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 WRITE_ONCE(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_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 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 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 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 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 }
233 rt_mutex_waiter_less(struct rt_mutex_waiter *left,
234 struct rt_mutex_waiter *right)
236 if (left->prio < right->prio)
240 * If both waiters have dl_prio(), we check the deadlines of the
242 * If left waiter has a dl_prio(), and we didn't return 1 above,
243 * then right waiter has a dl_prio() too.
245 if (dl_prio(left->prio))
246 return dl_time_before(left->deadline, right->deadline);
252 rt_mutex_waiter_equal(struct rt_mutex_waiter *left,
253 struct rt_mutex_waiter *right)
255 if (left->prio != right->prio)
259 * If both waiters have dl_prio(), we check the deadlines of the
261 * If left waiter has a dl_prio(), and we didn't return 0 above,
262 * then right waiter has a dl_prio() too.
264 if (dl_prio(left->prio))
265 return left->deadline == right->deadline;
271 rt_mutex_enqueue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
273 struct rb_node **link = &lock->waiters.rb_root.rb_node;
274 struct rb_node *parent = NULL;
275 struct rt_mutex_waiter *entry;
276 bool leftmost = true;
280 entry = rb_entry(parent, struct rt_mutex_waiter, tree_entry);
281 if (rt_mutex_waiter_less(waiter, entry)) {
282 link = &parent->rb_left;
284 link = &parent->rb_right;
289 rb_link_node(&waiter->tree_entry, parent, link);
290 rb_insert_color_cached(&waiter->tree_entry, &lock->waiters, leftmost);
294 rt_mutex_dequeue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
296 if (RB_EMPTY_NODE(&waiter->tree_entry))
299 rb_erase_cached(&waiter->tree_entry, &lock->waiters);
300 RB_CLEAR_NODE(&waiter->tree_entry);
304 rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
306 struct rb_node **link = &task->pi_waiters.rb_root.rb_node;
307 struct rb_node *parent = NULL;
308 struct rt_mutex_waiter *entry;
309 bool leftmost = true;
313 entry = rb_entry(parent, struct rt_mutex_waiter, pi_tree_entry);
314 if (rt_mutex_waiter_less(waiter, entry)) {
315 link = &parent->rb_left;
317 link = &parent->rb_right;
322 rb_link_node(&waiter->pi_tree_entry, parent, link);
323 rb_insert_color_cached(&waiter->pi_tree_entry, &task->pi_waiters, leftmost);
327 rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
329 if (RB_EMPTY_NODE(&waiter->pi_tree_entry))
332 rb_erase_cached(&waiter->pi_tree_entry, &task->pi_waiters);
333 RB_CLEAR_NODE(&waiter->pi_tree_entry);
336 static void rt_mutex_adjust_prio(struct task_struct *p)
338 struct task_struct *pi_task = NULL;
340 lockdep_assert_held(&p->pi_lock);
342 if (task_has_pi_waiters(p))
343 pi_task = task_top_pi_waiter(p)->task;
345 rt_mutex_setprio(p, pi_task);
349 * Deadlock detection is conditional:
351 * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
352 * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
354 * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
355 * conducted independent of the detect argument.
357 * If the waiter argument is NULL this indicates the deboost path and
358 * deadlock detection is disabled independent of the detect argument
359 * and the config settings.
361 static bool rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
362 enum rtmutex_chainwalk chwalk)
365 * This is just a wrapper function for the following call,
366 * because debug_rt_mutex_detect_deadlock() smells like a magic
367 * debug feature and I wanted to keep the cond function in the
368 * main source file along with the comments instead of having
369 * two of the same in the headers.
371 return debug_rt_mutex_detect_deadlock(waiter, chwalk);
375 * Max number of times we'll walk the boosting chain:
377 int max_lock_depth = 1024;
379 static inline struct rt_mutex *task_blocked_on_lock(struct task_struct *p)
381 return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
385 * Adjust the priority chain. Also used for deadlock detection.
386 * Decreases task's usage by one - may thus free the task.
388 * @task: the task owning the mutex (owner) for which a chain walk is
390 * @chwalk: do we have to carry out deadlock detection?
391 * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
392 * things for a task that has just got its priority adjusted, and
393 * is waiting on a mutex)
394 * @next_lock: the mutex on which the owner of @orig_lock was blocked before
395 * we dropped its pi_lock. Is never dereferenced, only used for
396 * comparison to detect lock chain changes.
397 * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
398 * its priority to the mutex owner (can be NULL in the case
399 * depicted above or if the top waiter is gone away and we are
400 * actually deboosting the owner)
401 * @top_task: the current top waiter
403 * Returns 0 or -EDEADLK.
