projects/linux·blob assoc-array-fixes · refs

480 lines · 14.5 KiB

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#include <linux/wait_bit.h>
9
10void __init_waitqueue_head(struct wait_queue_head *wq_head, const char *name, struct lock_class_key *key)
11{
12 spin_lock_init(&wq_head->lock);
13 lockdep_set_class_and_name(&wq_head->lock, key, name);
14 INIT_LIST_HEAD(&wq_head->head);
15}
16
17EXPORT_SYMBOL(__init_waitqueue_head);
18
19void add_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
20{
21 unsigned long flags;
22
23 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
24 spin_lock_irqsave(&wq_head->lock, flags);
25 __add_wait_queue(wq_head, wq_entry);
26 spin_unlock_irqrestore(&wq_head->lock, flags);
27}
28EXPORT_SYMBOL(add_wait_queue);
29
30void add_wait_queue_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
31{
32 unsigned long flags;
33
34 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
35 spin_lock_irqsave(&wq_head->lock, flags);
36 __add_wait_queue_entry_tail(wq_head, wq_entry);
37 spin_unlock_irqrestore(&wq_head->lock, flags);
38}
39EXPORT_SYMBOL(add_wait_queue_exclusive);
40
41void add_wait_queue_priority(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
42{
43 unsigned long flags;
44
45 wq_entry->flags |= WQ_FLAG_PRIORITY;
46 spin_lock_irqsave(&wq_head->lock, flags);
47 __add_wait_queue(wq_head, wq_entry);
48 spin_unlock_irqrestore(&wq_head->lock, flags);
49}
50EXPORT_SYMBOL_GPL(add_wait_queue_priority);
51
52int add_wait_queue_priority_exclusive(struct wait_queue_head *wq_head,
53 struct wait_queue_entry *wq_entry)
54{
55 struct list_head *head = &wq_head->head;
56
57 wq_entry->flags |= WQ_FLAG_EXCLUSIVE | WQ_FLAG_PRIORITY;
58
59 guard(spinlock_irqsave)(&wq_head->lock);
60
61 if (!list_empty(head) &&
62 (list_first_entry(head, typeof(*wq_entry), entry)->flags & WQ_FLAG_PRIORITY))
63 return -EBUSY;
64
65 list_add(&wq_entry->entry, head);
66 return 0;
67}
68EXPORT_SYMBOL_GPL(add_wait_queue_priority_exclusive);
69
70void remove_wait_queue(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
71{
72 unsigned long flags;
73
74 spin_lock_irqsave(&wq_head->lock, flags);
75 __remove_wait_queue(wq_head, wq_entry);
76 spin_unlock_irqrestore(&wq_head->lock, flags);
77}
78EXPORT_SYMBOL(remove_wait_queue);
79
80/*
81 * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just
82 * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve
83 * number) then we wake that number of exclusive tasks, and potentially all
84 * the non-exclusive tasks. Normally, exclusive tasks will be at the end of
85 * the list and any non-exclusive tasks will be woken first. A priority task
86 * may be at the head of the list, and can consume the event without any other
87 * tasks being woken if it's also an exclusive task.
88 *
89 * There are circumstances in which we can try to wake a task which has already
90 * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns
91 * zero in this (rare) case, and we handle it by continuing to scan the queue.
92 */
93static int __wake_up_common(struct wait_queue_head *wq_head, unsigned int mode,
94 int nr_exclusive, int wake_flags, void *key)
95{
96 wait_queue_entry_t *curr, *next;
97
98 lockdep_assert_held(&wq_head->lock);
99
100 curr = list_first_entry(&wq_head->head, wait_queue_entry_t, entry);
101
102 if (&curr->entry == &wq_head->head)
103 return nr_exclusive;
104
105 list_for_each_entry_safe_from(curr, next, &wq_head->head, entry) {
106 unsigned flags = curr->flags;
107 int ret;
108
109 ret = curr->func(curr, mode, wake_flags, key);
110 if (ret < 0)
111 break;
112 if (ret && (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive)
113 break;
114 }
115
116 return nr_exclusive;
117}
118
119static int __wake_up_common_lock(struct wait_queue_head *wq_head, unsigned int mode,
120 int nr_exclusive, int wake_flags, void *key)
121{
122 unsigned long flags;
123 int remaining;
124
125 spin_lock_irqsave(&wq_head->lock, flags);
126 remaining = __wake_up_common(wq_head, mode, nr_exclusive, wake_flags,
127 key);
128 spin_unlock_irqrestore(&wq_head->lock, flags);
129
130 return nr_exclusive - remaining;
131}
132
133/**
134 * __wake_up - wake up threads blocked on a waitqueue.
