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
;
270 #define __node_2_waiter(node) \
271 rb_entry((node), struct rt_mutex_waiter, tree_entry)
273 static inline bool __waiter_less(struct rb_node
*a
, const struct rb_node
*b
)
275 return rt_mutex_waiter_less(__node_2_waiter(a
), __node_2_waiter(b
));
279 rt_mutex_enqueue(struct rt_mutex
*lock
, struct rt_mutex_waiter
*waiter
)
281 rb_add_cached(&waiter
->tree_entry
, &lock
->waiters
, __waiter_less
);
285 rt_mutex_dequeue(struct rt_mutex
*lock
, struct rt_mutex_waiter
*waiter
)
287 if (RB_EMPTY_NODE(&waiter
->tree_entry
))
290 rb_erase_cached(&waiter
->tree_entry
, &lock
->waiters
);
291 RB_CLEAR_NODE(&waiter
->tree_entry
);
294 #define __node_2_pi_waiter(node) \
295 rb_entry((node), struct rt_mutex_waiter, pi_tree_entry)
297 static inline bool __pi_waiter_less(struct rb_node
*a
, const struct rb_node
*b
)
299 return rt_mutex_waiter_less(__node_2_pi_waiter(a
), __node_2_pi_waiter(b
));
303 rt_mutex_enqueue_pi(struct task_struct
*task
, struct rt_mutex_waiter
*waiter
)
305 rb_add_cached(&waiter
->pi_tree_entry
, &task
->pi_waiters
, __pi_waiter_less
);
309 rt_mutex_dequeue_pi(struct task_struct
*task
, struct rt_mutex_waiter
*waiter
)
311 if (RB_EMPTY_NODE(&waiter
->pi_tree_entry
))
314 rb_erase_cached(&waiter
->pi_tree_entry
, &task
->pi_waiters
);
315 RB_CLEAR_NODE(&waiter
->pi_tree_entry
);
318 static void rt_mutex_adjust_prio(struct task_struct
*p
)
320 struct task_struct
*pi_task
= NULL
;
322 lockdep_assert_held(&p
->pi_lock
);
324 if (task_has_pi_waiters(p
))
325 pi_task
= task_top_pi_waiter(p
)->task
;
327 rt_mutex_setprio(p
, pi_task
);
331 * Deadlock detection is conditional:
333 * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
334 * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
336 * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
337 * conducted independent of the detect argument.
339 * If the waiter argument is NULL this indicates the deboost path and
340 * deadlock detection is disabled independent of the detect argument
341 * and the config settings.
343 static bool 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 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 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
781 static int try_to_take_rt_mutex(struct rt_mutex
*lock
, struct task_struct
*task
,
782 struct rt_mutex_waiter
*waiter
)
784 lockdep_assert_held(&lock
->wait_lock
);
787 * Before testing whether we can acquire @lock, we set the
788 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
789 * other tasks which try to modify @lock into the slow path
790 * and they serialize on @lock->wait_lock.
792 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
793 * as explained at the top of this file if and only if:
795 * - There is a lock owner. The caller must fixup the
796 * transient state if it does a trylock or leaves the lock
797 * function due to a signal or timeout.
799 * - @task acquires the lock and there are no other
800 * waiters. This is undone in rt_mutex_set_owner(@task) at
801 * the end of this function.
803 mark_rt_mutex_waiters(lock
);
806 * If @lock has an owner, give up.
808 if (rt_mutex_owner(lock
))
812 * If @waiter != NULL, @task has already enqueued the waiter
813 * into @lock waiter tree. If @waiter == NULL then this is a
818 * If waiter is not the highest priority waiter of
821 if (waiter
!= rt_mutex_top_waiter(lock
))
825 * We can acquire the lock. Remove the waiter from the
828 rt_mutex_dequeue(lock
, waiter
);
832 * If the lock has waiters already we check whether @task is
833 * eligible to take over the lock.
835 * If there are no other waiters, @task can acquire
836 * the lock. @task->pi_blocked_on is NULL, so it does
837 * not need to be dequeued.
