1 // SPDX-License-Identifier: GPL-2.0-only
3 * RT-Mutexes: simple blocking mutual exclusion locks with PI support
5 * started by Ingo Molnar and Thomas Gleixner.
7 * Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
8 * Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
9 * Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
10 * Copyright (C) 2006 Esben Nielsen
12 * See Documentation/locking/rt-mutex-design.rst for details.
14 #include <linux/spinlock.h>
15 #include <linux/export.h>
16 #include <linux/sched/signal.h>
17 #include <linux/sched/rt.h>
18 #include <linux/sched/deadline.h>
19 #include <linux/sched/wake_q.h>
20 #include <linux/sched/debug.h>
21 #include <linux/timer.h>
23 #include "rtmutex_common.h"
26 * lock->owner state tracking:
28 * lock->owner holds the task_struct pointer of the owner. Bit 0
29 * is used to keep track of the "lock has waiters" state.
32 * NULL 0 lock is free (fast acquire possible)
33 * NULL 1 lock is free and has waiters and the top waiter
34 * is going to take the lock*
35 * taskpointer 0 lock is held (fast release possible)
36 * taskpointer 1 lock is held and has waiters**
38 * The fast atomic compare exchange based acquire and release is only
39 * possible when bit 0 of lock->owner is 0.
41 * (*) It also can be a transitional state when grabbing the lock
42 * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
43 * we need to set the bit0 before looking at the lock, and the owner may be
44 * NULL in this small time, hence this can be a transitional state.
46 * (**) There is a small time when bit 0 is set but there are no
47 * waiters. This can happen when grabbing the lock in the slow path.
48 * To prevent a cmpxchg of the owner releasing the lock, we need to
49 * set this bit before looking at the lock.
53 rt_mutex_set_owner(struct rt_mutex
*lock
, struct task_struct
*owner
)
55 unsigned long val
= (unsigned long)owner
;
57 if (rt_mutex_has_waiters(lock
))
58 val
|= RT_MUTEX_HAS_WAITERS
;
60 lock
->owner
= (struct task_struct
*)val
;
63 static inline void clear_rt_mutex_waiters(struct rt_mutex
*lock
)
65 lock
->owner
= (struct task_struct
*)
66 ((unsigned long)lock
->owner
& ~RT_MUTEX_HAS_WAITERS
);
69 static void fixup_rt_mutex_waiters(struct rt_mutex
*lock
)
71 unsigned long owner
, *p
= (unsigned long *) &lock
->owner
;
73 if (rt_mutex_has_waiters(lock
))
77 * The rbtree has no waiters enqueued, now make sure that the
78 * lock->owner still has the waiters bit set, otherwise the
79 * following can happen:
85 * l->owner = T1 | HAS_WAITERS;
93 * l->owner = T1 | HAS_WAITERS;
98 * signal(->T2) signal(->T3)
105 * ==> wait list is empty
109 * fixup_rt_mutex_waiters()
110 * if (wait_list_empty(l) {
112 * owner = l->owner & ~HAS_WAITERS;
116 * rt_mutex_unlock(l) fixup_rt_mutex_waiters()
117 * if (wait_list_empty(l) {
118 * owner = l->owner & ~HAS_WAITERS;
119 * cmpxchg(l->owner, T1, NULL)
120 * ===> Success (l->owner = NULL)
126 * With the check for the waiter bit in place T3 on CPU2 will not
127 * overwrite. All tasks fiddling with the waiters bit are
128 * serialized by l->lock, so nothing else can modify the waiters
129 * bit. If the bit is set then nothing can change l->owner either
130 * so the simple RMW is safe. The cmpxchg() will simply fail if it
131 * happens in the middle of the RMW because the waiters bit is
134 owner
= READ_ONCE(*p
);
135 if (owner
& RT_MUTEX_HAS_WAITERS
)
136 WRITE_ONCE(*p
, owner
& ~RT_MUTEX_HAS_WAITERS
);
140 * We can speed up the acquire/release, if there's no debugging state to be
143 #ifndef CONFIG_DEBUG_RT_MUTEXES
144 # define rt_mutex_cmpxchg_relaxed(l,c,n) (cmpxchg_relaxed(&l->owner, c, n) == c)
145 # define rt_mutex_cmpxchg_acquire(l,c,n) (cmpxchg_acquire(&l->owner, c, n) == c)
146 # define rt_mutex_cmpxchg_release(l,c,n) (cmpxchg_release(&l->owner, c, n) == c)
149 * Callers must hold the ->wait_lock -- which is the whole purpose as we force
150 * all future threads that attempt to [Rmw] the lock to the slowpath. As such
151 * relaxed semantics suffice.
153 static inline void mark_rt_mutex_waiters(struct rt_mutex
*lock
)
155 unsigned long owner
, *p
= (unsigned long *) &lock
->owner
;
159 } while (cmpxchg_relaxed(p
, owner
,
160 owner
| RT_MUTEX_HAS_WAITERS
) != owner
);
164 * Safe fastpath aware unlock:
165 * 1) Clear the waiters bit
166 * 2) Drop lock->wait_lock
167 * 3) Try to unlock the lock with cmpxchg
169 static inline bool unlock_rt_mutex_safe(struct rt_mutex
*lock
,
171 __releases(lock
->wait_lock
)
173 struct task_struct
*owner
= rt_mutex_owner(lock
);
175 clear_rt_mutex_waiters(lock
);
176 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
178 * If a new waiter comes in between the unlock and the cmpxchg
179 * we have two situations:
183 * cmpxchg(p, owner, 0) == owner
184 * mark_rt_mutex_waiters(lock);
190 * mark_rt_mutex_waiters(lock);
192 * cmpxchg(p, owner, 0) != owner
201 return rt_mutex_cmpxchg_release(lock
, owner
, NULL
);
205 # define rt_mutex_cmpxchg_relaxed(l,c,n) (0)
206 # define rt_mutex_cmpxchg_acquire(l,c,n) (0)
207 # define rt_mutex_cmpxchg_release(l,c,n) (0)
209 static inline void mark_rt_mutex_waiters(struct rt_mutex
*lock
)
211 lock
->owner
= (struct task_struct
*)
212 ((unsigned long)lock
->owner
| RT_MUTEX_HAS_WAITERS
);
216 * Simple slow path only version: lock->owner is protected by lock->wait_lock.
218 static inline bool unlock_rt_mutex_safe(struct rt_mutex
*lock
,
220 __releases(lock
->wait_lock
)
223 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
229 * Only use with rt_mutex_waiter_{less,equal}()
231 #define task_to_waiter(p) \
232 &(struct rt_mutex_waiter){ .prio = (p)->prio, .deadline = (p)->dl.deadline }
235 rt_mutex_waiter_less(struct rt_mutex_waiter
*left
,
236 struct rt_mutex_waiter
*right
)
238 if (left
->prio
< right
->prio
)
242 * If both waiters have dl_prio(), we check the deadlines of the
244 * If left waiter has a dl_prio(), and we didn't return 1 above,
245 * then right waiter has a dl_prio() too.