405 * Chain walk basics and protection scope
407 * [R] refcount on task
408 * [P] task->pi_lock held
409 * [L] rtmutex->wait_lock held
411 * Step Description Protected by
412 * function arguments:
414 * @orig_lock if != NULL @top_task is blocked on it
415 * @next_lock Unprotected. Cannot be
416 * dereferenced. Only used for
418 * @orig_waiter if != NULL @top_task is blocked on it
419 * @top_task current, or in case of proxy
420 * locking protected by calling
423 * loop_sanity_check();
425 * [1] lock(task->pi_lock); [R] acquire [P]
426 * [2] waiter = task->pi_blocked_on; [P]
427 * [3] check_exit_conditions_1(); [P]
428 * [4] lock = waiter->lock; [P]
429 * [5] if (!try_lock(lock->wait_lock)) { [P] try to acquire [L]
430 * unlock(task->pi_lock); release [P]
433 * [6] check_exit_conditions_2(); [P] + [L]
434 * [7] requeue_lock_waiter(lock, waiter); [P] + [L]
435 * [8] unlock(task->pi_lock); release [P]
436 * put_task_struct(task); release [R]
437 * [9] check_exit_conditions_3(); [L]
438 * [10] task = owner(lock); [L]
439 * get_task_struct(task); [L] acquire [R]
440 * lock(task->pi_lock); [L] acquire [P]
441 * [11] requeue_pi_waiter(tsk, waiters(lock));[P] + [L]
442 * [12] check_exit_conditions_4(); [P] + [L]
443 * [13] unlock(task->pi_lock); release [P]
444 * unlock(lock->wait_lock); release [L]
447 static int rt_mutex_adjust_prio_chain(struct task_struct *task,
448 enum rtmutex_chainwalk chwalk,
449 struct rt_mutex *orig_lock,
450 struct rt_mutex *next_lock,
451 struct rt_mutex_waiter *orig_waiter,
452 struct task_struct *top_task)
454 struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
455 struct rt_mutex_waiter *prerequeue_top_waiter;
456 int ret = 0, depth = 0;
457 struct rt_mutex *lock;
458 bool detect_deadlock;
461 detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
464 * The (de)boosting is a step by step approach with a lot of
465 * pitfalls. We want this to be preemptible and we want hold a
466 * maximum of two locks per step. So we have to check
467 * carefully whether things change under us.
471 * We limit the lock chain length for each invocation.
473 if (++depth > max_lock_depth) {
477 * Print this only once. If the admin changes the limit,
478 * print a new message when reaching the limit again.
480 if (prev_max != max_lock_depth) {
481 prev_max = max_lock_depth;
482 printk(KERN_WARNING "Maximum lock depth %d reached "
483 "task: %s (%d)\n", max_lock_depth,
484 top_task->comm, task_pid_nr(top_task));
486 put_task_struct(task);
492 * We are fully preemptible here and only hold the refcount on
493 * @task. So everything can have changed under us since the
494 * caller or our own code below (goto retry/again) dropped all
499 * [1] Task cannot go away as we did a get_task() before !
501 raw_spin_lock_irq(&task->pi_lock);
504 * [2] Get the waiter on which @task is blocked on.
506 waiter = task->pi_blocked_on;
509 * [3] check_exit_conditions_1() protected by task->pi_lock.
513 * Check whether the end of the boosting chain has been
514 * reached or the state of the chain has changed while we
521 * Check the orig_waiter state. After we dropped the locks,
522 * the previous owner of the lock might have released the lock.
524 if (orig_waiter && !rt_mutex_owner(orig_lock))
528 * We dropped all locks after taking a refcount on @task, so
529 * the task might have moved on in the lock chain or even left
530 * the chain completely and blocks now on an unrelated lock or
533 * We stored the lock on which @task was blocked in @next_lock,
534 * so we can detect the chain change.
536 if (next_lock != waiter->lock)
540 * Drop out, when the task has no waiters. Note,
541 * top_waiter can be NULL, when we are in the deboosting
545 if (!task_has_pi_waiters(task))
548 * If deadlock detection is off, we stop here if we
549 * are not the top pi waiter of the task. If deadlock
550 * detection is enabled we continue, but stop the
551 * requeueing in the chain walk.
553 if (top_waiter != task_top_pi_waiter(task)) {
554 if (!detect_deadlock)
562 * If the waiter priority is the same as the task priority
563 * then there is no further priority adjustment necessary. If
564 * deadlock detection is off, we stop the chain walk. If its
565 * enabled we continue, but stop the requeueing in the chain
568 if (rt_mutex_waiter_equal(waiter, task_to_waiter(task))) {
569 if (!detect_deadlock)
576 * [4] Get the next lock
580 * [5] We need to trylock here as we are holding task->pi_lock,
581 * which is the reverse lock order versus the other rtmutex
584 if (!raw_spin_trylock(&lock->wait_lock)) {
585 raw_spin_unlock_irq(&task->pi_lock);
591 * [6] check_exit_conditions_2() protected by task->pi_lock and
594 * Deadlock detection. If the lock is the same as the original
595 * lock which caused us to walk the lock chain or if the
596 * current lock is owned by the task which initiated the chain
597 * walk, we detected a deadlock.
599 if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
600 debug_rt_mutex_deadlock(chwalk, orig_waiter, lock);
601 raw_spin_unlock(&lock->wait_lock);
607 * If we just follow the lock chain for deadlock detection, no
608 * need to do all the requeue operations. To avoid a truckload
609 * of conditionals around the various places below, just do the
610 * minimum chain walk checks.
614 * No requeue[7] here. Just release @task [8]
616 raw_spin_unlock(&task->pi_lock);
617 put_task_struct(task);
620 * [9] check_exit_conditions_3 protected by lock->wait_lock.
621 * If there is no owner of the lock, end of chain.
623 if (!rt_mutex_owner(lock)) {
624 raw_spin_unlock_irq(&lock->wait_lock);
628 /* [10] Grab the next task, i.e. owner of @lock */
629 task = get_task_struct(rt_mutex_owner(lock));
630 raw_spin_lock(&task->pi_lock);
633 * No requeue [11] here. We just do deadlock detection.
635 * [12] Store whether owner is blocked
636 * itself. Decision is made after dropping the locks
638 next_lock = task_blocked_on_lock(task);
640 * Get the top waiter for the next iteration
642 top_waiter = rt_mutex_top_waiter(lock);
644 /* [13] Drop locks */
645 raw_spin_unlock(&task->pi_lock);
646 raw_spin_unlock_irq(&lock->wait_lock);
648 /* If owner is not blocked, end of chain. */
655 * Store the current top waiter before doing the requeue
656 * operation on @lock. We need it for the boost/deboost
659 prerequeue_top_waiter = rt_mutex_top_waiter(lock);
661 /* [7] Requeue the waiter in the lock waiter tree. */
662 rt_mutex_dequeue(lock, waiter);
665 * Update the waiter prio fields now that we're dequeued.