839 if (rt_mutex_has_waiters(lock
)) {
841 * If @task->prio is greater than or equal to
842 * the top waiter priority (kernel view),
845 if (!rt_mutex_waiter_less(task_to_waiter(task
),
846 rt_mutex_top_waiter(lock
)))
850 * The current top waiter stays enqueued. We
851 * don't have to change anything in the lock
856 * No waiters. Take the lock without the
857 * pi_lock dance.@task->pi_blocked_on is NULL
858 * and we have no waiters to enqueue in @task
866 * Clear @task->pi_blocked_on. Requires protection by
867 * @task->pi_lock. Redundant operation for the @waiter == NULL
868 * case, but conditionals are more expensive than a redundant
871 raw_spin_lock(&task
->pi_lock
);
872 task
->pi_blocked_on
= NULL
;
874 * Finish the lock acquisition. @task is the new owner. If
875 * other waiters exist we have to insert the highest priority
876 * waiter into @task->pi_waiters tree.
878 if (rt_mutex_has_waiters(lock
))
879 rt_mutex_enqueue_pi(task
, rt_mutex_top_waiter(lock
));
880 raw_spin_unlock(&task
->pi_lock
);
884 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
885 * are still waiters or clears it.
887 rt_mutex_set_owner(lock
, task
);
893 * Task blocks on lock.
895 * Prepare waiter and propagate pi chain
897 * This must be called with lock->wait_lock held and interrupts disabled
899 static int task_blocks_on_rt_mutex(struct rt_mutex
*lock
,
900 struct rt_mutex_waiter
*waiter
,
901 struct task_struct
*task
,
902 enum rtmutex_chainwalk chwalk
)
904 struct task_struct
*owner
= rt_mutex_owner(lock
);
905 struct rt_mutex_waiter
*top_waiter
= waiter
;
906 struct rt_mutex
*next_lock
;
907 int chain_walk
= 0, res
;
909 lockdep_assert_held(&lock
->wait_lock
);
912 * Early deadlock detection. We really don't want the task to
913 * enqueue on itself just to untangle the mess later. It's not
914 * only an optimization. We drop the locks, so another waiter
915 * can come in before the chain walk detects the deadlock. So
916 * the other will detect the deadlock and return -EDEADLOCK,
917 * which is wrong, as the other waiter is not in a deadlock
923 raw_spin_lock(&task
->pi_lock
);
926 waiter
->prio
= task
->prio
;
927 waiter
->deadline
= task
->dl
.deadline
;
929 /* Get the top priority waiter on the lock */
930 if (rt_mutex_has_waiters(lock
))
931 top_waiter
= rt_mutex_top_waiter(lock
);
932 rt_mutex_enqueue(lock
, waiter
);
934 task
->pi_blocked_on
= waiter
;
936 raw_spin_unlock(&task
->pi_lock
);
941 raw_spin_lock(&owner
->pi_lock
);
942 if (waiter
== rt_mutex_top_waiter(lock
)) {
943 rt_mutex_dequeue_pi(owner
, top_waiter
);
944 rt_mutex_enqueue_pi(owner
, waiter
);
946 rt_mutex_adjust_prio(owner
);
947 if (owner
->pi_blocked_on
)
949 } else if (rt_mutex_cond_detect_deadlock(waiter
, chwalk
)) {
953 /* Store the lock on which owner is blocked or NULL */
954 next_lock
= task_blocked_on_lock(owner
);
956 raw_spin_unlock(&owner
->pi_lock
);
958 * Even if full deadlock detection is on, if the owner is not
959 * blocked itself, we can avoid finding this out in the chain
962 if (!chain_walk
|| !next_lock
)
966 * The owner can't disappear while holding a lock,
967 * so the owner struct is protected by wait_lock.
968 * Gets dropped in rt_mutex_adjust_prio_chain()!
970 get_task_struct(owner
);
972 raw_spin_unlock_irq(&lock
->wait_lock
);
974 res
= rt_mutex_adjust_prio_chain(owner
, chwalk
, lock
,
975 next_lock
, waiter
, task
);
977 raw_spin_lock_irq(&lock
->wait_lock
);
983 * Remove the top waiter from the current tasks pi waiter tree and
986 * Called with lock->wait_lock held and interrupts disabled.
988 static void mark_wakeup_next_waiter(struct wake_q_head
*wake_q
,
989 struct rt_mutex
*lock
)
991 struct rt_mutex_waiter
*waiter
;
993 raw_spin_lock(¤t
->pi_lock
);
995 waiter
= rt_mutex_top_waiter(lock
);
998 * Remove it from current->pi_waiters and deboost.
