1 // SPDX-License-Identifier: GPL-2.0-only
3 * kernel/locking/mutex.c
5 * Mutexes: blocking mutual exclusion locks
7 * Started by Ingo Molnar:
9 * Copyright (C) 2004, 2005, 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
11 * Many thanks to Arjan van de Ven, Thomas Gleixner, Steven Rostedt and
12 * David Howells for suggestions and improvements.
14 * - Adaptive spinning for mutexes by Peter Zijlstra. (Ported to mainline
15 * from the -rt tree, where it was originally implemented for rtmutexes
16 * by Steven Rostedt, based on work by Gregory Haskins, Peter Morreale
19 * Also see Documentation/locking/mutex-design.rst.
21 #include <linux/mutex.h>
22 #include <linux/ww_mutex.h>
23 #include <linux/sched/signal.h>
24 #include <linux/sched/rt.h>
25 #include <linux/sched/wake_q.h>
26 #include <linux/sched/debug.h>
27 #include <linux/export.h>
28 #include <linux/spinlock.h>
29 #include <linux/interrupt.h>
30 #include <linux/debug_locks.h>
31 #include <linux/osq_lock.h>
35 #ifdef CONFIG_DEBUG_MUTEXES
36 # define MUTEX_WARN_ON(cond) DEBUG_LOCKS_WARN_ON(cond)
38 # define MUTEX_WARN_ON(cond)
42 __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
44 atomic_long_set(&lock->owner, 0);
45 spin_lock_init(&lock->wait_lock);
46 INIT_LIST_HEAD(&lock->wait_list);
47 #ifdef CONFIG_MUTEX_SPIN_ON_OWNER
48 osq_lock_init(&lock->osq);
51 debug_mutex_init(lock, name, key);
53 EXPORT_SYMBOL(__mutex_init);
56 * @owner: contains: 'struct task_struct *' to the current lock owner,
57 * NULL means not owned. Since task_struct pointers are aligned at
58 * at least L1_CACHE_BYTES, we have low bits to store extra state.
60 * Bit0 indicates a non-empty waiter list; unlock must issue a wakeup.
61 * Bit1 indicates unlock needs to hand the lock to the top-waiter
62 * Bit2 indicates handoff has been done and we're waiting for pickup.
64 #define MUTEX_FLAG_WAITERS 0x01
65 #define MUTEX_FLAG_HANDOFF 0x02
66 #define MUTEX_FLAG_PICKUP 0x04
68 #define MUTEX_FLAGS 0x07
71 * Internal helper function; C doesn't allow us to hide it :/
73 * DO NOT USE (outside of mutex code).
75 static inline struct task_struct *__mutex_owner(struct mutex *lock)
77 return (struct task_struct *)(atomic_long_read(&lock->owner) & ~MUTEX_FLAGS);
80 static inline struct task_struct *__owner_task(unsigned long owner)
82 return (struct task_struct *)(owner & ~MUTEX_FLAGS);
85 bool mutex_is_locked(struct mutex *lock)
87 return __mutex_owner(lock) != NULL;
89 EXPORT_SYMBOL(mutex_is_locked);
91 static inline unsigned long __owner_flags(unsigned long owner)
93 return owner & MUTEX_FLAGS;
96 static inline struct task_struct *__mutex_trylock_common(struct mutex *lock, bool handoff)
98 unsigned long owner, curr = (unsigned long)current;
100 owner = atomic_long_read(&lock->owner);
101 for (;;) { /* must loop, can race against a flag */
102 unsigned long flags = __owner_flags(owner);
103 unsigned long task = owner & ~MUTEX_FLAGS;
106 if (flags & MUTEX_FLAG_PICKUP) {
109 flags &= ~MUTEX_FLAG_PICKUP;
110 } else if (handoff) {
111 if (flags & MUTEX_FLAG_HANDOFF)
113 flags |= MUTEX_FLAG_HANDOFF;
118 MUTEX_WARN_ON(flags & (MUTEX_FLAG_HANDOFF | MUTEX_FLAG_PICKUP));
122 if (atomic_long_try_cmpxchg_acquire(&lock->owner, &owner, task | flags)) {
129 return __owner_task(owner);
133 * Trylock or set HANDOFF
135 static inline bool __mutex_trylock_or_handoff(struct mutex *lock, bool handoff)
137 return !__mutex_trylock_common(lock, handoff);
141 * Actual trylock that will work on any unlocked state.
143 static inline bool __mutex_trylock(struct mutex *lock)
145 return !__mutex_trylock_common(lock, false);
148 #ifndef CONFIG_DEBUG_LOCK_ALLOC
150 * Lockdep annotations are contained to the slow paths for simplicity.
151 * There is nothing that would stop spreading the lockdep annotations outwards
156 * Optimistic trylock that only works in the uncontended case. Make sure to
157 * follow with a __mutex_trylock() before failing.