247 if (dl_prio(left
->prio
))
248 return dl_time_before(left
->deadline
, right
->deadline
);
254 rt_mutex_waiter_equal(struct rt_mutex_waiter
*left
,
255 struct rt_mutex_waiter
*right
)
257 if (left
->prio
!= right
->prio
)
261 * If both waiters have dl_prio(), we check the deadlines of the
263 * If left waiter has a dl_prio(), and we didn't return 0 above,
264 * then right waiter has a dl_prio() too.
266 if (dl_prio(left
->prio
))
267 return left
->deadline
== right
->deadline
;
273 rt_mutex_enqueue(struct rt_mutex
*lock
, struct rt_mutex_waiter
*waiter
)
275 struct rb_node
**link
= &lock
->waiters
.rb_root
.rb_node
;
276 struct rb_node
*parent
= NULL
;
277 struct rt_mutex_waiter
*entry
;
278 bool leftmost
= true;
282 entry
= rb_entry(parent
, struct rt_mutex_waiter
, tree_entry
);
283 if (rt_mutex_waiter_less(waiter
, entry
)) {
284 link
= &parent
->rb_left
;
286 link
= &parent
->rb_right
;
291 rb_link_node(&waiter
->tree_entry
, parent
, link
);
292 rb_insert_color_cached(&waiter
->tree_entry
, &lock
->waiters
, leftmost
);
296 rt_mutex_dequeue(struct rt_mutex
*lock
, struct rt_mutex_waiter
*waiter
)
298 if (RB_EMPTY_NODE(&waiter
->tree_entry
))
301 rb_erase_cached(&waiter
->tree_entry
, &lock
->waiters
);
302 RB_CLEAR_NODE(&waiter
->tree_entry
);
306 rt_mutex_enqueue_pi(struct task_struct
*task
, struct rt_mutex_waiter
*waiter
)
308 struct rb_node
**link
= &task
->pi_waiters
.rb_root
.rb_node
;
309 struct rb_node
*parent
= NULL
;
310 struct rt_mutex_waiter
*entry
;
311 bool leftmost
= true;
315 entry
= rb_entry(parent
, struct rt_mutex_waiter
, pi_tree_entry
);
316 if (rt_mutex_waiter_less(waiter
, entry
)) {
317 link
= &parent
->rb_left
;
319 link
= &parent
->rb_right
;
324 rb_link_node(&waiter
->pi_tree_entry
, parent
, link
);
325 rb_insert_color_cached(&waiter
->pi_tree_entry
, &task
->pi_waiters
, leftmost
);
329 rt_mutex_dequeue_pi(struct task_struct
*task
, struct rt_mutex_waiter
*waiter
)
331 if (RB_EMPTY_NODE(&waiter
->pi_tree_entry
))
334 rb_erase_cached(&waiter
->pi_tree_entry
, &task
->pi_waiters
);
335 RB_CLEAR_NODE(&waiter
->pi_tree_entry
);
338 static void rt_mutex_adjust_prio(struct task_struct
*p
)
340 struct task_struct
*pi_task
= NULL
;
342 lockdep_assert_held(&p
->pi_lock
);
344 if (task_has_pi_waiters(p
))
345 pi_task
= task_top_pi_waiter(p
)->task
;
347 rt_mutex_setprio(p
, pi_task
);
351 * Deadlock detection is conditional:
353 * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
354 * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
356 * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
357 * conducted independent of the detect argument.
359 * If the waiter argument is NULL this indicates the deboost path and
360 * deadlock detection is disabled independent of the detect argument
361 * and the config settings.
363 static bool rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter
*waiter
,
364 enum rtmutex_chainwalk chwalk
)
367 * This is just a wrapper function for the following call,
368 * because debug_rt_mutex_detect_deadlock() smells like a magic
369 * debug feature and I wanted to keep the cond function in the
370 * main source file along with the comments instead of having
371 * two of the same in the headers.
373 return debug_rt_mutex_detect_deadlock(waiter
, chwalk
);
377 * Max number of times we'll walk the boosting chain:
379 int max_lock_depth
= 1024;
381 static inline struct rt_mutex
*task_blocked_on_lock(struct task_struct
*p
)
383 return p
->pi_blocked_on
? p
->pi_blocked_on
->lock
: NULL
;
387 * Adjust the priority chain. Also used for deadlock detection.
388 * Decreases task's usage by one - may thus free the task.
390 * @task: the task owning the mutex (owner) for which a chain walk is
392 * @chwalk: do we have to carry out deadlock detection?
393 * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
394 * things for a task that has just got its priority adjusted, and
395 * is waiting on a mutex)
396 * @next_lock: the mutex on which the owner of @orig_lock was blocked before
397 * we dropped its pi_lock. Is never dereferenced, only used for
398 * comparison to detect lock chain changes.
399 * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
400 * its priority to the mutex owner (can be NULL in the case
401 * depicted above or if the top waiter is gone away and we are
402 * actually deboosting the owner)
403 * @top_task: the current top waiter
405 * Returns 0 or -EDEADLK.
407 * Chain walk basics and protection scope
409 * [R] refcount on task
410 * [P] task->pi_lock held
411 * [L] rtmutex->wait_lock held
413 * Step Description Protected by
414 * function arguments:
416 * @orig_lock if != NULL @top_task is blocked on it
417 * @next_lock Unprotected. Cannot be
418 * dereferenced. Only used for
420 * @orig_waiter if != NULL @top_task is blocked on it
421 * @top_task current, or in case of proxy
422 * locking protected by calling
425 * loop_sanity_check();
427 * [1] lock(task->pi_lock); [R] acquire [P]
428 * [2] waiter = task->pi_blocked_on; [P]
429 * [3] check_exit_conditions_1(); [P]
430 * [4] lock = waiter->lock; [P]
431 * [5] if (!try_lock(lock->wait_lock)) { [P] try to acquire [L]
432 * unlock(task->pi_lock); release [P]
435 * [6] check_exit_conditions_2(); [P] + [L]
436 * [7] requeue_lock_waiter(lock, waiter); [P] + [L]
437 * [8] unlock(task->pi_lock); release [P]
438 * put_task_struct(task); release [R]
439 * [9] check_exit_conditions_3(); [L]
440 * [10] task = owner(lock); [L]
441 * get_task_struct(task); [L] acquire [R]
442 * lock(task->pi_lock); [L] acquire [P]
443 * [11] requeue_pi_waiter(tsk, waiters(lock));[P] + [L]
444 * [12] check_exit_conditions_4(); [P] + [L]
445 * [13] unlock(task->pi_lock); release [P]
446 * unlock(lock->wait_lock); release [L]
449 static int rt_mutex_adjust_prio_chain(struct task_struct
*task
,
450 enum rtmutex_chainwalk chwalk
,
451 struct rt_mutex
*orig_lock
,
452 struct rt_mutex
*next_lock
,
453 struct rt_mutex_waiter
*orig_waiter
,
454 struct task_struct
*top_task
)
456 struct rt_mutex_waiter
*waiter
, *top_waiter
= orig_waiter
;
457 struct rt_mutex_waiter
*prerequeue_top_waiter
;
458 int ret
= 0, depth
= 0;
459 struct rt_mutex
*lock
;
460 bool detect_deadlock
;
463 detect_deadlock
= rt_mutex_cond_detect_deadlock(orig_waiter
, chwalk
);
466 * The (de)boosting is a step by step approach with a lot of
467 * pitfalls. We want this to be preemptible and we want hold a
468 * maximum of two locks per step. So we have to check
469 * carefully whether things change under us.
