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 if (!rt_mutex_owner(lock)) {
468 * If the requeue above changed the top waiter, then we need
469 * to wake the new top waiter up to try to get the lock.
472 if (top_waiter != rt_mutex_top_waiter(lock))
473 wake_up_process(rt_mutex_top_waiter(lock)->task);
474 raw_spin_unlock(&lock->wait_lock);
477 put_task_struct(task);
479 /* Grab the next task */
480 task = rt_mutex_owner(lock);
481 get_task_struct(task);
482 raw_spin_lock_irqsave(&task->pi_lock, flags);
484 if (waiter == rt_mutex_top_waiter(lock)) {
485 /* Boost the owner */
486 rt_mutex_dequeue_pi(task, top_waiter);
487 rt_mutex_enqueue_pi(task, waiter);
488 __rt_mutex_adjust_prio(task);
490 } else if (top_waiter == waiter) {
491 /* Deboost the owner */
492 rt_mutex_dequeue_pi(task, waiter);
493 waiter = rt_mutex_top_waiter(lock);
494 rt_mutex_enqueue_pi(task, waiter);
495 __rt_mutex_adjust_prio(task);
499 * Check whether the task which owns the current lock is pi
500 * blocked itself. If yes we store a pointer to the lock for
501 * the lock chain change detection above. After we dropped
502 * task->pi_lock next_lock cannot be dereferenced anymore.
504 next_lock = task_blocked_on_lock(task);
506 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
508 top_waiter = rt_mutex_top_waiter(lock);
509 raw_spin_unlock(&lock->wait_lock);
512 * We reached the end of the lock chain. Stop right here. No
513 * point to go back just to figure that out.
518 if (!detect_deadlock && waiter != top_waiter)
524 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
526 put_task_struct(task);
532 * Try to take an rt-mutex
534 * Must be called with lock->wait_lock held.
536 * @lock: The lock to be acquired.
537 * @task: The task which wants to acquire the lock
538 * @waiter: The waiter that is queued to the lock's wait list if the
539 * callsite called task_blocked_on_lock(), otherwise NULL
541 static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
542 struct rt_mutex_waiter *waiter)
547 * Before testing whether we can acquire @lock, we set the
548 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
549 * other tasks which try to modify @lock into the slow path
550 * and they serialize on @lock->wait_lock.
552 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
553 * as explained at the top of this file if and only if:
555 * - There is a lock owner. The caller must fixup the
556 * transient state if it does a trylock or leaves the lock
557 * function due to a signal or timeout.
559 * - @task acquires the lock and there are no other
560 * waiters. This is undone in rt_mutex_set_owner(@task) at
561 * the end of this function.
563 mark_rt_mutex_waiters(lock);
566 * If @lock has an owner, give up.
568 if (rt_mutex_owner(lock))
572 * If @waiter != NULL, @task has already enqueued the waiter
573 * into @lock waiter list. If @waiter == NULL then this is a
578 * If waiter is not the highest priority waiter of
581 if (waiter != rt_mutex_top_waiter(lock))
585 * We can acquire the lock. Remove the waiter from the
588 rt_mutex_dequeue(lock, waiter);
592 * If the lock has waiters already we check whether @task is
593 * eligible to take over the lock.
595 * If there are no other waiters, @task can acquire
596 * the lock. @task->pi_blocked_on is NULL, so it does
597 * not need to be dequeued.
599 if (rt_mutex_has_waiters(lock)) {
601 * If @task->prio is greater than or equal to
602 * the top waiter priority (kernel view),
605 if (task->prio >= rt_mutex_top_waiter(lock)->prio)
609 * The current top waiter stays enqueued. We
610 * don't have to change anything in the lock
615 * No waiters. Take the lock without the
616 * pi_lock dance.@task->pi_blocked_on is NULL
617 * and we have no waiters to enqueue in @task
625 * Clear @task->pi_blocked_on. Requires protection by
626 * @task->pi_lock. Redundant operation for the @waiter == NULL
627 * case, but conditionals are more expensive than a redundant
630 raw_spin_lock_irqsave(&task->pi_lock, flags);
631 task->pi_blocked_on = NULL;
633 * Finish the lock acquisition. @task is the new owner. If
634 * other waiters exist we have to insert the highest priority
635 * waiter into @task->pi_waiters list.
