projects/linux·blob v4.3 · refs

1677 lines · 43.5 KiB

1/*
2 * RT-Mutexes: simple blocking mutual exclusion locks with PI support
3 *
4 * started by Ingo Molnar and Thomas Gleixner.
5 *
6 * Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <[email protected]>
7 * Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <[email protected]>
8 * Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
9 * Copyright (C) 2006 Esben Nielsen
10 *
11 * See Documentation/locking/rt-mutex-design.txt for details.
12 */
13#include <linux/spinlock.h>
14#include <linux/export.h>
15#include <linux/sched.h>
16#include <linux/sched/rt.h>
17#include <linux/sched/deadline.h>
18#include <linux/timer.h>
19
20#include "rtmutex_common.h"
21
22/*
23 * lock->owner state tracking:
24 *
25 * lock->owner holds the task_struct pointer of the owner. Bit 0
26 * is used to keep track of the "lock has waiters" state.
27 *
28 * owner bit0
29 * NULL 0 lock is free (fast acquire possible)
30 * NULL 1 lock is free and has waiters and the top waiter
31 * is going to take the lock*
32 * taskpointer 0 lock is held (fast release possible)
33 * taskpointer 1 lock is held and has waiters**
34 *
35 * The fast atomic compare exchange based acquire and release is only
36 * possible when bit 0 of lock->owner is 0.
37 *
38 * (*) It also can be a transitional state when grabbing the lock
39 * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
40 * we need to set the bit0 before looking at the lock, and the owner may be
41 * NULL in this small time, hence this can be a transitional state.
42 *
43 * (**) There is a small time when bit 0 is set but there are no
44 * waiters. This can happen when grabbing the lock in the slow path.
45 * To prevent a cmpxchg of the owner releasing the lock, we need to
46 * set this bit before looking at the lock.
47 */
48
49static void
50rt_mutex_set_owner(struct rt_mutex *lock, struct task_struct *owner)
51{
52 unsigned long val = (unsigned long)owner;
53
54 if (rt_mutex_has_waiters(lock))
55 val |= RT_MUTEX_HAS_WAITERS;
56
57 lock->owner = (struct task_struct *)val;
58}
59
60static inline void clear_rt_mutex_waiters(struct rt_mutex *lock)
61{
62 lock->owner = (struct task_struct *)
63 ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
64}
65
66static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
67{
68 if (!rt_mutex_has_waiters(lock))
69 clear_rt_mutex_waiters(lock);
70}
71
72/*
73 * We can speed up the acquire/release, if there's no debugging state to be
74 * set up.
75 */
76#ifndef CONFIG_DEBUG_RT_MUTEXES
77# define rt_mutex_cmpxchg(l,c,n) (cmpxchg(&l->owner, c, n) == c)
78static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
79{
80 unsigned long owner, *p = (unsigned long *) &lock->owner;
81
82 do {
83 owner = *p;
84 } while (cmpxchg(p, owner, owner | RT_MUTEX_HAS_WAITERS) != owner);
85}
86
87/*
88 * Safe fastpath aware unlock:
89 * 1) Clear the waiters bit
90 * 2) Drop lock->wait_lock
91 * 3) Try to unlock the lock with cmpxchg
92 */
93static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock)
94 __releases(lock->wait_lock)
95{
96 struct task_struct *owner = rt_mutex_owner(lock);
97
98 clear_rt_mutex_waiters(lock);
99 raw_spin_unlock(&lock->wait_lock);
100 /*
101 * If a new waiter comes in between the unlock and the cmpxchg
102 * we have two situations:
103 *
104 * unlock(wait_lock);
105 * lock(wait_lock);
106 * cmpxchg(p, owner, 0) == owner
107 * mark_rt_mutex_waiters(lock);
108 * acquire(lock);
109 * or:
110 *
111 * unlock(wait_lock);
112 * lock(wait_lock);
113 * mark_rt_mutex_waiters(lock);
114 *
115 * cmpxchg(p, owner, 0) != owner
116 * enqueue_waiter();
117 * unlock(wait_lock);
118 * lock(wait_lock);
119 * wake waiter();
120 * unlock(wait_lock);
121 * lock(wait_lock);
122 * acquire(lock);
123 */
124 return rt_mutex_cmpxchg(lock, owner, NULL);
125}
126
127#else
128# define rt_mutex_cmpxchg(l,c,n) (0)
129static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
130{
131 lock->owner = (struct task_struct *)
132 ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
133}
134
135/*
136 * Simple slow path only version: lock->owner is protected by lock->wait_lock.
137 */
138static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock)
139 __releases(lock->wait_lock)
140{
141 lock->owner = NULL;
142 raw_spin_unlock(&lock->wait_lock);
143 return true;
144}
145#endif
146
147static inline int
148rt_mutex_waiter_less(struct rt_mutex_waiter *left,
149 struct rt_mutex_waiter *right)
150{
151 if (left->prio < right->prio)
152 return 1;
153
154 /*
155 * If both waiters have dl_prio(), we check the deadlines of the
156 * associated tasks.
157 * If left waiter has a dl_prio(), and we didn't return 1 above,
158 * then right waiter has a dl_prio() too.
