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.
52 static __always_inline void
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 WRITE_ONCE(lock->owner, (struct task_struct *)val);
63 static __always_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 __always_inline 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_acquire(l,c,n) (cmpxchg_acquire(&l->owner, c, n) == c)
145 # define rt_mutex_cmpxchg_release(l,c,n) (cmpxchg_release(&l->owner, c, n) == c)
148 * Callers must hold the ->wait_lock -- which is the whole purpose as we force
149 * all future threads that attempt to [Rmw] the lock to the slowpath. As such
150 * relaxed semantics suffice.
152 static __always_inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
154 unsigned long owner, *p = (unsigned long *) &lock->owner;
158 } while (cmpxchg_relaxed(p, owner,
159 owner | RT_MUTEX_HAS_WAITERS) != owner);
163 * Safe fastpath aware unlock:
164 * 1) Clear the waiters bit
165 * 2) Drop lock->wait_lock
166 * 3) Try to unlock the lock with cmpxchg
168 static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex *lock,
170 __releases(lock->wait_lock)
172 struct task_struct *owner = rt_mutex_owner(lock);
174 clear_rt_mutex_waiters(lock);
175 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
177 * If a new waiter comes in between the unlock and the cmpxchg
178 * we have two situations:
182 * cmpxchg(p, owner, 0) == owner
183 * mark_rt_mutex_waiters(lock);
189 * mark_rt_mutex_waiters(lock);
191 * cmpxchg(p, owner, 0) != owner
200 return rt_mutex_cmpxchg_release(lock, owner, NULL);
204 # define rt_mutex_cmpxchg_acquire(l,c,n) (0)
205 # define rt_mutex_cmpxchg_release(l,c,n) (0)
207 static __always_inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
209 lock->owner = (struct task_struct *)
210 ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
214 * Simple slow path only version: lock->owner is protected by lock->wait_lock.
216 static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex *lock,
218 __releases(lock->wait_lock)
221 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
227 * Only use with rt_mutex_waiter_{less,equal}()
229 #define task_to_waiter(p) \
230 &(struct rt_mutex_waiter){ .prio = (p)->prio, .deadline = (p)->dl.deadline }
232 static __always_inline int rt_mutex_waiter_less(struct rt_mutex_waiter *left,
233 struct rt_mutex_waiter *right)
235 if (left->prio < right->prio)
239 * If both waiters have dl_prio(), we check the deadlines of the
241 * If left waiter has a dl_prio(), and we didn't return 1 above,
242 * then right waiter has a dl_prio() too.
244 if (dl_prio(left->prio))
245 return dl_time_before(left->deadline, right->deadline);
250 static __always_inline int rt_mutex_waiter_equal(struct rt_mutex_waiter *left,
251 struct rt_mutex_waiter *right)
253 if (left->prio != right->prio)
257 * If both waiters have dl_prio(), we check the deadlines of the
259 * If left waiter has a dl_prio(), and we didn't return 0 above,
260 * then right waiter has a dl_prio() too.
262 if (dl_prio(left->prio))
263 return left->deadline == right->deadline;
268 #define __node_2_waiter(node) \
269 rb_entry((node), struct rt_mutex_waiter, tree_entry)
271 static __always_inline bool __waiter_less(struct rb_node *a, const struct rb_node *b)
273 return rt_mutex_waiter_less(__node_2_waiter(a), __node_2_waiter(b));
276 static __always_inline void
277 rt_mutex_enqueue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
279 rb_add_cached(&waiter->tree_entry, &lock->waiters, __waiter_less);
282 static __always_inline void
283 rt_mutex_dequeue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
285 if (RB_EMPTY_NODE(&waiter->tree_entry))
288 rb_erase_cached(&waiter->tree_entry, &lock->waiters);
289 RB_CLEAR_NODE(&waiter->tree_entry);
292 #define __node_2_pi_waiter(node) \
293 rb_entry((node), struct rt_mutex_waiter, pi_tree_entry)
295 static __always_inline bool
296 __pi_waiter_less(struct rb_node *a, const struct rb_node *b)
298 return rt_mutex_waiter_less(__node_2_pi_waiter(a), __node_2_pi_waiter(b));
301 static __always_inline void
302 rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
304 rb_add_cached(&waiter->pi_tree_entry, &task->pi_waiters, __pi_waiter_less);
307 static __always_inline void
308 rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
310 if (RB_EMPTY_NODE(&waiter->pi_tree_entry))
313 rb_erase_cached(&waiter->pi_tree_entry, &task->pi_waiters);
314 RB_CLEAR_NODE(&waiter->pi_tree_entry);
317 static __always_inline void rt_mutex_adjust_prio(struct task_struct *p)
319 struct task_struct *pi_task = NULL;
321 lockdep_assert_held(&p->pi_lock);
323 if (task_has_pi_waiters(p))
324 pi_task = task_top_pi_waiter(p)->task;
326 rt_mutex_setprio(p, pi_task);
330 * Deadlock detection is conditional:
332 * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
333 * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
335 * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
336 * conducted independent of the detect argument.
338 * If the waiter argument is NULL this indicates the deboost path and
339 * deadlock detection is disabled independent of the detect argument
340 * and the config settings.
342 static __always_inline bool
343 rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
344 enum rtmutex_chainwalk chwalk)
346 if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEX))
347 return waiter != NULL;
348 return chwalk == RT_MUTEX_FULL_CHAINWALK;
352 * Max number of times we'll walk the boosting chain:
354 int max_lock_depth = 1024;
356 static __always_inline struct rt_mutex *task_blocked_on_lock(struct task_struct *p)
358 return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
362 * Adjust the priority chain. Also used for deadlock detection.