135 * @wq_head: the waitqueue
136 * @mode: which threads
137 * @nr_exclusive: how many wake-one or wake-many threads to wake up
138 * @key: is directly passed to the wakeup function
139 *
140 * If this function wakes up a task, it executes a full memory barrier
141 * before accessing the task state. Returns the number of exclusive
142 * tasks that were awaken.
143 */
144int __wake_up(struct wait_queue_head *wq_head, unsigned int mode,
145 int nr_exclusive, void *key)
146{
147 return __wake_up_common_lock(wq_head, mode, nr_exclusive, 0, key);
148}
149EXPORT_SYMBOL(__wake_up);
150
151void __wake_up_on_current_cpu(struct wait_queue_head *wq_head, unsigned int mode, void *key)
152{
153 __wake_up_common_lock(wq_head, mode, 1, WF_CURRENT_CPU, key);
154}
155
156/*
157 * Same as __wake_up but called with the spinlock in wait_queue_head_t held.
158 */
159void __wake_up_locked(struct wait_queue_head *wq_head, unsigned int mode, int nr)
160{
161 __wake_up_common(wq_head, mode, nr, 0, NULL);
162}
163EXPORT_SYMBOL_GPL(__wake_up_locked);
164
165void __wake_up_locked_key(struct wait_queue_head *wq_head, unsigned int mode, void *key)
166{
167 __wake_up_common(wq_head, mode, 1, 0, key);
168}
169EXPORT_SYMBOL_GPL(__wake_up_locked_key);
170
171/**
172 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
173 * @wq_head: the waitqueue
174 * @mode: which threads
175 * @key: opaque value to be passed to wakeup targets
176 *
177 * The sync wakeup differs that the waker knows that it will schedule
178 * away soon, so while the target thread will be woken up, it will not
179 * be migrated to another CPU - ie. the two threads are 'synchronized'
180 * with each other. This can prevent needless bouncing between CPUs.
181 *
182 * On UP it can prevent extra preemption.
183 *
184 * If this function wakes up a task, it executes a full memory barrier before
185 * accessing the task state.
186 */
187void __wake_up_sync_key(struct wait_queue_head *wq_head, unsigned int mode,
188 void *key)
189{
190 if (unlikely(!wq_head))
191 return;
192
193 __wake_up_common_lock(wq_head, mode, 1, WF_SYNC, key);
194}
195EXPORT_SYMBOL_GPL(__wake_up_sync_key);
196
197/**
198 * __wake_up_locked_sync_key - wake up a thread blocked on a locked waitqueue.
199 * @wq_head: the waitqueue
200 * @mode: which threads
201 * @key: opaque value to be passed to wakeup targets
202 *
203 * The sync wakeup differs in that the waker knows that it will schedule
204 * away soon, so while the target thread will be woken up, it will not
205 * be migrated to another CPU - ie. the two threads are 'synchronized'
206 * with each other. This can prevent needless bouncing between CPUs.
207 *
208 * On UP it can prevent extra preemption.
209 *
210 * If this function wakes up a task, it executes a full memory barrier before
211 * accessing the task state.
212 */
213void __wake_up_locked_sync_key(struct wait_queue_head *wq_head,
214 unsigned int mode, void *key)
215{
216 __wake_up_common(wq_head, mode, 1, WF_SYNC, key);
217}
218EXPORT_SYMBOL_GPL(__wake_up_locked_sync_key);
219
220/*
221 * __wake_up_sync - see __wake_up_sync_key()
222 */
223void __wake_up_sync(struct wait_queue_head *wq_head, unsigned int mode)
224{
225 __wake_up_sync_key(wq_head, mode, NULL);
226}
227EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
228
229void __wake_up_pollfree(struct wait_queue_head *wq_head)
230{
231 __wake_up(wq_head, TASK_NORMAL, 0, poll_to_key(EPOLLHUP | POLLFREE));
232 /* POLLFREE must have cleared the queue. */
233 WARN_ON_ONCE(waitqueue_active(wq_head));
234}
235
236/*
237 * Note: we use "set_current_state()" _after_ the wait-queue add,
238 * because we need a memory barrier there on SMP, so that any
239 * wake-function that tests for the wait-queue being active
240 * will be guaranteed to see waitqueue addition _or_ subsequent
241 * tests in this thread will see the wakeup having taken place.