637 if (rt_mutex_has_waiters(lock))
638 rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
639 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
642 /* We got the lock. */
643 debug_rt_mutex_lock(lock);
646 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
647 * are still waiters or clears it.
649 rt_mutex_set_owner(lock, task);
651 rt_mutex_deadlock_account_lock(lock, task);
657 * Task blocks on lock.
659 * Prepare waiter and propagate pi chain
661 * This must be called with lock->wait_lock held.
663 static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
664 struct rt_mutex_waiter *waiter,
665 struct task_struct *task,
668 struct task_struct *owner = rt_mutex_owner(lock);
669 struct rt_mutex_waiter *top_waiter = waiter;
670 struct rt_mutex *next_lock;
671 int chain_walk = 0, res;
675 * Early deadlock detection. We really don't want the task to
676 * enqueue on itself just to untangle the mess later. It's not
677 * only an optimization. We drop the locks, so another waiter
678 * can come in before the chain walk detects the deadlock. So
679 * the other will detect the deadlock and return -EDEADLOCK,
680 * which is wrong, as the other waiter is not in a deadlock
686 raw_spin_lock_irqsave(&task->pi_lock, flags);
687 __rt_mutex_adjust_prio(task);
690 waiter->prio = task->prio;
692 /* Get the top priority waiter on the lock */
693 if (rt_mutex_has_waiters(lock))
694 top_waiter = rt_mutex_top_waiter(lock);
695 rt_mutex_enqueue(lock, waiter);
697 task->pi_blocked_on = waiter;
699 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
704 raw_spin_lock_irqsave(&owner->pi_lock, flags);
705 if (waiter == rt_mutex_top_waiter(lock)) {
706 rt_mutex_dequeue_pi(owner, top_waiter);
707 rt_mutex_enqueue_pi(owner, waiter);
709 __rt_mutex_adjust_prio(owner);
710 if (owner->pi_blocked_on)
712 } else if (debug_rt_mutex_detect_deadlock(waiter, detect_deadlock)) {
716 /* Store the lock on which owner is blocked or NULL */
717 next_lock = task_blocked_on_lock(owner);
719 raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
721 * Even if full deadlock detection is on, if the owner is not
722 * blocked itself, we can avoid finding this out in the chain
725 if (!chain_walk || !next_lock)
729 * The owner can't disappear while holding a lock,
730 * so the owner struct is protected by wait_lock.
731 * Gets dropped in rt_mutex_adjust_prio_chain()!
733 get_task_struct(owner);
735 raw_spin_unlock(&lock->wait_lock);
737 res = rt_mutex_adjust_prio_chain(owner, detect_deadlock, lock,
738 next_lock, waiter, task);
740 raw_spin_lock(&lock->wait_lock);
746 * Wake up the next waiter on the lock.
748 * Remove the top waiter from the current tasks pi waiter list and
751 * Called with lock->wait_lock held.
753 static void wakeup_next_waiter(struct rt_mutex *lock)
755 struct rt_mutex_waiter *waiter;
758 raw_spin_lock_irqsave(¤t->pi_lock, flags);
760 waiter = rt_mutex_top_waiter(lock);
763 * Remove it from current->pi_waiters. We do not adjust a
764 * possible priority boost right now. We execute wakeup in the
765 * boosted mode and go back to normal after releasing
768 rt_mutex_dequeue_pi(current, waiter);
771 * As we are waking up the top waiter, and the waiter stays
772 * queued on the lock until it gets the lock, this lock
773 * obviously has waiters. Just set the bit here and this has
774 * the added benefit of forcing all new tasks into the
775 * slow path making sure no task of lower priority than
776 * the top waiter can steal this lock.
778 lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
780 raw_spin_unlock_irqrestore(¤t->pi_lock, flags);
783 * It's safe to dereference waiter as it cannot go away as
784 * long as we hold lock->wait_lock. The waiter task needs to
785 * acquire it in order to dequeue the waiter.
