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> | |
12 | ||
f07fdec5 | 13 | void __init_waitqueue_head(wait_queue_head_t *q, const char *name, struct lock_class_key *key) |
21d71f51 IM |
14 | { |
15 | spin_lock_init(&q->lock); | |
f07fdec5 | 16 | lockdep_set_class_and_name(&q->lock, key, name); |
21d71f51 IM |
17 | INIT_LIST_HEAD(&q->task_list); |
18 | } | |
eb4542b9 | 19 | |
2fc39111 | 20 | EXPORT_SYMBOL(__init_waitqueue_head); |
eb4542b9 | 21 | |
7ad5b3a5 | 22 | void add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait) |
1da177e4 LT |
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 | } | |
31 | EXPORT_SYMBOL(add_wait_queue); | |
32 | ||
7ad5b3a5 | 33 | void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait) |
1da177e4 LT |
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 | } | |
42 | EXPORT_SYMBOL(add_wait_queue_exclusive); | |
43 | ||
7ad5b3a5 | 44 | void remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait) |
1da177e4 LT |
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 | } | |
52 | EXPORT_SYMBOL(remove_wait_queue); | |
53 | ||
54 | ||
55 | /* | |
56 | * Note: we use "set_current_state()" _after_ the wait-queue add, | |
57 | * because we need a memory barrier there on SMP, so that any | |
58 | * wake-function that tests for the wait-queue being active | |
59 | * will be guaranteed to see waitqueue addition _or_ subsequent | |
60 | * tests in this thread will see the wakeup having taken place. | |
61 | * | |
62 | * The spin_unlock() itself is semi-permeable and only protects | |
63 | * one way (it only protects stuff inside the critical region and | |
64 | * stops them from bleeding out - it would still allow subsequent | |
59c51591 | 65 | * loads to move into the critical region). |
1da177e4 | 66 | */ |
7ad5b3a5 | 67 | void |
1da177e4 LT |
68 | prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state) |
69 | { | |
70 | unsigned long flags; | |
71 | ||
72 | wait->flags &= ~WQ_FLAG_EXCLUSIVE; | |
73 | spin_lock_irqsave(&q->lock, flags); | |
74 | if (list_empty(&wait->task_list)) | |
75 | __add_wait_queue(q, wait); | |
a25d644f | 76 | set_current_state(state); |
1da177e4 LT |
77 | spin_unlock_irqrestore(&q->lock, flags); |
78 | } | |
79 | EXPORT_SYMBOL(prepare_to_wait); | |
80 | ||
7ad5b3a5 | 81 | void |
1da177e4 LT |
82 | prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state) |
83 | { | |
84 | unsigned long flags; | |
85 | ||
86 | wait->flags |= WQ_FLAG_EXCLUSIVE; | |
87 | spin_lock_irqsave(&q->lock, flags); | |
88 | if (list_empty(&wait->task_list)) | |
89 | __add_wait_queue_tail(q, wait); | |
a25d644f | 90 | set_current_state(state); |
1da177e4 LT |
91 | spin_unlock_irqrestore(&q->lock, flags); |
92 | } | |
93 | EXPORT_SYMBOL(prepare_to_wait_exclusive); | |
94 | ||
c2d81644 ON |
95 | long prepare_to_wait_event(wait_queue_head_t *q, wait_queue_t *wait, int state) |
96 | { | |
97 | unsigned long flags; | |
98 | ||
99 | if (signal_pending_state(state, current)) | |
100 | return -ERESTARTSYS; | |
101 | ||
102 | wait->private = current; | |
103 | wait->func = autoremove_wake_function; | |
104 | ||
105 | spin_lock_irqsave(&q->lock, flags); | |
106 | if (list_empty(&wait->task_list)) { | |
107 | if (wait->flags & WQ_FLAG_EXCLUSIVE) | |
108 | __add_wait_queue_tail(q, wait); | |
109 | else | |
110 | __add_wait_queue(q, wait); | |
111 | } | |
112 | set_current_state(state); | |
113 | spin_unlock_irqrestore(&q->lock, flags); | |
114 | ||
115 | return 0; | |
116 | } | |
117 | EXPORT_SYMBOL(prepare_to_wait_event); | |
118 | ||
ee2f154a | 119 | /** |
777c6c5f JW |
120 | * finish_wait - clean up after waiting in a queue |
121 | * @q: waitqueue waited on | |