667 * These values can have changed through either:
669 * sys_sched_set_scheduler() / sys_sched_setattr()
673 * DL CBS enforcement advancing the effective deadline.
675 * Even though pi_waiters also uses these fields, and that tree is only
676 * updated in [11], we can do this here, since we hold [L], which
677 * serializes all pi_waiters access and rb_erase() does not care about
678 * the values of the node being removed.
680 waiter->prio = task->prio;
681 waiter->deadline = task->dl.deadline;
683 rt_mutex_enqueue(lock, waiter);
685 /* [8] Release the task */
686 raw_spin_unlock(&task->pi_lock);
687 put_task_struct(task);
690 * [9] check_exit_conditions_3 protected by lock->wait_lock.
692 * We must abort the chain walk if there is no lock owner even
693 * in the dead lock detection case, as we have nothing to
694 * follow here. This is the end of the chain we are walking.
696 if (!rt_mutex_owner(lock)) {
698 * If the requeue [7] above changed the top waiter,
699 * then we need to wake the new top waiter up to try
702 if (prerequeue_top_waiter != rt_mutex_top_waiter(lock))
703 wake_up_process(rt_mutex_top_waiter(lock)->task);
704 raw_spin_unlock_irq(&lock->wait_lock);
708 /* [10] Grab the next task, i.e. the owner of @lock */
709 task = get_task_struct(rt_mutex_owner(lock));
710 raw_spin_lock(&task->pi_lock);
712 /* [11] requeue the pi waiters if necessary */
713 if (waiter == rt_mutex_top_waiter(lock)) {
715 * The waiter became the new top (highest priority)
716 * waiter on the lock. Replace the previous top waiter
717 * in the owner tasks pi waiters tree with this waiter
718 * and adjust the priority of the owner.
720 rt_mutex_dequeue_pi(task, prerequeue_top_waiter);
721 rt_mutex_enqueue_pi(task, waiter);
722 rt_mutex_adjust_prio(task);
724 } else if (prerequeue_top_waiter == waiter) {
726 * The waiter was the top waiter on the lock, but is
727 * no longer the top prority waiter. Replace waiter in
728 * the owner tasks pi waiters tree with the new top
729 * (highest priority) waiter and adjust the priority
731 * The new top waiter is stored in @waiter so that
732 * @waiter == @top_waiter evaluates to true below and
733 * we continue to deboost the rest of the chain.
735 rt_mutex_dequeue_pi(task, waiter);
736 waiter = rt_mutex_top_waiter(lock);
737 rt_mutex_enqueue_pi(task, waiter);
738 rt_mutex_adjust_prio(task);
741 * Nothing changed. No need to do any priority
747 * [12] check_exit_conditions_4() protected by task->pi_lock
748 * and lock->wait_lock. The actual decisions are made after we
751 * Check whether the task which owns the current lock is pi
752 * blocked itself. If yes we store a pointer to the lock for
753 * the lock chain change detection above. After we dropped
754 * task->pi_lock next_lock cannot be dereferenced anymore.
756 next_lock = task_blocked_on_lock(task);
758 * Store the top waiter of @lock for the end of chain walk
761 top_waiter = rt_mutex_top_waiter(lock);
763 /* [13] Drop the locks */
764 raw_spin_unlock(&task->pi_lock);
765 raw_spin_unlock_irq(&lock->wait_lock);
768 * Make the actual exit decisions [12], based on the stored
771 * We reached the end of the lock chain. Stop right here. No
772 * point to go back just to figure that out.
778 * If the current waiter is not the top waiter on the lock,
779 * then we can stop the chain walk here if we are not in full
780 * deadlock detection mode.
782 if (!detect_deadlock && waiter != top_waiter)
788 raw_spin_unlock_irq(&task->pi_lock);
790 put_task_struct(task);
796 * Try to take an rt-mutex
798 * Must be called with lock->wait_lock held and interrupts disabled
800 * @lock: The lock to be acquired.
801 * @task: The task which wants to acquire the lock
802 * @waiter: The waiter that is queued to the lock's wait tree if the
803 * callsite called task_blocked_on_lock(), otherwise NULL
805 static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
806 struct rt_mutex_waiter *waiter)
808 lockdep_assert_held(&lock->wait_lock);
811 * Before testing whether we can acquire @lock, we set the
812 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
813 * other tasks which try to modify @lock into the slow path
814 * and they serialize on @lock->wait_lock.
816 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
817 * as explained at the top of this file if and only if:
819 * - There is a lock owner. The caller must fixup the
820 * transient state if it does a trylock or leaves the lock
821 * function due to a signal or timeout.
823 * - @task acquires the lock and there are no other
824 * waiters. This is undone in rt_mutex_set_owner(@task) at
825 * the end of this function.
827 mark_rt_mutex_waiters(lock);
830 * If @lock has an owner, give up.
832 if (rt_mutex_owner(lock))
836 * If @waiter != NULL, @task has already enqueued the waiter
837 * into @lock waiter tree. If @waiter == NULL then this is a
842 * If waiter is not the highest priority waiter of
845 if (waiter != rt_mutex_top_waiter(lock))
849 * We can acquire the lock. Remove the waiter from the
852 rt_mutex_dequeue(lock, waiter);
856 * If the lock has waiters already we check whether @task is
857 * eligible to take over the lock.
859 * If there are no other waiters, @task can acquire
860 * the lock. @task->pi_blocked_on is NULL, so it does
861 * not need to be dequeued.