242 *
243 * The spin_unlock() itself is semi-permeable and only protects
244 * one way (it only protects stuff inside the critical region and
245 * stops them from bleeding out - it would still allow subsequent
246 * loads to move into the critical region).
247 */
248void
249prepare_to_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
250{
251 unsigned long flags;
252
253 wq_entry->flags &= ~WQ_FLAG_EXCLUSIVE;
254 spin_lock_irqsave(&wq_head->lock, flags);
255 if (list_empty(&wq_entry->entry))
256 __add_wait_queue(wq_head, wq_entry);
257 set_current_state(state);
258 spin_unlock_irqrestore(&wq_head->lock, flags);
259}
260EXPORT_SYMBOL(prepare_to_wait);
261
262/* Returns true if we are the first waiter in the queue, false otherwise. */
263bool
264prepare_to_wait_exclusive(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
265{
266 unsigned long flags;
267 bool was_empty = false;
268
269 wq_entry->flags |= WQ_FLAG_EXCLUSIVE;
270 spin_lock_irqsave(&wq_head->lock, flags);
271 if (list_empty(&wq_entry->entry)) {
272 was_empty = list_empty(&wq_head->head);
273 __add_wait_queue_entry_tail(wq_head, wq_entry);
274 }
275 set_current_state(state);
276 spin_unlock_irqrestore(&wq_head->lock, flags);
277 return was_empty;
278}
279EXPORT_SYMBOL(prepare_to_wait_exclusive);
280
281void init_wait_entry(struct wait_queue_entry *wq_entry, int flags)
282{
283 wq_entry->flags = flags;
284 wq_entry->private = current;
285 wq_entry->func = autoremove_wake_function;
286 INIT_LIST_HEAD(&wq_entry->entry);
287}
288EXPORT_SYMBOL(init_wait_entry);
289
290long prepare_to_wait_event(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry, int state)
291{
292 unsigned long flags;
293 long ret = 0;
294
295 spin_lock_irqsave(&wq_head->lock, flags);
296 if (signal_pending_state(state, current)) {
297 /*
298 * Exclusive waiter must not fail if it was selected by wakeup,
299 * it should "consume" the condition we were waiting for.
300 *
301 * The caller will recheck the condition and return success if
302 * we were already woken up, we can not miss the event because
303 * wakeup locks/unlocks the same wq_head->lock.
304 *
305 * But we need to ensure that set-condition + wakeup after that
306 * can't see us, it should wake up another exclusive waiter if
307 * we fail.
308 */
309 list_del_init(&wq_entry->entry);
310 ret = -ERESTARTSYS;
311 } else {
312 if (list_empty(&wq_entry->entry)) {
313 if (wq_entry->flags & WQ_FLAG_EXCLUSIVE)
314 __add_wait_queue_entry_tail(wq_head, wq_entry);
315 else
316 __add_wait_queue(wq_head, wq_entry);
317 }
318 set_current_state(state);
319 }
320 spin_unlock_irqrestore(&wq_head->lock, flags);
321
322 return ret;
323}
324EXPORT_SYMBOL(prepare_to_wait_event);
325
326/*
327 * Note! These two wait functions are entered with the
328 * wait-queue lock held (and interrupts off in the _irq
329 * case), so there is no race with testing the wakeup
330 * condition in the caller before they add the wait
331 * entry to the wake queue.
332 */
333int do_wait_intr(wait_queue_head_t *wq, wait_queue_entry_t *wait)
334{
335 if (likely(list_empty(&wait->entry)))
336 __add_wait_queue_entry_tail(wq, wait);
337
338 set_current_state(TASK_INTERRUPTIBLE);
339 if (signal_pending(current))
340 return -ERESTARTSYS;
341
342 spin_unlock(&wq->lock);
343 schedule();
344 spin_lock(&wq->lock);
345
346 return 0;
347}
348EXPORT_SYMBOL(do_wait_intr);
349
350int do_wait_intr_irq(wait_queue_head_t *wq, wait_queue_entry_t *wait)
351{
352 if (likely(list_empty(&wait->entry)))
353 __add_wait_queue_entry_tail(wq, wait);
354
355 set_current_state(TASK_INTERRUPTIBLE);
356 if (signal_pending(current))
357 return -ERESTARTSYS;
358
359 spin_unlock_irq(&wq->lock);
360 schedule();
361 spin_lock_irq(&wq->lock);
362
363 return 0;
364}
365EXPORT_SYMBOL(do_wait_intr_irq);
366
367/**
368 * finish_wait - clean up after waiting in a queue
369 * @wq_head: waitqueue waited on
370 * @wq_entry: wait descriptor
371 *
372 * Sets current thread back to running state and removes
373 * the wait descriptor from the given waitqueue if still
374 * queued.
