1/*
2 * kernel/locking/mutex.c
3 *
4 * Mutexes: blocking mutual exclusion locks
5 *
6 * Started by Ingo Molnar:
7 *
8 *  Copyright (C) 2004, 2005, 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
9 *
10 * Many thanks to Arjan van de Ven, Thomas Gleixner, Steven Rostedt and
11 * David Howells for suggestions and improvements.
12 *
13 *  - Adaptive spinning for mutexes by Peter Zijlstra. (Ported to mainline
14 *    from the -rt tree, where it was originally implemented for rtmutexes
15 *    by Steven Rostedt, based on work by Gregory Haskins, Peter Morreale
16 *    and Sven Dietrich.
17 *
18 * Also see Documentation/locking/mutex-design.txt.
19 */
20#include <linux/mutex.h>
21#include <linux/ww_mutex.h>
22#include <linux/sched.h>
23#include <linux/sched/rt.h>
24#include <linux/export.h>
25#include <linux/spinlock.h>
26#include <linux/interrupt.h>
27#include <linux/debug_locks.h>
28#include "mcs_spinlock.h"
29
30/*
31 * In the DEBUG case we are using the "NULL fastpath" for mutexes,
32 * which forces all calls into the slowpath:
33 */
34#ifdef CONFIG_DEBUG_MUTEXES
35# include "mutex-debug.h"
36# include <asm-generic/mutex-null.h>
37/*
38 * Must be 0 for the debug case so we do not do the unlock outside of the
39 * wait_lock region. debug_mutex_unlock() will do the actual unlock in this
40 * case.
41 */
42# undef __mutex_slowpath_needs_to_unlock
43# define  __mutex_slowpath_needs_to_unlock()	0
44#else
45# include "mutex.h"
46# include <asm/mutex.h>
47#endif
48
49void
50__mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
51{
52	atomic_set(&lock->count, 1);
53	spin_lock_init(&lock->wait_lock);
54	INIT_LIST_HEAD(&lock->wait_list);
55	mutex_clear_owner(lock);
56#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
57	osq_lock_init(&lock->osq);
58#endif
59
60	debug_mutex_init(lock, name, key);
61}
62
63EXPORT_SYMBOL(__mutex_init);
64
65#ifndef CONFIG_DEBUG_LOCK_ALLOC
66/*
67 * We split the mutex lock/unlock logic into separate fastpath and
68 * slowpath functions, to reduce the register pressure on the fastpath.
69 * We also put the fastpath first in the kernel image, to make sure the
70 * branch is predicted by the CPU as default-untaken.
71 */
72__visible void __sched __mutex_lock_slowpath(atomic_t *lock_count);
73
74/**
75 * mutex_lock - acquire the mutex
76 * @lock: the mutex to be acquired
77 *
78 * Lock the mutex exclusively for this task. If the mutex is not
79 * available right now, it will sleep until it can get it.
80 *
81 * The mutex must later on be released by the same task that
82 * acquired it. Recursive locking is not allowed. The task
83 * may not exit without first unlocking the mutex. Also, kernel
84 * memory where the mutex resides mutex must not be freed with
85 * the mutex still locked. The mutex must first be initialized
86 * (or statically defined) before it can be locked. memset()-ing
87 * the mutex to 0 is not allowed.
88 *
89 * ( The CONFIG_DEBUG_MUTEXES .config option turns on debugging
90 *   checks that will enforce the restrictions and will also do
91 *   deadlock debugging. )
92 *
93 * This function is similar to (but not equivalent to) down().
94 */
95void __sched mutex_lock(struct mutex *lock)
96{
97	might_sleep();
98	/*
99	 * The locking fastpath is the 1->0 transition from
100	 * 'unlocked' into 'locked' state.
101	 */
102	__mutex_fastpath_lock(&lock->count, __mutex_lock_slowpath);
103	mutex_set_owner(lock);
104}
105
106EXPORT_SYMBOL(mutex_lock);
107#endif
108
109static __always_inline void ww_mutex_lock_acquired(struct ww_mutex *ww,
110						   struct ww_acquire_ctx *ww_ctx)
111{
112#ifdef CONFIG_DEBUG_MUTEXES
113	/*
114	 * If this WARN_ON triggers, you used ww_mutex_lock to acquire,
115	 * but released with a normal mutex_unlock in this call.