787 wake_up_process(waiter->task);
791 * Remove a waiter from a lock and give up
793 * Must be called with lock->wait_lock held and
794 * have just failed to try_to_take_rt_mutex().
796 static void remove_waiter(struct rt_mutex *lock,
797 struct rt_mutex_waiter *waiter)
799 int first = (waiter == rt_mutex_top_waiter(lock));
800 struct task_struct *owner = rt_mutex_owner(lock);
801 struct rt_mutex *next_lock = NULL;
804 raw_spin_lock_irqsave(¤t->pi_lock, flags);
805 rt_mutex_dequeue(lock, waiter);
806 current->pi_blocked_on = NULL;
807 raw_spin_unlock_irqrestore(¤t->pi_lock, flags);
814 raw_spin_lock_irqsave(&owner->pi_lock, flags);
816 rt_mutex_dequeue_pi(owner, waiter);
818 if (rt_mutex_has_waiters(lock)) {
819 struct rt_mutex_waiter *next;
821 next = rt_mutex_top_waiter(lock);
822 rt_mutex_enqueue_pi(owner, next);
824 __rt_mutex_adjust_prio(owner);
826 /* Store the lock on which owner is blocked or NULL */
827 next_lock = task_blocked_on_lock(owner);
829 raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
835 /* gets dropped in rt_mutex_adjust_prio_chain()! */
836 get_task_struct(owner);
838 raw_spin_unlock(&lock->wait_lock);
840 rt_mutex_adjust_prio_chain(owner, 0, lock, next_lock, NULL, current);
842 raw_spin_lock(&lock->wait_lock);
846 * Recheck the pi chain, in case we got a priority setting
848 * Called from sched_setscheduler
850 void rt_mutex_adjust_pi(struct task_struct *task)
852 struct rt_mutex_waiter *waiter;
853 struct rt_mutex *next_lock;
856 raw_spin_lock_irqsave(&task->pi_lock, flags);
858 waiter = task->pi_blocked_on;
859 if (!waiter || (waiter->prio == task->prio &&
860 !dl_prio(task->prio))) {
861 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
864 next_lock = waiter->lock;
865 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
867 /* gets dropped in rt_mutex_adjust_prio_chain()! */
868 get_task_struct(task);
870 rt_mutex_adjust_prio_chain(task, 0, NULL, next_lock, NULL, task);
874 * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
875 * @lock: the rt_mutex to take
876 * @state: the state the task should block in (TASK_INTERRUPTIBLE
877 * or TASK_UNINTERRUPTIBLE)
878 * @timeout: the pre-initialized and started timer, or NULL for none
879 * @waiter: the pre-initialized rt_mutex_waiter
881 * lock->wait_lock must be held by the caller.
884 __rt_mutex_slowlock(struct rt_mutex *lock, int state,
885 struct hrtimer_sleeper *timeout,
886 struct rt_mutex_waiter *waiter)
891 /* Try to acquire the lock: */
892 if (try_to_take_rt_mutex(lock, current, waiter))
896 * TASK_INTERRUPTIBLE checks for signals and
897 * timeout. Ignored otherwise.
899 if (unlikely(state == TASK_INTERRUPTIBLE)) {
900 /* Signal pending? */
901 if (signal_pending(current))
903 if (timeout && !timeout->task)
909 raw_spin_unlock(&lock->wait_lock);
911 debug_rt_mutex_print_deadlock(waiter);
913 schedule_rt_mutex(lock);
915 raw_spin_lock(&lock->wait_lock);
916 set_current_state(state);
922 static void rt_mutex_handle_deadlock(int res, int detect_deadlock,
923 struct rt_mutex_waiter *w)
926 * If the result is not -EDEADLOCK or the caller requested
927 * deadlock detection, nothing to do here.
929 if (res != -EDEADLOCK || detect_deadlock)
933 * Yell lowdly and stop the task right here.