375 */
376void finish_wait(struct wait_queue_head *wq_head, struct wait_queue_entry *wq_entry)
377{
378 unsigned long flags;
379
380 __set_current_state(TASK_RUNNING);
381 /*
382 * We can check for list emptiness outside the lock
383 * IFF:
384 * - we use the "careful" check that verifies both
385 * the next and prev pointers, so that there cannot
386 * be any half-pending updates in progress on other
387 * CPU's that we haven't seen yet (and that might
388 * still change the stack area.
389 * and
390 * - all other users take the lock (ie we can only
391 * have _one_ other CPU that looks at or modifies
392 * the list).
393 */
394 if (!list_empty_careful(&wq_entry->entry)) {
395 spin_lock_irqsave(&wq_head->lock, flags);
396 list_del_init(&wq_entry->entry);
397 spin_unlock_irqrestore(&wq_head->lock, flags);
398 }
399}
400EXPORT_SYMBOL(finish_wait);
401
402int autoremove_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
403{
404 int ret = default_wake_function(wq_entry, mode, sync, key);
405
406 if (ret)
407 list_del_init_careful(&wq_entry->entry);
408
409 return ret;
410}
411EXPORT_SYMBOL(autoremove_wake_function);
412
413/*
414 * DEFINE_WAIT_FUNC(wait, woken_wake_func);
415 *
416 * add_wait_queue(&wq_head, &wait);
417 * for (;;) {
418 * if (condition)
419 * break;
420 *
421 * // in wait_woken() // in woken_wake_function()
422 *
423 * p->state = mode; wq_entry->flags |= WQ_FLAG_WOKEN;
424 * smp_mb(); // A try_to_wake_up():
425 * if (!(wq_entry->flags & WQ_FLAG_WOKEN)) <full barrier>
426 * schedule() if (p->state & mode)
427 * p->state = TASK_RUNNING; p->state = TASK_RUNNING;
428 * wq_entry->flags &= ~WQ_FLAG_WOKEN; ~~~~~~~~~~~~~~~~~~
429 * smp_mb(); // B condition = true;
430 * } smp_mb(); // C
431 * remove_wait_queue(&wq_head, &wait); wq_entry->flags |= WQ_FLAG_WOKEN;
432 */
433long wait_woken(struct wait_queue_entry *wq_entry, unsigned mode, long timeout)
434{
435 /*
436 * The below executes an smp_mb(), which matches with the full barrier
437 * executed by the try_to_wake_up() in woken_wake_function() such that
438 * either we see the store to wq_entry->flags in woken_wake_function()
439 * or woken_wake_function() sees our store to current->state.
440 */
441 set_current_state(mode); /* A */
442 if (!(wq_entry->flags & WQ_FLAG_WOKEN) && !kthread_should_stop_or_park())
443 timeout = schedule_timeout(timeout);
444 __set_current_state(TASK_RUNNING);
445
446 /*
447 * The below executes an smp_mb(), which matches with the smp_mb() (C)
448 * in woken_wake_function() such that either we see the wait condition
449 * being true or the store to wq_entry->flags in woken_wake_function()
450 * follows ours in the coherence order.
451 */
452 smp_store_mb(wq_entry->flags, wq_entry->flags & ~WQ_FLAG_WOKEN); /* B */
453
454 return timeout;
455}
456EXPORT_SYMBOL(wait_woken);
457
458int woken_wake_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *key)
459{
460 /* Pairs with the smp_store_mb() in wait_woken(). */
461 smp_mb(); /* C */
462 wq_entry->flags |= WQ_FLAG_WOKEN;
463
464 return default_wake_function(wq_entry, mode, sync, key);
465}
466EXPORT_SYMBOL(woken_wake_function);
467
468int woken_wake_bit_function(struct wait_queue_entry *wq_entry, unsigned mode, int sync, void *arg)
469{
470 struct wait_bit_key *key = __var_wake_key(wq_entry, arg);
471 if (!key)
472 return 0;
473
474 /* Pairs with the smp_store_mb() in wait_woken(). */
475 smp_mb(); /* C */
476 wq_entry->flags |= WQ_FLAG_WOKEN;
477
478 return default_wake_function(wq_entry, mode, sync, key);
479}
480EXPORT_SYMBOL(woken_wake_bit_function);