159 static __always_inline bool __mutex_trylock_fast(struct mutex *lock)
161 unsigned long curr = (unsigned long)current;
162 unsigned long zero = 0UL;
164 if (atomic_long_try_cmpxchg_acquire(&lock->owner, &zero, curr))
170 static __always_inline bool __mutex_unlock_fast(struct mutex *lock)
172 unsigned long curr = (unsigned long)current;
174 return atomic_long_try_cmpxchg_release(&lock->owner, &curr, 0UL);
178 static inline void __mutex_set_flag(struct mutex *lock, unsigned long flag)
180 atomic_long_or(flag, &lock->owner);
183 static inline void __mutex_clear_flag(struct mutex *lock, unsigned long flag)
185 atomic_long_andnot(flag, &lock->owner);
188 static inline bool __mutex_waiter_is_first(struct mutex *lock, struct mutex_waiter *waiter)
190 return list_first_entry(&lock->wait_list, struct mutex_waiter, list) == waiter;
194 * Add @waiter to a given location in the lock wait_list and set the
195 * FLAG_WAITERS flag if it's the first waiter.
198 __mutex_add_waiter(struct mutex *lock, struct mutex_waiter *waiter,
199 struct list_head *list)
201 debug_mutex_add_waiter(lock, waiter, current);
203 list_add_tail(&waiter->list, list);
204 if (__mutex_waiter_is_first(lock, waiter))
205 __mutex_set_flag(lock, MUTEX_FLAG_WAITERS);
209 __mutex_remove_waiter(struct mutex *lock, struct mutex_waiter *waiter)
211 list_del(&waiter->list);
212 if (likely(list_empty(&lock->wait_list)))
213 __mutex_clear_flag(lock, MUTEX_FLAGS);
215 debug_mutex_remove_waiter(lock, waiter, current);
219 * Give up ownership to a specific task, when @task = NULL, this is equivalent
220 * to a regular unlock. Sets PICKUP on a handoff, clears HANDOFF, preserves
221 * WAITERS. Provides RELEASE semantics like a regular unlock, the
222 * __mutex_trylock() provides a matching ACQUIRE semantics for the handoff.
224 static void __mutex_handoff(struct mutex *lock, struct task_struct *task)
226 unsigned long owner = atomic_long_read(&lock->owner);
231 MUTEX_WARN_ON(__owner_task(owner) != current);
232 MUTEX_WARN_ON(owner & MUTEX_FLAG_PICKUP);
234 new = (owner & MUTEX_FLAG_WAITERS);
235 new |= (unsigned long)task;
237 new |= MUTEX_FLAG_PICKUP;
239 if (atomic_long_try_cmpxchg_release(&lock->owner, &owner, new))
244 #ifndef CONFIG_DEBUG_LOCK_ALLOC
246 * We split the mutex lock/unlock logic into separate fastpath and
247 * slowpath functions, to reduce the register pressure on the fastpath.
248 * We also put the fastpath first in the kernel image, to make sure the
249 * branch is predicted by the CPU as default-untaken.
251 static void __sched __mutex_lock_slowpath(struct mutex *lock);
254 * mutex_lock - acquire the mutex
255 * @lock: the mutex to be acquired
257 * Lock the mutex exclusively for this task. If the mutex is not
258 * available right now, it will sleep until it can get it.
260 * The mutex must later on be released by the same task that
261 * acquired it. Recursive locking is not allowed. The task
262 * may not exit without first unlocking the mutex. Also, kernel
263 * memory where the mutex resides must not be freed with
264 * the mutex still locked. The mutex must first be initialized
265 * (or statically defined) before it can be locked. memset()-ing
266 * the mutex to 0 is not allowed.
268 * (The CONFIG_DEBUG_MUTEXES .config option turns on debugging
269 * checks that will enforce the restrictions and will also do
270 * deadlock debugging)
272 * This function is similar to (but not equivalent to) down().
274 void __sched mutex_lock(struct mutex *lock)
278 if (!__mutex_trylock_fast(lock))
279 __mutex_lock_slowpath(lock);
281 EXPORT_SYMBOL(mutex_lock);
286 * The newer transactions are killed when:
287 * It (the new transaction) makes a request for a lock being held
288 * by an older transaction.
291 * The newer transactions are wounded when:
292 * An older transaction makes a request for a lock being held by
293 * the newer transaction.
297 * Associate the ww_mutex @ww with the context @ww_ctx under which we acquired
300 static __always_inline void
301 ww_mutex_lock_acquired(struct ww_mutex *ww, struct ww_acquire_ctx *ww_ctx)
303 #ifdef CONFIG_DEBUG_MUTEXES
305 * If this WARN_ON triggers, you used ww_mutex_lock to acquire,
306 * but released with a normal mutex_unlock in this call.
308 * This should never happen, always use ww_mutex_unlock.
310 DEBUG_LOCKS_WARN_ON(ww->ctx);
313 * Not quite done after calling ww_acquire_done() ?