935 rt_mutex_print_deadlock(w);
937 set_current_state(TASK_INTERRUPTIBLE);
943 * Slow path lock function:
946 rt_mutex_slowlock(struct rt_mutex *lock, int state,
947 struct hrtimer_sleeper *timeout,
950 struct rt_mutex_waiter waiter;
953 debug_rt_mutex_init_waiter(&waiter);
954 RB_CLEAR_NODE(&waiter.pi_tree_entry);
955 RB_CLEAR_NODE(&waiter.tree_entry);
957 raw_spin_lock(&lock->wait_lock);
959 /* Try to acquire the lock again: */
960 if (try_to_take_rt_mutex(lock, current, NULL)) {
961 raw_spin_unlock(&lock->wait_lock);
965 set_current_state(state);
967 /* Setup the timer, when timeout != NULL */
968 if (unlikely(timeout)) {
969 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
970 if (!hrtimer_active(&timeout->timer))
971 timeout->task = NULL;
974 ret = task_blocks_on_rt_mutex(lock, &waiter, current, detect_deadlock);
977 ret = __rt_mutex_slowlock(lock, state, timeout, &waiter);
979 set_current_state(TASK_RUNNING);
982 remove_waiter(lock, &waiter);
983 rt_mutex_handle_deadlock(ret, detect_deadlock, &waiter);
987 * try_to_take_rt_mutex() sets the waiter bit
988 * unconditionally. We might have to fix that up.
990 fixup_rt_mutex_waiters(lock);
992 raw_spin_unlock(&lock->wait_lock);
994 /* Remove pending timer: */
995 if (unlikely(timeout))
996 hrtimer_cancel(&timeout->timer);
998 debug_rt_mutex_free_waiter(&waiter);
1004 * Slow path try-lock function:
1006 static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
1011 * If the lock already has an owner we fail to get the lock.
1012 * This can be done without taking the @lock->wait_lock as
1013 * it is only being read, and this is a trylock anyway.
1015 if (rt_mutex_owner(lock))
1019 * The mutex has currently no owner. Lock the wait lock and
1020 * try to acquire the lock.
1022 raw_spin_lock(&lock->wait_lock);
1024 ret = try_to_take_rt_mutex(lock, current, NULL);
1027 * try_to_take_rt_mutex() sets the lock waiters bit
1028 * unconditionally. Clean this up.
1030 fixup_rt_mutex_waiters(lock);
1032 raw_spin_unlock(&lock->wait_lock);
1038 * Slow path to release a rt-mutex:
1041 rt_mutex_slowunlock(struct rt_mutex *lock)
1043 raw_spin_lock(&lock->wait_lock);
1045 debug_rt_mutex_unlock(lock);
1047 rt_mutex_deadlock_account_unlock(current);
1050 * We must be careful here if the fast path is enabled. If we
1051 * have no waiters queued we cannot set owner to NULL here
1054 * foo->lock->owner = NULL;
1055 * rtmutex_lock(foo->lock); <- fast path
1056 * free = atomic_dec_and_test(foo->refcnt);
1057 * rtmutex_unlock(foo->lock); <- fast path
1060 * raw_spin_unlock(foo->lock->wait_lock);
1062 * So for the fastpath enabled kernel:
1064 * Nothing can set the waiters bit as long as we hold
1065 * lock->wait_lock. So we do the following sequence:
1067 * owner = rt_mutex_owner(lock);
1068 * clear_rt_mutex_waiters(lock);
1069 * raw_spin_unlock(&lock->wait_lock);
1070 * if (cmpxchg(&lock->owner, owner, 0) == owner)
1074 * The fastpath disabled variant is simple as all access to
1075 * lock->owner is serialized by lock->wait_lock:
1077 * lock->owner = NULL;
1078 * raw_spin_unlock(&lock->wait_lock);
1080 while (!rt_mutex_has_waiters(lock)) {
1081 /* Drops lock->wait_lock ! */
1082 if (unlock_rt_mutex_safe(lock) == true)
1084 /* Relock the rtmutex and try again */
1085 raw_spin_lock(&lock->wait_lock);
1089 * The wakeup next waiter path does not suffer from the above
1090 * race. See the comments there.
1092 wakeup_next_waiter(lock);
1094 raw_spin_unlock(&lock->wait_lock);
1096 /* Undo pi boosting if necessary: */
1097 rt_mutex_adjust_prio(current);
1101 * debug aware fast / slowpath lock,trylock,unlock
1103 * The atomic acquire/release ops are compiled away, when either the
1104 * architecture does not support cmpxchg or when debugging is enabled.