116	 *
117	 * This should never happen, always use ww_mutex_unlock.
118	 */
119	DEBUG_LOCKS_WARN_ON(ww->ctx);
120
121	/*
122	 * Not quite done after calling ww_acquire_done() ?
123	 */
124	DEBUG_LOCKS_WARN_ON(ww_ctx->done_acquire);
125
126	if (ww_ctx->contending_lock) {
127		/*
128		 * After -EDEADLK you tried to
129		 * acquire a different ww_mutex? Bad!
130		 */
131		DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock != ww);
132
133		/*
134		 * You called ww_mutex_lock after receiving -EDEADLK,
135		 * but 'forgot' to unlock everything else first?
136		 */
137		DEBUG_LOCKS_WARN_ON(ww_ctx->acquired > 0);
138		ww_ctx->contending_lock = NULL;
139	}
140
141	/*
142	 * Naughty, using a different class will lead to undefined behavior!
143	 */
144	DEBUG_LOCKS_WARN_ON(ww_ctx->ww_class != ww->ww_class);
145#endif
146	ww_ctx->acquired++;
147}
148
149/*
150 * after acquiring lock with fastpath or when we lost out in contested
151 * slowpath, set ctx and wake up any waiters so they can recheck.
152 *
153 * This function is never called when CONFIG_DEBUG_LOCK_ALLOC is set,
154 * as the fastpath and opportunistic spinning are disabled in that case.
155 */
156static __always_inline void
157ww_mutex_set_context_fastpath(struct ww_mutex *lock,
158			       struct ww_acquire_ctx *ctx)
159{
160	unsigned long flags;
161	struct mutex_waiter *cur;
162
163	ww_mutex_lock_acquired(lock, ctx);
164
165	lock->ctx = ctx;
166
167	/*
168	 * The lock->ctx update should be visible on all cores before
169	 * the atomic read is done, otherwise contended waiters might be
170	 * missed. The contended waiters will either see ww_ctx == NULL
171	 * and keep spinning, or it will acquire wait_lock, add itself
172	 * to waiter list and sleep.
173	 */
174	smp_mb(); /* ^^^ */
175
176	/*
177	 * Check if lock is contended, if not there is nobody to wake up
178	 */
179	if (likely(atomic_read(&lock->base.count) == 0))
180		return;
181
182	/*
183	 * Uh oh, we raced in fastpath, wake up everyone in this case,
184	 * so they can see the new lock->ctx.
185	 */
186	spin_lock_mutex(&lock->base.wait_lock, flags);
187	list_for_each_entry(cur, &lock->base.wait_list, list) {
188		debug_mutex_wake_waiter(&lock->base, cur);
189		wake_up_process(cur->task);
190	}
191	spin_unlock_mutex(&lock->base.wait_lock, flags);
192}
193
194
195#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
196/*
197 * In order to avoid a stampede of mutex spinners from acquiring the mutex
198 * more or less simultaneously, the spinners need to acquire a MCS lock
199 * first before spinning on the owner field.
200 *
201 */
202
203/*
204 * Mutex spinning code migrated from kernel/sched/core.c
205 */
206
207static inline bool owner_running(struct mutex *lock, struct task_struct *owner)
208{
209	if (lock->owner != owner)
210		return false;
211
212	/*
213	 * Ensure we emit the owner->on_cpu, dereference _after_ checking
214	 * lock->owner still matches owner, if that fails, owner might
215	 * point to free()d memory, if it still matches, the rcu_read_lock()
216	 * ensures the memory stays valid.
217	 */
218	barrier();
219
220	return owner->on_cpu;
221}
222
223/*
224 * Look out! "owner" is an entirely speculative pointer
225 * access and not reliable.
226 */
227static noinline
228int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner)
229{
230	rcu_read_lock();
231	while (owner_running(lock, owner)) {
232		if (need_resched())
233			break;
234
235		cpu_relax_lowlatency();
236	}
237	rcu_read_unlock();
238
239	/*
240	 * We break out the loop above on need_resched() and when the
241	 * owner changed, which is a sign for heavy contention. Return
242	 * success only when lock->owner is NULL.