863 if (rt_mutex_has_waiters(lock)) {
865 * If @task->prio is greater than or equal to
866 * the top waiter priority (kernel view),
869 if (!rt_mutex_waiter_less(task_to_waiter(task),
870 rt_mutex_top_waiter(lock)))
874 * The current top waiter stays enqueued. We
875 * don't have to change anything in the lock
880 * No waiters. Take the lock without the
881 * pi_lock dance.@task->pi_blocked_on is NULL
882 * and we have no waiters to enqueue in @task
890 * Clear @task->pi_blocked_on. Requires protection by
891 * @task->pi_lock. Redundant operation for the @waiter == NULL
892 * case, but conditionals are more expensive than a redundant
895 raw_spin_lock(&task->pi_lock);
896 task->pi_blocked_on = NULL;
898 * Finish the lock acquisition. @task is the new owner. If
899 * other waiters exist we have to insert the highest priority
900 * waiter into @task->pi_waiters tree.
902 if (rt_mutex_has_waiters(lock))
903 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
904 raw_spin_unlock(&task->pi_lock);
907 /* We got the lock. */
908 debug_rt_mutex_lock(lock);
911 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
912 * are still waiters or clears it.
914 rt_mutex_set_owner(lock, task);
920 * Task blocks on lock.
922 * Prepare waiter and propagate pi chain
924 * This must be called with lock->wait_lock held and interrupts disabled
926 static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
927 struct rt_mutex_waiter *waiter,
928 struct task_struct *task,
929 enum rtmutex_chainwalk chwalk)
931 struct task_struct *owner = rt_mutex_owner(lock);
932 struct rt_mutex_waiter *top_waiter = waiter;
933 struct rt_mutex *next_lock;
934 int chain_walk = 0, res;
936 lockdep_assert_held(&lock->wait_lock);
939 * Early deadlock detection. We really don't want the task to
940 * enqueue on itself just to untangle the mess later. It's not
941 * only an optimization. We drop the locks, so another waiter
942 * can come in before the chain walk detects the deadlock. So
943 * the other will detect the deadlock and return -EDEADLOCK,
944 * which is wrong, as the other waiter is not in a deadlock
950 raw_spin_lock(&task->pi_lock);
953 waiter->prio = task->prio;
954 waiter->deadline = task->dl.deadline;
956 /* Get the top priority waiter on the lock */
957 if (rt_mutex_has_waiters(lock))
958 top_waiter = rt_mutex_top_waiter(lock);
959 rt_mutex_enqueue(lock, waiter);
961 task->pi_blocked_on = waiter;
963 raw_spin_unlock(&task->pi_lock);
968 raw_spin_lock(&owner->pi_lock);
969 if (waiter == rt_mutex_top_waiter(lock)) {
970 rt_mutex_dequeue_pi(owner, top_waiter);
971 rt_mutex_enqueue_pi(owner, waiter);
973 rt_mutex_adjust_prio(owner);
974 if (owner->pi_blocked_on)
976 } else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
980 /* Store the lock on which owner is blocked or NULL */
981 next_lock = task_blocked_on_lock(owner);
983 raw_spin_unlock(&owner->pi_lock);
985 * Even if full deadlock detection is on, if the owner is not
986 * blocked itself, we can avoid finding this out in the chain
989 if (!chain_walk || !next_lock)
993 * The owner can't disappear while holding a lock,
994 * so the owner struct is protected by wait_lock.
995 * Gets dropped in rt_mutex_adjust_prio_chain()!
997 get_task_struct(owner);
999 raw_spin_unlock_irq(&lock->wait_lock);
1001 res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
1002 next_lock, waiter, task);
1004 raw_spin_lock_irq(&lock->wait_lock);
1010 * Remove the top waiter from the current tasks pi waiter tree and
1013 * Called with lock->wait_lock held and interrupts disabled.
1015 static void mark_wakeup_next_waiter(struct wake_q_head *wake_q,
1016 struct rt_mutex *lock)
1018 struct rt_mutex_waiter *waiter;
1020 raw_spin_lock(¤t->pi_lock);
1022 waiter = rt_mutex_top_waiter(lock);
1025 * Remove it from current->pi_waiters and deboost.
1027 * We must in fact deboost here in order to ensure we call
1028 * rt_mutex_setprio() to update p->pi_top_task before the
1031 rt_mutex_dequeue_pi(current, waiter);
1032 rt_mutex_adjust_prio(current);
1035 * As we are waking up the top waiter, and the waiter stays
1036 * queued on the lock until it gets the lock, this lock
1037 * obviously has waiters. Just set the bit here and this has
1038 * the added benefit of forcing all new tasks into the
1039 * slow path making sure no task of lower priority than
1040 * the top waiter can steal this lock.
1042 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
1045 * We deboosted before waking the top waiter task such that we don't
1046 * run two tasks with the 'same' priority (and ensure the
1047 * p->pi_top_task pointer points to a blocked task). This however can
1048 * lead to priority inversion if we would get preempted after the
1049 * deboost but before waking our donor task, hence the preempt_disable()
1052 * Pairs with preempt_enable() in rt_mutex_postunlock();
1055 wake_q_add(wake_q, waiter->task);
1056 raw_spin_unlock(¤t->pi_lock);
1060 * Remove a waiter from a lock and give up
1062 * Must be called with lock->wait_lock held and interrupts disabled. I must
1063 * have just failed to try_to_take_rt_mutex().
