]> git.karo-electronics.de Git - karo-tx-linux.git/blob - kernel/sched/wait.c
ARM: dts: armada-385-linksys: disk-activity trigger for all
[karo-tx-linux.git] / kernel / sched / wait.c
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/signal.h>
9 #include <linux/sched/debug.h>
10 #include <linux/mm.h>
11 #include <linux/wait.h>
12 #include <linux/hash.h>
13 #include <linux/kthread.h>
14
15 void __init_waitqueue_head(wait_queue_head_t *q, const char *name, struct lock_class_key *key)
16 {
17         spin_lock_init(&q->lock);
18         lockdep_set_class_and_name(&q->lock, key, name);
19         INIT_LIST_HEAD(&q->task_list);
20 }
21
22 EXPORT_SYMBOL(__init_waitqueue_head);
23
24 void add_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
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
35 void add_wait_queue_exclusive(wait_queue_head_t *q, wait_queue_t *wait)
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
46 void remove_wait_queue(wait_queue_head_t *q, wait_queue_t *wait)
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
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
110 void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
111 {
112         __wake_up_common(q, mode, 1, 0, key);
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
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
170  * loads to move into the critical region).
171  */
172 void
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);
181         set_current_state(state);
182         spin_unlock_irqrestore(&q->lock, flags);
183 }
184 EXPORT_SYMBOL(prepare_to_wait);
185
186 void
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);
195         set_current_state(state);
196         spin_unlock_irqrestore(&q->lock, flags);
197 }
198 EXPORT_SYMBOL(prepare_to_wait_exclusive);
199
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
209 long prepare_to_wait_event(wait_queue_head_t *q, wait_queue_t *wait, int state)
210 {
211         unsigned long flags;
212         long ret = 0;
213
214         spin_lock_irqsave(&q->lock, flags);
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);
238         }
239         spin_unlock_irqrestore(&q->lock, flags);
240
241         return ret;
242 }
243 EXPORT_SYMBOL(prepare_to_wait_event);
244
245 /**
246  * finish_wait - clean up after waiting in a queue
247  * @q: waitqueue waited on
248  * @wait: wait descriptor
249  *
250  * Sets current thread back to running state and removes
251  * the wait descriptor from the given waitqueue if still
252  * queued.
253  */
254 void finish_wait(wait_queue_head_t *q, wait_queue_t *wait)
255 {
256         unsigned long flags;
257
258         __set_current_state(TASK_RUNNING);
259         /*
260          * We can check for list emptiness outside the lock
261          * IFF:
262          *  - we use the "careful" check that verifies both
263          *    the next and prev pointers, so that there cannot
264          *    be any half-pending updates in progress on other
265          *    CPU's that we haven't seen yet (and that might
266          *    still change the stack area.
267          * and
268          *  - all other users take the lock (ie we can only
269          *    have _one_ other CPU that looks at or modifies
270          *    the list).
271          */
272         if (!list_empty_careful(&wait->task_list)) {
273                 spin_lock_irqsave(&q->lock, flags);
274                 list_del_init(&wait->task_list);
275                 spin_unlock_irqrestore(&q->lock, flags);
276         }
277 }
278 EXPORT_SYMBOL(finish_wait);
279
280 int autoremove_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
281 {
282         int ret = default_wake_function(wait, mode, sync, key);
283
284         if (ret)
285                 list_del_init(&wait->task_list);
286         return ret;
287 }
288 EXPORT_SYMBOL(autoremove_wake_function);
289
290 static inline bool is_kthread_should_stop(void)
291 {
292         return (current->flags & PF_KTHREAD) && kthread_should_stop();
293 }
294
295 /*
296  * DEFINE_WAIT_FUNC(wait, woken_wake_func);
297  *
298  * add_wait_queue(&wq, &wait);
299  * for (;;) {
300  *     if (condition)
301  *         break;
302  *
303  *     p->state = mode;                         condition = true;
304  *     smp_mb(); // A                           smp_wmb(); // C
305  *     if (!wait->flags & WQ_FLAG_WOKEN)        wait->flags |= WQ_FLAG_WOKEN;
306  *         schedule()                           try_to_wake_up();
307  *     p->state = TASK_RUNNING;             ~~~~~~~~~~~~~~~~~~
308  *     wait->flags &= ~WQ_FLAG_WOKEN;           condition = true;
309  *     smp_mb() // B                            smp_wmb(); // C
310  *                                              wait->flags |= WQ_FLAG_WOKEN;
311  * }
312  * remove_wait_queue(&wq, &wait);
313  *
314  */
315 long wait_woken(wait_queue_t *wait, unsigned mode, long timeout)
316 {
317         set_current_state(mode); /* A */
318         /*
319          * The above implies an smp_mb(), which matches with the smp_wmb() from
320          * woken_wake_function() such that if we observe WQ_FLAG_WOKEN we must
321          * also observe all state before the wakeup.
