2 * RT-Mutexes: simple blocking mutual exclusion locks with PI support
4 * started by Ingo Molnar and Thomas Gleixner.
6 * Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
7 * Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
8 * Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
9 * Copyright (C) 2006 Esben Nielsen
11 * See Documentation/rt-mutex-design.txt for details.
13 #include <linux/spinlock.h>
14 #include <linux/export.h>
15 #include <linux/sched.h>
16 #include <linux/sched/rt.h>
17 #include <linux/sched/deadline.h>
18 #include <linux/timer.h>
20 #include "rtmutex_common.h"
23 * lock->owner state tracking:
25 * lock->owner holds the task_struct pointer of the owner. Bit 0
26 * is used to keep track of the "lock has waiters" state.
29 * NULL 0 lock is free (fast acquire possible)
30 * NULL 1 lock is free and has waiters and the top waiter
31 * is going to take the lock*
32 * taskpointer 0 lock is held (fast release possible)
33 * taskpointer 1 lock is held and has waiters**
35 * The fast atomic compare exchange based acquire and release is only
36 * possible when bit 0 of lock->owner is 0.
38 * (*) It also can be a transitional state when grabbing the lock
39 * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
40 * we need to set the bit0 before looking at the lock, and the owner may be
41 * NULL in this small time, hence this can be a transitional state.
43 * (**) There is a small time when bit 0 is set but there are no
44 * waiters. This can happen when grabbing the lock in the slow path.
45 * To prevent a cmpxchg of the owner releasing the lock, we need to
46 * set this bit before looking at the lock.
50 rt_mutex_set_owner(struct rt_mutex *lock, struct task_struct *owner)
52 unsigned long val = (unsigned long)owner;
54 if (rt_mutex_has_waiters(lock))
55 val |= RT_MUTEX_HAS_WAITERS;
57 lock->owner = (struct task_struct *)val;
60 static inline void clear_rt_mutex_waiters(struct rt_mutex *lock)
62 lock->owner = (struct task_struct *)
63 ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
66 static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
68 if (!rt_mutex_has_waiters(lock))
69 clear_rt_mutex_waiters(lock);
73 * We can speed up the acquire/release, if the architecture
74 * supports cmpxchg and if there's no debugging state to be set up
76 #if defined(__HAVE_ARCH_CMPXCHG) && !defined(CONFIG_DEBUG_RT_MUTEXES)
77 # define rt_mutex_cmpxchg(l,c,n) (cmpxchg(&l->owner, c, n) == c)
78 static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
80 unsigned long owner, *p = (unsigned long *) &lock->owner;
84 } while (cmpxchg(p, owner, owner | RT_MUTEX_HAS_WAITERS) != owner);
88 * Safe fastpath aware unlock:
89 * 1) Clear the waiters bit
90 * 2) Drop lock->wait_lock
91 * 3) Try to unlock the lock with cmpxchg
93 static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock)
94 __releases(lock->wait_lock)
96 struct task_struct *owner = rt_mutex_owner(lock);
98 clear_rt_mutex_waiters(lock);
99 raw_spin_unlock(&lock->wait_lock);
101 * If a new waiter comes in between the unlock and the cmpxchg
102 * we have two situations:
106 * cmpxchg(p, owner, 0) == owner
107 * mark_rt_mutex_waiters(lock);
113 * mark_rt_mutex_waiters(lock);
115 * cmpxchg(p, owner, 0) != owner
124 return rt_mutex_cmpxchg(lock, owner, NULL);
128 # define rt_mutex_cmpxchg(l,c,n) (0)
129 static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
131 lock->owner = (struct task_struct *)
132 ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
136 * Simple slow path only version: lock->owner is protected by lock->wait_lock.
138 static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock)
139 __releases(lock->wait_lock)
142 raw_spin_unlock(&lock->wait_lock);
148 rt_mutex_waiter_less(struct rt_mutex_waiter *left,
149 struct rt_mutex_waiter *right)
151 if (left->prio < right->prio)
155 * If both waiters have dl_prio(), we check the deadlines of the
157 * If left waiter has a dl_prio(), and we didn't return 1 above,
158 * then right waiter has a dl_prio() too.