363 * Decreases task's usage by one - may thus free the task.
365 * @task: the task owning the mutex (owner) for which a chain walk is
367 * @chwalk: do we have to carry out deadlock detection?
368 * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
369 * things for a task that has just got its priority adjusted, and
370 * is waiting on a mutex)
371 * @next_lock: the mutex on which the owner of @orig_lock was blocked before
372 * we dropped its pi_lock. Is never dereferenced, only used for
373 * comparison to detect lock chain changes.
374 * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
375 * its priority to the mutex owner (can be NULL in the case
376 * depicted above or if the top waiter is gone away and we are
377 * actually deboosting the owner)
378 * @top_task: the current top waiter
380 * Returns 0 or -EDEADLK.
382 * Chain walk basics and protection scope
384 * [R] refcount on task
385 * [P] task->pi_lock held
386 * [L] rtmutex->wait_lock held
388 * Step Description Protected by
389 * function arguments:
391 * @orig_lock if != NULL @top_task is blocked on it
392 * @next_lock Unprotected. Cannot be
393 * dereferenced. Only used for
395 * @orig_waiter if != NULL @top_task is blocked on it
396 * @top_task current, or in case of proxy
397 * locking protected by calling
400 * loop_sanity_check();
402 * [1] lock(task->pi_lock); [R] acquire [P]
403 * [2] waiter = task->pi_blocked_on; [P]
404 * [3] check_exit_conditions_1(); [P]
405 * [4] lock = waiter->lock; [P]
406 * [5] if (!try_lock(lock->wait_lock)) { [P] try to acquire [L]
407 * unlock(task->pi_lock); release [P]
410 * [6] check_exit_conditions_2(); [P] + [L]
411 * [7] requeue_lock_waiter(lock, waiter); [P] + [L]
412 * [8] unlock(task->pi_lock); release [P]
413 * put_task_struct(task); release [R]
414 * [9] check_exit_conditions_3(); [L]
415 * [10] task = owner(lock); [L]
416 * get_task_struct(task); [L] acquire [R]
417 * lock(task->pi_lock); [L] acquire [P]
418 * [11] requeue_pi_waiter(tsk, waiters(lock));[P] + [L]
419 * [12] check_exit_conditions_4(); [P] + [L]
420 * [13] unlock(task->pi_lock); release [P]
421 * unlock(lock->wait_lock); release [L]
424 static int __sched rt_mutex_adjust_prio_chain(struct task_struct *task,
425 enum rtmutex_chainwalk chwalk,
426 struct rt_mutex *orig_lock,
427 struct rt_mutex *next_lock,
428 struct rt_mutex_waiter *orig_waiter,
429 struct task_struct *top_task)
431 struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
432 struct rt_mutex_waiter *prerequeue_top_waiter;
433 int ret = 0, depth = 0;
434 struct rt_mutex *lock;
435 bool detect_deadlock;
438 detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
441 * The (de)boosting is a step by step approach with a lot of
442 * pitfalls. We want this to be preemptible and we want hold a
443 * maximum of two locks per step. So we have to check
444 * carefully whether things change under us.
448 * We limit the lock chain length for each invocation.
450 if (++depth > max_lock_depth) {
454 * Print this only once. If the admin changes the limit,
455 * print a new message when reaching the limit again.
457 if (prev_max != max_lock_depth) {
458 prev_max = max_lock_depth;
459 printk(KERN_WARNING "Maximum lock depth %d reached "
460 "task: %s (%d)\n", max_lock_depth,
461 top_task->comm, task_pid_nr(top_task));
463 put_task_struct(task);
469 * We are fully preemptible here and only hold the refcount on
470 * @task. So everything can have changed under us since the
471 * caller or our own code below (goto retry/again) dropped all
476 * [1] Task cannot go away as we did a get_task() before !
478 raw_spin_lock_irq(&task->pi_lock);
481 * [2] Get the waiter on which @task is blocked on.
483 waiter = task->pi_blocked_on;
486 * [3] check_exit_conditions_1() protected by task->pi_lock.
490 * Check whether the end of the boosting chain has been
491 * reached or the state of the chain has changed while we
498 * Check the orig_waiter state. After we dropped the locks,
499 * the previous owner of the lock might have released the lock.
501 if (orig_waiter && !rt_mutex_owner(orig_lock))
505 * We dropped all locks after taking a refcount on @task, so
506 * the task might have moved on in the lock chain or even left
507 * the chain completely and blocks now on an unrelated lock or
510 * We stored the lock on which @task was blocked in @next_lock,
511 * so we can detect the chain change.
513 if (next_lock != waiter->lock)
517 * Drop out, when the task has no waiters. Note,
518 * top_waiter can be NULL, when we are in the deboosting
522 if (!task_has_pi_waiters(task))
525 * If deadlock detection is off, we stop here if we
526 * are not the top pi waiter of the task. If deadlock
527 * detection is enabled we continue, but stop the
528 * requeueing in the chain walk.
530 if (top_waiter != task_top_pi_waiter(task)) {
531 if (!detect_deadlock)
539 * If the waiter priority is the same as the task priority
540 * then there is no further priority adjustment necessary. If
541 * deadlock detection is off, we stop the chain walk. If its
542 * enabled we continue, but stop the requeueing in the chain
545 if (rt_mutex_waiter_equal(waiter, task_to_waiter(task))) {
546 if (!detect_deadlock)
553 * [4] Get the next lock
557 * [5] We need to trylock here as we are holding task->pi_lock,
558 * which is the reverse lock order versus the other rtmutex
561 if (!raw_spin_trylock(&lock->wait_lock)) {
562 raw_spin_unlock_irq(&task->pi_lock);
568 * [6] check_exit_conditions_2() protected by task->pi_lock and
571 * Deadlock detection. If the lock is the same as the original
572 * lock which caused us to walk the lock chain or if the
573 * current lock is owned by the task which initiated the chain
574 * walk, we detected a deadlock.