1107 rt_mutex_fastlock(struct rt_mutex *lock, int state,
1108 int detect_deadlock,
1109 int (*slowfn)(struct rt_mutex *lock, int state,
1110 struct hrtimer_sleeper *timeout,
1111 int detect_deadlock))
1113 if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1114 rt_mutex_deadlock_account_lock(lock, current);
1117 return slowfn(lock, state, NULL, detect_deadlock);
1121 rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
1122 struct hrtimer_sleeper *timeout, int detect_deadlock,
1123 int (*slowfn)(struct rt_mutex *lock, int state,
1124 struct hrtimer_sleeper *timeout,
1125 int detect_deadlock))
1127 if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1128 rt_mutex_deadlock_account_lock(lock, current);
1131 return slowfn(lock, state, timeout, detect_deadlock);
1135 rt_mutex_fasttrylock(struct rt_mutex *lock,
1136 int (*slowfn)(struct rt_mutex *lock))
1138 if (likely(rt_mutex_cmpxchg(lock, NULL, current))) {
1139 rt_mutex_deadlock_account_lock(lock, current);
1142 return slowfn(lock);
1146 rt_mutex_fastunlock(struct rt_mutex *lock,
1147 void (*slowfn)(struct rt_mutex *lock))
1149 if (likely(rt_mutex_cmpxchg(lock, current, NULL)))
1150 rt_mutex_deadlock_account_unlock(current);
1156 * rt_mutex_lock - lock a rt_mutex
1158 * @lock: the rt_mutex to be locked
1160 void __sched rt_mutex_lock(struct rt_mutex *lock)
1164 rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, 0, rt_mutex_slowlock);
1166 EXPORT_SYMBOL_GPL(rt_mutex_lock);
1169 * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
1171 * @lock: the rt_mutex to be locked
1172 * @detect_deadlock: deadlock detection on/off
1176 * -EINTR when interrupted by a signal
1177 * -EDEADLK when the lock would deadlock (when deadlock detection is on)
1179 int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock,
1180 int detect_deadlock)
1184 return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE,
1185 detect_deadlock, rt_mutex_slowlock);
1187 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
1190 * rt_mutex_timed_lock - lock a rt_mutex interruptible
1191 * the timeout structure is provided
1194 * @lock: the rt_mutex to be locked
1195 * @timeout: timeout structure or NULL (no timeout)
1196 * @detect_deadlock: deadlock detection on/off
1200 * -EINTR when interrupted by a signal
1201 * -ETIMEDOUT when the timeout expired
1202 * -EDEADLK when the lock would deadlock (when deadlock detection is on)
1205 rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout,
1206 int detect_deadlock)
1210 return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
1211 detect_deadlock, rt_mutex_slowlock);
1213 EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
1216 * rt_mutex_trylock - try to lock a rt_mutex
1218 * @lock: the rt_mutex to be locked
1220 * Returns 1 on success and 0 on contention
1222 int __sched rt_mutex_trylock(struct rt_mutex *lock)
1224 return rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
1226 EXPORT_SYMBOL_GPL(rt_mutex_trylock);
1229 * rt_mutex_unlock - unlock a rt_mutex
1231 * @lock: the rt_mutex to be unlocked
1233 void __sched rt_mutex_unlock(struct rt_mutex *lock)
1235 rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
1237 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
1240 * rt_mutex_destroy - mark a mutex unusable
1241 * @lock: the mutex to be destroyed
1243 * This function marks the mutex uninitialized, and any subsequent
1244 * use of the mutex is forbidden. The mutex must not be locked when
1245 * this function is called.
1247 void rt_mutex_destroy(struct rt_mutex *lock)
1249 WARN_ON(rt_mutex_is_locked(lock));
1250 #ifdef CONFIG_DEBUG_RT_MUTEXES
1255 EXPORT_SYMBOL_GPL(rt_mutex_destroy);
1258 * __rt_mutex_init - initialize the rt lock
1260 * @lock: the rt lock to be initialized
1262 * Initialize the rt lock to unlocked state.