159 */
160 if (dl_prio(left->prio))
161 return (left->task->dl.deadline < right->task->dl.deadline);
162
163 return 0;
164}
165
166static void
167rt_mutex_enqueue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
168{
169 struct rb_node **link = &lock->waiters.rb_node;
170 struct rb_node *parent = NULL;
171 struct rt_mutex_waiter *entry;
172 int leftmost = 1;
173
174 while (*link) {
175 parent = *link;
176 entry = rb_entry(parent, struct rt_mutex_waiter, tree_entry);
177 if (rt_mutex_waiter_less(waiter, entry)) {
178 link = &parent->rb_left;
179 } else {
180 link = &parent->rb_right;
181 leftmost = 0;
182 }
183 }
184
185 if (leftmost)
186 lock->waiters_leftmost = &waiter->tree_entry;
187
188 rb_link_node(&waiter->tree_entry, parent, link);
189 rb_insert_color(&waiter->tree_entry, &lock->waiters);
190}
191
192static void
193rt_mutex_dequeue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
194{
195 if (RB_EMPTY_NODE(&waiter->tree_entry))
196 return;
197
198 if (lock->waiters_leftmost == &waiter->tree_entry)
199 lock->waiters_leftmost = rb_next(&waiter->tree_entry);
200
201 rb_erase(&waiter->tree_entry, &lock->waiters);
202 RB_CLEAR_NODE(&waiter->tree_entry);
203}
204
205static void
206rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
207{
208 struct rb_node **link = &task->pi_waiters.rb_node;
209 struct rb_node *parent = NULL;
210 struct rt_mutex_waiter *entry;
211 int leftmost = 1;
212
213 while (*link) {
214 parent = *link;
215 entry = rb_entry(parent, struct rt_mutex_waiter, pi_tree_entry);
216 if (rt_mutex_waiter_less(waiter, entry)) {
217 link = &parent->rb_left;
218 } else {
219 link = &parent->rb_right;
220 leftmost = 0;
221 }
222 }
223
224 if (leftmost)
225 task->pi_waiters_leftmost = &waiter->pi_tree_entry;
226
227 rb_link_node(&waiter->pi_tree_entry, parent, link);
228 rb_insert_color(&waiter->pi_tree_entry, &task->pi_waiters);
229}
230
231static void
232rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
233{
234 if (RB_EMPTY_NODE(&waiter->pi_tree_entry))
235 return;
236
237 if (task->pi_waiters_leftmost == &waiter->pi_tree_entry)
238 task->pi_waiters_leftmost = rb_next(&waiter->pi_tree_entry);
239
240 rb_erase(&waiter->pi_tree_entry, &task->pi_waiters);
241 RB_CLEAR_NODE(&waiter->pi_tree_entry);
242}
243
244/*
245 * Calculate task priority from the waiter tree priority
246 *
247 * Return task->normal_prio when the waiter tree is empty or when
248 * the waiter is not allowed to do priority boosting
249 */
250int rt_mutex_getprio(struct task_struct *task)
251{
252 if (likely(!task_has_pi_waiters(task)))
253 return task->normal_prio;
254
255 return min(task_top_pi_waiter(task)->prio,
256 task->normal_prio);
257}
258
259struct task_struct *rt_mutex_get_top_task(struct task_struct *task)
260{
261 if (likely(!task_has_pi_waiters(task)))
262 return NULL;
263
264 return task_top_pi_waiter(task)->task;
265}
266
267/*
268 * Called by sched_setscheduler() to get the priority which will be
269 * effective after the change.
270 */
271int rt_mutex_get_effective_prio(struct task_struct *task, int newprio)
272{
273 if (!task_has_pi_waiters(task))
274 return newprio;
275
276 if (task_top_pi_waiter(task)->task->prio <= newprio)
277 return task_top_pi_waiter(task)->task->prio;
278 return newprio;
279}
280
281/*
282 * Adjust the priority of a task, after its pi_waiters got modified.
283 *
284 * This can be both boosting and unboosting. task->pi_lock must be held.
285 */
286static void __rt_mutex_adjust_prio(struct task_struct *task)
287{
288 int prio = rt_mutex_getprio(task);
289
290 if (task->prio != prio || dl_prio(prio))
291 rt_mutex_setprio(task, prio);
292}
293
294/*
295 * Adjust task priority (undo boosting). Called from the exit path of
296 * rt_mutex_slowunlock() and rt_mutex_slowlock().
297 *
298 * (Note: We do this outside of the protection of lock->wait_lock to
299 * allow the lock to be taken while or before we readjust the priority
300 * of task. We do not use the spin_xx_mutex() variants here as we are
301 * outside of the debug path.)
302 */
303void rt_mutex_adjust_prio(struct task_struct *task)
304{
305 unsigned long flags;
306
307 raw_spin_lock_irqsave(&task->pi_lock, flags);
308 __rt_mutex_adjust_prio(task);
309 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
310}
311
312/*
313 * Deadlock detection is conditional:
314 *
315 * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
316 * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
317 *
318 * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
319 * conducted independent of the detect argument.
320 *
321 * If the waiter argument is NULL this indicates the deboost path and
322 * deadlock detection is disabled independent of the detect argument
323 * and the config settings.
324 */
325static bool rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
326 enum rtmutex_chainwalk chwalk)
327{
328 /*
329 * This is just a wrapper function for the following call,
330 * because debug_rt_mutex_detect_deadlock() smells like a magic
331 * debug feature and I wanted to keep the cond function in the
332 * main source file along with the comments instead of having
333 * two of the same in the headers.
334 */
335 return debug_rt_mutex_detect_deadlock(waiter, chwalk);
336}
337
338/*
339 * Max number of times we'll walk the boosting chain:
340 */
341int max_lock_depth = 1024;
342
343static inline struct rt_mutex *task_blocked_on_lock(struct task_struct *p)
344{
345 return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
346}
347
348/*
349 * Adjust the priority chain. Also used for deadlock detection.
350 * Decreases task's usage by one - may thus free the task.
351 *
352 * @task: the task owning the mutex (owner) for which a chain walk is
353 * probably needed
354 * @chwalk: do we have to carry out deadlock detection?
355 * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
356 * things for a task that has just got its priority adjusted, and
357 * is waiting on a mutex)
358 * @next_lock: the mutex on which the owner of @orig_lock was blocked before
359 * we dropped its pi_lock. Is never dereferenced, only used for
360 * comparison to detect lock chain changes.
361 * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
362 * its priority to the mutex owner (can be NULL in the case
363 * depicted above or if the top waiter is gone away and we are
364 * actually deboosting the owner)
365 * @top_task: the current top waiter
366 *
367 * Returns 0 or -EDEADLK.
368 *
369 * Chain walk basics and protection scope
370 *
371 * [R] refcount on task
372 * [P] task->pi_lock held
373 * [L] rtmutex->wait_lock held
374 *
375 * Step Description Protected by
376 * function arguments:
377 * @task [R]
378 * @orig_lock if != NULL @top_task is blocked on it
379 * @next_lock Unprotected. Cannot be
380 * dereferenced. Only used for
381 * comparison.
382 * @orig_waiter if != NULL @top_task is blocked on it
383 * @top_task current, or in case of proxy
384 * locking protected by calling
385 * code
386 * again:
387 * loop_sanity_check();
388 * retry:
389 * [1] lock(task->pi_lock); [R] acquire [P]
390 * [2] waiter = task->pi_blocked_on; [P]
391 * [3] check_exit_conditions_1(); [P]
392 * [4] lock = waiter->lock; [P]
393 * [5] if (!try_lock(lock->wait_lock)) { [P] try to acquire [L]
394 * unlock(task->pi_lock); release [P]
395 * goto retry;
396 * }
397 * [6] check_exit_conditions_2(); [P] + [L]
398 * [7] requeue_lock_waiter(lock, waiter); [P] + [L]
399 * [8] unlock(task->pi_lock); release [P]
400 * put_task_struct(task); release [R]
401 * [9] check_exit_conditions_3(); [L]
402 * [10] task = owner(lock); [L]
403 * get_task_struct(task); [L] acquire [R]
404 * lock(task->pi_lock); [L] acquire [P]
405 * [11] requeue_pi_waiter(tsk, waiters(lock));[P] + [L]
406 * [12] check_exit_conditions_4(); [P] + [L]
407 * [13] unlock(task->pi_lock); release [P]
408 * unlock(lock->wait_lock); release [L]
409 * goto again;
410 */
411static int rt_mutex_adjust_prio_chain(struct task_struct *task,
412 enum rtmutex_chainwalk chwalk,
413 struct rt_mutex *orig_lock,
414 struct rt_mutex *next_lock,
415 struct rt_mutex_waiter *orig_waiter,
416 struct task_struct *top_task)
417{
418 struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
419 struct rt_mutex_waiter *prerequeue_top_waiter;
420 int ret = 0, depth = 0;
421 struct rt_mutex *lock;
422 bool detect_deadlock;
423 unsigned long flags;
424 bool requeue = true;
425
426 detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
427
428 /*
429 * The (de)boosting is a step by step approach with a lot of
430 * pitfalls. We want this to be preemptible and we want hold a
431 * maximum of two locks per step. So we have to check
432 * carefully whether things change under us.