160 if (dl_prio(left->prio))
161 return (left->task->dl.deadline < right->task->dl.deadline);
167 rt_mutex_enqueue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
169 struct rb_node **link = &lock->waiters.rb_node;
170 struct rb_node *parent = NULL;
171 struct rt_mutex_waiter *entry;
176 entry = rb_entry(parent, struct rt_mutex_waiter, tree_entry);
177 if (rt_mutex_waiter_less(waiter, entry)) {
178 link = &parent->rb_left;
180 link = &parent->rb_right;
186 lock->waiters_leftmost = &waiter->tree_entry;
188 rb_link_node(&waiter->tree_entry, parent, link);
189 rb_insert_color(&waiter->tree_entry, &lock->waiters);
193 rt_mutex_dequeue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
195 if (RB_EMPTY_NODE(&waiter->tree_entry))
198 if (lock->waiters_leftmost == &waiter->tree_entry)
199 lock->waiters_leftmost = rb_next(&waiter->tree_entry);
201 rb_erase(&waiter->tree_entry, &lock->waiters);
202 RB_CLEAR_NODE(&waiter->tree_entry);
206 rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
208 struct rb_node **link = &task->pi_waiters.rb_node;
209 struct rb_node *parent = NULL;
210 struct rt_mutex_waiter *entry;
215 entry = rb_entry(parent, struct rt_mutex_waiter, pi_tree_entry);
216 if (rt_mutex_waiter_less(waiter, entry)) {
217 link = &parent->rb_left;
219 link = &parent->rb_right;
225 task->pi_waiters_leftmost = &waiter->pi_tree_entry;
227 rb_link_node(&waiter->pi_tree_entry, parent, link);
228 rb_insert_color(&waiter->pi_tree_entry, &task->pi_waiters);
232 rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
234 if (RB_EMPTY_NODE(&waiter->pi_tree_entry))
237 if (task->pi_waiters_leftmost == &waiter->pi_tree_entry)
238 task->pi_waiters_leftmost = rb_next(&waiter->pi_tree_entry);
240 rb_erase(&waiter->pi_tree_entry, &task->pi_waiters);
241 RB_CLEAR_NODE(&waiter->pi_tree_entry);
245 * Calculate task priority from the waiter tree priority
247 * Return task->normal_prio when the waiter tree is empty or when
248 * the waiter is not allowed to do priority boosting
250 int rt_mutex_getprio(struct task_struct *task)
252 if (likely(!task_has_pi_waiters(task)))
253 return task->normal_prio;
255 return min(task_top_pi_waiter(task)->prio,
259 struct task_struct *rt_mutex_get_top_task(struct task_struct *task)
261 if (likely(!task_has_pi_waiters(task)))
264 return task_top_pi_waiter(task)->task;
268 * Called by sched_setscheduler() to check whether the priority change
269 * is overruled by a possible priority boosting.
271 int rt_mutex_check_prio(struct task_struct *task, int newprio)
273 if (!task_has_pi_waiters(task))
276 return task_top_pi_waiter(task)->task->prio <= newprio;
280 * Adjust the priority of a task, after its pi_waiters got modified.
282 * This can be both boosting and unboosting. task->pi_lock must be held.
284 static void __rt_mutex_adjust_prio(struct task_struct *task)
286 int prio = rt_mutex_getprio(task);
288 if (task->prio != prio || dl_prio(prio))
289 rt_mutex_setprio(task, prio);
293 * Adjust task priority (undo boosting). Called from the exit path of
294 * rt_mutex_slowunlock() and rt_mutex_slowlock().
296 * (Note: We do this outside of the protection of lock->wait_lock to
297 * allow the lock to be taken while or before we readjust the priority
298 * of task. We do not use the spin_xx_mutex() variants here as we are
299 * outside of the debug path.)
301 static void rt_mutex_adjust_prio(struct task_struct *task)
305 raw_spin_lock_irqsave(&task->pi_lock, flags);
306 __rt_mutex_adjust_prio(task);
307 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
311 * Max number of times we'll walk the boosting chain:
313 int max_lock_depth = 1024;
315 static inline struct rt_mutex *task_blocked_on_lock(struct task_struct *p)
317 return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
321 * Adjust the priority chain. Also used for deadlock detection.
322 * Decreases task's usage by one - may thus free the task.
324 * @task: the task owning the mutex (owner) for which a chain walk is
326 * @deadlock_detect: do we have to carry out deadlock detection?
327 * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
328 * things for a task that has just got its priority adjusted, and
329 * is waiting on a mutex)
330 * @next_lock: the mutex on which the owner of @orig_lock was blocked before
331 * we dropped its pi_lock. Is never dereferenced, only used for
332 * comparison to detect lock chain changes.
333 * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
334 * its priority to the mutex owner (can be NULL in the case
335 * depicted above or if the top waiter is gone away and we are
336 * actually deboosting the owner)
337 * @top_task: the current top waiter
339 * Returns 0 or -EDEADLK.