315 DEBUG_LOCKS_WARN_ON(ww_ctx->done_acquire);
317 if (ww_ctx->contending_lock) {
319 * After -EDEADLK you tried to
320 * acquire a different ww_mutex? Bad!
322 DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock != ww);
325 * You called ww_mutex_lock after receiving -EDEADLK,
326 * but 'forgot' to unlock everything else first?
328 DEBUG_LOCKS_WARN_ON(ww_ctx->acquired > 0);
329 ww_ctx->contending_lock = NULL;
333 * Naughty, using a different class will lead to undefined behavior!
335 DEBUG_LOCKS_WARN_ON(ww_ctx->ww_class != ww->ww_class);
342 * Determine if context @a is 'after' context @b. IOW, @a is a younger
343 * transaction than @b and depending on algorithm either needs to wait for
346 static inline bool __sched
347 __ww_ctx_stamp_after(struct ww_acquire_ctx *a, struct ww_acquire_ctx *b)
350 return (signed long)(a->stamp - b->stamp) > 0;
354 * Wait-Die; wake a younger waiter context (when locks held) such that it can
357 * Among waiters with context, only the first one can have other locks acquired
358 * already (ctx->acquired > 0), because __ww_mutex_add_waiter() and
359 * __ww_mutex_check_kill() wake any but the earliest context.
362 __ww_mutex_die(struct mutex *lock, struct mutex_waiter *waiter,
363 struct ww_acquire_ctx *ww_ctx)
365 if (!ww_ctx->is_wait_die)
368 if (waiter->ww_ctx->acquired > 0 &&
369 __ww_ctx_stamp_after(waiter->ww_ctx, ww_ctx)) {
370 debug_mutex_wake_waiter(lock, waiter);
371 wake_up_process(waiter->task);
378 * Wound-Wait; wound a younger @hold_ctx if it holds the lock.
380 * Wound the lock holder if there are waiters with older transactions than
381 * the lock holders. Even if multiple waiters may wound the lock holder,
382 * it's sufficient that only one does.
384 static bool __ww_mutex_wound(struct mutex *lock,
385 struct ww_acquire_ctx *ww_ctx,
386 struct ww_acquire_ctx *hold_ctx)
388 struct task_struct *owner = __mutex_owner(lock);
390 lockdep_assert_held(&lock->wait_lock);
393 * Possible through __ww_mutex_add_waiter() when we race with
394 * ww_mutex_set_context_fastpath(). In that case we'll get here again
395 * through __ww_mutex_check_waiters().
401 * Can have !owner because of __mutex_unlock_slowpath(), but if owner,
402 * it cannot go away because we'll have FLAG_WAITERS set and hold
408 if (ww_ctx->acquired > 0 && __ww_ctx_stamp_after(hold_ctx, ww_ctx)) {
409 hold_ctx->wounded = 1;
412 * wake_up_process() paired with set_current_state()
413 * inserts sufficient barriers to make sure @owner either sees
414 * it's wounded in __ww_mutex_check_kill() or has a
415 * wakeup pending to re-read the wounded state.
417 if (owner != current)
418 wake_up_process(owner);
427 * We just acquired @lock under @ww_ctx, if there are later contexts waiting
428 * behind us on the wait-list, check if they need to die, or wound us.
430 * See __ww_mutex_add_waiter() for the list-order construction; basically the
431 * list is ordered by stamp, smallest (oldest) first.
433 * This relies on never mixing wait-die/wound-wait on the same wait-list;
434 * which is currently ensured by that being a ww_class property.
436 * The current task must not be on the wait list.
439 __ww_mutex_check_waiters(struct mutex *lock, struct ww_acquire_ctx *ww_ctx)
441 struct mutex_waiter *cur;
443 lockdep_assert_held(&lock->wait_lock);
445 list_for_each_entry(cur, &lock->wait_list, list) {
449 if (__ww_mutex_die(lock, cur, ww_ctx) ||
450 __ww_mutex_wound(lock, cur->ww_ctx, ww_ctx))
456 * After acquiring lock with fastpath, where we do not hold wait_lock, set ctx
457 * and wake up any waiters so they can recheck.
459 static __always_inline void
460 ww_mutex_set_context_fastpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
462 ww_mutex_lock_acquired(lock, ctx);
465 * The lock->ctx update should be visible on all cores before
466 * the WAITERS check is done, otherwise contended waiters might be
467 * missed. The contended waiters will either see ww_ctx == NULL
468 * and keep spinning, or it will acquire wait_lock, add itself
469 * to waiter list and sleep.
471 smp_mb(); /* See comments above and below. */
474 * [W] ww->ctx = ctx [W] MUTEX_FLAG_WAITERS
476 * [R] MUTEX_FLAG_WAITERS [R] ww->ctx
478 * The memory barrier above pairs with the memory barrier in
479 * __ww_mutex_add_waiter() and makes sure we either observe ww->ctx
480 * and/or !empty list.