473 * We limit the lock chain length for each invocation.
475 if (++depth
> max_lock_depth
) {
479 * Print this only once. If the admin changes the limit,
480 * print a new message when reaching the limit again.
482 if (prev_max
!= max_lock_depth
) {
483 prev_max
= max_lock_depth
;
484 printk(KERN_WARNING
"Maximum lock depth %d reached "
485 "task: %s (%d)\n", max_lock_depth
,
486 top_task
->comm
, task_pid_nr(top_task
));
488 put_task_struct(task
);
494 * We are fully preemptible here and only hold the refcount on
495 * @task. So everything can have changed under us since the
496 * caller or our own code below (goto retry/again) dropped all
501 * [1] Task cannot go away as we did a get_task() before !
503 raw_spin_lock_irq(&task
->pi_lock
);
506 * [2] Get the waiter on which @task is blocked on.
508 waiter
= task
->pi_blocked_on
;
511 * [3] check_exit_conditions_1() protected by task->pi_lock.
515 * Check whether the end of the boosting chain has been
516 * reached or the state of the chain has changed while we
523 * Check the orig_waiter state. After we dropped the locks,
524 * the previous owner of the lock might have released the lock.
526 if (orig_waiter
&& !rt_mutex_owner(orig_lock
))
530 * We dropped all locks after taking a refcount on @task, so
531 * the task might have moved on in the lock chain or even left
532 * the chain completely and blocks now on an unrelated lock or
535 * We stored the lock on which @task was blocked in @next_lock,
536 * so we can detect the chain change.
538 if (next_lock
!= waiter
->lock
)
542 * Drop out, when the task has no waiters. Note,
543 * top_waiter can be NULL, when we are in the deboosting
547 if (!task_has_pi_waiters(task
))
550 * If deadlock detection is off, we stop here if we
551 * are not the top pi waiter of the task. If deadlock
552 * detection is enabled we continue, but stop the
553 * requeueing in the chain walk.
555 if (top_waiter
!= task_top_pi_waiter(task
)) {
556 if (!detect_deadlock
)
564 * If the waiter priority is the same as the task priority
565 * then there is no further priority adjustment necessary. If
566 * deadlock detection is off, we stop the chain walk. If its
567 * enabled we continue, but stop the requeueing in the chain
570 if (rt_mutex_waiter_equal(waiter
, task_to_waiter(task
))) {
571 if (!detect_deadlock
)
578 * [4] Get the next lock
582 * [5] We need to trylock here as we are holding task->pi_lock,
583 * which is the reverse lock order versus the other rtmutex
586 if (!raw_spin_trylock(&lock
->wait_lock
)) {
587 raw_spin_unlock_irq(&task
->pi_lock
);
593 * [6] check_exit_conditions_2() protected by task->pi_lock and
596 * Deadlock detection. If the lock is the same as the original
597 * lock which caused us to walk the lock chain or if the
598 * current lock is owned by the task which initiated the chain
599 * walk, we detected a deadlock.
601 if (lock
== orig_lock
|| rt_mutex_owner(lock
) == top_task
) {
602 debug_rt_mutex_deadlock(chwalk
, orig_waiter
, lock
);
603 raw_spin_unlock(&lock
->wait_lock
);
609 * If we just follow the lock chain for deadlock detection, no
610 * need to do all the requeue operations. To avoid a truckload
611 * of conditionals around the various places below, just do the
612 * minimum chain walk checks.
616 * No requeue[7] here. Just release @task [8]
618 raw_spin_unlock(&task
->pi_lock
);
619 put_task_struct(task
);
622 * [9] check_exit_conditions_3 protected by lock->wait_lock.
623 * If there is no owner of the lock, end of chain.
625 if (!rt_mutex_owner(lock
)) {
626 raw_spin_unlock_irq(&lock
->wait_lock
);
630 /* [10] Grab the next task, i.e. owner of @lock */
631 task
= get_task_struct(rt_mutex_owner(lock
));
632 raw_spin_lock(&task
->pi_lock
);
635 * No requeue [11] here. We just do deadlock detection.
637 * [12] Store whether owner is blocked
638 * itself. Decision is made after dropping the locks
640 next_lock
= task_blocked_on_lock(task
);
642 * Get the top waiter for the next iteration
644 top_waiter
= rt_mutex_top_waiter(lock
);
646 /* [13] Drop locks */
647 raw_spin_unlock(&task
->pi_lock
);
648 raw_spin_unlock_irq(&lock
->wait_lock
);
650 /* If owner is not blocked, end of chain. */
657 * Store the current top waiter before doing the requeue
658 * operation on @lock. We need it for the boost/deboost
661 prerequeue_top_waiter
= rt_mutex_top_waiter(lock
);
663 /* [7] Requeue the waiter in the lock waiter tree. */
664 rt_mutex_dequeue(lock
, waiter
);
667 * Update the waiter prio fields now that we're dequeued.
669 * These values can have changed through either:
671 * sys_sched_set_scheduler() / sys_sched_setattr()
675 * DL CBS enforcement advancing the effective deadline.
677 * Even though pi_waiters also uses these fields, and that tree is only
678 * updated in [11], we can do this here, since we hold [L], which
679 * serializes all pi_waiters access and rb_erase() does not care about
680 * the values of the node being removed.
682 waiter
->prio
= task
->prio
;
683 waiter
->deadline
= task
->dl
.deadline
;
685 rt_mutex_enqueue(lock
, waiter
);
687 /* [8] Release the task */
688 raw_spin_unlock(&task
->pi_lock
);
689 put_task_struct(task
);
692 * [9] check_exit_conditions_3 protected by lock->wait_lock.
694 * We must abort the chain walk if there is no lock owner even
695 * in the dead lock detection case, as we have nothing to
696 * follow here. This is the end of the chain we are walking.
698 if (!rt_mutex_owner(lock
)) {
700 * If the requeue [7] above changed the top waiter,
701 * then we need to wake the new top waiter up to try
704 if (prerequeue_top_waiter
!= rt_mutex_top_waiter(lock
))
705 wake_up_process(rt_mutex_top_waiter(lock
)->task
);
706 raw_spin_unlock_irq(&lock
->wait_lock
);
710 /* [10] Grab the next task, i.e. the owner of @lock */
711 task
= get_task_struct(rt_mutex_owner(lock
));
712 raw_spin_lock(&task
->pi_lock
);
714 /* [11] requeue the pi waiters if necessary */
715 if (waiter
== rt_mutex_top_waiter(lock
)) {
717 * The waiter became the new top (highest priority)
718 * waiter on the lock. Replace the previous top waiter
719 * in the owner tasks pi waiters tree with this waiter
720 * and adjust the priority of the owner.