341 static int rt_mutex_adjust_prio_chain(struct task_struct *task,
343 struct rt_mutex *orig_lock,
344 struct rt_mutex *next_lock,
345 struct rt_mutex_waiter *orig_waiter,
346 struct task_struct *top_task)
348 struct rt_mutex *lock;
349 struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
350 int detect_deadlock, ret = 0, depth = 0;
353 detect_deadlock = debug_rt_mutex_detect_deadlock(orig_waiter,
357 * The (de)boosting is a step by step approach with a lot of
358 * pitfalls. We want this to be preemptible and we want hold a
359 * maximum of two locks per step. So we have to check
360 * carefully whether things change under us.
363 if (++depth > max_lock_depth) {
367 * Print this only once. If the admin changes the limit,
368 * print a new message when reaching the limit again.
370 if (prev_max != max_lock_depth) {
371 prev_max = max_lock_depth;
372 printk(KERN_WARNING "Maximum lock depth %d reached "
373 "task: %s (%d)\n", max_lock_depth,
374 top_task->comm, task_pid_nr(top_task));
376 put_task_struct(task);
382 * Task can not go away as we did a get_task() before !
384 raw_spin_lock_irqsave(&task->pi_lock, flags);
386 waiter = task->pi_blocked_on;
388 * Check whether the end of the boosting chain has been
389 * reached or the state of the chain has changed while we
396 * Check the orig_waiter state. After we dropped the locks,
397 * the previous owner of the lock might have released the lock.
399 if (orig_waiter && !rt_mutex_owner(orig_lock))
403 * We dropped all locks after taking a refcount on @task, so
404 * the task might have moved on in the lock chain or even left
405 * the chain completely and blocks now on an unrelated lock or
408 * We stored the lock on which @task was blocked in @next_lock,
409 * so we can detect the chain change.
411 if (next_lock != waiter->lock)
415 * Drop out, when the task has no waiters. Note,
416 * top_waiter can be NULL, when we are in the deboosting
420 if (!task_has_pi_waiters(task))
423 * If deadlock detection is off, we stop here if we
424 * are not the top pi waiter of the task.
426 if (!detect_deadlock && top_waiter != task_top_pi_waiter(task))
431 * When deadlock detection is off then we check, if further
432 * priority adjustment is necessary.
434 if (!detect_deadlock && waiter->prio == task->prio)
438 if (!raw_spin_trylock(&lock->wait_lock)) {
439 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
445 * Deadlock detection. If the lock is the same as the original
446 * lock which caused us to walk the lock chain or if the
447 * current lock is owned by the task which initiated the chain
448 * walk, we detected a deadlock.
450 if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
451 debug_rt_mutex_deadlock(deadlock_detect, orig_waiter, lock);
452 raw_spin_unlock(&lock->wait_lock);
457 top_waiter = rt_mutex_top_waiter(lock);
459 /* Requeue the waiter */
460 rt_mutex_dequeue(lock, waiter);
461 waiter->prio = task->prio;
462 rt_mutex_enqueue(lock, waiter);
464 /* Release the task */
465 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
466 put_task_struct(task);
468 if (!rt_mutex_owner(lock)) {
470 * If the requeue above changed the top waiter, then we need
471 * to wake the new top waiter up to try to get the lock.
474 if (top_waiter != rt_mutex_top_waiter(lock))
475 wake_up_process(rt_mutex_top_waiter(lock)->task);
476 raw_spin_unlock(&lock->wait_lock);
480 /* Grab the next task */
481 task = rt_mutex_owner(lock);
482 get_task_struct(task);
483 raw_spin_lock_irqsave(&task->pi_lock, flags);
485 if (waiter == rt_mutex_top_waiter(lock)) {
486 /* Boost the owner */
487 rt_mutex_dequeue_pi(task, top_waiter);
488 rt_mutex_enqueue_pi(task, waiter);
489 __rt_mutex_adjust_prio(task);
491 } else if (top_waiter == waiter) {
492 /* Deboost the owner */
493 rt_mutex_dequeue_pi(task, waiter);
494 waiter = rt_mutex_top_waiter(lock);
495 rt_mutex_enqueue_pi(task, waiter);
496 __rt_mutex_adjust_prio(task);
500 * Check whether the task which owns the current lock is pi
501 * blocked itself. If yes we store a pointer to the lock for
502 * the lock chain change detection above. After we dropped
503 * task->pi_lock next_lock cannot be dereferenced anymore.
505 next_lock = task_blocked_on_lock(task);
507 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
509 top_waiter = rt_mutex_top_waiter(lock);
510 raw_spin_unlock(&lock->wait_lock);
513 * We reached the end of the lock chain. Stop right here. No
514 * point to go back just to figure that out.