576 if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
577 raw_spin_unlock(&lock->wait_lock);
583 * If we just follow the lock chain for deadlock detection, no
584 * need to do all the requeue operations. To avoid a truckload
585 * of conditionals around the various places below, just do the
586 * minimum chain walk checks.
590 * No requeue[7] here. Just release @task [8]
592 raw_spin_unlock(&task->pi_lock);
593 put_task_struct(task);
596 * [9] check_exit_conditions_3 protected by lock->wait_lock.
597 * If there is no owner of the lock, end of chain.
599 if (!rt_mutex_owner(lock)) {
600 raw_spin_unlock_irq(&lock->wait_lock);
604 /* [10] Grab the next task, i.e. owner of @lock */
605 task = get_task_struct(rt_mutex_owner(lock));
606 raw_spin_lock(&task->pi_lock);
609 * No requeue [11] here. We just do deadlock detection.
611 * [12] Store whether owner is blocked
612 * itself. Decision is made after dropping the locks
614 next_lock = task_blocked_on_lock(task);
616 * Get the top waiter for the next iteration
618 top_waiter = rt_mutex_top_waiter(lock);
620 /* [13] Drop locks */
621 raw_spin_unlock(&task->pi_lock);
622 raw_spin_unlock_irq(&lock->wait_lock);
624 /* If owner is not blocked, end of chain. */
631 * Store the current top waiter before doing the requeue
632 * operation on @lock. We need it for the boost/deboost
635 prerequeue_top_waiter = rt_mutex_top_waiter(lock);
637 /* [7] Requeue the waiter in the lock waiter tree. */
638 rt_mutex_dequeue(lock, waiter);
641 * Update the waiter prio fields now that we're dequeued.
643 * These values can have changed through either:
645 * sys_sched_set_scheduler() / sys_sched_setattr()
649 * DL CBS enforcement advancing the effective deadline.
651 * Even though pi_waiters also uses these fields, and that tree is only
652 * updated in [11], we can do this here, since we hold [L], which
653 * serializes all pi_waiters access and rb_erase() does not care about
654 * the values of the node being removed.
656 waiter->prio = task->prio;
657 waiter->deadline = task->dl.deadline;
659 rt_mutex_enqueue(lock, waiter);
661 /* [8] Release the task */
662 raw_spin_unlock(&task->pi_lock);
663 put_task_struct(task);
666 * [9] check_exit_conditions_3 protected by lock->wait_lock.
668 * We must abort the chain walk if there is no lock owner even
669 * in the dead lock detection case, as we have nothing to
670 * follow here. This is the end of the chain we are walking.
672 if (!rt_mutex_owner(lock)) {
674 * If the requeue [7] above changed the top waiter,
675 * then we need to wake the new top waiter up to try
678 if (prerequeue_top_waiter != rt_mutex_top_waiter(lock))
679 wake_up_process(rt_mutex_top_waiter(lock)->task);
680 raw_spin_unlock_irq(&lock->wait_lock);
684 /* [10] Grab the next task, i.e. the owner of @lock */
685 task = get_task_struct(rt_mutex_owner(lock));
686 raw_spin_lock(&task->pi_lock);
688 /* [11] requeue the pi waiters if necessary */
689 if (waiter == rt_mutex_top_waiter(lock)) {
691 * The waiter became the new top (highest priority)
692 * waiter on the lock. Replace the previous top waiter
693 * in the owner tasks pi waiters tree with this waiter
694 * and adjust the priority of the owner.
696 rt_mutex_dequeue_pi(task, prerequeue_top_waiter);
697 rt_mutex_enqueue_pi(task, waiter);
698 rt_mutex_adjust_prio(task);
700 } else if (prerequeue_top_waiter == waiter) {
702 * The waiter was the top waiter on the lock, but is
703 * no longer the top priority waiter. Replace waiter in
704 * the owner tasks pi waiters tree with the new top
705 * (highest priority) waiter and adjust the priority
707 * The new top waiter is stored in @waiter so that
708 * @waiter == @top_waiter evaluates to true below and
709 * we continue to deboost the rest of the chain.
711 rt_mutex_dequeue_pi(task, waiter);
712 waiter = rt_mutex_top_waiter(lock);
713 rt_mutex_enqueue_pi(task, waiter);
714 rt_mutex_adjust_prio(task);
717 * Nothing changed. No need to do any priority
723 * [12] check_exit_conditions_4() protected by task->pi_lock
724 * and lock->wait_lock. The actual decisions are made after we
727 * Check whether the task which owns the current lock is pi
728 * blocked itself. If yes we store a pointer to the lock for
729 * the lock chain change detection above. After we dropped
730 * task->pi_lock next_lock cannot be dereferenced anymore.
732 next_lock = task_blocked_on_lock(task);
734 * Store the top waiter of @lock for the end of chain walk
737 top_waiter = rt_mutex_top_waiter(lock);
739 /* [13] Drop the locks */
740 raw_spin_unlock(&task->pi_lock);
741 raw_spin_unlock_irq(&lock->wait_lock);
744 * Make the actual exit decisions [12], based on the stored
747 * We reached the end of the lock chain. Stop right here. No
748 * point to go back just to figure that out.