1065 static void remove_waiter(struct rt_mutex *lock,
1066 struct rt_mutex_waiter *waiter)
1068 bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
1069 struct task_struct *owner = rt_mutex_owner(lock);
1070 struct rt_mutex *next_lock;
1072 lockdep_assert_held(&lock->wait_lock);
1074 raw_spin_lock(¤t->pi_lock);
1075 rt_mutex_dequeue(lock, waiter);
1076 current->pi_blocked_on = NULL;
1077 raw_spin_unlock(¤t->pi_lock);
1080 * Only update priority if the waiter was the highest priority
1081 * waiter of the lock and there is an owner to update.
1083 if (!owner || !is_top_waiter)
1086 raw_spin_lock(&owner->pi_lock);
1088 rt_mutex_dequeue_pi(owner, waiter);
1090 if (rt_mutex_has_waiters(lock))
1091 rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock));
1093 rt_mutex_adjust_prio(owner);
1095 /* Store the lock on which owner is blocked or NULL */
1096 next_lock = task_blocked_on_lock(owner);
1098 raw_spin_unlock(&owner->pi_lock);
1101 * Don't walk the chain, if the owner task is not blocked
1107 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1108 get_task_struct(owner);
1110 raw_spin_unlock_irq(&lock->wait_lock);
1112 rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
1113 next_lock, NULL, current);
1115 raw_spin_lock_irq(&lock->wait_lock);
1119 * Recheck the pi chain, in case we got a priority setting
1121 * Called from sched_setscheduler
1123 void rt_mutex_adjust_pi(struct task_struct *task)
1125 struct rt_mutex_waiter *waiter;
1126 struct rt_mutex *next_lock;
1127 unsigned long flags;
1129 raw_spin_lock_irqsave(&task->pi_lock, flags);
1131 waiter = task->pi_blocked_on;
1132 if (!waiter || rt_mutex_waiter_equal(waiter, task_to_waiter(task))) {
1133 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
1136 next_lock = waiter->lock;
1137 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
1139 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1140 get_task_struct(task);
1142 rt_mutex_adjust_prio_chain(task, RT_MUTEX_MIN_CHAINWALK, NULL,
1143 next_lock, NULL, task);
1146 void rt_mutex_init_waiter(struct rt_mutex_waiter *waiter)
1148 debug_rt_mutex_init_waiter(waiter);
1149 RB_CLEAR_NODE(&waiter->pi_tree_entry);
1150 RB_CLEAR_NODE(&waiter->tree_entry);
1151 waiter->task = NULL;
1155 * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
1156 * @lock: the rt_mutex to take
1157 * @state: the state the task should block in (TASK_INTERRUPTIBLE
1158 * or TASK_UNINTERRUPTIBLE)
1159 * @timeout: the pre-initialized and started timer, or NULL for none
1160 * @waiter: the pre-initialized rt_mutex_waiter
1162 * Must be called with lock->wait_lock held and interrupts disabled
1165 __rt_mutex_slowlock(struct rt_mutex *lock, int state,
1166 struct hrtimer_sleeper *timeout,
1167 struct rt_mutex_waiter *waiter)
1172 /* Try to acquire the lock: */
1173 if (try_to_take_rt_mutex(lock, current, waiter))
1177 * TASK_INTERRUPTIBLE checks for signals and
1178 * timeout. Ignored otherwise.
1180 if (likely(state == TASK_INTERRUPTIBLE)) {
1181 /* Signal pending? */
1182 if (signal_pending(current))
1184 if (timeout && !timeout->task)
1190 raw_spin_unlock_irq(&lock->wait_lock);
1192 debug_rt_mutex_print_deadlock(waiter);
1196 raw_spin_lock_irq(&lock->wait_lock);
1197 set_current_state(state);
1200 __set_current_state(TASK_RUNNING);
1204 static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
1205 struct rt_mutex_waiter *w)
1208 * If the result is not -EDEADLOCK or the caller requested
1209 * deadlock detection, nothing to do here.
1211 if (res != -EDEADLOCK || detect_deadlock)
1215 * Yell lowdly and stop the task right here.
1217 rt_mutex_print_deadlock(w);
1219 set_current_state(TASK_INTERRUPTIBLE);
1225 * Slow path lock function:
1228 rt_mutex_slowlock(struct rt_mutex *lock, int state,
1229 struct hrtimer_sleeper *timeout,
1230 enum rtmutex_chainwalk chwalk)
1232 struct rt_mutex_waiter waiter;
1233 unsigned long flags;
1236 rt_mutex_init_waiter(&waiter);
1239 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1240 * be called in early boot if the cmpxchg() fast path is disabled
1241 * (debug, no architecture support). In this case we will acquire the
1242 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1243 * enable interrupts in that early boot case. So we need to use the
1244 * irqsave/restore variants.
1246 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1248 /* Try to acquire the lock again: */
1249 if (try_to_take_rt_mutex(lock, current, NULL)) {
1250 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1254 set_current_state(state);
1256 /* Setup the timer, when timeout != NULL */
1257 if (unlikely(timeout))
1258 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
1260 ret = task_blocks_on_rt_mutex(lock, &waiter, current, chwalk);
1263 /* sleep on the mutex */
1264 ret = __rt_mutex_slowlock(lock, state, timeout, &waiter);
1266 if (unlikely(ret)) {
1267 __set_current_state(TASK_RUNNING);
1268 remove_waiter(lock, &waiter);
1269 rt_mutex_handle_deadlock(ret, chwalk, &waiter);
1273 * try_to_take_rt_mutex() sets the waiter bit
1274 * unconditionally. We might have to fix that up.
1276 fixup_rt_mutex_waiters(lock);
1278 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1280 /* Remove pending timer: */
1281 if (unlikely(timeout))
1282 hrtimer_cancel(&timeout->timer);
1284 debug_rt_mutex_free_waiter(&waiter);
1289 static inline int __rt_mutex_slowtrylock(struct rt_mutex *lock)
1291 int ret = try_to_take_rt_mutex(lock, current, NULL);
1294 * try_to_take_rt_mutex() sets the lock waiters bit
1295 * unconditionally. Clean this up.