519 if (!detect_deadlock && waiter != top_waiter)
525 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
527 put_task_struct(task);
533 * Try to take an rt-mutex
535 * Must be called with lock->wait_lock held.
537 * @lock: The lock to be acquired.
538 * @task: The task which wants to acquire the lock
539 * @waiter: The waiter that is queued to the lock's wait list if the
540 * callsite called task_blocked_on_lock(), otherwise NULL
542 static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
543 struct rt_mutex_waiter *waiter)
548 * Before testing whether we can acquire @lock, we set the
549 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
550 * other tasks which try to modify @lock into the slow path
551 * and they serialize on @lock->wait_lock.
553 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
554 * as explained at the top of this file if and only if:
556 * - There is a lock owner. The caller must fixup the
557 * transient state if it does a trylock or leaves the lock
558 * function due to a signal or timeout.
560 * - @task acquires the lock and there are no other
561 * waiters. This is undone in rt_mutex_set_owner(@task) at
562 * the end of this function.
564 mark_rt_mutex_waiters(lock);
567 * If @lock has an owner, give up.
569 if (rt_mutex_owner(lock))
573 * If @waiter != NULL, @task has already enqueued the waiter
574 * into @lock waiter list. If @waiter == NULL then this is a
579 * If waiter is not the highest priority waiter of
582 if (waiter != rt_mutex_top_waiter(lock))
586 * We can acquire the lock. Remove the waiter from the
589 rt_mutex_dequeue(lock, waiter);
593 * If the lock has waiters already we check whether @task is
594 * eligible to take over the lock.
596 * If there are no other waiters, @task can acquire
597 * the lock. @task->pi_blocked_on is NULL, so it does
598 * not need to be dequeued.
600 if (rt_mutex_has_waiters(lock)) {
602 * If @task->prio is greater than or equal to
603 * the top waiter priority (kernel view),
606 if (task->prio >= rt_mutex_top_waiter(lock)->prio)
610 * The current top waiter stays enqueued. We
611 * don't have to change anything in the lock
616 * No waiters. Take the lock without the
617 * pi_lock dance.@task->pi_blocked_on is NULL
618 * and we have no waiters to enqueue in @task
626 * Clear @task->pi_blocked_on. Requires protection by
627 * @task->pi_lock. Redundant operation for the @waiter == NULL
628 * case, but conditionals are more expensive than a redundant
631 raw_spin_lock_irqsave(&task->pi_lock, flags);
632 task->pi_blocked_on = NULL;
634 * Finish the lock acquisition. @task is the new owner. If
635 * other waiters exist we have to insert the highest priority
636 * waiter into @task->pi_waiters list.
638 if (rt_mutex_has_waiters(lock))
639 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
640 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
643 /* We got the lock. */
644 debug_rt_mutex_lock(lock);
647 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
648 * are still waiters or clears it.
650 rt_mutex_set_owner(lock, task);
652 rt_mutex_deadlock_account_lock(lock, task);
658 * Task blocks on lock.
660 * Prepare waiter and propagate pi chain
662 * This must be called with lock->wait_lock held.
664 static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
665 struct rt_mutex_waiter *waiter,
666 struct task_struct *task,
669 struct task_struct *owner = rt_mutex_owner(lock);
670 struct rt_mutex_waiter *top_waiter = waiter;
671 struct rt_mutex *next_lock;
672 int chain_walk = 0, res;
676 * Early deadlock detection. We really don't want the task to
677 * enqueue on itself just to untangle the mess later. It's not
678 * only an optimization. We drop the locks, so another waiter
679 * can come in before the chain walk detects the deadlock. So
680 * the other will detect the deadlock and return -EDEADLOCK,
681 * which is wrong, as the other waiter is not in a deadlock
687 raw_spin_lock_irqsave(&task->pi_lock, flags);
688 __rt_mutex_adjust_prio(task);
691 waiter->prio = task->prio;
693 /* Get the top priority waiter on the lock */
694 if (rt_mutex_has_waiters(lock))
695 top_waiter = rt_mutex_top_waiter(lock);
696 rt_mutex_enqueue(lock, waiter);
698 task->pi_blocked_on = waiter;
700 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
705 raw_spin_lock_irqsave(&owner->pi_lock, flags);
706 if (waiter == rt_mutex_top_waiter(lock)) {
707 rt_mutex_dequeue_pi(owner, top_waiter);
708 rt_mutex_enqueue_pi(owner, waiter);
710 __rt_mutex_adjust_prio(owner);
711 if (owner->pi_blocked_on)
713 } else if (debug_rt_mutex_detect_deadlock(waiter, detect_deadlock)) {
717 /* Store the lock on which owner is blocked or NULL */
718 next_lock = task_blocked_on_lock(owner);
720 raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
722 * Even if full deadlock detection is on, if the owner is not
723 * blocked itself, we can avoid finding this out in the chain
726 if (!chain_walk || !next_lock)
730 * The owner can't disappear while holding a lock,
731 * so the owner struct is protected by wait_lock.