243	 */
244	return lock->owner == NULL;
245}
246
247/*
248 * Initial check for entering the mutex spinning loop
249 */
250static inline int mutex_can_spin_on_owner(struct mutex *lock)
251{
252	struct task_struct *owner;
253	int retval = 1;
254
255	if (need_resched())
256		return 0;
257
258	rcu_read_lock();
259	owner = ACCESS_ONCE(lock->owner);
260	if (owner)
261		retval = owner->on_cpu;
262	rcu_read_unlock();
263	/*
264	 * if lock->owner is not set, the mutex owner may have just acquired
265	 * it and not set the owner yet or the mutex has been released.
266	 */
267	return retval;
268}
269
270/*
271 * Atomically try to take the lock when it is available
272 */
273static inline bool mutex_try_to_acquire(struct mutex *lock)
274{
275	return !mutex_is_locked(lock) &&
276		(atomic_cmpxchg(&lock->count, 1, 0) == 1);
277}
278
279/*
280 * Optimistic spinning.
281 *
282 * We try to spin for acquisition when we find that the lock owner
283 * is currently running on a (different) CPU and while we don't
284 * need to reschedule. The rationale is that if the lock owner is
285 * running, it is likely to release the lock soon.
286 *
287 * Since this needs the lock owner, and this mutex implementation
288 * doesn't track the owner atomically in the lock field, we need to
289 * track it non-atomically.
290 *
291 * We can't do this for DEBUG_MUTEXES because that relies on wait_lock
292 * to serialize everything.
293 *
294 * The mutex spinners are queued up using MCS lock so that only one
295 * spinner can compete for the mutex. However, if mutex spinning isn't
296 * going to happen, there is no point in going through the lock/unlock
297 * overhead.
298 *
299 * Returns true when the lock was taken, otherwise false, indicating
300 * that we need to jump to the slowpath and sleep.
301 */
302static bool mutex_optimistic_spin(struct mutex *lock,
303				  struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
304{
305	struct task_struct *task = current;
306
307	if (!mutex_can_spin_on_owner(lock))
308		goto done;
309
310	if (!osq_lock(&lock->osq))
311		goto done;
312
313	while (true) {
314		struct task_struct *owner;
315
316		if (use_ww_ctx && ww_ctx->acquired > 0) {
317			struct ww_mutex *ww;
318
319			ww = container_of(lock, struct ww_mutex, base);
320			/*
321			 * If ww->ctx is set the contents are undefined, only
322			 * by acquiring wait_lock there is a guarantee that
323			 * they are not invalid when reading.
324			 *
325			 * As such, when deadlock detection needs to be
326			 * performed the optimistic spinning cannot be done.
327			 */
328			if (ACCESS_ONCE(ww->ctx))
329				break;
330		}
331
332		/*
333		 * If there's an owner, wait for it to either
334		 * release the lock or go to sleep.
335		 */
336		owner = ACCESS_ONCE(lock->owner);
337		if (owner && !mutex_spin_on_owner(lock, owner))
338			break;
339
340		/* Try to acquire the mutex if it is unlocked. */
341		if (mutex_try_to_acquire(lock)) {
342			lock_acquired(&lock->dep_map, ip);
343
344			if (use_ww_ctx) {
345				struct ww_mutex *ww;
346				ww = container_of(lock, struct ww_mutex, base);
347
348				ww_mutex_set_context_fastpath(ww, ww_ctx);
349			}
350
351			mutex_set_owner(lock);
352			osq_unlock(&lock->osq);
353			return true;
354		}
355
356		/*
357		 * When there's no owner, we might have preempted between the
358		 * owner acquiring the lock and setting the owner field. If
359		 * we're an RT task that will live-lock because we won't let
360		 * the owner complete.
361		 */
362		if (!owner && (need_resched() || rt_task(task)))
363			break;
364
365		/*
366		 * The cpu_relax() call is a compiler barrier which forces
367		 * everything in this loop to be re-loaded. We don't need
368		 * memory barriers as we'll eventually observe the right
369		 * values at the cost of a few extra spins.
370		 */
371		cpu_relax_lowlatency();
372	}
373
374	osq_unlock(&lock->osq);
375done:
376	/*
377	 * If we fell out of the spin path because of need_resched(),
378	 * reschedule now, before we try-lock the mutex. This avoids getting
379	 * scheduled out right after we obtained the mutex.