1264 * Initializing of a locked rt lock is not allowed
1266 void __rt_mutex_init(struct rt_mutex *lock, const char *name)
1269 raw_spin_lock_init(&lock->wait_lock);
1270 lock->waiters = RB_ROOT;
1271 lock->waiters_leftmost = NULL;
1273 debug_rt_mutex_init(lock, name);
1275 EXPORT_SYMBOL_GPL(__rt_mutex_init);
1278 * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
1281 * @lock: the rt_mutex to be locked
1282 * @proxy_owner:the task to set as owner
1284 * No locking. Caller has to do serializing itself
1285 * Special API call for PI-futex support
1287 void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
1288 struct task_struct *proxy_owner)
1290 __rt_mutex_init(lock, NULL);
1291 debug_rt_mutex_proxy_lock(lock, proxy_owner);
1292 rt_mutex_set_owner(lock, proxy_owner);
1293 rt_mutex_deadlock_account_lock(lock, proxy_owner);
1297 * rt_mutex_proxy_unlock - release a lock on behalf of owner
1299 * @lock: the rt_mutex to be locked
1301 * No locking. Caller has to do serializing itself
1302 * Special API call for PI-futex support
1304 void rt_mutex_proxy_unlock(struct rt_mutex *lock,
1305 struct task_struct *proxy_owner)
1307 debug_rt_mutex_proxy_unlock(lock);
1308 rt_mutex_set_owner(lock, NULL);
1309 rt_mutex_deadlock_account_unlock(proxy_owner);
1313 * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1314 * @lock: the rt_mutex to take
1315 * @waiter: the pre-initialized rt_mutex_waiter
1316 * @task: the task to prepare
1317 * @detect_deadlock: perform deadlock detection (1) or not (0)
1320 * 0 - task blocked on lock
1321 * 1 - acquired the lock for task, caller should wake it up
1324 * Special API call for FUTEX_REQUEUE_PI support.
1326 int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
1327 struct rt_mutex_waiter *waiter,
1328 struct task_struct *task, int detect_deadlock)
1332 raw_spin_lock(&lock->wait_lock);
1334 if (try_to_take_rt_mutex(lock, task, NULL)) {
1335 raw_spin_unlock(&lock->wait_lock);
1339 /* We enforce deadlock detection for futexes */
1340 ret = task_blocks_on_rt_mutex(lock, waiter, task, 1);
1342 if (ret && !rt_mutex_owner(lock)) {
1344 * Reset the return value. We might have
1345 * returned with -EDEADLK and the owner
1346 * released the lock while we were walking the
1347 * pi chain. Let the waiter sort it out.
1353 remove_waiter(lock, waiter);
1355 raw_spin_unlock(&lock->wait_lock);
1357 debug_rt_mutex_print_deadlock(waiter);
1363 * rt_mutex_next_owner - return the next owner of the lock
1365 * @lock: the rt lock query
1367 * Returns the next owner of the lock or NULL
1369 * Caller has to serialize against other accessors to the lock
1372 * Special API call for PI-futex support
1374 struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock)
1376 if (!rt_mutex_has_waiters(lock))
1379 return rt_mutex_top_waiter(lock)->task;
1383 * rt_mutex_finish_proxy_lock() - Complete lock acquisition
1384 * @lock: the rt_mutex we were woken on
1385 * @to: the timeout, null if none. hrtimer should already have
1387 * @waiter: the pre-initialized rt_mutex_waiter
1388 * @detect_deadlock: perform deadlock detection (1) or not (0)
1390 * Complete the lock acquisition started our behalf by another thread.
1394 * <0 - error, one of -EINTR, -ETIMEDOUT, or -EDEADLK
1396 * Special API call for PI-futex requeue support
1398 int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
1399 struct hrtimer_sleeper *to,
1400 struct rt_mutex_waiter *waiter,
1401 int detect_deadlock)
1405 raw_spin_lock(&lock->wait_lock);
1407 set_current_state(TASK_INTERRUPTIBLE);
1409 ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
1411 set_current_state(TASK_RUNNING);
1414 remove_waiter(lock, waiter);
1417 * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1418 * have to fix that up.
1420 fixup_rt_mutex_waiters(lock);
1422 raw_spin_unlock(&lock->wait_lock);