1000 * We must in fact deboost here in order to ensure we call
1001 * rt_mutex_setprio() to update p->pi_top_task before the
1004 rt_mutex_dequeue_pi(current
, waiter
);
1005 rt_mutex_adjust_prio(current
);
1008 * As we are waking up the top waiter, and the waiter stays
1009 * queued on the lock until it gets the lock, this lock
1010 * obviously has waiters. Just set the bit here and this has
1011 * the added benefit of forcing all new tasks into the
1012 * slow path making sure no task of lower priority than
1013 * the top waiter can steal this lock.
1015 lock
->owner
= (void *) RT_MUTEX_HAS_WAITERS
;
1018 * We deboosted before waking the top waiter task such that we don't
1019 * run two tasks with the 'same' priority (and ensure the
1020 * p->pi_top_task pointer points to a blocked task). This however can
1021 * lead to priority inversion if we would get preempted after the
1022 * deboost but before waking our donor task, hence the preempt_disable()
1025 * Pairs with preempt_enable() in rt_mutex_postunlock();
1028 wake_q_add(wake_q
, waiter
->task
);
1029 raw_spin_unlock(¤t
->pi_lock
);
1033 * Remove a waiter from a lock and give up
1035 * Must be called with lock->wait_lock held and interrupts disabled. I must
1036 * have just failed to try_to_take_rt_mutex().
1038 static void remove_waiter(struct rt_mutex
*lock
,
1039 struct rt_mutex_waiter
*waiter
)
1041 bool is_top_waiter
= (waiter
== rt_mutex_top_waiter(lock
));
1042 struct task_struct
*owner
= rt_mutex_owner(lock
);
1043 struct rt_mutex
*next_lock
;
1045 lockdep_assert_held(&lock
->wait_lock
);
1047 raw_spin_lock(¤t
->pi_lock
);
1048 rt_mutex_dequeue(lock
, waiter
);
1049 current
->pi_blocked_on
= NULL
;
1050 raw_spin_unlock(¤t
->pi_lock
);
1053 * Only update priority if the waiter was the highest priority
1054 * waiter of the lock and there is an owner to update.
1056 if (!owner
|| !is_top_waiter
)
1059 raw_spin_lock(&owner
->pi_lock
);
1061 rt_mutex_dequeue_pi(owner
, waiter
);
1063 if (rt_mutex_has_waiters(lock
))
1064 rt_mutex_enqueue_pi(owner
, rt_mutex_top_waiter(lock
));
1066 rt_mutex_adjust_prio(owner
);
1068 /* Store the lock on which owner is blocked or NULL */
1069 next_lock
= task_blocked_on_lock(owner
);
1071 raw_spin_unlock(&owner
->pi_lock
);
1074 * Don't walk the chain, if the owner task is not blocked
1080 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1081 get_task_struct(owner
);
1083 raw_spin_unlock_irq(&lock
->wait_lock
);
1085 rt_mutex_adjust_prio_chain(owner
, RT_MUTEX_MIN_CHAINWALK
, lock
,
1086 next_lock
, NULL
, current
);
1088 raw_spin_lock_irq(&lock
->wait_lock
);
1092 * Recheck the pi chain, in case we got a priority setting
1094 * Called from sched_setscheduler
1096 void rt_mutex_adjust_pi(struct task_struct
*task
)
1098 struct rt_mutex_waiter
*waiter
;
1099 struct rt_mutex
*next_lock
;
1100 unsigned long flags
;
1102 raw_spin_lock_irqsave(&task
->pi_lock
, flags
);
1104 waiter
= task
->pi_blocked_on
;
1105 if (!waiter
|| rt_mutex_waiter_equal(waiter
, task_to_waiter(task
))) {
1106 raw_spin_unlock_irqrestore(&task
->pi_lock
, flags
);
1109 next_lock
= waiter
->lock
;
1110 raw_spin_unlock_irqrestore(&task
->pi_lock
, flags
);
1112 /* gets dropped in rt_mutex_adjust_prio_chain()! */
1113 get_task_struct(task
);
1115 rt_mutex_adjust_prio_chain(task
, RT_MUTEX_MIN_CHAINWALK
, NULL
,
1116 next_lock
, NULL
, task
);
1119 void rt_mutex_init_waiter(struct rt_mutex_waiter
*waiter
)
1121 debug_rt_mutex_init_waiter(waiter
);
1122 RB_CLEAR_NODE(&waiter
->pi_tree_entry
);
1123 RB_CLEAR_NODE(&waiter
->tree_entry
);
1124 waiter
->task
= NULL
;
1128 * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
1129 * @lock: the rt_mutex to take
1130 * @state: the state the task should block in (TASK_INTERRUPTIBLE
1131 * or TASK_UNINTERRUPTIBLE)
1132 * @timeout: the pre-initialized and started timer, or NULL for none
1133 * @waiter: the pre-initialized rt_mutex_waiter
1135 * Must be called with lock->wait_lock held and interrupts disabled
1138 __rt_mutex_slowlock(struct rt_mutex
*lock
, 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
))
1150 * TASK_INTERRUPTIBLE checks for signals and
1151 * timeout. Ignored otherwise.