722 rt_mutex_dequeue_pi(task
, prerequeue_top_waiter
);
723 rt_mutex_enqueue_pi(task
, waiter
);
724 rt_mutex_adjust_prio(task
);
726 } else if (prerequeue_top_waiter
== waiter
) {
728 * The waiter was the top waiter on the lock, but is
729 * no longer the top prority waiter. Replace waiter in
730 * the owner tasks pi waiters tree with the new top
731 * (highest priority) waiter and adjust the priority
733 * The new top waiter is stored in @waiter so that
734 * @waiter == @top_waiter evaluates to true below and
735 * we continue to deboost the rest of the chain.
737 rt_mutex_dequeue_pi(task
, waiter
);
738 waiter
= rt_mutex_top_waiter(lock
);
739 rt_mutex_enqueue_pi(task
, waiter
);
740 rt_mutex_adjust_prio(task
);
743 * Nothing changed. No need to do any priority
749 * [12] check_exit_conditions_4() protected by task->pi_lock
750 * and lock->wait_lock. The actual decisions are made after we
753 * Check whether the task which owns the current lock is pi
754 * blocked itself. If yes we store a pointer to the lock for
755 * the lock chain change detection above. After we dropped
756 * task->pi_lock next_lock cannot be dereferenced anymore.
758 next_lock
= task_blocked_on_lock(task
);
760 * Store the top waiter of @lock for the end of chain walk
763 top_waiter
= rt_mutex_top_waiter(lock
);
765 /* [13] Drop the locks */
766 raw_spin_unlock(&task
->pi_lock
);
767 raw_spin_unlock_irq(&lock
->wait_lock
);
770 * Make the actual exit decisions [12], based on the stored
773 * We reached the end of the lock chain. Stop right here. No
774 * point to go back just to figure that out.
780 * If the current waiter is not the top waiter on the lock,
781 * then we can stop the chain walk here if we are not in full
782 * deadlock detection mode.
784 if (!detect_deadlock
&& waiter
!= top_waiter
)
790 raw_spin_unlock_irq(&task
->pi_lock
);
792 put_task_struct(task
);
798 * Try to take an rt-mutex
800 * Must be called with lock->wait_lock held and interrupts disabled
802 * @lock: The lock to be acquired.
803 * @task: The task which wants to acquire the lock
804 * @waiter: The waiter that is queued to the lock's wait tree if the
805 * callsite called task_blocked_on_lock(), otherwise NULL
807 static int try_to_take_rt_mutex(struct rt_mutex
*lock
, struct task_struct
*task
,
808 struct rt_mutex_waiter
*waiter
)
810 lockdep_assert_held(&lock
->wait_lock
);
813 * Before testing whether we can acquire @lock, we set the
814 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
815 * other tasks which try to modify @lock into the slow path
816 * and they serialize on @lock->wait_lock.
818 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
819 * as explained at the top of this file if and only if:
821 * - There is a lock owner. The caller must fixup the
822 * transient state if it does a trylock or leaves the lock
823 * function due to a signal or timeout.
825 * - @task acquires the lock and there are no other
826 * waiters. This is undone in rt_mutex_set_owner(@task) at
827 * the end of this function.
829 mark_rt_mutex_waiters(lock
);
832 * If @lock has an owner, give up.
834 if (rt_mutex_owner(lock
))
838 * If @waiter != NULL, @task has already enqueued the waiter
839 * into @lock waiter tree. If @waiter == NULL then this is a
844 * If waiter is not the highest priority waiter of
847 if (waiter
!= rt_mutex_top_waiter(lock
))
851 * We can acquire the lock. Remove the waiter from the
854 rt_mutex_dequeue(lock
, waiter
);
858 * If the lock has waiters already we check whether @task is
859 * eligible to take over the lock.
861 * If there are no other waiters, @task can acquire
862 * the lock. @task->pi_blocked_on is NULL, so it does
863 * not need to be dequeued.
865 if (rt_mutex_has_waiters(lock
)) {
867 * If @task->prio is greater than or equal to
868 * the top waiter priority (kernel view),
871 if (!rt_mutex_waiter_less(task_to_waiter(task
),
872 rt_mutex_top_waiter(lock
)))
876 * The current top waiter stays enqueued. We
877 * don't have to change anything in the lock
882 * No waiters. Take the lock without the
883 * pi_lock dance.@task->pi_blocked_on is NULL
884 * and we have no waiters to enqueue in @task
892 * Clear @task->pi_blocked_on. Requires protection by
893 * @task->pi_lock. Redundant operation for the @waiter == NULL
894 * case, but conditionals are more expensive than a redundant
897 raw_spin_lock(&task
->pi_lock
);
898 task
->pi_blocked_on
= NULL
;
900 * Finish the lock acquisition. @task is the new owner. If
901 * other waiters exist we have to insert the highest priority
902 * waiter into @task->pi_waiters tree.
904 if (rt_mutex_has_waiters(lock
))
905 rt_mutex_enqueue_pi(task
, rt_mutex_top_waiter(lock
));
906 raw_spin_unlock(&task
->pi_lock
);
909 /* We got the lock. */
910 debug_rt_mutex_lock(lock
);
913 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
914 * are still waiters or clears it.
916 rt_mutex_set_owner(lock
, task
);
922 * Task blocks on lock.
924 * Prepare waiter and propagate pi chain
926 * This must be called with lock->wait_lock held and interrupts disabled
928 static int task_blocks_on_rt_mutex(struct rt_mutex
*lock
,
929 struct rt_mutex_waiter
*waiter
,
930 struct task_struct
*task
,
931 enum rtmutex_chainwalk chwalk
)
933 struct task_struct
*owner
= rt_mutex_owner(lock
);
934 struct rt_mutex_waiter
*top_waiter
= waiter
;
935 struct rt_mutex
*next_lock
;
936 int chain_walk
= 0, res
;
938 lockdep_assert_held(&lock
->wait_lock
);
941 * Early deadlock detection. We really don't want the task to
942 * enqueue on itself just to untangle the mess later. It's not
943 * only an optimization. We drop the locks, so another waiter
944 * can come in before the chain walk detects the deadlock. So
945 * the other will detect the deadlock and return -EDEADLOCK,
946 * which is wrong, as the other waiter is not in a deadlock
952 raw_spin_lock(&task
->pi_lock
);
955 waiter
->prio
= task
->prio
;
956 waiter
->deadline
= task
->dl
.deadline
;
958 /* Get the top priority waiter on the lock */
959 if (rt_mutex_has_waiters(lock
))
960 top_waiter
= rt_mutex_top_waiter(lock
);
961 rt_mutex_enqueue(lock
, waiter
);
963 task
->pi_blocked_on
= waiter
;
965 raw_spin_unlock(&task
->pi_lock
);
970 raw_spin_lock(&owner
->pi_lock
);
971 if (waiter
== rt_mutex_top_waiter(lock
)) {
972 rt_mutex_dequeue_pi(owner
, top_waiter
);
973 rt_mutex_enqueue_pi(owner
, waiter
);
975 rt_mutex_adjust_prio(owner
);
976 if (owner
->pi_blocked_on
)
978 } else if (rt_mutex_cond_detect_deadlock(waiter
, chwalk
)) {
982 /* Store the lock on which owner is blocked or NULL */
983 next_lock
= task_blocked_on_lock(owner
);
985 raw_spin_unlock(&owner
->pi_lock
);
987 * Even if full deadlock detection is on, if the owner is not
988 * blocked itself, we can avoid finding this out in the chain
991 if (!chain_walk
|| !next_lock
)
995 * The owner can't disappear while holding a lock,
996 * so the owner struct is protected by wait_lock.
