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