754 * If the current waiter is not the top waiter on the lock,
755 * then we can stop the chain walk here if we are not in full
756 * deadlock detection mode.
758 if (!detect_deadlock && waiter != top_waiter)
764 raw_spin_unlock_irq(&task->pi_lock);
766 put_task_struct(task);
772 * Try to take an rt-mutex
774 * Must be called with lock->wait_lock held and interrupts disabled
776 * @lock: The lock to be acquired.
777 * @task: The task which wants to acquire the lock
778 * @waiter: The waiter that is queued to the lock's wait tree if the
779 * callsite called task_blocked_on_lock(), otherwise NULL
782 try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
783 struct rt_mutex_waiter *waiter)
785 lockdep_assert_held(&lock->wait_lock);
788 * Before testing whether we can acquire @lock, we set the
789 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
790 * other tasks which try to modify @lock into the slow path
791 * and they serialize on @lock->wait_lock.
793 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
794 * as explained at the top of this file if and only if:
796 * - There is a lock owner. The caller must fixup the
797 * transient state if it does a trylock or leaves the lock
798 * function due to a signal or timeout.
800 * - @task acquires the lock and there are no other
801 * waiters. This is undone in rt_mutex_set_owner(@task) at
802 * the end of this function.
804 mark_rt_mutex_waiters(lock);
807 * If @lock has an owner, give up.
809 if (rt_mutex_owner(lock))
813 * If @waiter != NULL, @task has already enqueued the waiter
814 * into @lock waiter tree. If @waiter == NULL then this is a
819 * If waiter is not the highest priority waiter of
822 if (waiter != rt_mutex_top_waiter(lock))
826 * We can acquire the lock. Remove the waiter from the
829 rt_mutex_dequeue(lock, waiter);
833 * If the lock has waiters already we check whether @task is
834 * eligible to take over the lock.
836 * If there are no other waiters, @task can acquire
837 * the lock. @task->pi_blocked_on is NULL, so it does
838 * not need to be dequeued.
840 if (rt_mutex_has_waiters(lock)) {
842 * If @task->prio is greater than or equal to
843 * the top waiter priority (kernel view),
846 if (!rt_mutex_waiter_less(task_to_waiter(task),
847 rt_mutex_top_waiter(lock)))
851 * The current top waiter stays enqueued. We
852 * don't have to change anything in the lock
857 * No waiters. Take the lock without the
858 * pi_lock dance.@task->pi_blocked_on is NULL
859 * and we have no waiters to enqueue in @task
867 * Clear @task->pi_blocked_on. Requires protection by
868 * @task->pi_lock. Redundant operation for the @waiter == NULL
869 * case, but conditionals are more expensive than a redundant
872 raw_spin_lock(&task->pi_lock);
873 task->pi_blocked_on = NULL;
875 * Finish the lock acquisition. @task is the new owner. If
876 * other waiters exist we have to insert the highest priority
877 * waiter into @task->pi_waiters tree.
879 if (rt_mutex_has_waiters(lock))
880 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
881 raw_spin_unlock(&task->pi_lock);
885 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
886 * are still waiters or clears it.
888 rt_mutex_set_owner(lock, task);
894 * Task blocks on lock.
896 * Prepare waiter and propagate pi chain
898 * This must be called with lock->wait_lock held and interrupts disabled
900 static int __sched task_blocks_on_rt_mutex(struct rt_mutex *lock,
901 struct rt_mutex_waiter *waiter,
902 struct task_struct *task,
903 enum rtmutex_chainwalk chwalk)
905 struct task_struct *owner = rt_mutex_owner(lock);
906 struct rt_mutex_waiter *top_waiter = waiter;
907 struct rt_mutex *next_lock;
908 int chain_walk = 0, res;
910 lockdep_assert_held(&lock->wait_lock);
913 * Early deadlock detection. We really don't want the task to
914 * enqueue on itself just to untangle the mess later. It's not
915 * only an optimization. We drop the locks, so another waiter
916 * can come in before the chain walk detects the deadlock. So
917 * the other will detect the deadlock and return -EDEADLOCK,
918 * which is wrong, as the other waiter is not in a deadlock
924 raw_spin_lock(&task->pi_lock);
927 waiter->prio = task->prio;
928 waiter->deadline = task->dl.deadline;
930 /* Get the top priority waiter on the lock */
931 if (rt_mutex_has_waiters(lock))
932 top_waiter = rt_mutex_top_waiter(lock);
933 rt_mutex_enqueue(lock, waiter);
935 task->pi_blocked_on = waiter;
937 raw_spin_unlock(&task->pi_lock);
942 raw_spin_lock(&owner->pi_lock);
943 if (waiter == rt_mutex_top_waiter(lock)) {
944 rt_mutex_dequeue_pi(owner, top_waiter);
945 rt_mutex_enqueue_pi(owner, waiter);
947 rt_mutex_adjust_prio(owner);
948 if (owner->pi_blocked_on)
950 } else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
954 /* Store the lock on which owner is blocked or NULL */
955 next_lock = task_blocked_on_lock(owner);
957 raw_spin_unlock(&owner->pi_lock);
959 * Even if full deadlock detection is on, if the owner is not
960 * blocked itself, we can avoid finding this out in the chain
963 if (!chain_walk || !next_lock)
967 * The owner can't disappear while holding a lock,
968 * so the owner struct is protected by wait_lock.
969 * Gets dropped in rt_mutex_adjust_prio_chain()!
971 get_task_struct(owner);
973 raw_spin_unlock_irq(&lock->wait_lock);
975 res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
976 next_lock, waiter, task);
978 raw_spin_lock_irq(&lock->wait_lock);
984 * Remove the top waiter from the current tasks pi waiter tree and
987 * Called with lock->wait_lock held and interrupts disabled.