433 */
434 again:
435 /*
436 * We limit the lock chain length for each invocation.
437 */
438 if (++depth > max_lock_depth) {
439 static int prev_max;
440
441 /*
442 * Print this only once. If the admin changes the limit,
443 * print a new message when reaching the limit again.
444 */
445 if (prev_max != max_lock_depth) {
446 prev_max = max_lock_depth;
447 printk(KERN_WARNING "Maximum lock depth %d reached "
448 "task: %s (%d)\n", max_lock_depth,
449 top_task->comm, task_pid_nr(top_task));
450 }
451 put_task_struct(task);
452
453 return -EDEADLK;
454 }
455
456 /*
457 * We are fully preemptible here and only hold the refcount on
458 * @task. So everything can have changed under us since the
459 * caller or our own code below (goto retry/again) dropped all
460 * locks.
461 */
462 retry:
463 /*
464 * [1] Task cannot go away as we did a get_task() before !
465 */
466 raw_spin_lock_irqsave(&task->pi_lock, flags);
467
468 /*
469 * [2] Get the waiter on which @task is blocked on.
470 */
471 waiter = task->pi_blocked_on;
472
473 /*
474 * [3] check_exit_conditions_1() protected by task->pi_lock.
475 */
476
477 /*
478 * Check whether the end of the boosting chain has been
479 * reached or the state of the chain has changed while we
480 * dropped the locks.
481 */
482 if (!waiter)
483 goto out_unlock_pi;
484
485 /*
486 * Check the orig_waiter state. After we dropped the locks,
487 * the previous owner of the lock might have released the lock.
488 */
489 if (orig_waiter && !rt_mutex_owner(orig_lock))
490 goto out_unlock_pi;
491
492 /*
493 * We dropped all locks after taking a refcount on @task, so
494 * the task might have moved on in the lock chain or even left
495 * the chain completely and blocks now on an unrelated lock or
496 * on @orig_lock.
497 *
498 * We stored the lock on which @task was blocked in @next_lock,
499 * so we can detect the chain change.
500 */
501 if (next_lock != waiter->lock)
502 goto out_unlock_pi;
503
504 /*
505 * Drop out, when the task has no waiters. Note,
506 * top_waiter can be NULL, when we are in the deboosting
507 * mode!
508 */
509 if (top_waiter) {
510 if (!task_has_pi_waiters(task))
511 goto out_unlock_pi;
512 /*
513 * If deadlock detection is off, we stop here if we
514 * are not the top pi waiter of the task. If deadlock
515 * detection is enabled we continue, but stop the
516 * requeueing in the chain walk.
517 */
518 if (top_waiter != task_top_pi_waiter(task)) {
519 if (!detect_deadlock)
520 goto out_unlock_pi;
521 else
522 requeue = false;
523 }
524 }
525
526 /*
527 * If the waiter priority is the same as the task priority
528 * then there is no further priority adjustment necessary. If
529 * deadlock detection is off, we stop the chain walk. If its
530 * enabled we continue, but stop the requeueing in the chain
531 * walk.
532 */
533 if (waiter->prio == task->prio) {
534 if (!detect_deadlock)
535 goto out_unlock_pi;
536 else
537 requeue = false;
538 }
539
540 /*
541 * [4] Get the next lock
542 */
543 lock = waiter->lock;
544 /*
545 * [5] We need to trylock here as we are holding task->pi_lock,
546 * which is the reverse lock order versus the other rtmutex
547 * operations.
548 */
549 if (!raw_spin_trylock(&lock->wait_lock)) {
550 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
551 cpu_relax();
552 goto retry;
553 }
554
555 /*
556 * [6] check_exit_conditions_2() protected by task->pi_lock and
557 * lock->wait_lock.
558 *
559 * Deadlock detection. If the lock is the same as the original
560 * lock which caused us to walk the lock chain or if the
561 * current lock is owned by the task which initiated the chain
562 * walk, we detected a deadlock.
563 */
564 if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
565 debug_rt_mutex_deadlock(chwalk, orig_waiter, lock);
566 raw_spin_unlock(&lock->wait_lock);
567 ret = -EDEADLK;
568 goto out_unlock_pi;
569 }
570
571 /*
572 * If we just follow the lock chain for deadlock detection, no
573 * need to do all the requeue operations. To avoid a truckload
574 * of conditionals around the various places below, just do the
575 * minimum chain walk checks.
576 */
577 if (!requeue) {
578 /*
579 * No requeue[7] here. Just release @task [8]
580 */
581 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
582 put_task_struct(task);
583
584 /*
585 * [9] check_exit_conditions_3 protected by lock->wait_lock.
586 * If there is no owner of the lock, end of chain.
587 */
588 if (!rt_mutex_owner(lock)) {
589 raw_spin_unlock(&lock->wait_lock);
590 return 0;
591 }
592
593 /* [10] Grab the next task, i.e. owner of @lock */
594 task = rt_mutex_owner(lock);
595 get_task_struct(task);
596 raw_spin_lock_irqsave(&task->pi_lock, flags);
597
598 /*
599 * No requeue [11] here. We just do deadlock detection.
600 *
601 * [12] Store whether owner is blocked
602 * itself. Decision is made after dropping the locks
603 */
604 next_lock = task_blocked_on_lock(task);
605 /*
606 * Get the top waiter for the next iteration
607 */
608 top_waiter = rt_mutex_top_waiter(lock);
609
610 /* [13] Drop locks */
611 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
612 raw_spin_unlock(&lock->wait_lock);
613
614 /* If owner is not blocked, end of chain. */
615 if (!next_lock)
616 goto out_put_task;
617 goto again;
618 }
619
620 /*
621 * Store the current top waiter before doing the requeue
622 * operation on @lock. We need it for the boost/deboost
623 * decision below.