997 * Gets dropped in rt_mutex_adjust_prio_chain()!
999 get_task_struct(owner
);
1001 raw_spin_unlock_irq(&lock
->wait_lock
);
1003 res
= rt_mutex_adjust_prio_chain(owner
, chwalk
, lock
,
1004 next_lock
, waiter
, task
);
1006 raw_spin_lock_irq(&lock
->wait_lock
);
1012 * Remove the top waiter from the current tasks pi waiter tree and
1015 * Called with lock->wait_lock held and interrupts disabled.
1017 static void mark_wakeup_next_waiter(struct wake_q_head
*wake_q
,
1018 struct rt_mutex
*lock
)
1020 struct rt_mutex_waiter
*waiter
;
1022 raw_spin_lock(¤t
->pi_lock
);
1024 waiter
= rt_mutex_top_waiter(lock
);
1027 * Remove it from current->pi_waiters and deboost.
1029 * We must in fact deboost here in order to ensure we call
1030 * rt_mutex_setprio() to update p->pi_top_task before the
1033 rt_mutex_dequeue_pi(current
, waiter
);
1034 rt_mutex_adjust_prio(current
);
1037 * As we are waking up the top waiter, and the waiter stays
1038 * queued on the lock until it gets the lock, this lock
1039 * obviously has waiters. Just set the bit here and this has
1040 * the added benefit of forcing all new tasks into the
1041 * slow path making sure no task of lower priority than
1042 * the top waiter can steal this lock.
1044 lock
->owner
= (void *) RT_MUTEX_HAS_WAITERS
;
1047 * We deboosted before waking the top waiter task such that we don't
1048 * run two tasks with the 'same' priority (and ensure the
1049 * p->pi_top_task pointer points to a blocked task). This however can
1050 * lead to priority inversion if we would get preempted after the
1051 * deboost but before waking our donor task, hence the preempt_disable()
1054 * Pairs with preempt_enable() in rt_mutex_postunlock();
1057 wake_q_add(wake_q
, waiter
->task
);
1058 raw_spin_unlock(¤t
->pi_lock
);
1062 * Remove a waiter from a lock and give up
1064 * Must be called with lock->wait_lock held and interrupts disabled. I must
1065 * have just failed to try_to_take_rt_mutex().
1067 static void remove_waiter(struct rt_mutex
*lock
,
1068 struct rt_mutex_waiter
*waiter
)
1070 bool is_top_waiter
= (waiter
== rt_mutex_top_waiter(lock
));
1071 struct task_struct
*owner
= rt_mutex_owner(lock
);
1072 struct rt_mutex
*next_lock
;
1074 lockdep_assert_held(&lock
->wait_lock
);
1076 raw_spin_lock(¤t
->pi_lock
);
1077 rt_mutex_dequeue(lock
, waiter
);
1078 current
->pi_blocked_on
= NULL
;
1079 raw_spin_unlock(¤t
->pi_lock
);
1082 * Only update priority if the waiter was the highest priority
1083 * waiter of the lock and there is an owner to update.
1085 if (!owner
|| !is_top_waiter
)
1088 raw_spin_lock(&owner
->pi_lock
);
1090 rt_mutex_dequeue_pi(owner
, waiter
);
1092 if (rt_mutex_has_waiters(lock
))
1093 rt_mutex_enqueue_pi(owner
, rt_mutex_top_waiter(lock
));
1095 rt_mutex_adjust_prio(owner
);
1097 /* Store the lock on which owner is blocked or NULL */
1098 next_lock
= task_blocked_on_lock(owner
);
1100 raw_spin_unlock(&owner
->pi_lock
);
1103 * Don't walk the chain, if the owner task is not blocked
1109 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1110 get_task_struct(owner
);
1112 raw_spin_unlock_irq(&lock
->wait_lock
);
1114 rt_mutex_adjust_prio_chain(owner
, RT_MUTEX_MIN_CHAINWALK
, lock
,
1115 next_lock
, NULL
, current
);
1117 raw_spin_lock_irq(&lock
->wait_lock
);
1121 * Recheck the pi chain, in case we got a priority setting
1123 * Called from sched_setscheduler
1125 void rt_mutex_adjust_pi(struct task_struct
*task
)
1127 struct rt_mutex_waiter
*waiter
;
1128 struct rt_mutex
*next_lock
;
1129 unsigned long flags
;
1131 raw_spin_lock_irqsave(&task
->pi_lock
, flags
);
1133 waiter
= task
->pi_blocked_on
;
1134 if (!waiter
|| rt_mutex_waiter_equal(waiter
, task_to_waiter(task
))) {
1135 raw_spin_unlock_irqrestore(&task
->pi_lock
, flags
);
1138 next_lock
= waiter
->lock
;
1139 raw_spin_unlock_irqrestore(&task
->pi_lock
, flags
);
1141 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1142 get_task_struct(task
);
1144 rt_mutex_adjust_prio_chain(task
, RT_MUTEX_MIN_CHAINWALK
, NULL
,
1145 next_lock
, NULL
, task
);
1148 void rt_mutex_init_waiter(struct rt_mutex_waiter
*waiter
)
1150 debug_rt_mutex_init_waiter(waiter
);
1151 RB_CLEAR_NODE(&waiter
->pi_tree_entry
);
1152 RB_CLEAR_NODE(&waiter
->tree_entry
);
1153 waiter
->task
= NULL
;
1157 * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
1158 * @lock: the rt_mutex to take
1159 * @state: the state the task should block in (TASK_INTERRUPTIBLE
1160 * or TASK_UNINTERRUPTIBLE)
1161 * @timeout: the pre-initialized and started timer, or NULL for none
1162 * @waiter: the pre-initialized rt_mutex_waiter
1164 * Must be called with lock->wait_lock held and interrupts disabled
1167 __rt_mutex_slowlock(struct rt_mutex
*lock
, int state
,
1168 struct hrtimer_sleeper
*timeout
,
1169 struct rt_mutex_waiter
*waiter
)
1174 /* Try to acquire the lock: */
1175 if (try_to_take_rt_mutex(lock
, current
, waiter
))
1179 * TASK_INTERRUPTIBLE checks for signals and
1180 * timeout. Ignored otherwise.
1182 if (likely(state
== TASK_INTERRUPTIBLE
)) {
1183 /* Signal pending? */
1184 if (signal_pending(current
))
1186 if (timeout
&& !timeout
->task
)
1192 raw_spin_unlock_irq(&lock
->wait_lock
);
1194 debug_rt_mutex_print_deadlock(waiter
);
1198 raw_spin_lock_irq(&lock
->wait_lock
);
1199 set_current_state(state
);
1202 __set_current_state(TASK_RUNNING
);
1206 static void rt_mutex_handle_deadlock(int res
, int detect_deadlock
,
1207 struct rt_mutex_waiter
*w
)
1210 * If the result is not -EDEADLOCK or the caller requested
1211 * deadlock detection, nothing to do here.