732 * Gets dropped in rt_mutex_adjust_prio_chain()!
734 get_task_struct(owner);
736 raw_spin_unlock(&lock->wait_lock);
738 res = rt_mutex_adjust_prio_chain(owner, detect_deadlock, lock,
739 next_lock, waiter, task);
741 raw_spin_lock(&lock->wait_lock);
747 * Wake up the next waiter on the lock.
749 * Remove the top waiter from the current tasks pi waiter list and
752 * Called with lock->wait_lock held.
754 static void wakeup_next_waiter(struct rt_mutex *lock)
756 struct rt_mutex_waiter *waiter;
759 raw_spin_lock_irqsave(¤t->pi_lock, flags);
761 waiter = rt_mutex_top_waiter(lock);
764 * Remove it from current->pi_waiters. We do not adjust a
765 * possible priority boost right now. We execute wakeup in the
766 * boosted mode and go back to normal after releasing
769 rt_mutex_dequeue_pi(current, waiter);
772 * As we are waking up the top waiter, and the waiter stays
773 * queued on the lock until it gets the lock, this lock
774 * obviously has waiters. Just set the bit here and this has
775 * the added benefit of forcing all new tasks into the
776 * slow path making sure no task of lower priority than
777 * the top waiter can steal this lock.
779 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
781 raw_spin_unlock_irqrestore(¤t->pi_lock, flags);
784 * It's safe to dereference waiter as it cannot go away as
785 * long as we hold lock->wait_lock. The waiter task needs to
786 * acquire it in order to dequeue the waiter.
788 wake_up_process(waiter->task);
792 * Remove a waiter from a lock and give up
794 * Must be called with lock->wait_lock held and
795 * have just failed to try_to_take_rt_mutex().
797 static void remove_waiter(struct rt_mutex *lock,
798 struct rt_mutex_waiter *waiter)
800 int first = (waiter == rt_mutex_top_waiter(lock));
801 struct task_struct *owner = rt_mutex_owner(lock);
802 struct rt_mutex *next_lock = NULL;
805 raw_spin_lock_irqsave(¤t->pi_lock, flags);
806 rt_mutex_dequeue(lock, waiter);
807 current->pi_blocked_on = NULL;
808 raw_spin_unlock_irqrestore(¤t->pi_lock, flags);
815 raw_spin_lock_irqsave(&owner->pi_lock, flags);
817 rt_mutex_dequeue_pi(owner, waiter);
819 if (rt_mutex_has_waiters(lock)) {
820 struct rt_mutex_waiter *next;
822 next = rt_mutex_top_waiter(lock);
823 rt_mutex_enqueue_pi(owner, next);
825 __rt_mutex_adjust_prio(owner);
827 /* Store the lock on which owner is blocked or NULL */
828 next_lock = task_blocked_on_lock(owner);
830 raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
836 /* gets dropped in rt_mutex_adjust_prio_chain()! */
837 get_task_struct(owner);
839 raw_spin_unlock(&lock->wait_lock);
841 rt_mutex_adjust_prio_chain(owner, 0, lock, next_lock, NULL, current);
843 raw_spin_lock(&lock->wait_lock);
847 * Recheck the pi chain, in case we got a priority setting
849 * Called from sched_setscheduler
851 void rt_mutex_adjust_pi(struct task_struct *task)
853 struct rt_mutex_waiter *waiter;
854 struct rt_mutex *next_lock;
857 raw_spin_lock_irqsave(&task->pi_lock, flags);
859 waiter = task->pi_blocked_on;
860 if (!waiter || (waiter->prio == task->prio &&
861 !dl_prio(task->prio))) {
862 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
865 next_lock = waiter->lock;
866 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
868 /* gets dropped in rt_mutex_adjust_prio_chain()! */
869 get_task_struct(task);
871 rt_mutex_adjust_prio_chain(task, 0, NULL, next_lock, NULL, task);
875 * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
876 * @lock: the rt_mutex to take
877 * @state: the state the task should block in (TASK_INTERRUPTIBLE
878 * or TASK_UNINTERRUPTIBLE)
879 * @timeout: the pre-initialized and started timer, or NULL for none
880 * @waiter: the pre-initialized rt_mutex_waiter
882 * lock->wait_lock must be held by the caller.