624 */
625 prerequeue_top_waiter = rt_mutex_top_waiter(lock);
626
627 /* [7] Requeue the waiter in the lock waiter tree. */
628 rt_mutex_dequeue(lock, waiter);
629 waiter->prio = task->prio;
630 rt_mutex_enqueue(lock, waiter);
631
632 /* [8] Release the task */
633 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
634 put_task_struct(task);
635
636 /*
637 * [9] check_exit_conditions_3 protected by lock->wait_lock.
638 *
639 * We must abort the chain walk if there is no lock owner even
640 * in the dead lock detection case, as we have nothing to
641 * follow here. This is the end of the chain we are walking.
642 */
643 if (!rt_mutex_owner(lock)) {
644 /*
645 * If the requeue [7] above changed the top waiter,
646 * then we need to wake the new top waiter up to try
647 * to get the lock.
648 */
649 if (prerequeue_top_waiter != rt_mutex_top_waiter(lock))
650 wake_up_process(rt_mutex_top_waiter(lock)->task);
651 raw_spin_unlock(&lock->wait_lock);
652 return 0;
653 }
654
655 /* [10] Grab the next task, i.e. the owner of @lock */
656 task = rt_mutex_owner(lock);
657 get_task_struct(task);
658 raw_spin_lock_irqsave(&task->pi_lock, flags);
659
660 /* [11] requeue the pi waiters if necessary */
661 if (waiter == rt_mutex_top_waiter(lock)) {
662 /*
663 * The waiter became the new top (highest priority)
664 * waiter on the lock. Replace the previous top waiter
665 * in the owner tasks pi waiters tree with this waiter
666 * and adjust the priority of the owner.
667 */
668 rt_mutex_dequeue_pi(task, prerequeue_top_waiter);
669 rt_mutex_enqueue_pi(task, waiter);
670 __rt_mutex_adjust_prio(task);
671
672 } else if (prerequeue_top_waiter == waiter) {
673 /*
674 * The waiter was the top waiter on the lock, but is
675 * no longer the top prority waiter. Replace waiter in
676 * the owner tasks pi waiters tree with the new top
677 * (highest priority) waiter and adjust the priority
678 * of the owner.
679 * The new top waiter is stored in @waiter so that
680 * @waiter == @top_waiter evaluates to true below and
681 * we continue to deboost the rest of the chain.
682 */
683 rt_mutex_dequeue_pi(task, waiter);
684 waiter = rt_mutex_top_waiter(lock);
685 rt_mutex_enqueue_pi(task, waiter);
686 __rt_mutex_adjust_prio(task);
687 } else {
688 /*
689 * Nothing changed. No need to do any priority
690 * adjustment.
691 */
692 }
693
694 /*
695 * [12] check_exit_conditions_4() protected by task->pi_lock
696 * and lock->wait_lock. The actual decisions are made after we
697 * dropped the locks.
698 *
699 * Check whether the task which owns the current lock is pi
700 * blocked itself. If yes we store a pointer to the lock for
701 * the lock chain change detection above. After we dropped
702 * task->pi_lock next_lock cannot be dereferenced anymore.
703 */
704 next_lock = task_blocked_on_lock(task);
705 /*
706 * Store the top waiter of @lock for the end of chain walk
707 * decision below.
708 */
709 top_waiter = rt_mutex_top_waiter(lock);
710
711 /* [13] Drop the locks */
712 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
713 raw_spin_unlock(&lock->wait_lock);
714
715 /*
716 * Make the actual exit decisions [12], based on the stored
717 * values.
718 *
719 * We reached the end of the lock chain. Stop right here. No
720 * point to go back just to figure that out.
721 */
722 if (!next_lock)
723 goto out_put_task;
724
725 /*
726 * If the current waiter is not the top waiter on the lock,
727 * then we can stop the chain walk here if we are not in full
728 * deadlock detection mode.
729 */
730 if (!detect_deadlock && waiter != top_waiter)
731 goto out_put_task;
732
733 goto again;
734
735 out_unlock_pi:
736 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
737 out_put_task:
738 put_task_struct(task);
739
740 return ret;
741}
742
743/*
744 * Try to take an rt-mutex
745 *
746 * Must be called with lock->wait_lock held.
747 *
748 * @lock: The lock to be acquired.
749 * @task: The task which wants to acquire the lock
750 * @waiter: The waiter that is queued to the lock's wait tree if the
751 * callsite called task_blocked_on_lock(), otherwise NULL
752 */
753static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
754 struct rt_mutex_waiter *waiter)
755{
756 unsigned long flags;
757
758 /*
759 * Before testing whether we can acquire @lock, we set the
760 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
761 * other tasks which try to modify @lock into the slow path
762 * and they serialize on @lock->wait_lock.
763 *
764 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
765 * as explained at the top of this file if and only if:
766 *
767 * - There is a lock owner. The caller must fixup the
768 * transient state if it does a trylock or leaves the lock
769 * function due to a signal or timeout.
770 *
771 * - @task acquires the lock and there are no other
772 * waiters. This is undone in rt_mutex_set_owner(@task) at
773 * the end of this function.
774 */
775 mark_rt_mutex_waiters(lock);
776
777 /*
778 * If @lock has an owner, give up.
779 */
780 if (rt_mutex_owner(lock))
781 return 0;
782
783 /*
784 * If @waiter != NULL, @task has already enqueued the waiter
785 * into @lock waiter tree. If @waiter == NULL then this is a
786 * trylock attempt.
787 */
788 if (waiter) {
789 /*
790 * If waiter is not the highest priority waiter of
791 * @lock, give up.
792 */
793 if (waiter != rt_mutex_top_waiter(lock))
794 return 0;
795
796 /*
797 * We can acquire the lock. Remove the waiter from the
798 * lock waiters tree.
799 */
800 rt_mutex_dequeue(lock, waiter);
801
802 } else {
803 /*
804 * If the lock has waiters already we check whether @task is
805 * eligible to take over the lock.
806 *
807 * If there are no other waiters, @task can acquire
808 * the lock. @task->pi_blocked_on is NULL, so it does
809 * not need to be dequeued.
810 */
811 if (rt_mutex_has_waiters(lock)) {
812 /*
813 * If @task->prio is greater than or equal to
814 * the top waiter priority (kernel view),
815 * @task lost.
816 */
817 if (task->prio >= rt_mutex_top_waiter(lock)->prio)
818 return 0;
819
820 /*
821 * The current top waiter stays enqueued. We
822 * don't have to change anything in the lock
823 * waiters order.
824 */
825 } else {
826 /*
827 * No waiters. Take the lock without the
828 * pi_lock dance.@task->pi_blocked_on is NULL
829 * and we have no waiters to enqueue in @task
830 * pi waiters tree.