1213 if (res
!= -EDEADLOCK
|| detect_deadlock
)
1217 * Yell lowdly and stop the task right here.
1219 rt_mutex_print_deadlock(w
);
1221 set_current_state(TASK_INTERRUPTIBLE
);
1227 * Slow path lock function:
1230 rt_mutex_slowlock(struct rt_mutex
*lock
, int state
,
1231 struct hrtimer_sleeper
*timeout
,
1232 enum rtmutex_chainwalk chwalk
)
1234 struct rt_mutex_waiter waiter
;
1235 unsigned long flags
;
1238 rt_mutex_init_waiter(&waiter
);
1241 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1242 * be called in early boot if the cmpxchg() fast path is disabled
1243 * (debug, no architecture support). In this case we will acquire the
1244 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1245 * enable interrupts in that early boot case. So we need to use the
1246 * irqsave/restore variants.
1248 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1250 /* Try to acquire the lock again: */
1251 if (try_to_take_rt_mutex(lock
, current
, NULL
)) {
1252 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1256 set_current_state(state
);
1258 /* Setup the timer, when timeout != NULL */
1259 if (unlikely(timeout
))
1260 hrtimer_start_expires(&timeout
->timer
, HRTIMER_MODE_ABS
);
1262 ret
= task_blocks_on_rt_mutex(lock
, &waiter
, current
, chwalk
);
1265 /* sleep on the mutex */
1266 ret
= __rt_mutex_slowlock(lock
, state
, timeout
, &waiter
);
1268 if (unlikely(ret
)) {
1269 __set_current_state(TASK_RUNNING
);
1270 remove_waiter(lock
, &waiter
);
1271 rt_mutex_handle_deadlock(ret
, chwalk
, &waiter
);
1275 * try_to_take_rt_mutex() sets the waiter bit
1276 * unconditionally. We might have to fix that up.
1278 fixup_rt_mutex_waiters(lock
);
1280 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1282 /* Remove pending timer: */
1283 if (unlikely(timeout
))
1284 hrtimer_cancel(&timeout
->timer
);
1286 debug_rt_mutex_free_waiter(&waiter
);
1291 static inline int __rt_mutex_slowtrylock(struct rt_mutex
*lock
)
1293 int ret
= try_to_take_rt_mutex(lock
, current
, NULL
);
1296 * try_to_take_rt_mutex() sets the lock waiters bit
1297 * unconditionally. Clean this up.
1299 fixup_rt_mutex_waiters(lock
);
1305 * Slow path try-lock function:
1307 static inline int rt_mutex_slowtrylock(struct rt_mutex
*lock
)
1309 unsigned long flags
;
1313 * If the lock already has an owner we fail to get the lock.
1314 * This can be done without taking the @lock->wait_lock as
1315 * it is only being read, and this is a trylock anyway.
1317 if (rt_mutex_owner(lock
))
1321 * The mutex has currently no owner. Lock the wait lock and try to
1322 * acquire the lock. We use irqsave here to support early boot calls.
1324 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1326 ret
= __rt_mutex_slowtrylock(lock
);
1328 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1334 * Slow path to release a rt-mutex.
1336 * Return whether the current task needs to call rt_mutex_postunlock().
1338 static bool __sched
rt_mutex_slowunlock(struct rt_mutex
*lock
,
1339 struct wake_q_head
*wake_q
)
1341 unsigned long flags
;
1343 /* irqsave required to support early boot calls */
1344 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1346 debug_rt_mutex_unlock(lock
);
1349 * We must be careful here if the fast path is enabled. If we
1350 * have no waiters queued we cannot set owner to NULL here
1353 * foo->lock->owner = NULL;
1354 * rtmutex_lock(foo->lock); <- fast path
1355 * free = atomic_dec_and_test(foo->refcnt);
1356 * rtmutex_unlock(foo->lock); <- fast path
1359 * raw_spin_unlock(foo->lock->wait_lock);
1361 * So for the fastpath enabled kernel:
1363 * Nothing can set the waiters bit as long as we hold
1364 * lock->wait_lock. So we do the following sequence:
1366 * owner = rt_mutex_owner(lock);
1367 * clear_rt_mutex_waiters(lock);
1368 * raw_spin_unlock(&lock->wait_lock);
1369 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1373 * The fastpath disabled variant is simple as all access to
1374 * lock->owner is serialized by lock->wait_lock:
1376 * lock->owner = NULL;
1377 * raw_spin_unlock(&lock->wait_lock);
1379 while (!rt_mutex_has_waiters(lock
)) {
1380 /* Drops lock->wait_lock ! */
1381 if (unlock_rt_mutex_safe(lock
, flags
) == true)
1383 /* Relock the rtmutex and try again */
1384 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1388 * The wakeup next waiter path does not suffer from the above
1389 * race. See the comments there.
1391 * Queue the next waiter for wakeup once we release the wait_lock.
1393 mark_wakeup_next_waiter(wake_q
, lock
);
1394 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1396 return true; /* call rt_mutex_postunlock() */
1400 * debug aware fast / slowpath lock,trylock,unlock
1402 * The atomic acquire/release ops are compiled away, when either the
1403 * architecture does not support cmpxchg or when debugging is enabled.
1406 rt_mutex_fastlock(struct rt_mutex
*lock
, int state
,
1407 int (*slowfn
)(struct rt_mutex
*lock
, int state
,
1408 struct hrtimer_sleeper
*timeout
,
1409 enum rtmutex_chainwalk chwalk
))
1411 if (likely(rt_mutex_cmpxchg_acquire(lock
, NULL
, current
)))
1414 return slowfn(lock
, state
, NULL
, RT_MUTEX_MIN_CHAINWALK
);
1418 rt_mutex_timed_fastlock(struct rt_mutex
*lock
, int state
,
1419 struct hrtimer_sleeper
*timeout
,
1420 enum rtmutex_chainwalk chwalk
,
1421 int (*slowfn
)(struct rt_mutex
*lock
, int state
,
1422 struct hrtimer_sleeper
*timeout
,
1423 enum rtmutex_chainwalk chwalk
))
1425 if (chwalk
== RT_MUTEX_MIN_CHAINWALK
&&
1426 likely(rt_mutex_cmpxchg_acquire(lock
, NULL
, current
)))
1429 return slowfn(lock
, state
, timeout
, chwalk
);
1433 rt_mutex_fasttrylock(struct rt_mutex
*lock
,
1434 int (*slowfn
)(struct rt_mutex
*lock
))
1436 if (likely(rt_mutex_cmpxchg_acquire(lock
, NULL
, current
)))
1439 return slowfn(lock
);
1443 * Performs the wakeup of the the top-waiter and re-enables preemption.