482 if (likely(!(atomic_long_read(&lock->base.owner) & MUTEX_FLAG_WAITERS)))
486 * Uh oh, we raced in fastpath, check if any of the waiters need to
489 spin_lock(&lock->base.wait_lock);
490 __ww_mutex_check_waiters(&lock->base, ctx);
491 spin_unlock(&lock->base.wait_lock);
494 #ifdef CONFIG_MUTEX_SPIN_ON_OWNER
497 * Trylock variant that returns the owning task on failure.
499 static inline struct task_struct *__mutex_trylock_or_owner(struct mutex *lock)
501 return __mutex_trylock_common(lock, false);
505 bool ww_mutex_spin_on_owner(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
506 struct mutex_waiter *waiter)
510 ww = container_of(lock, struct ww_mutex, base);
513 * If ww->ctx is set the contents are undefined, only
514 * by acquiring wait_lock there is a guarantee that
515 * they are not invalid when reading.
517 * As such, when deadlock detection needs to be
518 * performed the optimistic spinning cannot be done.
520 * Check this in every inner iteration because we may
521 * be racing against another thread's ww_mutex_lock.
523 if (ww_ctx->acquired > 0 && READ_ONCE(ww->ctx))
527 * If we aren't on the wait list yet, cancel the spin
528 * if there are waiters. We want to avoid stealing the
529 * lock from a waiter with an earlier stamp, since the
530 * other thread may already own a lock that we also
533 if (!waiter && (atomic_long_read(&lock->owner) & MUTEX_FLAG_WAITERS))
537 * Similarly, stop spinning if we are no longer the
540 if (waiter && !__mutex_waiter_is_first(lock, waiter))
547 * Look out! "owner" is an entirely speculative pointer access and not
550 * "noinline" so that this function shows up on perf profiles.
553 bool mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner,
554 struct ww_acquire_ctx *ww_ctx, struct mutex_waiter *waiter)
559 while (__mutex_owner(lock) == owner) {
561 * Ensure we emit the owner->on_cpu, dereference _after_
562 * checking lock->owner still matches owner. If that fails,
563 * owner might point to freed memory. If it still matches,
564 * the rcu_read_lock() ensures the memory stays valid.
569 * Use vcpu_is_preempted to detect lock holder preemption issue.
571 if (!owner->on_cpu || need_resched() ||
572 vcpu_is_preempted(task_cpu(owner))) {
577 if (ww_ctx && !ww_mutex_spin_on_owner(lock, ww_ctx, waiter)) {
590 * Initial check for entering the mutex spinning loop
592 static inline int mutex_can_spin_on_owner(struct mutex *lock)
594 struct task_struct *owner;
601 owner = __mutex_owner(lock);
604 * As lock holder preemption issue, we both skip spinning if task is not
605 * on cpu or its cpu is preempted
608 retval = owner->on_cpu && !vcpu_is_preempted(task_cpu(owner));
612 * If lock->owner is not set, the mutex has been released. Return true
613 * such that we'll trylock in the spin path, which is a faster option
614 * than the blocking slow path.
620 * Optimistic spinning.
622 * We try to spin for acquisition when we find that the lock owner
623 * is currently running on a (different) CPU and while we don't
624 * need to reschedule. The rationale is that if the lock owner is
625 * running, it is likely to release the lock soon.
627 * The mutex spinners are queued up using MCS lock so that only one
628 * spinner can compete for the mutex. However, if mutex spinning isn't
629 * going to happen, there is no point in going through the lock/unlock
632 * Returns true when the lock was taken, otherwise false, indicating
633 * that we need to jump to the slowpath and sleep.
635 * The waiter flag is set to true if the spinner is a waiter in the wait
636 * queue. The waiter-spinner will spin on the lock directly and concurrently
637 * with the spinner at the head of the OSQ, if present, until the owner is
640 static __always_inline bool
641 mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
642 struct mutex_waiter *waiter)
646 * The purpose of the mutex_can_spin_on_owner() function is
647 * to eliminate the overhead of osq_lock() and osq_unlock()
648 * in case spinning isn't possible. As a waiter-spinner
649 * is not going to take OSQ lock anyway, there is no need
650 * to call mutex_can_spin_on_owner().
652 if (!mutex_can_spin_on_owner(lock))
656 * In order to avoid a stampede of mutex spinners trying to
657 * acquire the mutex all at once, the spinners need to take a
658 * MCS (queued) lock first before spinning on the owner field.
660 if (!osq_lock(&lock->osq))
665 struct task_struct *owner;
667 /* Try to acquire the mutex... */
668 owner = __mutex_trylock_or_owner(lock);
673 * There's an owner, wait for it to either
674 * release the lock or go to sleep.
676 if (!mutex_spin_on_owner(lock, owner, ww_ctx, waiter))
680 * The cpu_relax() call is a compiler barrier which forces
681 * everything in this loop to be re-loaded. We don't need
682 * memory barriers as we'll eventually observe the right
683 * values at the cost of a few extra spins.