122 | * @wait: wait descriptor | |
123 | * | |
124 | * Sets current thread back to running state and removes | |
125 | * the wait descriptor from the given waitqueue if still | |
126 | * queued. | |
127 | */ | |
7ad5b3a5 | 128 | void finish_wait(wait_queue_head_t *q, wait_queue_t *wait) |
1da177e4 LT |
129 | { |
130 | unsigned long flags; | |
131 | ||
132 | __set_current_state(TASK_RUNNING); | |
133 | /* | |
134 | * We can check for list emptiness outside the lock | |
135 | * IFF: | |
136 | * - we use the "careful" check that verifies both | |
137 | * the next and prev pointers, so that there cannot | |
138 | * be any half-pending updates in progress on other | |
139 | * CPU's that we haven't seen yet (and that might | |
140 | * still change the stack area. | |
141 | * and | |
142 | * - all other users take the lock (ie we can only | |
143 | * have _one_ other CPU that looks at or modifies | |
144 | * the list). | |
145 | */ | |
146 | if (!list_empty_careful(&wait->task_list)) { | |
147 | spin_lock_irqsave(&q->lock, flags); | |
148 | list_del_init(&wait->task_list); | |
149 | spin_unlock_irqrestore(&q->lock, flags); | |
150 | } | |
151 | } | |
152 | EXPORT_SYMBOL(finish_wait); | |
153 | ||
ee2f154a | 154 | /** |
777c6c5f JW |
155 | * abort_exclusive_wait - abort exclusive waiting in a queue |
156 | * @q: waitqueue waited on | |
157 | * @wait: wait descriptor | |
ee2f154a | 158 | * @mode: runstate of the waiter to be woken |
777c6c5f JW |
159 | * @key: key to identify a wait bit queue or %NULL |
160 | * | |
161 | * Sets current thread back to running state and removes | |
162 | * the wait descriptor from the given waitqueue if still | |
163 | * queued. | |
164 | * | |
165 | * Wakes up the next waiter if the caller is concurrently | |
166 | * woken up through the queue. | |
167 | * | |
168 | * This prevents waiter starvation where an exclusive waiter | |
25985edc | 169 | * aborts and is woken up concurrently and no one wakes up |
777c6c5f JW |
170 | * the next waiter. |
171 | */ | |
172 | void abort_exclusive_wait(wait_queue_head_t *q, wait_queue_t *wait, | |
173 | unsigned int mode, void *key) | |
174 | { | |
175 | unsigned long flags; | |
176 | ||
177 | __set_current_state(TASK_RUNNING); | |
178 | spin_lock_irqsave(&q->lock, flags); | |
179 | if (!list_empty(&wait->task_list)) | |
180 | list_del_init(&wait->task_list); | |
181 | else if (waitqueue_active(q)) | |
78ddb08f | 182 | __wake_up_locked_key(q, mode, key); |
777c6c5f JW |
183 | spin_unlock_irqrestore(&q->lock, flags); |
184 | } | |
185 | EXPORT_SYMBOL(abort_exclusive_wait); | |
186 | ||
1da177e4 LT |
187 | int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key) |
188 | { | |
189 | int ret = default_wake_function(wait, mode, sync, key); | |
190 | ||
191 | if (ret) | |
192 | list_del_init(&wait->task_list); | |
193 | return ret; | |
194 | } | |
195 | EXPORT_SYMBOL(autoremove_wake_function); | |
196 | ||
197 | int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg) | |
198 | { | |
199 | struct wait_bit_key *key = arg; | |
200 | struct wait_bit_queue *wait_bit | |
201 | = container_of(wait, struct wait_bit_queue, wait); | |
202 | ||
203 | if (wait_bit->key.flags != key->flags || | |
204 | wait_bit->key.bit_nr != key->bit_nr || | |
205 | test_bit(key->bit_nr, key->flags)) | |
206 | return 0; | |
207 | else | |
208 | return autoremove_wake_function(wait, mode, sync, key); | |
209 | } | |
210 | EXPORT_SYMBOL(wake_bit_function); | |
211 | ||
212 | /* | |
213 | * To allow interruptible waiting and asynchronous (i.e. nonblocking) | |
214 | * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are | |
215 | * permitted return codes. Nonzero return codes halt waiting and return. | |