989 static void __sched mark_wakeup_next_waiter(struct wake_q_head *wake_q,
990 struct rt_mutex *lock)
992 struct rt_mutex_waiter *waiter;
994 raw_spin_lock(¤t->pi_lock);
996 waiter = rt_mutex_top_waiter(lock);
999 * Remove it from current->pi_waiters and deboost.
1001 * We must in fact deboost here in order to ensure we call
1002 * rt_mutex_setprio() to update p->pi_top_task before the
1005 rt_mutex_dequeue_pi(current, waiter);
1006 rt_mutex_adjust_prio(current);
1009 * As we are waking up the top waiter, and the waiter stays
1010 * queued on the lock until it gets the lock, this lock
1011 * obviously has waiters. Just set the bit here and this has
1012 * the added benefit of forcing all new tasks into the
1013 * slow path making sure no task of lower priority than
1014 * the top waiter can steal this lock.
1016 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
1019 * We deboosted before waking the top waiter task such that we don't
1020 * run two tasks with the 'same' priority (and ensure the
1021 * p->pi_top_task pointer points to a blocked task). This however can
1022 * lead to priority inversion if we would get preempted after the
1023 * deboost but before waking our donor task, hence the preempt_disable()
1026 * Pairs with preempt_enable() in rt_mutex_postunlock();
1029 wake_q_add(wake_q, waiter->task);
1030 raw_spin_unlock(¤t->pi_lock);
1034 * Remove a waiter from a lock and give up
1036 * Must be called with lock->wait_lock held and interrupts disabled. I must
1037 * have just failed to try_to_take_rt_mutex().
1039 static void __sched remove_waiter(struct rt_mutex *lock,
1040 struct rt_mutex_waiter *waiter)
1042 bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
1043 struct task_struct *owner = rt_mutex_owner(lock);
1044 struct rt_mutex *next_lock;
1046 lockdep_assert_held(&lock->wait_lock);
1048 raw_spin_lock(¤t->pi_lock);
1049 rt_mutex_dequeue(lock, waiter);
1050 current->pi_blocked_on = NULL;
1051 raw_spin_unlock(¤t->pi_lock);
1054 * Only update priority if the waiter was the highest priority
1055 * waiter of the lock and there is an owner to update.
1057 if (!owner || !is_top_waiter)
1060 raw_spin_lock(&owner->pi_lock);
1062 rt_mutex_dequeue_pi(owner, waiter);
1064 if (rt_mutex_has_waiters(lock))
1065 rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock));
1067 rt_mutex_adjust_prio(owner);
1069 /* Store the lock on which owner is blocked or NULL */
1070 next_lock = task_blocked_on_lock(owner);
1072 raw_spin_unlock(&owner->pi_lock);
1075 * Don't walk the chain, if the owner task is not blocked
1081 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1082 get_task_struct(owner);
1084 raw_spin_unlock_irq(&lock->wait_lock);
1086 rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
1087 next_lock, NULL, current);
1089 raw_spin_lock_irq(&lock->wait_lock);
1093 * Recheck the pi chain, in case we got a priority setting
1095 * Called from sched_setscheduler
1097 void __sched rt_mutex_adjust_pi(struct task_struct *task)
1099 struct rt_mutex_waiter *waiter;
1100 struct rt_mutex *next_lock;
1101 unsigned long flags;
1103 raw_spin_lock_irqsave(&task->pi_lock, flags);
1105 waiter = task->pi_blocked_on;
1106 if (!waiter || rt_mutex_waiter_equal(waiter, task_to_waiter(task))) {
1107 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
1110 next_lock = waiter->lock;
1111 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
1113 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1114 get_task_struct(task);
1116 rt_mutex_adjust_prio_chain(task, RT_MUTEX_MIN_CHAINWALK, NULL,
1117 next_lock, NULL, task);
1120 void __sched rt_mutex_init_waiter(struct rt_mutex_waiter *waiter)
1122 debug_rt_mutex_init_waiter(waiter);
1123 RB_CLEAR_NODE(&waiter->pi_tree_entry);
1124 RB_CLEAR_NODE(&waiter->tree_entry);
1125 waiter->task = NULL;
1129 * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
1130 * @lock: the rt_mutex to take
1131 * @state: the state the task should block in (TASK_INTERRUPTIBLE
1132 * or TASK_UNINTERRUPTIBLE)
1133 * @timeout: the pre-initialized and started timer, or NULL for none
1134 * @waiter: the pre-initialized rt_mutex_waiter
1136 * Must be called with lock->wait_lock held and interrupts disabled
1138 static int __sched __rt_mutex_slowlock(struct rt_mutex *lock, unsigned int state,
1139 struct hrtimer_sleeper *timeout,
1140 struct rt_mutex_waiter *waiter)
1145 /* Try to acquire the lock: */
1146 if (try_to_take_rt_mutex(lock, current, waiter))
1149 if (timeout && !timeout->task) {
1153 if (signal_pending_state(state, current)) {
1158 raw_spin_unlock_irq(&lock->wait_lock);
1162 raw_spin_lock_irq(&lock->wait_lock);
1163 set_current_state(state);
1166 __set_current_state(TASK_RUNNING);
1170 static void __sched rt_mutex_handle_deadlock(int res, int detect_deadlock,
1171 struct rt_mutex_waiter *w)
1174 * If the result is not -EDEADLOCK or the caller requested
1175 * deadlock detection, nothing to do here.
1177 if (res != -EDEADLOCK || detect_deadlock)
1181 * Yell loudly and stop the task right here.