831 */
832 goto takeit;
833 }
834 }
835
836 /*
837 * Clear @task->pi_blocked_on. Requires protection by
838 * @task->pi_lock. Redundant operation for the @waiter == NULL
839 * case, but conditionals are more expensive than a redundant
840 * store.
841 */
842 raw_spin_lock_irqsave(&task->pi_lock, flags);
843 task->pi_blocked_on = NULL;
844 /*
845 * Finish the lock acquisition. @task is the new owner. If
846 * other waiters exist we have to insert the highest priority
847 * waiter into @task->pi_waiters tree.
848 */
849 if (rt_mutex_has_waiters(lock))
850 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
851 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
852
853takeit:
854 /* We got the lock. */
855 debug_rt_mutex_lock(lock);
856
857 /*
858 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
859 * are still waiters or clears it.
860 */
861 rt_mutex_set_owner(lock, task);
862
863 rt_mutex_deadlock_account_lock(lock, task);
864
865 return 1;
866}
867
868/*
869 * Task blocks on lock.
870 *
871 * Prepare waiter and propagate pi chain
872 *
873 * This must be called with lock->wait_lock held.
874 */
875static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
876 struct rt_mutex_waiter *waiter,
877 struct task_struct *task,
878 enum rtmutex_chainwalk chwalk)
879{
880 struct task_struct *owner = rt_mutex_owner(lock);
881 struct rt_mutex_waiter *top_waiter = waiter;
882 struct rt_mutex *next_lock;
883 int chain_walk = 0, res;
884 unsigned long flags;
885
886 /*
887 * Early deadlock detection. We really don't want the task to
888 * enqueue on itself just to untangle the mess later. It's not
889 * only an optimization. We drop the locks, so another waiter
890 * can come in before the chain walk detects the deadlock. So
891 * the other will detect the deadlock and return -EDEADLOCK,
892 * which is wrong, as the other waiter is not in a deadlock
893 * situation.
894 */
895 if (owner == task)
896 return -EDEADLK;
897
898 raw_spin_lock_irqsave(&task->pi_lock, flags);
899 __rt_mutex_adjust_prio(task);
900 waiter->task = task;
901 waiter->lock = lock;
902 waiter->prio = task->prio;
903
904 /* Get the top priority waiter on the lock */
905 if (rt_mutex_has_waiters(lock))
906 top_waiter = rt_mutex_top_waiter(lock);
907 rt_mutex_enqueue(lock, waiter);
908
909 task->pi_blocked_on = waiter;
910
911 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
912
913 if (!owner)
914 return 0;
915
916 raw_spin_lock_irqsave(&owner->pi_lock, flags);
917 if (waiter == rt_mutex_top_waiter(lock)) {
918 rt_mutex_dequeue_pi(owner, top_waiter);
919 rt_mutex_enqueue_pi(owner, waiter);
920
921 __rt_mutex_adjust_prio(owner);
922 if (owner->pi_blocked_on)
923 chain_walk = 1;
924 } else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
925 chain_walk = 1;
926 }
927
928 /* Store the lock on which owner is blocked or NULL */
929 next_lock = task_blocked_on_lock(owner);
930
931 raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
932 /*
933 * Even if full deadlock detection is on, if the owner is not
934 * blocked itself, we can avoid finding this out in the chain
935 * walk.
936 */
937 if (!chain_walk || !next_lock)
938 return 0;
939
940 /*
941 * The owner can't disappear while holding a lock,
942 * so the owner struct is protected by wait_lock.
943 * Gets dropped in rt_mutex_adjust_prio_chain()!
944 */
945 get_task_struct(owner);
946
947 raw_spin_unlock(&lock->wait_lock);
948
949 res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
950 next_lock, waiter, task);
951
952 raw_spin_lock(&lock->wait_lock);
953
954 return res;
955}
956
957/*
958 * Remove the top waiter from the current tasks pi waiter tree and
959 * queue it up.
960 *
961 * Called with lock->wait_lock held.
962 */
963static void mark_wakeup_next_waiter(struct wake_q_head *wake_q,
964 struct rt_mutex *lock)
965{
966 struct rt_mutex_waiter *waiter;
967 unsigned long flags;
968
969 raw_spin_lock_irqsave(&current->pi_lock, flags);
970
971 waiter = rt_mutex_top_waiter(lock);
972
973 /*
974 * Remove it from current->pi_waiters. We do not adjust a
975 * possible priority boost right now. We execute wakeup in the
976 * boosted mode and go back to normal after releasing
977 * lock->wait_lock.
978 */
979 rt_mutex_dequeue_pi(current, waiter);
980
981 /*
982 * As we are waking up the top waiter, and the waiter stays
983 * queued on the lock until it gets the lock, this lock
984 * obviously has waiters. Just set the bit here and this has
985 * the added benefit of forcing all new tasks into the
986 * slow path making sure no task of lower priority than
987 * the top waiter can steal this lock.
988 */
989 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
990
991 raw_spin_unlock_irqrestore(&current->pi_lock, flags);
992
993 wake_q_add(wake_q, waiter->task);
994}
995
996/*
997 * Remove a waiter from a lock and give up
998 *
999 * Must be called with lock->wait_lock held and
1000 * have just failed to try_to_take_rt_mutex().
1001 */
1002static void remove_waiter(struct rt_mutex *lock,
1003 struct rt_mutex_waiter *waiter)
1004{
1005 bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
1006 struct task_struct *owner = rt_mutex_owner(lock);
1007 struct rt_mutex *next_lock;
1008 unsigned long flags;
1009
1010 raw_spin_lock_irqsave(&current->pi_lock, flags);
1011 rt_mutex_dequeue(lock, waiter);
1012 current->pi_blocked_on = NULL;
1013 raw_spin_unlock_irqrestore(&current->pi_lock, flags);
1014
1015 /*
1016 * Only update priority if the waiter was the highest priority
1017 * waiter of the lock and there is an owner to update.
1018 */
1019 if (!owner || !is_top_waiter)
1020 return;
1021
1022 raw_spin_lock_irqsave(&owner->pi_lock, flags);
1023
1024 rt_mutex_dequeue_pi(owner, waiter);
1025
1026 if (rt_mutex_has_waiters(lock))
1027 rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock));
1028
1029 __rt_mutex_adjust_prio(owner);
1030
1031 /* Store the lock on which owner is blocked or NULL */
1032 next_lock = task_blocked_on_lock(owner);
1033
1034 raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
1035
1036 /*
1037 * Don't walk the chain, if the owner task is not blocked
1038 * itself.