216 | */ | |
7ad5b3a5 | 217 | int __sched |
1da177e4 LT |
218 | __wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q, |
219 | int (*action)(void *), unsigned mode) | |
220 | { | |
221 | int ret = 0; | |
222 | ||
223 | do { | |
224 | prepare_to_wait(wq, &q->wait, mode); | |
225 | if (test_bit(q->key.bit_nr, q->key.flags)) | |
226 | ret = (*action)(q->key.flags); | |
227 | } while (test_bit(q->key.bit_nr, q->key.flags) && !ret); | |
228 | finish_wait(wq, &q->wait); | |
229 | return ret; | |
230 | } | |
231 | EXPORT_SYMBOL(__wait_on_bit); | |
232 | ||
7ad5b3a5 | 233 | int __sched out_of_line_wait_on_bit(void *word, int bit, |
1da177e4 LT |
234 | int (*action)(void *), unsigned mode) |
235 | { | |
236 | wait_queue_head_t *wq = bit_waitqueue(word, bit); | |
237 | DEFINE_WAIT_BIT(wait, word, bit); | |
238 | ||
239 | return __wait_on_bit(wq, &wait, action, mode); | |
240 | } | |
241 | EXPORT_SYMBOL(out_of_line_wait_on_bit); | |
242 | ||
7ad5b3a5 | 243 | int __sched |
1da177e4 LT |
244 | __wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q, |
245 | int (*action)(void *), unsigned mode) | |
246 | { | |
1da177e4 | 247 | do { |
777c6c5f JW |
248 | int ret; |
249 | ||
1da177e4 | 250 | prepare_to_wait_exclusive(wq, &q->wait, mode); |
777c6c5f JW |
251 | if (!test_bit(q->key.bit_nr, q->key.flags)) |
252 | continue; | |
253 | ret = action(q->key.flags); | |
254 | if (!ret) | |
255 | continue; | |
256 | abort_exclusive_wait(wq, &q->wait, mode, &q->key); | |
257 | return ret; | |
1da177e4 LT |
258 | } while (test_and_set_bit(q->key.bit_nr, q->key.flags)); |
259 | finish_wait(wq, &q->wait); | |
777c6c5f | 260 | return 0; |
1da177e4 LT |
261 | } |
262 | EXPORT_SYMBOL(__wait_on_bit_lock); | |
263 | ||
7ad5b3a5 | 264 | int __sched out_of_line_wait_on_bit_lock(void *word, int bit, |
1da177e4 LT |
265 | int (*action)(void *), unsigned mode) |
266 | { | |
267 | wait_queue_head_t *wq = bit_waitqueue(word, bit); | |
268 | DEFINE_WAIT_BIT(wait, word, bit); | |
269 | ||
270 | return __wait_on_bit_lock(wq, &wait, action, mode); | |
271 | } | |
272 | EXPORT_SYMBOL(out_of_line_wait_on_bit_lock); | |
273 | ||
7ad5b3a5 | 274 | void __wake_up_bit(wait_queue_head_t *wq, void *word, int bit) |
1da177e4 LT |
275 | { |
276 | struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit); | |
277 | if (waitqueue_active(wq)) | |
e64d66c8 | 278 | __wake_up(wq, TASK_NORMAL, 1, &key); |
1da177e4 LT |
279 | } |
280 | EXPORT_SYMBOL(__wake_up_bit); | |
281 | ||
282 | /** | |
283 | * wake_up_bit - wake up a waiter on a bit | |
284 | * @word: the word being waited on, a kernel virtual address | |
285 | * @bit: the bit of the word being waited on | |
286 | * | |
287 | * There is a standard hashed waitqueue table for generic use. This | |
288 | * is the part of the hashtable's accessor API that wakes up waiters | |
289 | * on a bit. For instance, if one were to have waiters on a bitflag, | |
290 | * one would call wake_up_bit() after clearing the bit. | |
291 | * | |
292 | * In order for this to function properly, as it uses waitqueue_active() | |
293 | * internally, some kind of memory barrier must be done prior to calling | |
294 | * this. Typically, this will be smp_mb__after_clear_bit(), but in some | |
295 | * cases where bitflags are manipulated non-atomically under a lock, one | |
296 | * may need to use a less regular barrier, such fs/inode.c's smp_mb(), | |
297 | * because spin_unlock() does not guarantee a memory barrier. | |
298 | */ | |
7ad5b3a5 | 299 | void wake_up_bit(void *word, int bit) |
1da177e4 LT |
300 | { |
301 | __wake_up_bit(bit_waitqueue(word, bit), word, bit); | |
302 | } | |
303 | EXPORT_SYMBOL(wake_up_bit); | |
304 | ||