1183 WARN(1, "rtmutex deadlock detected\n");
1185 set_current_state(TASK_INTERRUPTIBLE);
1191 * Slow path lock function:
1193 static int __sched rt_mutex_slowlock(struct rt_mutex *lock, unsigned int state,
1194 struct hrtimer_sleeper *timeout,
1195 enum rtmutex_chainwalk chwalk)
1197 struct rt_mutex_waiter waiter;
1198 unsigned long flags;
1201 rt_mutex_init_waiter(&waiter);
1204 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1205 * be called in early boot if the cmpxchg() fast path is disabled
1206 * (debug, no architecture support). In this case we will acquire the
1207 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1208 * enable interrupts in that early boot case. So we need to use the
1209 * irqsave/restore variants.
1211 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1213 /* Try to acquire the lock again: */
1214 if (try_to_take_rt_mutex(lock, current, NULL)) {
1215 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1219 set_current_state(state);
1221 /* Setup the timer, when timeout != NULL */
1222 if (unlikely(timeout))
1223 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1225 ret = task_blocks_on_rt_mutex(lock, &waiter, current, chwalk);
1228 /* sleep on the mutex */
1229 ret = __rt_mutex_slowlock(lock, state, timeout, &waiter);
1231 if (unlikely(ret)) {
1232 __set_current_state(TASK_RUNNING);
1233 remove_waiter(lock, &waiter);
1234 rt_mutex_handle_deadlock(ret, chwalk, &waiter);
1238 * try_to_take_rt_mutex() sets the waiter bit
1239 * unconditionally. We might have to fix that up.
1241 fixup_rt_mutex_waiters(lock);
1243 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1245 /* Remove pending timer: */
1246 if (unlikely(timeout))
1247 hrtimer_cancel(&timeout->timer);
1249 debug_rt_mutex_free_waiter(&waiter);
1254 static int __sched __rt_mutex_slowtrylock(struct rt_mutex *lock)
1256 int ret = try_to_take_rt_mutex(lock, current, NULL);
1259 * try_to_take_rt_mutex() sets the lock waiters bit
1260 * unconditionally. Clean this up.
1262 fixup_rt_mutex_waiters(lock);
1268 * Slow path try-lock function:
1270 static int __sched rt_mutex_slowtrylock(struct rt_mutex *lock)
1272 unsigned long flags;
1276 * If the lock already has an owner we fail to get the lock.
1277 * This can be done without taking the @lock->wait_lock as
1278 * it is only being read, and this is a trylock anyway.
1280 if (rt_mutex_owner(lock))
1284 * The mutex has currently no owner. Lock the wait lock and try to
1285 * acquire the lock. We use irqsave here to support early boot calls.
1287 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1289 ret = __rt_mutex_slowtrylock(lock);
1291 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1297 * Performs the wakeup of the top-waiter and re-enables preemption.
1299 void __sched rt_mutex_postunlock(struct wake_q_head *wake_q)
1303 /* Pairs with preempt_disable() in mark_wakeup_next_waiter() */
1308 * Slow path to release a rt-mutex.
1310 * Return whether the current task needs to call rt_mutex_postunlock().
1312 static void __sched rt_mutex_slowunlock(struct rt_mutex *lock)
1314 DEFINE_WAKE_Q(wake_q);
1315 unsigned long flags;
1317 /* irqsave required to support early boot calls */
1318 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1320 debug_rt_mutex_unlock(lock);
1323 * We must be careful here if the fast path is enabled. If we
1324 * have no waiters queued we cannot set owner to NULL here
1327 * foo->lock->owner = NULL;
1328 * rtmutex_lock(foo->lock); <- fast path
1329 * free = atomic_dec_and_test(foo->refcnt);
1330 * rtmutex_unlock(foo->lock); <- fast path
1333 * raw_spin_unlock(foo->lock->wait_lock);
1335 * So for the fastpath enabled kernel:
1337 * Nothing can set the waiters bit as long as we hold
1338 * lock->wait_lock. So we do the following sequence:
1340 * owner = rt_mutex_owner(lock);
1341 * clear_rt_mutex_waiters(lock);
1342 * raw_spin_unlock(&lock->wait_lock);
1343 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1347 * The fastpath disabled variant is simple as all access to
1348 * lock->owner is serialized by lock->wait_lock:
1350 * lock->owner = NULL;
1351 * raw_spin_unlock(&lock->wait_lock);
1353 while (!rt_mutex_has_waiters(lock)) {
1354 /* Drops lock->wait_lock ! */
1355 if (unlock_rt_mutex_safe(lock, flags) == true)
1357 /* Relock the rtmutex and try again */
1358 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1362 * The wakeup next waiter path does not suffer from the above
1363 * race. See the comments there.
1365 * Queue the next waiter for wakeup once we release the wait_lock.
1367 mark_wakeup_next_waiter(&wake_q, lock);
1368 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1370 rt_mutex_postunlock(&wake_q);
1374 * debug aware fast / slowpath lock,trylock,unlock
1376 * The atomic acquire/release ops are compiled away, when either the
1377 * architecture does not support cmpxchg or when debugging is enabled.