689 osq_unlock(&lock->osq);
696 osq_unlock(&lock->osq);
700 * If we fell out of the spin path because of need_resched(),
701 * reschedule now, before we try-lock the mutex. This avoids getting
702 * scheduled out right after we obtained the mutex.
704 if (need_resched()) {
706 * We _should_ have TASK_RUNNING here, but just in case
707 * we do not, make it so, otherwise we might get stuck.
709 __set_current_state(TASK_RUNNING);
710 schedule_preempt_disabled();
716 static __always_inline bool
717 mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
718 struct mutex_waiter *waiter)
724 static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip);
727 * mutex_unlock - release the mutex
728 * @lock: the mutex to be released
730 * Unlock a mutex that has been locked by this task previously.
732 * This function must not be used in interrupt context. Unlocking
733 * of a not locked mutex is not allowed.
735 * This function is similar to (but not equivalent to) up().
737 void __sched mutex_unlock(struct mutex *lock)
739 #ifndef CONFIG_DEBUG_LOCK_ALLOC
740 if (__mutex_unlock_fast(lock))
743 __mutex_unlock_slowpath(lock, _RET_IP_);
745 EXPORT_SYMBOL(mutex_unlock);
748 * ww_mutex_unlock - release the w/w mutex
749 * @lock: the mutex to be released
751 * Unlock a mutex that has been locked by this task previously with any of the
752 * ww_mutex_lock* functions (with or without an acquire context). It is
753 * forbidden to release the locks after releasing the acquire context.
755 * This function must not be used in interrupt context. Unlocking
756 * of a unlocked mutex is not allowed.
758 void __sched ww_mutex_unlock(struct ww_mutex *lock)
761 * The unlocking fastpath is the 0->1 transition from 'locked'
762 * into 'unlocked' state:
765 MUTEX_WARN_ON(!lock->ctx->acquired);
766 if (lock->ctx->acquired > 0)
767 lock->ctx->acquired--;
771 mutex_unlock(&lock->base);
773 EXPORT_SYMBOL(ww_mutex_unlock);
776 static __always_inline int __sched
777 __ww_mutex_kill(struct mutex *lock, struct ww_acquire_ctx *ww_ctx)
779 if (ww_ctx->acquired > 0) {
780 #ifdef CONFIG_DEBUG_MUTEXES
783 ww = container_of(lock, struct ww_mutex, base);
784 DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock);
785 ww_ctx->contending_lock = ww;
795 * Check the wound condition for the current lock acquire.
797 * Wound-Wait: If we're wounded, kill ourself.
799 * Wait-Die: If we're trying to acquire a lock already held by an older
800 * context, kill ourselves.
802 * Since __ww_mutex_add_waiter() orders the wait-list on stamp, we only have to
803 * look at waiters before us in the wait-list.
805 static inline int __sched
806 __ww_mutex_check_kill(struct mutex *lock, struct mutex_waiter *waiter,
807 struct ww_acquire_ctx *ctx)
809 struct ww_mutex *ww = container_of(lock, struct ww_mutex, base);
810 struct ww_acquire_ctx *hold_ctx = READ_ONCE(ww->ctx);
811 struct mutex_waiter *cur;
813 if (ctx->acquired == 0)
816 if (!ctx->is_wait_die) {
818 return __ww_mutex_kill(lock, ctx);
823 if (hold_ctx && __ww_ctx_stamp_after(ctx, hold_ctx))
824 return __ww_mutex_kill(lock, ctx);
827 * If there is a waiter in front of us that has a context, then its
828 * stamp is earlier than ours and we must kill ourself.
831 list_for_each_entry_continue_reverse(cur, &lock->wait_list, list) {
835 return __ww_mutex_kill(lock, ctx);
842 * Add @waiter to the wait-list, keep the wait-list ordered by stamp, smallest
843 * first. Such that older contexts are preferred to acquire the lock over
846 * Waiters without context are interspersed in FIFO order.
848 * Furthermore, for Wait-Die kill ourself immediately when possible (there are
849 * older contexts already waiting) to avoid unnecessary waiting and for
850 * Wound-Wait ensure we wound the owning context when it is younger.
852 static inline int __sched
853 __ww_mutex_add_waiter(struct mutex_waiter *waiter,
855 struct ww_acquire_ctx *ww_ctx)
857 struct mutex_waiter *cur;
858 struct list_head *pos;
862 __mutex_add_waiter(lock, waiter, &lock->wait_list);
866 is_wait_die = ww_ctx->is_wait_die;
869 * Add the waiter before the first waiter with a higher stamp.
870 * Waiters without a context are skipped to avoid starving
871 * them. Wait-Die waiters may die here. Wound-Wait waiters
872 * never die here, but they are sorted in stamp order and
873 * may wound the lock holder.