1297 fixup_rt_mutex_waiters(lock);
1303 * Slow path try-lock function:
1305 static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
1307 unsigned long flags;
1311 * If the lock already has an owner we fail to get the lock.
1312 * This can be done without taking the @lock->wait_lock as
1313 * it is only being read, and this is a trylock anyway.
1315 if (rt_mutex_owner(lock))
1319 * The mutex has currently no owner. Lock the wait lock and try to
1320 * acquire the lock. We use irqsave here to support early boot calls.
1322 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1324 ret = __rt_mutex_slowtrylock(lock);
1326 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1332 * Slow path to release a rt-mutex.
1334 * Return whether the current task needs to call rt_mutex_postunlock().
1336 static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
1337 struct wake_q_head *wake_q)
1339 unsigned long flags;
1341 /* irqsave required to support early boot calls */
1342 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1344 debug_rt_mutex_unlock(lock);
1347 * We must be careful here if the fast path is enabled. If we
1348 * have no waiters queued we cannot set owner to NULL here
1351 * foo->lock->owner = NULL;
1352 * rtmutex_lock(foo->lock); <- fast path
1353 * free = atomic_dec_and_test(foo->refcnt);
1354 * rtmutex_unlock(foo->lock); <- fast path
1357 * raw_spin_unlock(foo->lock->wait_lock);
1359 * So for the fastpath enabled kernel:
1361 * Nothing can set the waiters bit as long as we hold
1362 * lock->wait_lock. So we do the following sequence:
1364 * owner = rt_mutex_owner(lock);
1365 * clear_rt_mutex_waiters(lock);
1366 * raw_spin_unlock(&lock->wait_lock);
1367 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1371 * The fastpath disabled variant is simple as all access to
1372 * lock->owner is serialized by lock->wait_lock:
1374 * lock->owner = NULL;
1375 * raw_spin_unlock(&lock->wait_lock);
1377 while (!rt_mutex_has_waiters(lock)) {
1378 /* Drops lock->wait_lock ! */
1379 if (unlock_rt_mutex_safe(lock, flags) == true)
1381 /* Relock the rtmutex and try again */
1382 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1386 * The wakeup next waiter path does not suffer from the above
1387 * race. See the comments there.
1389 * Queue the next waiter for wakeup once we release the wait_lock.
1391 mark_wakeup_next_waiter(wake_q, lock);
1392 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1394 return true; /* call rt_mutex_postunlock() */
1398 * debug aware fast / slowpath lock,trylock,unlock
1400 * The atomic acquire/release ops are compiled away, when either the
1401 * architecture does not support cmpxchg or when debugging is enabled.
1404 rt_mutex_fastlock(struct rt_mutex *lock, int state,
1405 int (*slowfn)(struct rt_mutex *lock, int state,
1406 struct hrtimer_sleeper *timeout,
1407 enum rtmutex_chainwalk chwalk))
1409 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1412 return slowfn(lock, state, NULL, RT_MUTEX_MIN_CHAINWALK);
1416 rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
1417 struct hrtimer_sleeper *timeout,
1418 enum rtmutex_chainwalk chwalk,
1419 int (*slowfn)(struct rt_mutex *lock, int state,
1420 struct hrtimer_sleeper *timeout,
1421 enum rtmutex_chainwalk chwalk))
1423 if (chwalk == RT_MUTEX_MIN_CHAINWALK &&
1424 likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1427 return slowfn(lock, state, timeout, chwalk);
1431 rt_mutex_fasttrylock(struct rt_mutex *lock,
1432 int (*slowfn)(struct rt_mutex *lock))
1434 if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1437 return slowfn(lock);
1441 * Performs the wakeup of the the top-waiter and re-enables preemption.
1443 void rt_mutex_postunlock(struct wake_q_head *wake_q)
1447 /* Pairs with preempt_disable() in rt_mutex_slowunlock() */
1452 rt_mutex_fastunlock(struct rt_mutex *lock,
1453 bool (*slowfn)(struct rt_mutex *lock,
1454 struct wake_q_head *wqh))
1456 DEFINE_WAKE_Q(wake_q);
1458 if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
1461 if (slowfn(lock, &wake_q))
1462 rt_mutex_postunlock(&wake_q);
1465 static inline void __rt_mutex_lock(struct rt_mutex *lock, unsigned int subclass)
1469 mutex_acquire(&lock->dep_map, subclass, 0, _RET_IP_);
1470 rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, rt_mutex_slowlock);
1473 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1475 * rt_mutex_lock_nested - lock a rt_mutex
1477 * @lock: the rt_mutex to be locked
1478 * @subclass: the lockdep subclass
1480 void __sched rt_mutex_lock_nested(struct rt_mutex *lock, unsigned int subclass)
1482 __rt_mutex_lock(lock, subclass);
1484 EXPORT_SYMBOL_GPL(rt_mutex_lock_nested);
1486 #else /* !CONFIG_DEBUG_LOCK_ALLOC */
1489 * rt_mutex_lock - lock a rt_mutex
1491 * @lock: the rt_mutex to be locked
1493 void __sched rt_mutex_lock(struct rt_mutex *lock)
1495 __rt_mutex_lock(lock, 0);
1497 EXPORT_SYMBOL_GPL(rt_mutex_lock);
1501 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
1503 * @lock: the rt_mutex to be locked
1507 * -EINTR when interrupted by a signal
1509 int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock)
1515 mutex_acquire(&lock->dep_map, 0, 0, _RET_IP_);
1516 ret = rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE, rt_mutex_slowlock);
1518 mutex_release(&lock->dep_map, _RET_IP_);
1522 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
1525 * Futex variant, must not use fastpath.
