_lock); | |
491 | ||
7ad5b3a5 | 492 | void __wake_up_bit(wait_queue_head_t *wq, void *word, int bit) |
1da177e4 LT |
493 | { |
494 | struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit); | |
495 | if (waitqueue_active(wq)) | |
e64d66c8 | 496 | __wake_up(wq, TASK_NORMAL, 1, &key); |
1da177e4 LT |
497 | } |
498 | EXPORT_SYMBOL(__wake_up_bit); | |
499 | ||
500 | /** | |
501 | * wake_up_bit - wake up a waiter on a bit | |
502 | * @word: the word being waited on, a kernel virtual address | |
503 | * @bit: the bit of the word being waited on | |
504 | * | |
505 | * There is a standard hashed waitqueue table for generic use. This | |
506 | * is the part of the hashtable's accessor API that wakes up waiters | |
507 | * on a bit. For instance, if one were to have waiters on a bitflag, | |
508 | * one would call wake_up_bit() after clearing the bit. | |
509 | * | |
510 | * In order for this to function properly, as it uses waitqueue_active() | |
511 | * internally, some kind of memory barrier must be done prior to calling | |
4e857c58 | 512 | * this. Typically, this will be smp_mb__after_atomic(), but in some |
1da177e4 LT |
513 | * cases where bitflags are manipulated non-atomically under a lock, one |
514 | * may need to use a less regular barrier, such fs/inode.c's smp_mb(), | |
515 | * because spin_unlock() does not guarantee a memory barrier. | |
516 | */ | |
7ad5b3a5 | 517 | void wake_up_bit(void *word, int bit) |
1da177e4 LT |
518 | { |
519 | __wake_up_bit(bit_waitqueue(word, bit), word, bit); | |
520 | } | |
521 | EXPORT_SYMBOL(wake_up_bit); | |
522 | ||
cb65537e DH |
523 | /* |
524 | * Manipulate the atomic_t address to produce a better bit waitqueue table hash | |
525 | * index (we're keying off bit -1, but that would produce a horrible hash | |
526 | * value). | |
527 | */ | |
528 | static inline wait_queue_head_t *atomic_t_waitqueue(atomic_t *p) | |
529 | { | |
530 | if (BITS_PER_LONG == 64) { | |
531 | unsigned long q = (unsigned long)p; | |
532 | return bit_waitqueue((void *)(q & ~1), q & 1); | |
533 | } | |
534 | return bit_waitqueue(p, 0); | |
535 | } | |
536 | ||
537 | static int wake_atomic_t_function(wait_queue_t *wait, unsigned mode, int sync, | |
538 | void *arg) | |
539 | { | |
540 | struct wait_bit_key *key = arg; | |
541 | struct wait_bit_queue *wait_bit | |
542 | = container_of(wait, struct wait_bit_queue, wait); | |
543 | atomic_t *val = key->flags; | |
544 | ||
545 | if (wait_bit->key.flags != key->flags || | |
546 | wait_bit->key.bit_nr != key->bit_nr || | |
547 | atomic_read(val) != 0) | |
548 | return 0; | |
549 | return autoremove_wake_function(wait, mode, sync, key); | |
550 | } | |
551 | ||
552 | /* | |
553 | * To allow interruptible waiting and asynchronous (i.e. nonblocking) waiting, | |
554 | * the actions of __wait_on_atomic_t() are permitted return codes. Nonzero | |
555 | * return codes halt waiting and return. | |
556 | */ | |
557 | static __sched | |
558 | int __wait_on_atomic_t(wait_queue_head_t *wq, struct wait_bit_queue *q, | |
559 | int (*action)(atomic_t *), unsigned mode) | |
560 | { | |
561 | atomic_t *val; | |
562 | int ret = 0; | |
563 | ||
564 | do { | |
565 | prepare_to_wait(wq, &q->wait, mode); | |
566 | val = q->key.flags; | |