1153 if (likely(state
== TASK_INTERRUPTIBLE
)) {
1154 /* Signal pending? */
1155 if (signal_pending(current
))
1157 if (timeout
&& !timeout
->task
)
1163 raw_spin_unlock_irq(&lock
->wait_lock
);
1167 raw_spin_lock_irq(&lock
->wait_lock
);
1168 set_current_state(state
);
1171 __set_current_state(TASK_RUNNING
);
1175 static void rt_mutex_handle_deadlock(int res
, int detect_deadlock
,
1176 struct rt_mutex_waiter
*w
)
1179 * If the result is not -EDEADLOCK or the caller requested
1180 * deadlock detection, nothing to do here.
1182 if (res
!= -EDEADLOCK
|| detect_deadlock
)
1186 * Yell loudly and stop the task right here.
1188 WARN(1, "rtmutex deadlock detected\n");
1190 set_current_state(TASK_INTERRUPTIBLE
);
1196 * Slow path lock function:
1199 rt_mutex_slowlock(struct rt_mutex
*lock
, int state
,
1200 struct hrtimer_sleeper
*timeout
,
1201 enum rtmutex_chainwalk chwalk
)
1203 struct rt_mutex_waiter waiter
;
1204 unsigned long flags
;
1207 rt_mutex_init_waiter(&waiter
);
1210 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1211 * be called in early boot if the cmpxchg() fast path is disabled
1212 * (debug, no architecture support). In this case we will acquire the
1213 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1214 * enable interrupts in that early boot case. So we need to use the
1215 * irqsave/restore variants.
1217 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1219 /* Try to acquire the lock again: */
1220 if (try_to_take_rt_mutex(lock
, current
, NULL
)) {
1221 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1225 set_current_state(state
);
1227 /* Setup the timer, when timeout != NULL */
1228 if (unlikely(timeout
))
1229 hrtimer_start_expires(&timeout
->timer
, HRTIMER_MODE_ABS
);
1231 ret
= task_blocks_on_rt_mutex(lock
, &waiter
, current
, chwalk
);
1234 /* sleep on the mutex */
1235 ret
= __rt_mutex_slowlock(lock
, state
, timeout
, &waiter
);
1237 if (unlikely(ret
)) {
1238 __set_current_state(TASK_RUNNING
);
1239 remove_waiter(lock
, &waiter
);
1240 rt_mutex_handle_deadlock(ret
, chwalk
, &waiter
);
1244 * try_to_take_rt_mutex() sets the waiter bit
1245 * unconditionally. We might have to fix that up.
1247 fixup_rt_mutex_waiters(lock
);
1249 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1251 /* Remove pending timer: */
1252 if (unlikely(timeout
))
1253 hrtimer_cancel(&timeout
->timer
);
1255 debug_rt_mutex_free_waiter(&waiter
);
1260 static inline int __rt_mutex_slowtrylock(struct rt_mutex
*lock
)
1262 int ret
= try_to_take_rt_mutex(lock
, current
, NULL
);
1265 * try_to_take_rt_mutex() sets the lock waiters bit
1266 * unconditionally. Clean this up.