1445 void rt_mutex_postunlock(struct wake_q_head
*wake_q
)
1449 /* Pairs with preempt_disable() in rt_mutex_slowunlock() */
1454 rt_mutex_fastunlock(struct rt_mutex
*lock
,
1455 bool (*slowfn
)(struct rt_mutex
*lock
,
1456 struct wake_q_head
*wqh
))
1458 DEFINE_WAKE_Q(wake_q
);
1460 if (likely(rt_mutex_cmpxchg_release(lock
, current
, NULL
)))
1463 if (slowfn(lock
, &wake_q
))
1464 rt_mutex_postunlock(&wake_q
);
1467 static inline void __rt_mutex_lock(struct rt_mutex
*lock
, unsigned int subclass
)
1471 mutex_acquire(&lock
->dep_map
, subclass
, 0, _RET_IP_
);
1472 rt_mutex_fastlock(lock
, TASK_UNINTERRUPTIBLE
, rt_mutex_slowlock
);
1475 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1477 * rt_mutex_lock_nested - lock a rt_mutex
1479 * @lock: the rt_mutex to be locked
1480 * @subclass: the lockdep subclass
1482 void __sched
rt_mutex_lock_nested(struct rt_mutex
*lock
, unsigned int subclass
)
1484 __rt_mutex_lock(lock
, subclass
);
1486 EXPORT_SYMBOL_GPL(rt_mutex_lock_nested
);
1488 #else /* !CONFIG_DEBUG_LOCK_ALLOC */
1491 * rt_mutex_lock - lock a rt_mutex
1493 * @lock: the rt_mutex to be locked
1495 void __sched
rt_mutex_lock(struct rt_mutex
*lock
)
1497 __rt_mutex_lock(lock
, 0);
1499 EXPORT_SYMBOL_GPL(rt_mutex_lock
);
1503 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
1505 * @lock: the rt_mutex to be locked
1509 * -EINTR when interrupted by a signal
1511 int __sched
rt_mutex_lock_interruptible(struct rt_mutex
*lock
)
1517 mutex_acquire(&lock
->dep_map
, 0, 0, _RET_IP_
);
1518 ret
= rt_mutex_fastlock(lock
, TASK_INTERRUPTIBLE
, rt_mutex_slowlock
);
1520 mutex_release(&lock
->dep_map
, _RET_IP_
);
1524 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible
);
1527 * Futex variant, must not use fastpath.
1529 int __sched
rt_mutex_futex_trylock(struct rt_mutex
*lock
)
1531 return rt_mutex_slowtrylock(lock
);
1534 int __sched
__rt_mutex_futex_trylock(struct rt_mutex
*lock
)
1536 return __rt_mutex_slowtrylock(lock
);
1540 * rt_mutex_timed_lock - lock a rt_mutex interruptible
1541 * the timeout structure is provided
1544 * @lock: the rt_mutex to be locked
1545 * @timeout: timeout structure or NULL (no timeout)
1549 * -EINTR when interrupted by a signal
1550 * -ETIMEDOUT when the timeout expired
1553 rt_mutex_timed_lock(struct rt_mutex
*lock
, struct hrtimer_sleeper
*timeout
)
1559 mutex_acquire(&lock
->dep_map
, 0, 0, _RET_IP_
);
1560 ret
= rt_mutex_timed_fastlock(lock
, TASK_INTERRUPTIBLE
, timeout
,
1561 RT_MUTEX_MIN_CHAINWALK
,
1564 mutex_release(&lock
->dep_map
, _RET_IP_
);
1568 EXPORT_SYMBOL_GPL(rt_mutex_timed_lock
);
1571 * rt_mutex_trylock - try to lock a rt_mutex
1573 * @lock: the rt_mutex to be locked
1575 * This function can only be called in thread context. It's safe to
1576 * call it from atomic regions, but not from hard interrupt or soft
1577 * interrupt context.
1579 * Returns 1 on success and 0 on contention
1581 int __sched
rt_mutex_trylock(struct rt_mutex
*lock
)
1585 if (WARN_ON_ONCE(in_irq() || in_nmi() || in_serving_softirq()))
1588 ret
= rt_mutex_fasttrylock(lock
, rt_mutex_slowtrylock
);
1590 mutex_acquire(&lock
->dep_map
, 0, 1, _RET_IP_
);
1594 EXPORT_SYMBOL_GPL(rt_mutex_trylock
);
1597 * rt_mutex_unlock - unlock a rt_mutex
1599 * @lock: the rt_mutex to be unlocked
1601 void __sched
rt_mutex_unlock(struct rt_mutex
*lock
)
1603 mutex_release(&lock
->dep_map
, _RET_IP_
);
1604 rt_mutex_fastunlock(lock
, rt_mutex_slowunlock
);
1606 EXPORT_SYMBOL_GPL(rt_mutex_unlock
);
1609 * Futex variant, that since futex variants do not use the fast-path, can be
1610 * simple and will not need to retry.
1612 bool __sched
__rt_mutex_futex_unlock(struct rt_mutex
*lock
,
1613 struct wake_q_head
*wake_q
)
1615 lockdep_assert_held(&lock
->wait_lock
);
1617 debug_rt_mutex_unlock(lock
);
1619 if (!rt_mutex_has_waiters(lock
)) {
1621 return false; /* done */
1625 * We've already deboosted, mark_wakeup_next_waiter() will
1626 * retain preempt_disabled when we drop the wait_lock, to
1627 * avoid inversion prior to the wakeup. preempt_disable()
1628 * therein pairs with rt_mutex_postunlock().
1630 mark_wakeup_next_waiter(wake_q
, lock
);
1632 return true; /* call postunlock() */
1635 void __sched
rt_mutex_futex_unlock(struct rt_mutex
*lock
)
1637 DEFINE_WAKE_Q(wake_q
);
1638 unsigned long flags
;
1641 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1642 postunlock
= __rt_mutex_futex_unlock(lock
, &wake_q
);
1643 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1646 rt_mutex_postunlock(&wake_q
);
1650 * rt_mutex_destroy - mark a mutex unusable
1651 * @lock: the mutex to be destroyed
1653 * This function marks the mutex uninitialized, and any subsequent
1654 * use of the mutex is forbidden. The mutex must not be locked when
1655 * this function is called.
1657 void rt_mutex_destroy(struct rt_mutex
*lock
)
1659 WARN_ON(rt_mutex_is_locked(lock
));
1660 #ifdef CONFIG_DEBUG_RT_MUTEXES
1664 EXPORT_SYMBOL_GPL(rt_mutex_destroy
);
1667 * __rt_mutex_init - initialize the rt lock
1669 * @lock: the rt lock to be initialized
1671 * Initialize the rt lock to unlocked state.
1673 * Initializing of a locked rt lock is not allowed
1675 void __rt_mutex_init(struct rt_mutex
*lock
, const char *name
,
1676 struct lock_class_key
*key
)
1679 raw_spin_lock_init(&lock
->wait_lock
);
1680 lock
->waiters
= RB_ROOT_CACHED
;
1683 debug_rt_mutex_init(lock
, name
, key
);
1685 EXPORT_SYMBOL_GPL(__rt_mutex_init
);
1688 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
1691 * @lock: the rt_mutex to be locked
1692 * @proxy_owner:the task to set as owner
1694 * No locking. Caller has to do serializing itself
1696 * Special API call for PI-futex support. This initializes the rtmutex and
1697 * assigns it to @proxy_owner. Concurrent operations on the rtmutex are not
1698 * possible at this point because the pi_state which contains the rtmutex
1699 * is not yet visible to other tasks.