380	 */
381	if (need_resched())
382		schedule_preempt_disabled();
383
384	return false;
385}
386#else
387static bool mutex_optimistic_spin(struct mutex *lock,
388				  struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
389{
390	return false;
391}
392#endif
393
394__visible __used noinline
395void __sched __mutex_unlock_slowpath(atomic_t *lock_count);
396
397/**
398 * mutex_unlock - release the mutex
399 * @lock: the mutex to be released
400 *
401 * Unlock a mutex that has been locked by this task previously.
402 *
403 * This function must not be used in interrupt context. Unlocking
404 * of a not locked mutex is not allowed.
405 *
406 * This function is similar to (but not equivalent to) up().
407 */
408void __sched mutex_unlock(struct mutex *lock)
409{
410	/*
411	 * The unlocking fastpath is the 0->1 transition from 'locked'
412	 * into 'unlocked' state:
413	 */
414#ifndef CONFIG_DEBUG_MUTEXES
415	/*
416	 * When debugging is enabled we must not clear the owner before time,
417	 * the slow path will always be taken, and that clears the owner field
418	 * after verifying that it was indeed current.
419	 */
420	mutex_clear_owner(lock);
421#endif
422	__mutex_fastpath_unlock(&lock->count, __mutex_unlock_slowpath);
423}
424
425EXPORT_SYMBOL(mutex_unlock);
426
427/**
428 * ww_mutex_unlock - release the w/w mutex
429 * @lock: the mutex to be released
430 *
431 * Unlock a mutex that has been locked by this task previously with any of the
432 * ww_mutex_lock* functions (with or without an acquire context). It is
433 * forbidden to release the locks after releasing the acquire context.
434 *
435 * This function must not be used in interrupt context. Unlocking
436 * of a unlocked mutex is not allowed.
437 */
438void __sched ww_mutex_unlock(struct ww_mutex *lock)
439{
440	/*
441	 * The unlocking fastpath is the 0->1 transition from 'locked'
442	 * into 'unlocked' state:
443	 */
444	if (lock->ctx) {
445#ifdef CONFIG_DEBUG_MUTEXES
446		DEBUG_LOCKS_WARN_ON(!lock->ctx->acquired);
447#endif
448		if (lock->ctx->acquired > 0)
449			lock->ctx->acquired--;
450		lock->ctx = NULL;
451	}
452
453#ifndef CONFIG_DEBUG_MUTEXES
454	/*
455	 * When debugging is enabled we must not clear the owner before time,
456	 * the slow path will always be taken, and that clears the owner field
457	 * after verifying that it was indeed current.
458	 */
459	mutex_clear_owner(&lock->base);
460#endif
461	__mutex_fastpath_unlock(&lock->base.count, __mutex_unlock_slowpath);
462}
463EXPORT_SYMBOL(ww_mutex_unlock);
464
465static inline int __sched
466__mutex_lock_check_stamp(struct mutex *lock, struct ww_acquire_ctx *ctx)
467{
468	struct ww_mutex *ww = container_of(lock, struct ww_mutex, base);
469	struct ww_acquire_ctx *hold_ctx = ACCESS_ONCE(ww->ctx);
470
471	if (!hold_ctx)
472		return 0;
473
474	if (unlikely(ctx == hold_ctx))
475		return -EALREADY;
476
477	if (ctx->stamp - hold_ctx->stamp <= LONG_MAX &&
478	    (ctx->stamp != hold_ctx->stamp || ctx > hold_ctx)) {
479#ifdef CONFIG_DEBUG_MUTEXES
480		DEBUG_LOCKS_WARN_ON(ctx->contending_lock);
481		ctx->contending_lock = ww;
482#endif
483		return -EDEADLK;
484	}
485
486	return 0;
487}
488
489/*
490 * Lock a mutex (possibly interruptible), slowpath:
491 */
492static __always_inline int __sched
493__mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
494		    struct lockdep_map *nest_lock, unsigned long ip,
495		    struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
496{
497	struct task_struct *task = current;
498	struct mutex_waiter waiter;
499	unsigned long flags;
500	int ret;
501
502	preempt_disable();
503	mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);
504
505	if (mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx)) {
506		/* got the lock, yay! */
507		preempt_enable();
508		return 0;
509	}
510
511	spin_lock_mutex(&lock->wait_lock, flags);
512
513	/*
514	 * Once more, try to acquire the lock. Only try-lock the mutex if
515	 * it is unlocked to reduce unnecessary xchg() operations.