875 pos = &lock->wait_list;
876 list_for_each_entry_reverse(cur, &lock->wait_list, list) {
880 if (__ww_ctx_stamp_after(ww_ctx, cur->ww_ctx)) {
882 * Wait-Die: if we find an older context waiting, there
883 * is no point in queueing behind it, as we'd have to
884 * die the moment it would acquire the lock.
887 int ret = __ww_mutex_kill(lock, ww_ctx);
898 /* Wait-Die: ensure younger waiters die. */
899 __ww_mutex_die(lock, cur, ww_ctx);
902 __mutex_add_waiter(lock, waiter, pos);
905 * Wound-Wait: if we're blocking on a mutex owned by a younger context,
906 * wound that such that we might proceed.
909 struct ww_mutex *ww = container_of(lock, struct ww_mutex, base);
912 * See ww_mutex_set_context_fastpath(). Orders setting
913 * MUTEX_FLAG_WAITERS vs the ww->ctx load,
914 * such that either we or the fastpath will wound @ww->ctx.
917 __ww_mutex_wound(lock, ww_ctx, ww->ctx);
924 * Lock a mutex (possibly interruptible), slowpath:
926 static __always_inline int __sched
927 __mutex_lock_common(struct mutex *lock, unsigned int state, unsigned int subclass,
928 struct lockdep_map *nest_lock, unsigned long ip,
929 struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
931 struct mutex_waiter waiter;
940 MUTEX_WARN_ON(lock->magic != lock);
942 ww = container_of(lock, struct ww_mutex, base);
944 if (unlikely(ww_ctx == READ_ONCE(ww->ctx)))
948 * Reset the wounded flag after a kill. No other process can
949 * race and wound us here since they can't have a valid owner
950 * pointer if we don't have any locks held.
952 if (ww_ctx->acquired == 0)
957 mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);
959 if (__mutex_trylock(lock) ||
960 mutex_optimistic_spin(lock, ww_ctx, NULL)) {
961 /* got the lock, yay! */
962 lock_acquired(&lock->dep_map, ip);
964 ww_mutex_set_context_fastpath(ww, ww_ctx);
969 spin_lock(&lock->wait_lock);
971 * After waiting to acquire the wait_lock, try again.
973 if (__mutex_trylock(lock)) {
975 __ww_mutex_check_waiters(lock, ww_ctx);
980 debug_mutex_lock_common(lock, &waiter);
982 lock_contended(&lock->dep_map, ip);
985 /* add waiting tasks to the end of the waitqueue (FIFO): */
986 __mutex_add_waiter(lock, &waiter, &lock->wait_list);
989 #ifdef CONFIG_DEBUG_MUTEXES
990 waiter.ww_ctx = MUTEX_POISON_WW_CTX;
994 * Add in stamp order, waking up waiters that must kill
997 ret = __ww_mutex_add_waiter(&waiter, lock, ww_ctx);
1001 waiter.ww_ctx = ww_ctx;
1004 waiter.task = current;
1006 set_current_state(state);
1011 * Once we hold wait_lock, we're serialized against
1012 * mutex_unlock() handing the lock off to us, do a trylock
1013 * before testing the error conditions to make sure we pick up
1016 if (__mutex_trylock(lock))
1020 * Check for signals and kill conditions while holding
1021 * wait_lock. This ensures the lock cancellation is ordered
1022 * against mutex_unlock() and wake-ups do not go missing.
1024 if (signal_pending_state(state, current)) {
1030 ret = __ww_mutex_check_kill(lock, &waiter, ww_ctx);
1035 spin_unlock(&lock->wait_lock);
1036 schedule_preempt_disabled();
1038 first = __mutex_waiter_is_first(lock, &waiter);
1040 set_current_state(state);
1042 * Here we order against unlock; we must either see it change
1043 * state back to RUNNING and fall through the next schedule(),
1044 * or we must see its unlock and acquire.
1046 if (__mutex_trylock_or_handoff(lock, first) ||
1047 (first && mutex_optimistic_spin(lock, ww_ctx, &waiter)))
1050 spin_lock(&lock->wait_lock);
1052 spin_lock(&lock->wait_lock);
1054 __set_current_state(TASK_RUNNING);
1058 * Wound-Wait; we stole the lock (!first_waiter), check the
1059 * waiters as anyone might want to wound us.