1527 int __sched rt_mutex_futex_trylock(struct rt_mutex *lock)
1529 return rt_mutex_slowtrylock(lock);
1532 int __sched __rt_mutex_futex_trylock(struct rt_mutex *lock)
1534 return __rt_mutex_slowtrylock(lock);
1538 * rt_mutex_timed_lock - lock a rt_mutex interruptible
1539 * the timeout structure is provided
1542 * @lock: the rt_mutex to be locked
1543 * @timeout: timeout structure or NULL (no timeout)
1547 * -EINTR when interrupted by a signal
1548 * -ETIMEDOUT when the timeout expired
1551 rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout)
1557 mutex_acquire(&lock->dep_map, 0, 0, _RET_IP_);
1558 ret = rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
1559 RT_MUTEX_MIN_CHAINWALK,
1562 mutex_release(&lock->dep_map, _RET_IP_);
1566 EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
1569 * rt_mutex_trylock - try to lock a rt_mutex
1571 * @lock: the rt_mutex to be locked
1573 * This function can only be called in thread context. It's safe to
1574 * call it from atomic regions, but not from hard interrupt or soft
1575 * interrupt context.
1577 * Returns 1 on success and 0 on contention
1579 int __sched rt_mutex_trylock(struct rt_mutex *lock)
1583 if (WARN_ON_ONCE(in_irq() || in_nmi() || in_serving_softirq()))
1586 ret = rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
1588 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
1592 EXPORT_SYMBOL_GPL(rt_mutex_trylock);
1595 * rt_mutex_unlock - unlock a rt_mutex
1597 * @lock: the rt_mutex to be unlocked
1599 void __sched rt_mutex_unlock(struct rt_mutex *lock)
1601 mutex_release(&lock->dep_map, _RET_IP_);
1602 rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
1604 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
1607 * Futex variant, that since futex variants do not use the fast-path, can be
1608 * simple and will not need to retry.
1610 bool __sched __rt_mutex_futex_unlock(struct rt_mutex *lock,
1611 struct wake_q_head *wake_q)
1613 lockdep_assert_held(&lock->wait_lock);
1615 debug_rt_mutex_unlock(lock);
1617 if (!rt_mutex_has_waiters(lock)) {
1619 return false; /* done */
1623 * We've already deboosted, mark_wakeup_next_waiter() will
1624 * retain preempt_disabled when we drop the wait_lock, to
1625 * avoid inversion prior to the wakeup. preempt_disable()
1626 * therein pairs with rt_mutex_postunlock().
1628 mark_wakeup_next_waiter(wake_q, lock);
1630 return true; /* call postunlock() */
1633 void __sched rt_mutex_futex_unlock(struct rt_mutex *lock)
1635 DEFINE_WAKE_Q(wake_q);
1636 unsigned long flags;
1639 raw_spin_lock_irqsave(&lock->wait_lock, flags);
1640 postunlock = __rt_mutex_futex_unlock(lock, &wake_q);
1641 raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1644 rt_mutex_postunlock(&wake_q);
1648 * rt_mutex_destroy - mark a mutex unusable
1649 * @lock: the mutex to be destroyed
1651 * This function marks the mutex uninitialized, and any subsequent
1652 * use of the mutex is forbidden. The mutex must not be locked when
1653 * this function is called.
1655 void rt_mutex_destroy(struct rt_mutex *lock)
1657 WARN_ON(rt_mutex_is_locked(lock));
1658 #ifdef CONFIG_DEBUG_RT_MUTEXES
1662 EXPORT_SYMBOL_GPL(rt_mutex_destroy);
1665 * __rt_mutex_init - initialize the rt lock
1667 * @lock: the rt lock to be initialized
1669 * Initialize the rt lock to unlocked state.
1671 * Initializing of a locked rt lock is not allowed
1673 void __rt_mutex_init(struct rt_mutex *lock, const char *name,
1674 struct lock_class_key *key)
1677 raw_spin_lock_init(&lock->wait_lock);
1678 lock->waiters = RB_ROOT_CACHED;
1681 debug_rt_mutex_init(lock, name, key);
1683 EXPORT_SYMBOL_GPL(__rt_mutex_init);
1686 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
1689 * @lock: the rt_mutex to be locked
1690 * @proxy_owner:the task to set as owner
1692 * No locking. Caller has to do serializing itself
1694 * Special API call for PI-futex support. This initializes the rtmutex and
1695 * assigns it to @proxy_owner. Concurrent operations on the rtmutex are not
1696 * possible at this point because the pi_state which contains the rtmutex
1697 * is not yet visible to other tasks.
1699 void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
1700 struct task_struct *proxy_owner)
1702 __rt_mutex_init(lock, NULL, NULL);
1703 debug_rt_mutex_proxy_lock(lock, proxy_owner);
1704 rt_mutex_set_owner(lock, proxy_owner);
1708 * rt_mutex_proxy_unlock - release a lock on behalf of owner
1710 * @lock: the rt_mutex to be locked
1712 * No locking. Caller has to do serializing itself
1714 * Special API call for PI-futex support. This merrily cleans up the rtmutex
1715 * (debugging) state. Concurrent operations on this rt_mutex are not
1716 * possible because it belongs to the pi_state which is about to be freed
1717 * and it is not longer visible to other tasks.