1379 static __always_inline int __rt_mutex_lock(struct rt_mutex *lock, long state,
1380 unsigned int subclass)
1385 mutex_acquire(&lock->dep_map, subclass, 0, _RET_IP_);
1387 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1390 ret = rt_mutex_slowlock(lock, state, NULL, RT_MUTEX_MIN_CHAINWALK);
1392 mutex_release(&lock->dep_map, _RET_IP_);
1396 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1398 * rt_mutex_lock_nested - lock a rt_mutex
1400 * @lock: the rt_mutex to be locked
1401 * @subclass: the lockdep subclass
1403 void __sched rt_mutex_lock_nested(struct rt_mutex *lock, unsigned int subclass)
1405 __rt_mutex_lock(lock, TASK_UNINTERRUPTIBLE, subclass);
1407 EXPORT_SYMBOL_GPL(rt_mutex_lock_nested);
1409 #else /* !CONFIG_DEBUG_LOCK_ALLOC */
1412 * rt_mutex_lock - lock a rt_mutex
1414 * @lock: the rt_mutex to be locked
1416 void __sched rt_mutex_lock(struct rt_mutex *lock)
1418 __rt_mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0);
1420 EXPORT_SYMBOL_GPL(rt_mutex_lock);
1424 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
1426 * @lock: the rt_mutex to be locked
1430 * -EINTR when interrupted by a signal
1432 int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock)
1434 return __rt_mutex_lock(lock, TASK_INTERRUPTIBLE, 0);
1436 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
1439 * rt_mutex_trylock - try to lock a rt_mutex
1441 * @lock: the rt_mutex to be locked
1443 * This function can only be called in thread context. It's safe to call it
1444 * from atomic regions, but not from hard or soft interrupt context.
1450 int __sched rt_mutex_trylock(struct rt_mutex *lock)
1454 if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES) && WARN_ON_ONCE(!in_task()))
1458 * No lockdep annotation required because lockdep disables the fast
1461 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1464 ret = rt_mutex_slowtrylock(lock);
1466 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
1470 EXPORT_SYMBOL_GPL(rt_mutex_trylock);
1473 * rt_mutex_unlock - unlock a rt_mutex
1475 * @lock: the rt_mutex to be unlocked
1477 void __sched rt_mutex_unlock(struct rt_mutex *lock)
1479 mutex_release(&lock->dep_map, _RET_IP_);
1480 if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
1483 rt_mutex_slowunlock(lock);
1485 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
1488 * Futex variants, must not use fastpath.
1490 int __sched rt_mutex_futex_trylock(struct rt_mutex *lock)
1492 return rt_mutex_slowtrylock(lock);
1495 int __sched __rt_mutex_futex_trylock(struct rt_mutex *lock)
1497 return __rt_mutex_slowtrylock(lock);
1501 * __rt_mutex_futex_unlock - Futex variant, that since futex variants
1502 * do not use the fast-path, can be simple and will not need to retry.
1504 * @lock: The rt_mutex to be unlocked
1505 * @wake_q: The wake queue head from which to get the next lock waiter
1507 bool __sched __rt_mutex_futex_unlock(struct rt_mutex *lock,
1508 struct wake_q_head *wake_q)
1510 lockdep_assert_held(&lock->wait_lock);
1512 debug_rt_mutex_unlock(lock);
1514 if (!rt_mutex_has_waiters(lock)) {
1516 return false; /* done */
1520 * We've already deboosted, mark_wakeup_next_waiter() will
1521 * retain preempt_disabled when we drop the wait_lock, to
1522 * avoid inversion prior to the wakeup. preempt_disable()
1523 * therein pairs with rt_mutex_postunlock().
1525 mark_wakeup_next_waiter(wake_q, lock);
1527 return true; /* call postunlock() */
1530 void __sched rt_mutex_futex_unlock(struct rt_mutex *lock)
1532 DEFINE_WAKE_Q(wake_q);
1533 unsigned long flags;
1536 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1537 postunlock = __rt_mutex_futex_unlock(lock, &wake_q);
1538 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1541 rt_mutex_postunlock(&wake_q);
1545 * __rt_mutex_init - initialize the rt_mutex
1547 * @lock: The rt_mutex to be initialized
1548 * @name: The lock name used for debugging
1549 * @key: The lock class key used for debugging
1551 * Initialize the rt_mutex to unlocked state.
1553 * Initializing of a locked rt_mutex is not allowed
1555 void __sched __rt_mutex_init(struct rt_mutex *lock, const char *name,
1556 struct lock_class_key *key)
1558 debug_check_no_locks_freed((void *)lock, sizeof(*lock));
1559 lockdep_init_map(&lock->dep_map, name, key, 0);
1561 __rt_mutex_basic_init(lock);
1563 EXPORT_SYMBOL_GPL(__rt_mutex_init);
1566 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
1569 * @lock: the rt_mutex to be locked
1570 * @proxy_owner:the task to set as owner
1572 * No locking. Caller has to do serializing itself
1574 * Special API call for PI-futex support. This initializes the rtmutex and
1575 * assigns it to @proxy_owner. Concurrent operations on the rtmutex are not
1576 * possible at this point because the pi_state which contains the rtmutex
1577 * is not yet visible to other tasks.
1579 void __sched rt_mutex_init_proxy_locked(struct rt_mutex *lock,
1580 struct task_struct *proxy_owner)
1582 __rt_mutex_basic_init(lock);
1583 rt_mutex_set_owner(lock, proxy_owner);
1587 * rt_mutex_proxy_unlock - release a lock on behalf of owner
1589 * @lock: the rt_mutex to be locked
1591 * No locking. Caller has to do serializing itself
1593 * Special API call for PI-futex support. This merrily cleans up the rtmutex
1594 * (debugging) state. Concurrent operations on this rt_mutex are not
1595 * possible because it belongs to the pi_state which is about to be freed
1596 * and it is not longer visible to other tasks.