1061 if (!ww_ctx->is_wait_die &&
1062 !__mutex_waiter_is_first(lock, &waiter))
1063 __ww_mutex_check_waiters(lock, ww_ctx);
1066 __mutex_remove_waiter(lock, &waiter);
1068 debug_mutex_free_waiter(&waiter);
1071 /* got the lock - cleanup and rejoice! */
1072 lock_acquired(&lock->dep_map, ip);
1075 ww_mutex_lock_acquired(ww, ww_ctx);
1077 spin_unlock(&lock->wait_lock);
1082 __set_current_state(TASK_RUNNING);
1083 __mutex_remove_waiter(lock, &waiter);
1085 spin_unlock(&lock->wait_lock);
1086 debug_mutex_free_waiter(&waiter);
1087 mutex_release(&lock->dep_map, ip);
1093 __mutex_lock(struct mutex *lock, unsigned int state, unsigned int subclass,
1094 struct lockdep_map *nest_lock, unsigned long ip)
1096 return __mutex_lock_common(lock, state, subclass, nest_lock, ip, NULL, false);
1100 __ww_mutex_lock(struct mutex *lock, unsigned int state, unsigned int subclass,
1101 struct lockdep_map *nest_lock, unsigned long ip,
1102 struct ww_acquire_ctx *ww_ctx)
1104 return __mutex_lock_common(lock, state, subclass, nest_lock, ip, ww_ctx, true);
1107 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1109 mutex_lock_nested(struct mutex *lock, unsigned int subclass)
1111 __mutex_lock(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_);
1114 EXPORT_SYMBOL_GPL(mutex_lock_nested);
1117 _mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest)
1119 __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, nest, _RET_IP_);
1121 EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock);
1124 mutex_lock_killable_nested(struct mutex *lock, unsigned int subclass)
1126 return __mutex_lock(lock, TASK_KILLABLE, subclass, NULL, _RET_IP_);
1128 EXPORT_SYMBOL_GPL(mutex_lock_killable_nested);
1131 mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
1133 return __mutex_lock(lock, TASK_INTERRUPTIBLE, subclass, NULL, _RET_IP_);
1135 EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
1138 mutex_lock_io_nested(struct mutex *lock, unsigned int subclass)
1144 token = io_schedule_prepare();
1145 __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE,
1146 subclass, NULL, _RET_IP_, NULL, 0);
1147 io_schedule_finish(token);
1149 EXPORT_SYMBOL_GPL(mutex_lock_io_nested);
1152 ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1154 #ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH
1157 if (ctx->deadlock_inject_countdown-- == 0) {
1158 tmp = ctx->deadlock_inject_interval;
1159 if (tmp > UINT_MAX/4)
1162 tmp = tmp*2 + tmp + tmp/2;
1164 ctx->deadlock_inject_interval = tmp;
1165 ctx->deadlock_inject_countdown = tmp;
1166 ctx->contending_lock = lock;
1168 ww_mutex_unlock(lock);
1178 ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1183 ret = __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE,
1184 0, ctx ? &ctx->dep_map : NULL, _RET_IP_,
1186 if (!ret && ctx && ctx->acquired > 1)
1187 return ww_mutex_deadlock_injection(lock, ctx);
1191 EXPORT_SYMBOL_GPL(ww_mutex_lock);
1194 ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1199 ret = __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE,
1200 0, ctx ? &ctx->dep_map : NULL, _RET_IP_,
1203 if (!ret && ctx && ctx->acquired > 1)
1204 return ww_mutex_deadlock_injection(lock, ctx);
1208 EXPORT_SYMBOL_GPL(ww_mutex_lock_interruptible);
1213 * Release the lock, slowpath:
1215 static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip)
1217 struct task_struct *next = NULL;
1218 DEFINE_WAKE_Q(wake_q);
1219 unsigned long owner;
1221 mutex_release(&lock->dep_map, ip);
1224 * Release the lock before (potentially) taking the spinlock such that
1225 * other contenders can get on with things ASAP.
1227 * Except when HANDOFF, in that case we must not clear the owner field,
1228 * but instead set it to the top waiter.
1230 owner = atomic_long_read(&lock->owner);
1232 MUTEX_WARN_ON(__owner_task(owner) != current);
1233 MUTEX_WARN_ON(owner & MUTEX_FLAG_PICKUP);
1235 if (owner & MUTEX_FLAG_HANDOFF)
1238 if (atomic_long_try_cmpxchg_release(&lock->owner, &owner, __owner_flags(owner))) {
1239 if (owner & MUTEX_FLAG_WAITERS)
1246 spin_lock(&lock->wait_lock);
1247 debug_mutex_unlock(lock);
1248 if (!list_empty(&lock->wait_list)) {
1249 /* get the first entry from the wait-list: */
1250 struct mutex_waiter *waiter =
1251 list_first_entry(&lock->wait_list,
1252 struct mutex_waiter, list);
1254 next = waiter->task;
1256 debug_mutex_wake_waiter(lock, waiter);
1257 wake_q_add(&wake_q, next);
1260 if (owner & MUTEX_FLAG_HANDOFF)
1261 __mutex_handoff(lock, next);
1263 spin_unlock(&lock->wait_lock);
1268 #ifndef CONFIG_DEBUG_LOCK_ALLOC
1270 * Here come the less common (and hence less performance-critical) APIs:
1271 * mutex_lock_interruptible() and mutex_trylock().