1719 void rt_mutex_proxy_unlock(struct rt_mutex *lock,
1720 struct task_struct *proxy_owner)
1722 debug_rt_mutex_proxy_unlock(lock);
1723 rt_mutex_set_owner(lock, NULL);
1727 * __rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1728 * @lock: the rt_mutex to take
1729 * @waiter: the pre-initialized rt_mutex_waiter
1730 * @task: the task to prepare
1732 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
1733 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
1735 * NOTE: does _NOT_ remove the @waiter on failure; must either call
1736 * rt_mutex_wait_proxy_lock() or rt_mutex_cleanup_proxy_lock() after this.
1739 * 0 - task blocked on lock
1740 * 1 - acquired the lock for task, caller should wake it up
1743 * Special API call for PI-futex support.
1745 int __rt_mutex_start_proxy_lock(struct rt_mutex *lock,
1746 struct rt_mutex_waiter *waiter,
1747 struct task_struct *task)
1751 lockdep_assert_held(&lock->wait_lock);
1753 if (try_to_take_rt_mutex(lock, task, NULL))
1756 /* We enforce deadlock detection for futexes */
1757 ret = task_blocks_on_rt_mutex(lock, waiter, task,
1758 RT_MUTEX_FULL_CHAINWALK);
1760 if (ret && !rt_mutex_owner(lock)) {
1762 * Reset the return value. We might have
1763 * returned with -EDEADLK and the owner
1764 * released the lock while we were walking the
1765 * pi chain. Let the waiter sort it out.
1770 debug_rt_mutex_print_deadlock(waiter);
1776 * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1777 * @lock: the rt_mutex to take
1778 * @waiter: the pre-initialized rt_mutex_waiter
1779 * @task: the task to prepare
1781 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
1782 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
1784 * NOTE: unlike __rt_mutex_start_proxy_lock this _DOES_ remove the @waiter
1788 * 0 - task blocked on lock
1789 * 1 - acquired the lock for task, caller should wake it up
1792 * Special API call for PI-futex support.
1794 int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
1795 struct rt_mutex_waiter *waiter,
1796 struct task_struct *task)
1800 raw_spin_lock_irq(&lock->wait_lock);
1801 ret = __rt_mutex_start_proxy_lock(lock, waiter, task);
1803 remove_waiter(lock, waiter);
1804 raw_spin_unlock_irq(&lock->wait_lock);
1810 * rt_mutex_next_owner - return the next owner of the lock
1812 * @lock: the rt lock query
1814 * Returns the next owner of the lock or NULL
1816 * Caller has to serialize against other accessors to the lock
1819 * Special API call for PI-futex support
1821 struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock)
1823 if (!rt_mutex_has_waiters(lock))
1826 return rt_mutex_top_waiter(lock)->task;
1830 * rt_mutex_wait_proxy_lock() - Wait for lock acquisition
1831 * @lock: the rt_mutex we were woken on
1832 * @to: the timeout, null if none. hrtimer should already have
1834 * @waiter: the pre-initialized rt_mutex_waiter
1836 * Wait for the the lock acquisition started on our behalf by
1837 * rt_mutex_start_proxy_lock(). Upon failure, the caller must call
1838 * rt_mutex_cleanup_proxy_lock().
1842 * <0 - error, one of -EINTR, -ETIMEDOUT
1844 * Special API call for PI-futex support
1846 int rt_mutex_wait_proxy_lock(struct rt_mutex *lock,
1847 struct hrtimer_sleeper *to,
1848 struct rt_mutex_waiter *waiter)
1852 raw_spin_lock_irq(&lock->wait_lock);
1853 /* sleep on the mutex */
1854 set_current_state(TASK_INTERRUPTIBLE);
1855 ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
1857 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1858 * have to fix that up.
1860 fixup_rt_mutex_waiters(lock);
1861 raw_spin_unlock_irq(&lock->wait_lock);
1867 * rt_mutex_cleanup_proxy_lock() - Cleanup failed lock acquisition
1868 * @lock: the rt_mutex we were woken on
1869 * @waiter: the pre-initialized rt_mutex_waiter
1871 * Attempt to clean up after a failed __rt_mutex_start_proxy_lock() or
1872 * rt_mutex_wait_proxy_lock().
1874 * Unless we acquired the lock; we're still enqueued on the wait-list and can
1875 * in fact still be granted ownership until we're removed. Therefore we can
1876 * find we are in fact the owner and must disregard the
1877 * rt_mutex_wait_proxy_lock() failure.
1880 * true - did the cleanup, we done.
1881 * false - we acquired the lock after rt_mutex_wait_proxy_lock() returned,
1882 * caller should disregards its return value.
1884 * Special API call for PI-futex support
1886 bool rt_mutex_cleanup_proxy_lock(struct rt_mutex *lock,
1887 struct rt_mutex_waiter *waiter)
1889 bool cleanup = false;
1891 raw_spin_lock_irq(&lock->wait_lock);
1893 * Do an unconditional try-lock, this deals with the lock stealing
1894 * state where __rt_mutex_futex_unlock() -> mark_wakeup_next_waiter()
1895 * sets a NULL owner.
1897 * We're not interested in the return value, because the subsequent
1898 * test on rt_mutex_owner() will infer that. If the trylock succeeded,
1899 * we will own the lock and it will have removed the waiter. If we
1900 * failed the trylock, we're still not owner and we need to remove
1903 try_to_take_rt_mutex(lock, current, waiter);
1905 * Unless we're the owner; we're still enqueued on the wait_list.
1906 * So check if we became owner, if not, take us off the wait_list.
1908 if (rt_mutex_owner(lock) != current) {
1909 remove_waiter(lock, waiter);
1913 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1914 * have to fix that up.
1916 fixup_rt_mutex_waiters(lock);
1918 raw_spin_unlock_irq(&lock->wait_lock);