1039 */
1040 if (!next_lock)
1041 return;
1042
1043 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1044 get_task_struct(owner);
1045
1046 raw_spin_unlock(&lock->wait_lock);
1047
1048 rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
1049 next_lock, NULL, current);
1050
1051 raw_spin_lock(&lock->wait_lock);
1052}
1053
1054/*
1055 * Recheck the pi chain, in case we got a priority setting
1056 *
1057 * Called from sched_setscheduler
1058 */
1059void rt_mutex_adjust_pi(struct task_struct *task)
1060{
1061 struct rt_mutex_waiter *waiter;
1062 struct rt_mutex *next_lock;
1063 unsigned long flags;
1064
1065 raw_spin_lock_irqsave(&task->pi_lock, flags);
1066
1067 waiter = task->pi_blocked_on;
1068 if (!waiter || (waiter->prio == task->prio &&
1069 !dl_prio(task->prio))) {
1070 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
1071 return;
1072 }
1073 next_lock = waiter->lock;
1074 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
1075
1076 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1077 get_task_struct(task);
1078
1079 rt_mutex_adjust_prio_chain(task, RT_MUTEX_MIN_CHAINWALK, NULL,
1080 next_lock, NULL, task);
1081}
1082
1083/**
1084 * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
1085 * @lock: the rt_mutex to take
1086 * @state: the state the task should block in (TASK_INTERRUPTIBLE
1087 * or TASK_UNINTERRUPTIBLE)
1088 * @timeout: the pre-initialized and started timer, or NULL for none
1089 * @waiter: the pre-initialized rt_mutex_waiter
1090 *
1091 * lock->wait_lock must be held by the caller.
1092 */
1093static int __sched
1094__rt_mutex_slowlock(struct rt_mutex *lock, int state,
1095 struct hrtimer_sleeper *timeout,
1096 struct rt_mutex_waiter *waiter)
1097{
1098 int ret = 0;
1099
1100 for (;;) {
1101 /* Try to acquire the lock: */
1102 if (try_to_take_rt_mutex(lock, current, waiter))
1103 break;
1104
1105 /*
1106 * TASK_INTERRUPTIBLE checks for signals and
1107 * timeout. Ignored otherwise.
1108 */
1109 if (unlikely(state == TASK_INTERRUPTIBLE)) {
1110 /* Signal pending? */
1111 if (signal_pending(current))
1112 ret = -EINTR;
1113 if (timeout && !timeout->task)
1114 ret = -ETIMEDOUT;
1115 if (ret)
1116 break;
1117 }
1118
1119 raw_spin_unlock(&lock->wait_lock);
1120
1121 debug_rt_mutex_print_deadlock(waiter);
1122
1123 schedule();
1124
1125 raw_spin_lock(&lock->wait_lock);
1126 set_current_state(state);
1127 }
1128
1129 __set_current_state(TASK_RUNNING);
1130 return ret;
1131}
1132
1133static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
1134 struct rt_mutex_waiter *w)
1135{
1136 /*
1137 * If the result is not -EDEADLOCK or the caller requested
1138 * deadlock detection, nothing to do here.
1139 */
1140 if (res != -EDEADLOCK || detect_deadlock)
1141 return;
1142
1143 /*
1144 * Yell lowdly and stop the task right here.
1145 */
1146 rt_mutex_print_deadlock(w);
1147 while (1) {
1148 set_current_state(TASK_INTERRUPTIBLE);
1149 schedule();
1150 }
1151}
1152
1153/*
1154 * Slow path lock function:
1155 */
1156static int __sched
1157rt_mutex_slowlock(struct rt_mutex *lock, int state,
1158 struct hrtimer_sleeper *timeout,
1159 enum rtmutex_chainwalk chwalk)
1160{
1161 struct rt_mutex_waiter waiter;
1162 int ret = 0;
1163
1164 debug_rt_mutex_init_waiter(&waiter);
1165 RB_CLEAR_NODE(&waiter.pi_tree_entry);
1166 RB_CLEAR_NODE(&waiter.tree_entry);
1167
1168 raw_spin_lock(&lock->wait_lock);
1169
1170 /* Try to acquire the lock again: */
1171 if (try_to_take_rt_mutex(lock, current, NULL)) {
1172 raw_spin_unlock(&lock->wait_lock);
1173 return 0;
1174 }
1175
1176 set_current_state(state);
1177
1178 /* Setup the timer, when timeout != NULL */
1179 if (unlikely(timeout))
1180 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1181
1182 ret = task_blocks_on_rt_mutex(lock, &waiter, current, chwalk);
1183
1184 if (likely(!ret))
1185 /* sleep on the mutex */
1186 ret = __rt_mutex_slowlock(lock, state, timeout, &waiter);
1187
1188 if (unlikely(ret)) {
1189 __set_current_state(TASK_RUNNING);
1190 if (rt_mutex_has_waiters(lock))
1191 remove_waiter(lock, &waiter);
1192 rt_mutex_handle_deadlock(ret, chwalk, &waiter);
1193 }
1194
1195 /*
1196 * try_to_take_rt_mutex() sets the waiter bit
1197 * unconditionally. We might have to fix that up.
1198 */
1199 fixup_rt_mutex_waiters(lock);
1200
1201 raw_spin_unlock(&lock->wait_lock);
1202
1203 /* Remove pending timer: */
1204 if (unlikely(timeout))
1205 hrtimer_cancel(&timeout->timer);
1206
1207 debug_rt_mutex_free_waiter(&waiter);
1208
1209 return ret;
1210}
1211
1212/*
1213 * Slow path try-lock function:
1214 */
1215static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
1216{
1217 int ret;
1218
1219 /*
1220 * If the lock already has an owner we fail to get the lock.
1221 * This can be done without taking the @lock->wait_lock as
1222 * it is only being read, and this is a trylock anyway.
1223 */
1224 if (rt_mutex_owner(lock))
1225 return 0;
1226
1227 /*
1228 * The mutex has currently no owner. Lock the wait lock and
1229 * try to acquire the lock.
1230 */
1231 raw_spin_lock(&lock->wait_lock);
1232
1233 ret = try_to_take_rt_mutex(lock, current, NULL);
1234
1235 /*
1236 * try_to_take_rt_mutex() sets the lock waiters bit
1237 * unconditionally. Clean this up.
1238 */
1239 fixup_rt_mutex_waiters(lock);
1240
1241 raw_spin_unlock(&lock->wait_lock);
1242
1243 return ret;
1244}
1245
1246/*
1247 * Slow path to release a rt-mutex.
1248 * Return whether the current task needs to undo a potential priority boosting.