322          */
323         if (!(wait->flags & WQ_FLAG_WOKEN) && !is_kthread_should_stop())
324                 timeout = schedule_timeout(timeout);
325         __set_current_state(TASK_RUNNING);
326
327         /*
328          * The below implies an smp_mb(), it too pairs with the smp_wmb() from
329          * woken_wake_function() such that we must either observe the wait
330          * condition being true _OR_ WQ_FLAG_WOKEN such that we will not miss
331          * an event.
332          */
333         smp_store_mb(wait->flags, wait->flags & ~WQ_FLAG_WOKEN); /* B */
334
335         return timeout;
336 }
337 EXPORT_SYMBOL(wait_woken);
338
339 int woken_wake_function(wait_queue_t *wait, unsigned mode, int sync, void *key)
340 {
341         /*
342          * Although this function is called under waitqueue lock, LOCK
343          * doesn't imply write barrier and the users expects write
344          * barrier semantics on wakeup functions.  The following
345          * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
346          * and is paired with smp_store_mb() in wait_woken().
347          */
348         smp_wmb(); /* C */
349         wait->flags |= WQ_FLAG_WOKEN;
350
351         return default_wake_function(wait, mode, sync, key);
352 }
353 EXPORT_SYMBOL(woken_wake_function);
354
355 int wake_bit_function(wait_queue_t *wait, unsigned mode, int sync, void *arg)
356 {
357         struct wait_bit_key *key = arg;
358         struct wait_bit_queue *wait_bit
359                 = container_of(wait, struct wait_bit_queue, wait);
360
361         if (wait_bit->key.flags != key->flags ||
362                         wait_bit->key.bit_nr != key->bit_nr ||
363                         test_bit(key->bit_nr, key->flags))
364                 return 0;
365         else
366                 return autoremove_wake_function(wait, mode, sync, key);
367 }
368 EXPORT_SYMBOL(wake_bit_function);
369
370 /*
371  * To allow interruptible waiting and asynchronous (i.e. nonblocking)
372  * waiting, the actions of __wait_on_bit() and __wait_on_bit_lock() are
373  * permitted return codes. Nonzero return codes halt waiting and return.
374  */
375 int __sched
376 __wait_on_bit(wait_queue_head_t *wq, struct wait_bit_queue *q,
377               wait_bit_action_f *action, unsigned mode)
378 {
379         int ret = 0;
380
381         do {
382                 prepare_to_wait(wq, &q->wait, mode);
383                 if (test_bit(q->key.bit_nr, q->key.flags))
384                         ret = (*action)(&q->key, mode);
385         } while (test_bit(q->key.bit_nr, q->key.flags) && !ret);
386         finish_wait(wq, &q->wait);
387         return ret;
388 }
389 EXPORT_SYMBOL(__wait_on_bit);
390
391 int __sched out_of_line_wait_on_bit(void *word, int bit,
392                                     wait_bit_action_f *action, unsigned mode)
393 {
394         wait_queue_head_t *wq = bit_waitqueue(word, bit);
395         DEFINE_WAIT_BIT(wait, word, bit);
396
397         return __wait_on_bit(wq, &wait, action, mode);
398 }
399 EXPORT_SYMBOL(out_of_line_wait_on_bit);
400
401 int __sched out_of_line_wait_on_bit_timeout(
402         void *word, int bit, wait_bit_action_f *action,
403         unsigned mode, unsigned long timeout)
404 {
405         wait_queue_head_t *wq = bit_waitqueue(word, bit);
406         DEFINE_WAIT_BIT(wait, word, bit);
407
408         wait.key.timeout = jiffies + timeout;
409         return __wait_on_bit(wq, &wait, action, mode);
410 }
411 EXPORT_SYMBOL_GPL(out_of_line_wait_on_bit_timeout);
412
413 int __sched
414 __wait_on_bit_lock(wait_queue_head_t *wq, struct wait_bit_queue *q,
415                         wait_bit_action_f *action, unsigned mode)
416 {
417         int ret = 0;
418
419         for (;;) {
420                 prepare_to_wait_exclusive(wq, &q->wait, mode);
421                 if (test_bit(q->key.bit_nr, q->key.flags)) {
422                         ret = action(&q->key, mode);
423                         /*
424                          * See the comment in prepare_to_wait_event().