1598 void __sched rt_mutex_proxy_unlock(struct rt_mutex *lock)
1600 debug_rt_mutex_proxy_unlock(lock);
1601 rt_mutex_set_owner(lock, NULL);
1605 * __rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1606 * @lock: the rt_mutex to take
1607 * @waiter: the pre-initialized rt_mutex_waiter
1608 * @task: the task to prepare
1610 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
1611 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
1613 * NOTE: does _NOT_ remove the @waiter on failure; must either call
1614 * rt_mutex_wait_proxy_lock() or rt_mutex_cleanup_proxy_lock() after this.
1617 * 0 - task blocked on lock
1618 * 1 - acquired the lock for task, caller should wake it up
1621 * Special API call for PI-futex support.
1623 int __sched __rt_mutex_start_proxy_lock(struct rt_mutex *lock,
1624 struct rt_mutex_waiter *waiter,
1625 struct task_struct *task)
1629 lockdep_assert_held(&lock->wait_lock);
1631 if (try_to_take_rt_mutex(lock, task, NULL))
1634 /* We enforce deadlock detection for futexes */
1635 ret = task_blocks_on_rt_mutex(lock, waiter, task,
1636 RT_MUTEX_FULL_CHAINWALK);
1638 if (ret && !rt_mutex_owner(lock)) {
1640 * Reset the return value. We might have
1641 * returned with -EDEADLK and the owner
1642 * released the lock while we were walking the
1643 * pi chain. Let the waiter sort it out.
1652 * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1653 * @lock: the rt_mutex to take
1654 * @waiter: the pre-initialized rt_mutex_waiter
1655 * @task: the task to prepare
1657 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
1658 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
1660 * NOTE: unlike __rt_mutex_start_proxy_lock this _DOES_ remove the @waiter
1664 * 0 - task blocked on lock
1665 * 1 - acquired the lock for task, caller should wake it up
1668 * Special API call for PI-futex support.
1670 int __sched rt_mutex_start_proxy_lock(struct rt_mutex *lock,
1671 struct rt_mutex_waiter *waiter,
1672 struct task_struct *task)
1676 raw_spin_lock_irq(&lock->wait_lock);
1677 ret = __rt_mutex_start_proxy_lock(lock, waiter, task);
1679 remove_waiter(lock, waiter);
1680 raw_spin_unlock_irq(&lock->wait_lock);
1686 * rt_mutex_wait_proxy_lock() - Wait for lock acquisition
1687 * @lock: the rt_mutex we were woken on
1688 * @to: the timeout, null if none. hrtimer should already have
1690 * @waiter: the pre-initialized rt_mutex_waiter
1692 * Wait for the lock acquisition started on our behalf by
1693 * rt_mutex_start_proxy_lock(). Upon failure, the caller must call
1694 * rt_mutex_cleanup_proxy_lock().
1698 * <0 - error, one of -EINTR, -ETIMEDOUT
1700 * Special API call for PI-futex support
1702 int __sched rt_mutex_wait_proxy_lock(struct rt_mutex *lock,
1703 struct hrtimer_sleeper *to,
1704 struct rt_mutex_waiter *waiter)
1708 raw_spin_lock_irq(&lock->wait_lock);
1709 /* sleep on the mutex */
1710 set_current_state(TASK_INTERRUPTIBLE);
1711 ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
1713 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1714 * have to fix that up.
1716 fixup_rt_mutex_waiters(lock);
1717 raw_spin_unlock_irq(&lock->wait_lock);
1723 * rt_mutex_cleanup_proxy_lock() - Cleanup failed lock acquisition
1724 * @lock: the rt_mutex we were woken on
1725 * @waiter: the pre-initialized rt_mutex_waiter
1727 * Attempt to clean up after a failed __rt_mutex_start_proxy_lock() or
1728 * rt_mutex_wait_proxy_lock().
1730 * Unless we acquired the lock; we're still enqueued on the wait-list and can
1731 * in fact still be granted ownership until we're removed. Therefore we can
1732 * find we are in fact the owner and must disregard the
1733 * rt_mutex_wait_proxy_lock() failure.
1736 * true - did the cleanup, we done.
1737 * false - we acquired the lock after rt_mutex_wait_proxy_lock() returned,
1738 * caller should disregards its return value.
1740 * Special API call for PI-futex support
1742 bool __sched rt_mutex_cleanup_proxy_lock(struct rt_mutex *lock,
1743 struct rt_mutex_waiter *waiter)
1745 bool cleanup = false;
1747 raw_spin_lock_irq(&lock->wait_lock);
1749 * Do an unconditional try-lock, this deals with the lock stealing
1750 * state where __rt_mutex_futex_unlock() -> mark_wakeup_next_waiter()
1751 * sets a NULL owner.
1753 * We're not interested in the return value, because the subsequent
1754 * test on rt_mutex_owner() will infer that. If the trylock succeeded,
1755 * we will own the lock and it will have removed the waiter. If we
1756 * failed the trylock, we're still not owner and we need to remove
1759 try_to_take_rt_mutex(lock, current, waiter);
1761 * Unless we're the owner; we're still enqueued on the wait_list.
1762 * So check if we became owner, if not, take us off the wait_list.
1764 if (rt_mutex_owner(lock) != current) {
1765 remove_waiter(lock, waiter);
1769 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1770 * have to fix that up.
1772 fixup_rt_mutex_waiters(lock);
1774 raw_spin_unlock_irq(&lock->wait_lock);
1779 #ifdef CONFIG_DEBUG_RT_MUTEXES
1780 void rt_mutex_debug_task_free(struct task_struct *task)
1782 DEBUG_LOCKS_WARN_ON(!RB_EMPTY_ROOT(&task->pi_waiters.rb_root));
1783 DEBUG_LOCKS_WARN_ON(task->pi_blocked_on);