1268 fixup_rt_mutex_waiters(lock
);
1274 * Slow path try-lock function:
1276 static inline int rt_mutex_slowtrylock(struct rt_mutex
*lock
)
1278 unsigned long flags
;
1282 * If the lock already has an owner we fail to get the lock.
1283 * This can be done without taking the @lock->wait_lock as
1284 * it is only being read, and this is a trylock anyway.
1286 if (rt_mutex_owner(lock
))
1290 * The mutex has currently no owner. Lock the wait lock and try to
1291 * acquire the lock. We use irqsave here to support early boot calls.
1293 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1295 ret
= __rt_mutex_slowtrylock(lock
);
1297 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1303 * Slow path to release a rt-mutex.
1305 * Return whether the current task needs to call rt_mutex_postunlock().
1307 static bool __sched
rt_mutex_slowunlock(struct rt_mutex
*lock
,
1308 struct wake_q_head
*wake_q
)
1310 unsigned long flags
;
1312 /* irqsave required to support early boot calls */
1313 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1315 debug_rt_mutex_unlock(lock
);
1318 * We must be careful here if the fast path is enabled. If we
1319 * have no waiters queued we cannot set owner to NULL here
1322 * foo->lock->owner = NULL;
1323 * rtmutex_lock(foo->lock); <- fast path
1324 * free = atomic_dec_and_test(foo->refcnt);
1325 * rtmutex_unlock(foo->lock); <- fast path
1328 * raw_spin_unlock(foo->lock->wait_lock);
1330 * So for the fastpath enabled kernel:
1332 * Nothing can set the waiters bit as long as we hold
1333 * lock->wait_lock. So we do the following sequence:
1335 * owner = rt_mutex_owner(lock);
1336 * clear_rt_mutex_waiters(lock);
1337 * raw_spin_unlock(&lock->wait_lock);
1338 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1342 * The fastpath disabled variant is simple as all access to
1343 * lock->owner is serialized by lock->wait_lock:
1345 * lock->owner = NULL;
1346 * raw_spin_unlock(&lock->wait_lock);
1348 while (!rt_mutex_has_waiters(lock
)) {
1349 /* Drops lock->wait_lock ! */
1350 if (unlock_rt_mutex_safe(lock
, flags
) == true)
1352 /* Relock the rtmutex and try again */
1353 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1357 * The wakeup next waiter path does not suffer from the above
1358 * race. See the comments there.
1360 * Queue the next waiter for wakeup once we release the wait_lock.
1362 mark_wakeup_next_waiter(wake_q
, lock
);
1363 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1365 return true; /* call rt_mutex_postunlock() */
1369 * debug aware fast / slowpath lock,trylock,unlock
1371 * The atomic acquire/release ops are compiled away, when either the
1372 * architecture does not support cmpxchg or when debugging is enabled.
1375 rt_mutex_fastlock(struct rt_mutex
*lock
, int state
,
1376 int (*slowfn
)(struct rt_mutex
*lock
, int state
,
1377 struct hrtimer_sleeper
*timeout
,
1378 enum rtmutex_chainwalk chwalk
))
1380 if (likely(rt_mutex_cmpxchg_acquire(lock
, NULL
, current
)))
1383 return slowfn(lock
, state
, NULL
, RT_MUTEX_MIN_CHAINWALK
);
1387 rt_mutex_fasttrylock(struct rt_mutex
*lock
,
1388 int (*slowfn
)(struct rt_mutex
*lock
))
1390 if (likely(rt_mutex_cmpxchg_acquire(lock
, NULL
, current
)))
1393 return slowfn(lock
);
1397 * Performs the wakeup of the top-waiter and re-enables preemption.