1273 static noinline int __sched
1274 __mutex_lock_killable_slowpath(struct mutex *lock);
1276 static noinline int __sched
1277 __mutex_lock_interruptible_slowpath(struct mutex *lock);
1280 * mutex_lock_interruptible() - Acquire the mutex, interruptible by signals.
1281 * @lock: The mutex to be acquired.
1283 * Lock the mutex like mutex_lock(). If a signal is delivered while the
1284 * process is sleeping, this function will return without acquiring the
1287 * Context: Process context.
1288 * Return: 0 if the lock was successfully acquired or %-EINTR if a
1291 int __sched mutex_lock_interruptible(struct mutex *lock)
1295 if (__mutex_trylock_fast(lock))
1298 return __mutex_lock_interruptible_slowpath(lock);
1301 EXPORT_SYMBOL(mutex_lock_interruptible);
1304 * mutex_lock_killable() - Acquire the mutex, interruptible by fatal signals.
1305 * @lock: The mutex to be acquired.
1307 * Lock the mutex like mutex_lock(). If a signal which will be fatal to
1308 * the current process is delivered while the process is sleeping, this
1309 * function will return without acquiring the mutex.
1311 * Context: Process context.
1312 * Return: 0 if the lock was successfully acquired or %-EINTR if a
1313 * fatal signal arrived.
1315 int __sched mutex_lock_killable(struct mutex *lock)
1319 if (__mutex_trylock_fast(lock))
1322 return __mutex_lock_killable_slowpath(lock);
1324 EXPORT_SYMBOL(mutex_lock_killable);
1327 * mutex_lock_io() - Acquire the mutex and mark the process as waiting for I/O
1328 * @lock: The mutex to be acquired.
1330 * Lock the mutex like mutex_lock(). While the task is waiting for this
1331 * mutex, it will be accounted as being in the IO wait state by the
1334 * Context: Process context.
1336 void __sched mutex_lock_io(struct mutex *lock)
1340 token = io_schedule_prepare();
1342 io_schedule_finish(token);
1344 EXPORT_SYMBOL_GPL(mutex_lock_io);
1346 static noinline void __sched
1347 __mutex_lock_slowpath(struct mutex *lock)
1349 __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_);
1352 static noinline int __sched
1353 __mutex_lock_killable_slowpath(struct mutex *lock)
1355 return __mutex_lock(lock, TASK_KILLABLE, 0, NULL, _RET_IP_);
1358 static noinline int __sched
1359 __mutex_lock_interruptible_slowpath(struct mutex *lock)
1361 return __mutex_lock(lock, TASK_INTERRUPTIBLE, 0, NULL, _RET_IP_);
1364 static noinline int __sched
1365 __ww_mutex_lock_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1367 return __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE, 0, NULL,
1371 static noinline int __sched
1372 __ww_mutex_lock_interruptible_slowpath(struct ww_mutex *lock,
1373 struct ww_acquire_ctx *ctx)
1375 return __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE, 0, NULL,
1382 * mutex_trylock - try to acquire the mutex, without waiting
1383 * @lock: the mutex to be acquired
1385 * Try to acquire the mutex atomically. Returns 1 if the mutex
1386 * has been acquired successfully, and 0 on contention.
1388 * NOTE: this function follows the spin_trylock() convention, so
1389 * it is negated from the down_trylock() return values! Be careful
1390 * about this when converting semaphore users to mutexes.
1392 * This function must not be used in interrupt context. The
1393 * mutex must be released by the same task that acquired it.
1395 int __sched mutex_trylock(struct mutex *lock)
1399 MUTEX_WARN_ON(lock->magic != lock);
1401 locked = __mutex_trylock(lock);
1403 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
1407 EXPORT_SYMBOL(mutex_trylock);
1409 #ifndef CONFIG_DEBUG_LOCK_ALLOC
1411 ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1415 if (__mutex_trylock_fast(&lock->base)) {
1417 ww_mutex_set_context_fastpath(lock, ctx);
1421 return __ww_mutex_lock_slowpath(lock, ctx);
1423 EXPORT_SYMBOL(ww_mutex_lock);
1426 ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1430 if (__mutex_trylock_fast(&lock->base)) {
1432 ww_mutex_set_context_fastpath(lock, ctx);
1436 return __ww_mutex_lock_interruptible_slowpath(lock, ctx);
1438 EXPORT_SYMBOL(ww_mutex_lock_interruptible);
1443 * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0
1444 * @cnt: the atomic which we are to dec
1445 * @lock: the mutex to return holding if we dec to 0
1447 * return true and hold lock if we dec to 0, return false otherwise
1449 int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock)
1451 /* dec if we can't possibly hit 0 */
1452 if (atomic_add_unless(cnt, -1, 1))
1454 /* we might hit 0, so take the lock */
1456 if (!atomic_dec_and_test(cnt)) {
1457 /* when we actually did the dec, we didn't hit 0 */
1461 /* we hit 0, and we hold the lock */
1464 EXPORT_SYMBOL(atomic_dec_and_mutex_lock);