425                          * finish_wait() does not necessarily takes wq->lock,
426                          * but test_and_set_bit() implies mb() which pairs with
427                          * smp_mb__after_atomic() before wake_up_page().
428                          */
429                         if (ret)
430                                 finish_wait(wq, &q->wait);
431                 }
432                 if (!test_and_set_bit(q->key.bit_nr, q->key.flags)) {
433                         if (!ret)
434                                 finish_wait(wq, &q->wait);
435                         return 0;
436                 } else if (ret) {
437                         return ret;
438                 }
439         }
440 }
441 EXPORT_SYMBOL(__wait_on_bit_lock);
442
443 int __sched out_of_line_wait_on_bit_lock(void *word, int bit,
444                                          wait_bit_action_f *action, unsigned mode)
445 {
446         wait_queue_head_t *wq = bit_waitqueue(word, bit);
447         DEFINE_WAIT_BIT(wait, word, bit);
448
449         return __wait_on_bit_lock(wq, &wait, action, mode);
450 }
451 EXPORT_SYMBOL(out_of_line_wait_on_bit_lock);
452
453 void __wake_up_bit(wait_queue_head_t *wq, void *word, int bit)
454 {
455         struct wait_bit_key key = __WAIT_BIT_KEY_INITIALIZER(word, bit);
456         if (waitqueue_active(wq))
457                 __wake_up(wq, TASK_NORMAL, 1, &key);
458 }
459 EXPORT_SYMBOL(__wake_up_bit);
460
461 /**
462  * wake_up_bit - wake up a waiter on a bit
463  * @word: the word being waited on, a kernel virtual address
464  * @bit: the bit of the word being waited on
465  *
466  * There is a standard hashed waitqueue table for generic use. This
467  * is the part of the hashtable's accessor API that wakes up waiters
468  * on a bit. For instance, if one were to have waiters on a bitflag,
469  * one would call wake_up_bit() after clearing the bit.
470  *
471  * In order for this to function properly, as it uses waitqueue_active()
472  * internally, some kind of memory barrier must be done prior to calling
473  * this. Typically, this will be smp_mb__after_atomic(), but in some
474  * cases where bitflags are manipulated non-atomically under a lock, one
475  * may need to use a less regular barrier, such fs/inode.c's smp_mb(),
476  * because spin_unlock() does not guarantee a memory barrier.
477  */
478 void wake_up_bit(void *word, int bit)
479 {
480         __wake_up_bit(bit_waitqueue(word, bit), word, bit);
481 }
482 EXPORT_SYMBOL(wake_up_bit);
483
484 /*
485  * Manipulate the atomic_t address to produce a better bit waitqueue table hash
486  * index (we're keying off bit -1, but that would produce a horrible hash
487  * value).
488  */
489 static inline wait_queue_head_t *atomic_t_waitqueue(atomic_t *p)
490 {
491         if (BITS_PER_LONG == 64) {
492                 unsigned long q = (unsigned long)p;
493                 return bit_waitqueue((void *)(q & ~1), q & 1);
494         }
495         return bit_waitqueue(p, 0);
496 }
497
498 static int wake_atomic_t_function(wait_queue_t *wait, unsigned mode, int sync,
499                                   void *arg)
500 {
501         struct wait_bit_key *key = arg;
502         struct wait_bit_queue *wait_bit
503                 = container_of(wait, struct wait_bit_queue, wait);
504         atomic_t *val = key->flags;
505
506         if (wait_bit->key.flags != key->flags ||
507             wait_bit->key.bit_nr != key->bit_nr ||
508             atomic_read(val) != 0)
509                 return 0;
510         return autoremove_wake_function(wait, mode, sync, key);
511 }
512
513 /*
514  * To allow interruptible waiting and asynchronous (i.e. nonblocking) waiting,
515  * the actions of __wait_on_atomic_t() are permitted return codes.  Nonzero
516  * return codes halt waiting and return.