1399 void rt_mutex_postunlock(struct wake_q_head
*wake_q
)
1403 /* Pairs with preempt_disable() in rt_mutex_slowunlock() */
1408 rt_mutex_fastunlock(struct rt_mutex
*lock
,
1409 bool (*slowfn
)(struct rt_mutex
*lock
,
1410 struct wake_q_head
*wqh
))
1412 DEFINE_WAKE_Q(wake_q
);
1414 if (likely(rt_mutex_cmpxchg_release(lock
, current
, NULL
)))
1417 if (slowfn(lock
, &wake_q
))
1418 rt_mutex_postunlock(&wake_q
);
1421 static inline void __rt_mutex_lock(struct rt_mutex
*lock
, unsigned int subclass
)
1425 mutex_acquire(&lock
->dep_map
, subclass
, 0, _RET_IP_
);
1426 rt_mutex_fastlock(lock
, TASK_UNINTERRUPTIBLE
, rt_mutex_slowlock
);
1429 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1431 * rt_mutex_lock_nested - lock a rt_mutex
1433 * @lock: the rt_mutex to be locked
1434 * @subclass: the lockdep subclass
1436 void __sched
rt_mutex_lock_nested(struct rt_mutex
*lock
, unsigned int subclass
)
1438 __rt_mutex_lock(lock
, subclass
);
1440 EXPORT_SYMBOL_GPL(rt_mutex_lock_nested
);
1442 #else /* !CONFIG_DEBUG_LOCK_ALLOC */
1445 * rt_mutex_lock - lock a rt_mutex
1447 * @lock: the rt_mutex to be locked
1449 void __sched
rt_mutex_lock(struct rt_mutex
*lock
)
1451 __rt_mutex_lock(lock
, 0);
1453 EXPORT_SYMBOL_GPL(rt_mutex_lock
);
1457 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
1459 * @lock: the rt_mutex to be locked
1463 * -EINTR when interrupted by a signal
1465 int __sched
rt_mutex_lock_interruptible(struct rt_mutex
*lock
)
1471 mutex_acquire(&lock
->dep_map
, 0, 0, _RET_IP_
);
1472 ret
= rt_mutex_fastlock(lock
, TASK_INTERRUPTIBLE
, rt_mutex_slowlock
);
1474 mutex_release(&lock
->dep_map
, _RET_IP_
);
1478 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible
);
1481 * Futex variant, must not use fastpath.
1483 int __sched
rt_mutex_futex_trylock(struct rt_mutex
*lock
)
1485 return rt_mutex_slowtrylock(lock
);
1488 int __sched
__rt_mutex_futex_trylock(struct rt_mutex
*lock
)
1490 return __rt_mutex_slowtrylock(lock
);
1494 * rt_mutex_trylock - try to lock a rt_mutex
1496 * @lock: the rt_mutex to be locked
1498 * This function can only be called in thread context. It's safe to
1499 * call it from atomic regions, but not from hard interrupt or soft
1500 * interrupt context.
1502 * Returns 1 on success and 0 on contention
1504 int __sched
rt_mutex_trylock(struct rt_mutex
*lock
)
1508 if (WARN_ON_ONCE(in_irq() || in_nmi() || in_serving_softirq()))
1511 ret
= rt_mutex_fasttrylock(lock
, rt_mutex_slowtrylock
);
1513 mutex_acquire(&lock
->dep_map
, 0, 1, _RET_IP_
);
1517 EXPORT_SYMBOL_GPL(rt_mutex_trylock
);
1520 * rt_mutex_unlock - unlock a rt_mutex
1522 * @lock: the rt_mutex to be unlocked
1524 void __sched
rt_mutex_unlock(struct rt_mutex
*lock
)
1526 mutex_release(&lock
->dep_map
, _RET_IP_
);
1527 rt_mutex_fastunlock(lock
, rt_mutex_slowunlock
);
1529 EXPORT_SYMBOL_GPL(rt_mutex_unlock
);
1532 * __rt_mutex_futex_unlock - Futex variant, that since futex variants
1533 * do not use the fast-path, can be simple and will not need to retry.
1535 * @lock: The rt_mutex to be unlocked
1536 * @wake_q: The wake queue head from which to get the next lock waiter
1538 bool __sched
__rt_mutex_futex_unlock(struct rt_mutex
*lock
,
1539 struct wake_q_head
*wake_q
)
1541 lockdep_assert_held(&lock
->wait_lock
);
1543 debug_rt_mutex_unlock(lock
);
1545 if (!rt_mutex_has_waiters(lock
)) {
1547 return false; /* done */
1551 * We've already deboosted, mark_wakeup_next_waiter() will
1552 * retain preempt_disabled when we drop the wait_lock, to
1553 * avoid inversion prior to the wakeup. preempt_disable()
1554 * therein pairs with rt_mutex_postunlock().