885 __rt_mutex_slowlock(struct rt_mutex *lock, int state,
886 struct hrtimer_sleeper *timeout,
887 struct rt_mutex_waiter *waiter)
892 /* Try to acquire the lock: */
893 if (try_to_take_rt_mutex(lock, current, waiter))
897 * TASK_INTERRUPTIBLE checks for signals and
898 * timeout. Ignored otherwise.
900 if (unlikely(state == TASK_INTERRUPTIBLE)) {
901 /* Signal pending? */
902 if (signal_pending(current))
904 if (timeout && !timeout->task)
910 raw_spin_unlock(&lock->wait_lock);
912 debug_rt_mutex_print_deadlock(waiter);
914 schedule_rt_mutex(lock);
916 raw_spin_lock(&lock->wait_lock);
917 set_current_state(state);
923 static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
924 struct rt_mutex_waiter *w)
927 * If the result is not -EDEADLOCK or the caller requested
928 * deadlock detection, nothing to do here.
930 if (res != -EDEADLOCK || detect_deadlock)
934 * Yell lowdly and stop the task right here.
936 rt_mutex_print_deadlock(w);
938 set_current_state(TASK_INTERRUPTIBLE);
944 * Slow path lock function:
947 rt_mutex_slowlock(struct rt_mutex *lock, int state,
948 struct hrtimer_sleeper *timeout,
951 struct rt_mutex_waiter waiter;
954 debug_rt_mutex_init_waiter(&waiter);
955 RB_CLEAR_NODE(&waiter.pi_tree_entry);
956 RB_CLEAR_NODE(&waiter.tree_entry);
958 raw_spin_lock(&lock->wait_lock);
960 /* Try to acquire the lock again: */
961 if (try_to_take_rt_mutex(lock, current, NULL)) {
962 raw_spin_unlock(&lock->wait_lock);
966 set_current_state(state);
968 /* Setup the timer, when timeout != NULL */
969 if (unlikely(timeout)) {
970 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
971 if (!hrtimer_active(&timeout->timer))
972 timeout->task = NULL;
975 ret = task_blocks_on_rt_mutex(lock, &waiter, current, detect_deadlock);
978 ret = __rt_mutex_slowlock(lock, state, timeout, &waiter);
980 set_current_state(TASK_RUNNING);
983 remove_waiter(lock, &waiter);
984 rt_mutex_handle_deadlock(ret, detect_deadlock, &waiter);
988 * try_to_take_rt_mutex() sets the waiter bit
989 * unconditionally. We might have to fix that up.
991 fixup_rt_mutex_waiters(lock);
993 raw_spin_unlock(&lock->wait_lock);
995 /* Remove pending timer: */
996 if (unlikely(timeout))
997 hrtimer_cancel(&timeout->timer);
999 debug_rt_mutex_free_waiter(&waiter);
1005 * Slow path try-lock function:
1007 static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
1012 * If the lock already has an owner we fail to get the lock.
1013 * This can be done without taking the @lock->wait_lock as
1014 * it is only being read, and this is a trylock anyway.
1016 if (rt_mutex_owner(lock))
1020 * The mutex has currently no owner. Lock the wait lock and
1021 * try to acquire the lock.
1023 raw_spin_lock(&lock->wait_lock);
1025 ret = try_to_take_rt_mutex(lock, current, NULL);
1028 * try_to_take_rt_mutex() sets the lock waiters bit
1029 * unconditionally. Clean this up.
1031 fixup_rt_mutex_waiters(lock);
1033 raw_spin_unlock(&lock->wait_lock);
1039 * Slow path to release a rt-mutex:
1042 rt_mutex_slowunlock(struct rt_mutex *lock)
1044 raw_spin_lock(&lock->wait_lock);
1046 debug_rt_mutex_unlock(lock);
1048 rt_mutex_deadlock_account_unlock(current);
1051 * We must be careful here if the fast path is enabled. If we
1052 * have no waiters queued we cannot set owner to NULL here
1055 * foo->lock->owner = NULL;
1056 * rtmutex_lock(foo->lock); <- fast path
1057 * free = atomic_dec_and_test(foo->refcnt);
1058 * rtmutex_unlock(foo->lock); <- fast path
1061 * raw_spin_unlock(foo->lock->wait_lock);
1063 * So for the fastpath enabled kernel:
1065 * Nothing can set the waiters bit as long as we hold
1066 * lock->wait_lock. So we do the following sequence:
1068 * owner = rt_mutex_owner(lock);
1069 * clear_rt_mutex_waiters(lock);
1070 * raw_spin_unlock(&lock->wait_lock);
1071 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1075 * The fastpath disabled variant is simple as all access to
1076 * lock->owner is serialized by lock->wait_lock:
1078 * lock->owner = NULL;
1079 * raw_spin_unlock(&lock->wait_lock);
1081 while (!rt_mutex_has_waiters(lock)) {
1082 /* Drops lock->wait_lock ! */
1083 if (unlock_rt_mutex_safe(lock) == true)
1085 /* Relock the rtmutex and try again */
1086 raw_spin_lock(&lock->wait_lock);
1090 * The wakeup next waiter path does not suffer from the above
1091 * race. See the comments there.