517  */
518 static __sched
519 int __wait_on_atomic_t(wait_queue_head_t *wq, struct wait_bit_queue *q,
520                        int (*action)(atomic_t *), unsigned mode)
521 {
522         atomic_t *val;
523         int ret = 0;
524
525         do {
526                 prepare_to_wait(wq, &q->wait, mode);
527                 val = q->key.flags;
528                 if (atomic_read(val) == 0)
529                         break;
530                 ret = (*action)(val);
531         } while (!ret && atomic_read(val) != 0);
532         finish_wait(wq, &q->wait);
533         return ret;
534 }
535
536 #define DEFINE_WAIT_ATOMIC_T(name, p)                                   \
537         struct wait_bit_queue name = {                                  \
538                 .key = __WAIT_ATOMIC_T_KEY_INITIALIZER(p),              \
539                 .wait   = {                                             \
540                         .private        = current,                      \
541                         .func           = wake_atomic_t_function,       \
542                         .task_list      =                               \
543                                 LIST_HEAD_INIT((name).wait.task_list),  \
544                 },                                                      \
545         }
546
547 __sched int out_of_line_wait_on_atomic_t(atomic_t *p, int (*action)(atomic_t *),
548                                          unsigned mode)
549 {
550         wait_queue_head_t *wq = atomic_t_waitqueue(p);
551         DEFINE_WAIT_ATOMIC_T(wait, p);
552
553         return __wait_on_atomic_t(wq, &wait, action, mode);
554 }
555 EXPORT_SYMBOL(out_of_line_wait_on_atomic_t);
556
557 /**
558  * wake_up_atomic_t - Wake up a waiter on a atomic_t
559  * @p: The atomic_t being waited on, a kernel virtual address
560  *
561  * Wake up anyone waiting for the atomic_t to go to zero.
562  *
563  * Abuse the bit-waker function and its waitqueue hash table set (the atomic_t
564  * check is done by the waiter's wake function, not the by the waker itself).
565  */
566 void wake_up_atomic_t(atomic_t *p)
567 {
568         __wake_up_bit(atomic_t_waitqueue(p), p, WAIT_ATOMIC_T_BIT_NR);
569 }
570 EXPORT_SYMBOL(wake_up_atomic_t);
571
572 __sched int bit_wait(struct wait_bit_key *word, int mode)
573 {
574         schedule();
575         if (signal_pending_state(mode, current))
576                 return -EINTR;
577         return 0;
578 }
579 EXPORT_SYMBOL(bit_wait);
580
581 __sched int bit_wait_io(struct wait_bit_key *word, int mode)
582 {
583         io_schedule();
584         if (signal_pending_state(mode, current))
585                 return -EINTR;
586         return 0;
587 }
588 EXPORT_SYMBOL(bit_wait_io);
589
590 __sched int bit_wait_timeout(struct wait_bit_key *word, int mode)
591 {
592         unsigned long now = READ_ONCE(jiffies);
593         if (time_after_eq(now, word->timeout))
594                 return -EAGAIN;
595         schedule_timeout(word->timeout - now);
596         if (signal_pending_state(mode, current))
597                 return -EINTR;
598         return 0;
599 }
600 EXPORT_SYMBOL_GPL(bit_wait_timeout);
601
602 __sched int bit_wait_io_timeout(struct wait_bit_key *word, int mode)
603 {
604         unsigned long now = READ_ONCE(jiffies);
605         if (time_after_eq(now, word->timeout))
606                 return -EAGAIN;
607         io_schedule_timeout(word->timeout - now);
608         if (signal_pending_state(mode, current))
609                 return -EINTR;
610         return 0;
611 }
612 EXPORT_SYMBOL_GPL(bit_wait_io_timeout);