567 | if (atomic_read(val) == 0) | |
42577ca8 DH |
568 | break; |
569 | ret = (*action)(val); | |
cb65537e DH |
570 | } while (!ret && atomic_read(val) != 0); |
571 | finish_wait(wq, &q->wait); | |
572 | return ret; | |
573 | } | |
574 | ||
575 | #define DEFINE_WAIT_ATOMIC_T(name, p) \ | |
576 | struct wait_bit_queue name = { \ | |
577 | .key = __WAIT_ATOMIC_T_KEY_INITIALIZER(p), \ | |
578 | .wait = { \ | |
579 | .private = current, \ | |
580 | .func = wake_atomic_t_function, \ | |
581 | .task_list = \ | |
582 | LIST_HEAD_INIT((name).wait.task_list), \ | |
583 | }, \ | |
584 | } | |
585 | ||
586 | __sched int out_of_line_wait_on_atomic_t(atomic_t *p, int (*action)(atomic_t *), | |
587 | unsigned mode) | |
588 | { | |
589 | wait_queue_head_t *wq = atomic_t_waitqueue(p); | |
590 | DEFINE_WAIT_ATOMIC_T(wait, p); | |
591 | ||
592 | return __wait_on_atomic_t(wq, &wait, action, mode); | |
593 | } | |
594 | EXPORT_SYMBOL(out_of_line_wait_on_atomic_t); | |
595 | ||
596 | /** | |
597 | * wake_up_atomic_t - Wake up a waiter on a atomic_t | |
2203547f | 598 | * @p: The atomic_t being waited on, a kernel virtual address |
cb65537e DH |
599 | * |
600 | * Wake up anyone waiting for the atomic_t to go to zero. | |
601 | * | |
602 | * Abuse the bit-waker function and its waitqueue hash table set (the atomic_t | |
603 | * check is done by the waiter's wake function, not the by the waker itself). | |
604 | */ | |
605 | void wake_up_atomic_t(atomic_t *p) | |
606 | { | |
607 | __wake_up_bit(atomic_t_waitqueue(p), p, WAIT_ATOMIC_T_BIT_NR); | |
608 | } | |
609 | EXPORT_SYMBOL(wake_up_atomic_t); | |
74316201 | 610 | |
dfd01f02 | 611 | __sched int bit_wait(struct wait_bit_key *word, int mode) |
74316201 | 612 | { |
74316201 | 613 | schedule(); |
dfd01f02 | 614 | if (signal_pending_state(mode, current)) |
68985633 | 615 | return -EINTR; |
74316201 N |
616 | return 0; |
617 | } | |
618 | EXPORT_SYMBOL(bit_wait); | |
619 | ||
dfd01f02 | 620 | __sched int bit_wait_io(struct wait_bit_key *word, int mode) |
74316201 | 621 | { |
74316201 | 622 | io_schedule(); |
dfd01f02 | 623 | if (signal_pending_state(mode, current)) |
68985633 | 624 | return -EINTR; |
74316201 N |
625 | return 0; |
626 | } | |
627 | EXPORT_SYMBOL(bit_wait_io); | |
cbbce822 | 628 | |
dfd01f02 | 629 | __sched int bit_wait_timeout(struct wait_bit_key *word, int mode) |
cbbce822 | 630 | { |
316c1608 | 631 | unsigned long now = READ_ONCE(jiffies); |
cbbce822 N |
632 | if (time_after_eq(now, word->timeout)) |
633 | return -EAGAIN; | |
634 | schedule_timeout(word->timeout - now); | |
dfd01f02 | 635 | if (signal_pending_state(mode, current)) |
68985633 | 636 | return -EINTR; |
cbbce822 N |
637 | return 0; |
638 | } | |
639 | EXPORT_SYMBOL_GPL(bit_wait_timeout); | |
640 | ||
dfd01f02 | 641 | __sched int bit_wait_io_timeout(struct wait_bit_key *word, int mode) |
cbbce822 | 642 | { |
316c1608 | 643 | unsigned long now = READ_ONCE(jiffies); |
cbbce822 N |
644 | if (time_after_eq(now, word->timeout)) |
645 | return -EAGAIN; | |
646 | io_schedule_timeout(word->timeout - now); | |
dfd01f02 | 647 | if (signal_pending_state(mode, current)) |
68985633 | 648 | return -EINTR; |
cbbce822 N |
649 | return 0; |
650 | } | |
651 | EXPORT_SYMBOL_GPL(bit_wait_io_timeout); |