1093 wakeup_next_waiter(lock);
1095 raw_spin_unlock(&lock->wait_lock);
1097 /* Undo pi boosting if necessary: */
1098 rt_mutex_adjust_prio(current);
1102 * debug aware fast / slowpath lock,trylock,unlock
1104 * The atomic acquire/release ops are compiled away, when either the
1105 * architecture does not support cmpxchg or when debugging is enabled.
1108 rt_mutex_fastlock(struct rt_mutex *lock, int state,
1109 int detect_deadlock,
1110 int (*slowfn)(struct rt_mutex *lock, int state,
1111 struct hrtimer_sleeper *timeout,
1112 int detect_deadlock))
1114 if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1115 rt_mutex_deadlock_account_lock(lock, current);
1118 return slowfn(lock, state, NULL, detect_deadlock);
1122 rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
1123 struct hrtimer_sleeper *timeout, int detect_deadlock,
1124 int (*slowfn)(struct rt_mutex *lock, int state,
1125 struct hrtimer_sleeper *timeout,
1126 int detect_deadlock))
1128 if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1129 rt_mutex_deadlock_account_lock(lock, current);
1132 return slowfn(lock, state, timeout, detect_deadlock);
1136 rt_mutex_fasttrylock(struct rt_mutex *lock,
1137 int (*slowfn)(struct rt_mutex *lock))
1139 if (likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1140 rt_mutex_deadlock_account_lock(lock, current);
1143 return slowfn(lock);
1147 rt_mutex_fastunlock(struct rt_mutex *lock,
1148 void (*slowfn)(struct rt_mutex *lock))
1150 if (likely(rt_mutex_cmpxchg(lock, current, NULL)))
1151 rt_mutex_deadlock_account_unlock(current);
1157 * rt_mutex_lock - lock a rt_mutex
1159 * @lock: the rt_mutex to be locked
1161 void __sched rt_mutex_lock(struct rt_mutex *lock)
1165 rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, 0, rt_mutex_slowlock);
1167 EXPORT_SYMBOL_GPL(rt_mutex_lock);
1170 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
1172 * @lock: the rt_mutex to be locked
1173 * @detect_deadlock: deadlock detection on/off
1177 * -EINTR when interrupted by a signal
1178 * -EDEADLK when the lock would deadlock (when deadlock detection is on)
1180 int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock,
1181 int detect_deadlock)
1185 return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE,
1186 detect_deadlock, rt_mutex_slowlock);
1188 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
1191 * rt_mutex_timed_lock - lock a rt_mutex interruptible
1192 * the timeout structure is provided
1195 * @lock: the rt_mutex to be locked
1196 * @timeout: timeout structure or NULL (no timeout)
1197 * @detect_deadlock: deadlock detection on/off
1201 * -EINTR when interrupted by a signal
1202 * -ETIMEDOUT when the timeout expired
1203 * -EDEADLK when the lock would deadlock (when deadlock detection is on)
1206 rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout,
1207 int detect_deadlock)
1211 return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
1212 detect_deadlock, rt_mutex_slowlock);
1214 EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
1217 * rt_mutex_trylock - try to lock a rt_mutex
1219 * @lock: the rt_mutex to be locked
1221 * Returns 1 on success and 0 on contention
1223 int __sched rt_mutex_trylock(struct rt_mutex *lock)
1225 return rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
1227 EXPORT_SYMBOL_GPL(rt_mutex_trylock);
1230 * rt_mutex_unlock - unlock a rt_mutex
1232 * @lock: the rt_mutex to be unlocked
1234 void __sched rt_mutex_unlock(struct rt_mutex *lock)
1236 rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
1238 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
1241 * rt_mutex_destroy - mark a mutex unusable
1242 * @lock: the mutex to be destroyed
1244 * This function marks the mutex uninitialized, and any subsequent
1245 * use of the mutex is forbidden. The mutex must not be locked when
1246 * this function is called.