7ad5b3a5 | 305 | wait_queue_head_t *bit_waitqueue(void *word, int bit) |
1da177e4 LT |
306 | { |
307 | const int shift = BITS_PER_LONG == 32 ? 5 : 6; | |
308 | const struct zone *zone = page_zone(virt_to_page(word)); | |
309 | unsigned long val = (unsigned long)word << shift | bit; | |
310 | ||
311 | return &zone->wait_table[hash_long(val, zone->wait_table_bits)]; | |
312 | } | |
313 | EXPORT_SYMBOL(bit_waitqueue); | |
cb65537e DH |
314 | |
315 | /* | |
316 | * Manipulate the atomic_t address to produce a better bit waitqueue table hash | |
317 | * index (we're keying off bit -1, but that would produce a horrible hash | |
318 | * value). | |
319 | */ | |
320 | static inline wait_queue_head_t *atomic_t_waitqueue(atomic_t *p) | |
321 | { | |
322 | if (BITS_PER_LONG == 64) { | |
323 | unsigned long q = (unsigned long)p; | |
324 | return bit_waitqueue((void *)(q & ~1), q & 1); | |
325 | } | |
326 | return bit_waitqueue(p, 0); | |
327 | } | |
328 | ||
329 | static int wake_atomic_t_function(wait_queue_t *wait, unsigned mode, int sync, | |
330 | void *arg) | |
331 | { | |
332 | struct wait_bit_key *key = arg; | |
333 | struct wait_bit_queue *wait_bit | |
334 | = container_of(wait, struct wait_bit_queue, wait); | |
335 | atomic_t *val = key->flags; | |
336 | ||
337 | if (wait_bit->key.flags != key->flags || | |
338 | wait_bit->key.bit_nr != key->bit_nr || | |
339 | atomic_read(val) != 0) | |
340 | return 0; | |
341 | return autoremove_wake_function(wait, mode, sync, key); | |
342 | } | |
343 | ||
344 | /* | |
345 | * To allow interruptible waiting and asynchronous (i.e. nonblocking) waiting, | |
346 | * the actions of __wait_on_atomic_t() are permitted return codes. Nonzero | |
347 | * return codes halt waiting and return. | |
348 | */ | |
349 | static __sched | |
350 | int __wait_on_atomic_t(wait_queue_head_t *wq, struct wait_bit_queue *q, | |
351 | int (*action)(atomic_t *), unsigned mode) | |
352 | { | |
353 | atomic_t *val; | |
354 | int ret = 0; | |
355 | ||
356 | do { | |
357 | prepare_to_wait(wq, &q->wait, mode); | |
358 | val = q->key.flags; | |
359 | if (atomic_read(val) == 0) | |
42577ca8 DH |
360 | break; |
361 | ret = (*action)(val); | |
cb65537e DH |
362 | } while (!ret && atomic_read(val) != 0); |
363 | finish_wait(wq, &q->wait); | |
364 | return ret; | |
365 | } | |
366 | ||
367 | #define DEFINE_WAIT_ATOMIC_T(name, p) \ | |
368 | struct wait_bit_queue name = { \ | |
369 | .key = __WAIT_ATOMIC_T_KEY_INITIALIZER(p), \ | |
370 | .wait = { \ | |
371 | .private = current, \ | |
372 | .func = wake_atomic_t_function, \ | |
373 | .task_list = \ | |
374 | LIST_HEAD_INIT((name).wait.task_list), \ | |
375 | }, \ | |
376 | } | |
377 | ||
378 | __sched int out_of_line_wait_on_atomic_t(atomic_t *p, int (*action)(atomic_t *), | |
379 | unsigned mode) | |
380 | { | |
381 | wait_queue_head_t *wq = atomic_t_waitqueue(p); | |
382 | DEFINE_WAIT_ATOMIC_T(wait, p); | |
383 | ||
384 | return __wait_on_atomic_t(wq, &wait, action, mode); | |
385 | } | |
386 | EXPORT_SYMBOL(out_of_line_wait_on_atomic_t); | |
387 | ||
388 | /** | |
389 | * wake_up_atomic_t - Wake up a waiter on a atomic_t | |
2203547f | 390 | * @p: The atomic_t being waited on, a kernel virtual address |
cb65537e DH |
391 | * |
392 | * Wake up anyone waiting for the atomic_t to go to zero. | |
393 | * | |
394 | * Abuse the bit-waker function and its waitqueue hash table set (the atomic_t | |
395 | * check is done by the waiter's wake function, not the by the waker itself). | |
396 | */ | |
397 | void wake_up_atomic_t(atomic_t *p) | |
398 | { | |
399 | __wake_up_bit(atomic_t_waitqueue(p), p, WAIT_ATOMIC_T_BIT_NR); | |
400 | } | |
401 | EXPORT_SYMBOL(wake_up_atomic_t); |