1556 mark_wakeup_next_waiter(wake_q
, lock
);
1558 return true; /* call postunlock() */
1561 void __sched
rt_mutex_futex_unlock(struct rt_mutex
*lock
)
1563 DEFINE_WAKE_Q(wake_q
);
1564 unsigned long flags
;
1567 raw_spin_lock_irqsave(&lock
->wait_lock
, flags
);
1568 postunlock
= __rt_mutex_futex_unlock(lock
, &wake_q
);
1569 raw_spin_unlock_irqrestore(&lock
->wait_lock
, flags
);
1572 rt_mutex_postunlock(&wake_q
);
1576 * __rt_mutex_init - initialize the rt_mutex
1578 * @lock: The rt_mutex to be initialized
1579 * @name: The lock name used for debugging
1580 * @key: The lock class key used for debugging
1582 * Initialize the rt_mutex to unlocked state.
1584 * Initializing of a locked rt_mutex is not allowed
1586 void __rt_mutex_init(struct rt_mutex
*lock
, const char *name
,
1587 struct lock_class_key
*key
)
1589 debug_check_no_locks_freed((void *)lock
, sizeof(*lock
));
1590 lockdep_init_map(&lock
->dep_map
, name
, key
, 0);
1592 __rt_mutex_basic_init(lock
);
1594 EXPORT_SYMBOL_GPL(__rt_mutex_init
);
1597 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
1600 * @lock: the rt_mutex to be locked
1601 * @proxy_owner:the task to set as owner
1603 * No locking. Caller has to do serializing itself
1605 * Special API call for PI-futex support. This initializes the rtmutex and
1606 * assigns it to @proxy_owner. Concurrent operations on the rtmutex are not
1607 * possible at this point because the pi_state which contains the rtmutex
1608 * is not yet visible to other tasks.
1610 void rt_mutex_init_proxy_locked(struct rt_mutex
*lock
,
1611 struct task_struct
*proxy_owner
)
1613 __rt_mutex_basic_init(lock
);
1614 rt_mutex_set_owner(lock
, proxy_owner
);
1618 * rt_mutex_proxy_unlock - release a lock on behalf of owner
1620 * @lock: the rt_mutex to be locked
1622 * No locking. Caller has to do serializing itself
1624 * Special API call for PI-futex support. This merrily cleans up the rtmutex
1625 * (debugging) state. Concurrent operations on this rt_mutex are not
1626 * possible because it belongs to the pi_state which is about to be freed
1627 * and it is not longer visible to other tasks.
1629 void rt_mutex_proxy_unlock(struct rt_mutex
*lock
)
1631 debug_rt_mutex_proxy_unlock(lock
);
1632 rt_mutex_set_owner(lock
, NULL
);
1636 * __rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1637 * @lock: the rt_mutex to take
1638 * @waiter: the pre-initialized rt_mutex_waiter
1639 * @task: the task to prepare
1641 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
1642 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
1644 * NOTE: does _NOT_ remove the @waiter on failure; must either call
1645 * rt_mutex_wait_proxy_lock() or rt_mutex_cleanup_proxy_lock() after this.
1648 * 0 - task blocked on lock
1649 * 1 - acquired the lock for task, caller should wake it up
1652 * Special API call for PI-futex support.
1654 int __rt_mutex_start_proxy_lock(struct rt_mutex
*lock
,
1655 struct rt_mutex_waiter
*waiter
,
1656 struct task_struct
*task
)
1660 lockdep_assert_held(&lock
->wait_lock
);
1662 if (try_to_take_rt_mutex(lock
, task
, NULL
))
1665 /* We enforce deadlock detection for futexes */
1666 ret
= task_blocks_on_rt_mutex(lock
, waiter
, task
,
1667 RT_MUTEX_FULL_CHAINWALK
);
1669 if (ret
&& !rt_mutex_owner(lock
)) {
1671 * Reset the return value. We might have
1672 * returned with -EDEADLK and the owner
1673 * released the lock while we were walking the
1674 * pi chain. Let the waiter sort it out.
1683 * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1684 * @lock: the rt_mutex to take
1685 * @waiter: the pre-initialized rt_mutex_waiter
1686 * @task: the task to prepare
1688 * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock
1689 * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that.
1691 * NOTE: unlike __rt_mutex_start_proxy_lock this _DOES_ remove the @waiter
1695 * 0 - task blocked on lock
1696 * 1 - acquired the lock for task, caller should wake it up
1699 * Special API call for PI-futex support.