1248 void rt_mutex_destroy(struct rt_mutex *lock)
1250 WARN_ON(rt_mutex_is_locked(lock));
1251 #ifdef CONFIG_DEBUG_RT_MUTEXES
1256 EXPORT_SYMBOL_GPL(rt_mutex_destroy);
1259 * __rt_mutex_init - initialize the rt lock
1261 * @lock: the rt lock to be initialized
1263 * Initialize the rt lock to unlocked state.
1265 * Initializing of a locked rt lock is not allowed
1267 void __rt_mutex_init(struct rt_mutex *lock, const char *name)
1270 raw_spin_lock_init(&lock->wait_lock);
1271 lock->waiters = RB_ROOT;
1272 lock->waiters_leftmost = NULL;
1274 debug_rt_mutex_init(lock, name);
1276 EXPORT_SYMBOL_GPL(__rt_mutex_init);
1279 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
1282 * @lock: the rt_mutex to be locked
1283 * @proxy_owner:the task to set as owner
1285 * No locking. Caller has to do serializing itself
1286 * Special API call for PI-futex support
1288 void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
1289 struct task_struct *proxy_owner)
1291 __rt_mutex_init(lock, NULL);
1292 debug_rt_mutex_proxy_lock(lock, proxy_owner);
1293 rt_mutex_set_owner(lock, proxy_owner);
1294 rt_mutex_deadlock_account_lock(lock, proxy_owner);
1298 * rt_mutex_proxy_unlock - release a lock on behalf of owner
1300 * @lock: the rt_mutex to be locked
1302 * No locking. Caller has to do serializing itself
1303 * Special API call for PI-futex support
1305 void rt_mutex_proxy_unlock(struct rt_mutex *lock,
1306 struct task_struct *proxy_owner)
1308 debug_rt_mutex_proxy_unlock(lock);
1309 rt_mutex_set_owner(lock, NULL);
1310 rt_mutex_deadlock_account_unlock(proxy_owner);
1314 * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1315 * @lock: the rt_mutex to take
1316 * @waiter: the pre-initialized rt_mutex_waiter
1317 * @task: the task to prepare
1318 * @detect_deadlock: perform deadlock detection (1) or not (0)
1321 * 0 - task blocked on lock
1322 * 1 - acquired the lock for task, caller should wake it up
1325 * Special API call for FUTEX_REQUEUE_PI support.
1327 int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
1328 struct rt_mutex_waiter *waiter,
1329 struct task_struct *task, int detect_deadlock)
1333 raw_spin_lock(&lock->wait_lock);
1335 if (try_to_take_rt_mutex(lock, task, NULL)) {
1336 raw_spin_unlock(&lock->wait_lock);
1340 /* We enforce deadlock detection for futexes */
1341 ret = task_blocks_on_rt_mutex(lock, waiter, task, 1);
1343 if (ret && !rt_mutex_owner(lock)) {
1345 * Reset the return value. We might have
1346 * returned with -EDEADLK and the owner
1347 * released the lock while we were walking the
1348 * pi chain. Let the waiter sort it out.
1354 remove_waiter(lock, waiter);
1356 raw_spin_unlock(&lock->wait_lock);
1358 debug_rt_mutex_print_deadlock(waiter);
1364 * rt_mutex_next_owner - return the next owner of the lock
1366 * @lock: the rt lock query
1368 * Returns the next owner of the lock or NULL
1370 * Caller has to serialize against other accessors to the lock
1373 * Special API call for PI-futex support
1375 struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock)
1377 if (!rt_mutex_has_waiters(lock))
1380 return rt_mutex_top_waiter(lock)->task;
1384 * rt_mutex_finish_proxy_lock() - Complete lock acquisition
1385 * @lock: the rt_mutex we were woken on
1386 * @to: the timeout, null if none. hrtimer should already have
1388 * @waiter: the pre-initialized rt_mutex_waiter
1389 * @detect_deadlock: perform deadlock detection (1) or not (0)
1391 * Complete the lock acquisition started our behalf by another thread.
1395 * <0 - error, one of -EINTR, -ETIMEDOUT, or -EDEADLK
1397 * Special API call for PI-futex requeue support
1399 int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
1400 struct hrtimer_sleeper *to,
1401 struct rt_mutex_waiter *waiter,
1402 int detect_deadlock)
1406 raw_spin_lock(&lock->wait_lock);
1408 set_current_state(TASK_INTERRUPTIBLE);
1410 ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
1412 set_current_state(TASK_RUNNING);
1415 remove_waiter(lock, waiter);
1418 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1419 * have to fix that up.
1421 fixup_rt_mutex_waiters(lock);
1423 raw_spin_unlock(&lock->wait_lock);