1249 */
1250static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
1251 struct wake_q_head *wake_q)
1252{
1253 raw_spin_lock(&lock->wait_lock);
1254
1255 debug_rt_mutex_unlock(lock);
1256
1257 rt_mutex_deadlock_account_unlock(current);
1258
1259 /*
1260 * We must be careful here if the fast path is enabled. If we
1261 * have no waiters queued we cannot set owner to NULL here
1262 * because of:
1263 *
1264 * foo->lock->owner = NULL;
1265 * rtmutex_lock(foo->lock); <- fast path
1266 * free = atomic_dec_and_test(foo->refcnt);
1267 * rtmutex_unlock(foo->lock); <- fast path
1268 * if (free)
1269 * kfree(foo);
1270 * raw_spin_unlock(foo->lock->wait_lock);
1271 *
1272 * So for the fastpath enabled kernel:
1273 *
1274 * Nothing can set the waiters bit as long as we hold
1275 * lock->wait_lock. So we do the following sequence:
1276 *
1277 * owner = rt_mutex_owner(lock);
1278 * clear_rt_mutex_waiters(lock);
1279 * raw_spin_unlock(&lock->wait_lock);
1280 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1281 * return;
1282 * goto retry;
1283 *
1284 * The fastpath disabled variant is simple as all access to
1285 * lock->owner is serialized by lock->wait_lock:
1286 *
1287 * lock->owner = NULL;
1288 * raw_spin_unlock(&lock->wait_lock);
1289 */
1290 while (!rt_mutex_has_waiters(lock)) {
1291 /* Drops lock->wait_lock ! */
1292 if (unlock_rt_mutex_safe(lock) == true)
1293 return false;
1294 /* Relock the rtmutex and try again */
1295 raw_spin_lock(&lock->wait_lock);
1296 }
1297
1298 /*
1299 * The wakeup next waiter path does not suffer from the above
1300 * race. See the comments there.
1301 *
1302 * Queue the next waiter for wakeup once we release the wait_lock.
1303 */
1304 mark_wakeup_next_waiter(wake_q, lock);
1305
1306 raw_spin_unlock(&lock->wait_lock);
1307
1308 /* check PI boosting */
1309 return true;
1310}
1311
1312/*
1313 * debug aware fast / slowpath lock,trylock,unlock
1314 *
1315 * The atomic acquire/release ops are compiled away, when either the
1316 * architecture does not support cmpxchg or when debugging is enabled.
1317 */
1318static inline int
1319rt_mutex_fastlock(struct rt_mutex *lock, int state,
1320 int (*slowfn)(struct rt_mutex *lock, int state,
1321 struct hrtimer_sleeper *timeout,
1322 enum rtmutex_chainwalk chwalk))
1323{
1324 if (likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1325 rt_mutex_deadlock_account_lock(lock, current);
1326 return 0;
1327 } else
1328 return slowfn(lock, state, NULL, RT_MUTEX_MIN_CHAINWALK);
1329}
1330
1331static inline int
1332rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
1333 struct hrtimer_sleeper *timeout,
1334 enum rtmutex_chainwalk chwalk,
1335 int (*slowfn)(struct rt_mutex *lock, int state,
1336 struct hrtimer_sleeper *timeout,
1337 enum rtmutex_chainwalk chwalk))
1338{
1339 if (chwalk == RT_MUTEX_MIN_CHAINWALK &&
1340 likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1341 rt_mutex_deadlock_account_lock(lock, current);
1342 return 0;
1343 } else
1344 return slowfn(lock, state, timeout, chwalk);
1345}
1346
1347static inline int
1348rt_mutex_fasttrylock(struct rt_mutex *lock,
1349 int (*slowfn)(struct rt_mutex *lock))
1350{
1351 if (likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1352 rt_mutex_deadlock_account_lock(lock, current);
1353 return 1;
1354 }
1355 return slowfn(lock);
1356}
1357
1358static inline void
1359rt_mutex_fastunlock(struct rt_mutex *lock,
1360 bool (*slowfn)(struct rt_mutex *lock,
1361 struct wake_q_head *wqh))
1362{
1363 WAKE_Q(wake_q);
1364
1365 if (likely(rt_mutex_cmpxchg(lock, current, NULL))) {
1366 rt_mutex_deadlock_account_unlock(current);
1367
1368 } else {
1369 bool deboost = slowfn(lock, &wake_q);
1370
1371 wake_up_q(&wake_q);
1372
1373 /* Undo pi boosting if necessary: */
1374 if (deboost)
1375 rt_mutex_adjust_prio(current);
1376 }
1377}
1378
1379/**
1380 * rt_mutex_lock - lock a rt_mutex
1381 *
1382 * @lock: the rt_mutex to be locked
1383 */
1384void __sched rt_mutex_lock(struct rt_mutex *lock)
1385{
1386 might_sleep();
1387
1388 rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, rt_mutex_slowlock);
1389}
1390EXPORT_SYMBOL_GPL(rt_mutex_lock);
1391
1392/**
1393 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
1394 *
1395 * @lock: the rt_mutex to be locked
1396 *
1397 * Returns:
1398 * 0 on success
1399 * -EINTR when interrupted by a signal
1400 */
1401int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock)
1402{
1403 might_sleep();
1404
1405 return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE, rt_mutex_slowlock);
1406}
1407EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
1408
1409/*
1410 * Futex variant with full deadlock detection.
1411 */
1412int rt_mutex_timed_futex_lock(struct rt_mutex *lock,
1413 struct hrtimer_sleeper *timeout)
1414{
1415 might_sleep();
1416
1417 return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
1418 RT_MUTEX_FULL_CHAINWALK,
1419 rt_mutex_slowlock);
1420}
1421
1422/**
1423 * rt_mutex_timed_lock - lock a rt_mutex interruptible
1424 * the timeout structure is provided
1425 * by the caller
1426 *
1427 * @lock: the rt_mutex to be locked
1428 * @timeout: timeout structure or NULL (no timeout)
1429 *
1430 * Returns:
1431 * 0 on success
1432 * -EINTR when interrupted by a signal
1433 * -ETIMEDOUT when the timeout expired
1434 */
1435int
1436rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout)
1437{
1438 might_sleep();
1439
1440 return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
1441 RT_MUTEX_MIN_CHAINWALK,
1442 rt_mutex_slowlock);
1443}
1444EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
1445
1446/**
1447 * rt_mutex_trylock - try to lock a rt_mutex
1448 *
1449 * @lock: the rt_mutex to be locked
1450 *
1451 * This function can only be called in thread context. It's safe to
1452 * call it from atomic regions, but not from hard interrupt or soft
1453 * interrupt context.
1454 *
1455 * Returns 1 on success and 0 on contention
1456 */
1457int __sched rt_mutex_trylock(struct rt_mutex *lock)
1458{
1459 if (WARN_ON(in_irq() || in_nmi() || in_serving_softirq()))
1460 return 0;
1461
1462 return rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
1463}
1464EXPORT_SYMBOL_GPL(rt_mutex_trylock);
1465
1466/**
1467 * rt_mutex_unlock - unlock a rt_mutex
1468 *
1469 * @lock: the rt_mutex to be unlocked
1470 */
1471void __sched rt_mutex_unlock(struct rt_mutex *lock)
1472{
1473 rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
1474}
1475EXPORT_SYMBOL_GPL(rt_mutex_unlock);
1476
1477/**
1478 * rt_mutex_futex_unlock - Futex variant of rt_mutex_unlock
1479 * @lock: the rt_mutex to be unlocked
1480 *
1481 * Returns: true/false indicating whether priority adjustment is
1482 * required or not.