1701 void rt_mutex_init_proxy_locked(struct rt_mutex
*lock
,
1702 struct task_struct
*proxy_owner
)
1704 __rt_mutex_init(lock
, NULL
, NULL
);
1705 debug_rt_mutex_proxy_lock(lock
, proxy_owner
);
1706 rt_mutex_set_owner(lock
, proxy_owner
);
1710 * rt_mutex_proxy_unlock - release a lock on behalf of owner
1712 * @lock: the rt_mutex to be locked
1714 * No locking. Caller has to do serializing itself
1716 * Special API call for PI-futex support. This merrily cleans up the rtmutex
1717 * (debugging) state. Concurrent operations on this rt_mutex are not
1718 * possible because it belongs to the pi_state which is about to be freed
1719 * and it is not longer visible to other tasks.
1721 void rt_mutex_proxy_unlock(struct rt_mutex
*lock
,
1722 struct task_struct
*proxy_owner
)
1724 debug_rt_mutex_proxy_unlock(lock
);
1725 rt_mutex_set_owner(lock
, NULL
);
1729 * __rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1730 * @lock: the rt_mutex to take
1731 * @waiter: the pre-initialized rt_mutex_waiter
1732 * @task: the task to prepare
1734 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
1735 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
1737 * NOTE: does _NOT_ remove the @waiter on failure; must either call
1738 * rt_mutex_wait_proxy_lock() or rt_mutex_cleanup_proxy_lock() after this.
1741 * 0 - task blocked on lock
1742 * 1 - acquired the lock for task, caller should wake it up
1745 * Special API call for PI-futex support.
1747 int __rt_mutex_start_proxy_lock(struct rt_mutex
*lock
,
1748 struct rt_mutex_waiter
*waiter
,
1749 struct task_struct
*task
)
1753 lockdep_assert_held(&lock
->wait_lock
);
1755 if (try_to_take_rt_mutex(lock
, task
, NULL
))
1758 /* We enforce deadlock detection for futexes */
1759 ret
= task_blocks_on_rt_mutex(lock
, waiter
, task
,
1760 RT_MUTEX_FULL_CHAINWALK
);
1762 if (ret
&& !rt_mutex_owner(lock
)) {
1764 * Reset the return value. We might have
1765 * returned with -EDEADLK and the owner
1766 * released the lock while we were walking the
1767 * pi chain. Let the waiter sort it out.
1772 debug_rt_mutex_print_deadlock(waiter
);
1778 * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1779 * @lock: the rt_mutex to take
1780 * @waiter: the pre-initialized rt_mutex_waiter
1781 * @task: the task to prepare
1783 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
1784 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
1786 * NOTE: unlike __rt_mutex_start_proxy_lock this _DOES_ remove the @waiter
1790 * 0 - task blocked on lock
1791 * 1 - acquired the lock for task, caller should wake it up
1794 * Special API call for PI-futex support.
1796 int rt_mutex_start_proxy_lock(struct rt_mutex
*lock
,
1797 struct rt_mutex_waiter
*waiter
,
1798 struct task_struct
*task
)
1802 raw_spin_lock_irq(&lock
->wait_lock
);
1803 ret
= __rt_mutex_start_proxy_lock(lock
, waiter
, task
);
1805 remove_waiter(lock
, waiter
);
1806 raw_spin_unlock_irq(&lock
->wait_lock
);
1812 * rt_mutex_next_owner - return the next owner of the lock
1814 * @lock: the rt lock query
1816 * Returns the next owner of the lock or NULL
1818 * Caller has to serialize against other accessors to the lock
1821 * Special API call for PI-futex support
1823 struct task_struct
*rt_mutex_next_owner(struct rt_mutex
*lock
)
1825 if (!rt_mutex_has_waiters(lock
))
1828 return rt_mutex_top_waiter(lock
)->task
;
1832 * rt_mutex_wait_proxy_lock() - Wait for lock acquisition
1833 * @lock: the rt_mutex we were woken on
1834 * @to: the timeout, null if none. hrtimer should already have
1836 * @waiter: the pre-initialized rt_mutex_waiter
1838 * Wait for the the lock acquisition started on our behalf by
1839 * rt_mutex_start_proxy_lock(). Upon failure, the caller must call
1840 * rt_mutex_cleanup_proxy_lock().
1844 * <0 - error, one of -EINTR, -ETIMEDOUT
1846 * Special API call for PI-futex support
1848 int rt_mutex_wait_proxy_lock(struct rt_mutex
*lock
,
1849 struct hrtimer_sleeper
*to
,
1850 struct rt_mutex_waiter
*waiter
)
1854 raw_spin_lock_irq(&lock
->wait_lock
);
1855 /* sleep on the mutex */
1856 set_current_state(TASK_INTERRUPTIBLE
);
1857 ret
= __rt_mutex_slowlock(lock
, TASK_INTERRUPTIBLE
, to
, waiter
);
1859 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1860 * have to fix that up.
1862 fixup_rt_mutex_waiters(lock
);
1863 raw_spin_unlock_irq(&lock
->wait_lock
);
1869 * rt_mutex_cleanup_proxy_lock() - Cleanup failed lock acquisition
1870 * @lock: the rt_mutex we were woken on
1871 * @waiter: the pre-initialized rt_mutex_waiter
1873 * Attempt to clean up after a failed __rt_mutex_start_proxy_lock() or
1874 * rt_mutex_wait_proxy_lock().
1876 * Unless we acquired the lock; we're still enqueued on the wait-list and can
1877 * in fact still be granted ownership until we're removed. Therefore we can
1878 * find we are in fact the owner and must disregard the
1879 * rt_mutex_wait_proxy_lock() failure.
1882 * true - did the cleanup, we done.
1883 * false - we acquired the lock after rt_mutex_wait_proxy_lock() returned,
1884 * caller should disregards its return value.
1886 * Special API call for PI-futex support
1888 bool rt_mutex_cleanup_proxy_lock(struct rt_mutex
*lock
,
1889 struct rt_mutex_waiter
*waiter
)
1891 bool cleanup
= false;
1893 raw_spin_lock_irq(&lock
->wait_lock
);
1895 * Do an unconditional try-lock, this deals with the lock stealing
1896 * state where __rt_mutex_futex_unlock() -> mark_wakeup_next_waiter()
1897 * sets a NULL owner.
1899 * We're not interested in the return value, because the subsequent
1900 * test on rt_mutex_owner() will infer that. If the trylock succeeded,
1901 * we will own the lock and it will have removed the waiter. If we
1902 * failed the trylock, we're still not owner and we need to remove
1905 try_to_take_rt_mutex(lock
, current
, waiter
);
1907 * Unless we're the owner; we're still enqueued on the wait_list.
1908 * So check if we became owner, if not, take us off the wait_list.
1910 if (rt_mutex_owner(lock
) != current
) {
1911 remove_waiter(lock
, waiter
);
1915 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1916 * have to fix that up.
1918 fixup_rt_mutex_waiters(lock
);
1920 raw_spin_unlock_irq(&lock
->wait_lock
);