1701 int rt_mutex_start_proxy_lock(struct rt_mutex
*lock
,
1702 struct rt_mutex_waiter
*waiter
,
1703 struct task_struct
*task
)
1707 raw_spin_lock_irq(&lock
->wait_lock
);
1708 ret
= __rt_mutex_start_proxy_lock(lock
, waiter
, task
);
1710 remove_waiter(lock
, waiter
);
1711 raw_spin_unlock_irq(&lock
->wait_lock
);
1717 * rt_mutex_wait_proxy_lock() - Wait for lock acquisition
1718 * @lock: the rt_mutex we were woken on
1719 * @to: the timeout, null if none. hrtimer should already have
1721 * @waiter: the pre-initialized rt_mutex_waiter
1723 * Wait for the lock acquisition started on our behalf by
1724 * rt_mutex_start_proxy_lock(). Upon failure, the caller must call
1725 * rt_mutex_cleanup_proxy_lock().
1729 * <0 - error, one of -EINTR, -ETIMEDOUT
1731 * Special API call for PI-futex support
1733 int rt_mutex_wait_proxy_lock(struct rt_mutex
*lock
,
1734 struct hrtimer_sleeper
*to
,
1735 struct rt_mutex_waiter
*waiter
)
1739 raw_spin_lock_irq(&lock
->wait_lock
);
1740 /* sleep on the mutex */
1741 set_current_state(TASK_INTERRUPTIBLE
);
1742 ret
= __rt_mutex_slowlock(lock
, TASK_INTERRUPTIBLE
, to
, waiter
);
1744 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1745 * have to fix that up.
1747 fixup_rt_mutex_waiters(lock
);
1748 raw_spin_unlock_irq(&lock
->wait_lock
);
1754 * rt_mutex_cleanup_proxy_lock() - Cleanup failed lock acquisition
1755 * @lock: the rt_mutex we were woken on
1756 * @waiter: the pre-initialized rt_mutex_waiter
1758 * Attempt to clean up after a failed __rt_mutex_start_proxy_lock() or
1759 * rt_mutex_wait_proxy_lock().
1761 * Unless we acquired the lock; we're still enqueued on the wait-list and can
1762 * in fact still be granted ownership until we're removed. Therefore we can
1763 * find we are in fact the owner and must disregard the
1764 * rt_mutex_wait_proxy_lock() failure.
1767 * true - did the cleanup, we done.
1768 * false - we acquired the lock after rt_mutex_wait_proxy_lock() returned,
1769 * caller should disregards its return value.
1771 * Special API call for PI-futex support
1773 bool rt_mutex_cleanup_proxy_lock(struct rt_mutex
*lock
,
1774 struct rt_mutex_waiter
*waiter
)
1776 bool cleanup
= false;
1778 raw_spin_lock_irq(&lock
->wait_lock
);
1780 * Do an unconditional try-lock, this deals with the lock stealing
1781 * state where __rt_mutex_futex_unlock() -> mark_wakeup_next_waiter()
1782 * sets a NULL owner.
1784 * We're not interested in the return value, because the subsequent
1785 * test on rt_mutex_owner() will infer that. If the trylock succeeded,
1786 * we will own the lock and it will have removed the waiter. If we
1787 * failed the trylock, we're still not owner and we need to remove
1790 try_to_take_rt_mutex(lock
, current
, waiter
);
1792 * Unless we're the owner; we're still enqueued on the wait_list.
1793 * So check if we became owner, if not, take us off the wait_list.
1795 if (rt_mutex_owner(lock
) != current
) {
1796 remove_waiter(lock
, waiter
);
1800 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1801 * have to fix that up.
1803 fixup_rt_mutex_waiters(lock
);
1805 raw_spin_unlock_irq(&lock
->wait_lock
);
1810 #ifdef CONFIG_DEBUG_RT_MUTEXES
1811 void rt_mutex_debug_task_free(struct task_struct
*task
)
1813 DEBUG_LOCKS_WARN_ON(!RB_EMPTY_ROOT(&task
->pi_waiters
.rb_root
));
1814 DEBUG_LOCKS_WARN_ON(task
->pi_blocked_on
);