1483 */
1484bool __sched rt_mutex_futex_unlock(struct rt_mutex *lock,
1485 struct wake_q_head *wqh)
1486{
1487 if (likely(rt_mutex_cmpxchg(lock, current, NULL))) {
1488 rt_mutex_deadlock_account_unlock(current);
1489 return false;
1490 }
1491 return rt_mutex_slowunlock(lock, wqh);
1492}
1493
1494/**
1495 * rt_mutex_destroy - mark a mutex unusable
1496 * @lock: the mutex to be destroyed
1497 *
1498 * This function marks the mutex uninitialized, and any subsequent
1499 * use of the mutex is forbidden. The mutex must not be locked when
1500 * this function is called.
1501 */
1502void rt_mutex_destroy(struct rt_mutex *lock)
1503{
1504 WARN_ON(rt_mutex_is_locked(lock));
1505#ifdef CONFIG_DEBUG_RT_MUTEXES
1506 lock->magic = NULL;
1507#endif
1508}
1509
1510EXPORT_SYMBOL_GPL(rt_mutex_destroy);
1511
1512/**
1513 * __rt_mutex_init - initialize the rt lock
1514 *
1515 * @lock: the rt lock to be initialized
1516 *
1517 * Initialize the rt lock to unlocked state.
1518 *
1519 * Initializing of a locked rt lock is not allowed
1520 */
1521void __rt_mutex_init(struct rt_mutex *lock, const char *name)
1522{
1523 lock->owner = NULL;
1524 raw_spin_lock_init(&lock->wait_lock);
1525 lock->waiters = RB_ROOT;
1526 lock->waiters_leftmost = NULL;
1527
1528 debug_rt_mutex_init(lock, name);
1529}
1530EXPORT_SYMBOL_GPL(__rt_mutex_init);
1531
1532/**
1533 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
1534 * proxy owner
1535 *
1536 * @lock: the rt_mutex to be locked
1537 * @proxy_owner:the task to set as owner
1538 *
1539 * No locking. Caller has to do serializing itself
1540 * Special API call for PI-futex support
1541 */
1542void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
1543 struct task_struct *proxy_owner)
1544{
1545 __rt_mutex_init(lock, NULL);
1546 debug_rt_mutex_proxy_lock(lock, proxy_owner);
1547 rt_mutex_set_owner(lock, proxy_owner);
1548 rt_mutex_deadlock_account_lock(lock, proxy_owner);
1549}
1550
1551/**
1552 * rt_mutex_proxy_unlock - release a lock on behalf of owner
1553 *
1554 * @lock: the rt_mutex to be locked
1555 *
1556 * No locking. Caller has to do serializing itself
1557 * Special API call for PI-futex support
1558 */
1559void rt_mutex_proxy_unlock(struct rt_mutex *lock,
1560 struct task_struct *proxy_owner)
1561{
1562 debug_rt_mutex_proxy_unlock(lock);
1563 rt_mutex_set_owner(lock, NULL);
1564 rt_mutex_deadlock_account_unlock(proxy_owner);
1565}
1566
1567/**
1568 * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1569 * @lock: the rt_mutex to take
1570 * @waiter: the pre-initialized rt_mutex_waiter
1571 * @task: the task to prepare
1572 *
1573 * Returns:
1574 * 0 - task blocked on lock
1575 * 1 - acquired the lock for task, caller should wake it up
1576 * <0 - error
1577 *
1578 * Special API call for FUTEX_REQUEUE_PI support.
1579 */
1580int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
1581 struct rt_mutex_waiter *waiter,
1582 struct task_struct *task)
1583{
1584 int ret;
1585
1586 raw_spin_lock(&lock->wait_lock);
1587
1588 if (try_to_take_rt_mutex(lock, task, NULL)) {
1589 raw_spin_unlock(&lock->wait_lock);
1590 return 1;
1591 }
1592
1593 /* We enforce deadlock detection for futexes */
1594 ret = task_blocks_on_rt_mutex(lock, waiter, task,
1595 RT_MUTEX_FULL_CHAINWALK);
1596
1597 if (ret && !rt_mutex_owner(lock)) {
1598 /*
1599 * Reset the return value. We might have
1600 * returned with -EDEADLK and the owner
1601 * released the lock while we were walking the
1602 * pi chain. Let the waiter sort it out.
1603 */
1604 ret = 0;
1605 }
1606
1607 if (unlikely(ret))
1608 remove_waiter(lock, waiter);
1609
1610 raw_spin_unlock(&lock->wait_lock);
1611
1612 debug_rt_mutex_print_deadlock(waiter);
1613
1614 return ret;
1615}
1616
1617/**
1618 * rt_mutex_next_owner - return the next owner of the lock
1619 *
1620 * @lock: the rt lock query
1621 *
1622 * Returns the next owner of the lock or NULL
1623 *
1624 * Caller has to serialize against other accessors to the lock
1625 * itself.
1626 *
1627 * Special API call for PI-futex support
1628 */
1629struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock)
1630{
1631 if (!rt_mutex_has_waiters(lock))
1632 return NULL;
1633
1634 return rt_mutex_top_waiter(lock)->task;
1635}
1636
1637/**
1638 * rt_mutex_finish_proxy_lock() - Complete lock acquisition
1639 * @lock: the rt_mutex we were woken on
1640 * @to: the timeout, null if none. hrtimer should already have
1641 * been started.
1642 * @waiter: the pre-initialized rt_mutex_waiter
1643 *
1644 * Complete the lock acquisition started our behalf by another thread.
1645 *
1646 * Returns:
1647 * 0 - success
1648 * <0 - error, one of -EINTR, -ETIMEDOUT
1649 *
1650 * Special API call for PI-futex requeue support
1651 */
1652int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
1653 struct hrtimer_sleeper *to,
1654 struct rt_mutex_waiter *waiter)
1655{
1656 int ret;
1657
1658 raw_spin_lock(&lock->wait_lock);
1659
1660 set_current_state(TASK_INTERRUPTIBLE);
1661
1662 /* sleep on the mutex */
1663 ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
1664
1665 if (unlikely(ret))
1666 remove_waiter(lock, waiter);
1667
1668 /*
1669 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1670 * have to fix that up.
1671 */
1672 fixup_rt_mutex_waiters(lock);
1673
1674 raw_spin_unlock(&lock->wait_lock);
1675
1676 return ret;
1677}