516	 */
517	if (!mutex_is_locked(lock) && (atomic_xchg(&lock->count, 0) == 1))
518		goto skip_wait;
519
520	debug_mutex_lock_common(lock, &waiter);
521	debug_mutex_add_waiter(lock, &waiter, task_thread_info(task));
522
523	/* add waiting tasks to the end of the waitqueue (FIFO): */
524	list_add_tail(&waiter.list, &lock->wait_list);
525	waiter.task = task;
526
527	lock_contended(&lock->dep_map, ip);
528
529	for (;;) {
530		/*
531		 * Lets try to take the lock again - this is needed even if
532		 * we get here for the first time (shortly after failing to
533		 * acquire the lock), to make sure that we get a wakeup once
534		 * it's unlocked. Later on, if we sleep, this is the
535		 * operation that gives us the lock. We xchg it to -1, so
536		 * that when we release the lock, we properly wake up the
537		 * other waiters. We only attempt the xchg if the count is
538		 * non-negative in order to avoid unnecessary xchg operations:
539		 */
540		if (atomic_read(&lock->count) >= 0 &&
541		    (atomic_xchg(&lock->count, -1) == 1))
542			break;
543
544		/*
545		 * got a signal? (This code gets eliminated in the
546		 * TASK_UNINTERRUPTIBLE case.)
547		 */
548		if (unlikely(signal_pending_state(state, task))) {
549			ret = -EINTR;
550			goto err;
551		}
552
553		if (use_ww_ctx && ww_ctx->acquired > 0) {
554			ret = __mutex_lock_check_stamp(lock, ww_ctx);
555			if (ret)
556				goto err;
557		}
558
559		__set_task_state(task, state);
560
561		/* didn't get the lock, go to sleep: */
562		spin_unlock_mutex(&lock->wait_lock, flags);
563		schedule_preempt_disabled();
564		spin_lock_mutex(&lock->wait_lock, flags);
565	}
566	mutex_remove_waiter(lock, &waiter, current_thread_info());
567	/* set it to 0 if there are no waiters left: */
568	if (likely(list_empty(&lock->wait_list)))
569		atomic_set(&lock->count, 0);
570	debug_mutex_free_waiter(&waiter);
571
572skip_wait:
573	/* got the lock - cleanup and rejoice! */
574	lock_acquired(&lock->dep_map, ip);
575	mutex_set_owner(lock);
576
577	if (use_ww_ctx) {
578		struct ww_mutex *ww = container_of(lock, struct ww_mutex, base);
579		struct mutex_waiter *cur;
580
581		/*
582		 * This branch gets optimized out for the common case,
583		 * and is only important for ww_mutex_lock.
584		 */
585		ww_mutex_lock_acquired(ww, ww_ctx);
586		ww->ctx = ww_ctx;
587
588		/*
589		 * Give any possible sleeping processes the chance to wake up,
590		 * so they can recheck if they have to back off.
591		 */
592		list_for_each_entry(cur, &lock->wait_list, list) {
593			debug_mutex_wake_waiter(lock, cur);
594			wake_up_process(cur->task);
595		}
596	}
597
598	spin_unlock_mutex(&lock->wait_lock, flags);
599	preempt_enable();
600	return 0;
601
602err:
603	mutex_remove_waiter(lock, &waiter, task_thread_info(task));
604	spin_unlock_mutex(&lock->wait_lock, flags);
605	debug_mutex_free_waiter(&waiter);
606	mutex_release(&lock->dep_map, 1, ip);
607	preempt_enable();
608	return ret;
609}
610
611#ifdef CONFIG_DEBUG_LOCK_ALLOC
612void __sched
613mutex_lock_nested(struct mutex *lock, unsigned int subclass)
614{
615	might_sleep();
616	__mutex_lock_common(lock, TASK_UNINTERRUPTIBLE,
617			    subclass, NULL, _RET_IP_, NULL, 0);
618}
619
620EXPORT_SYMBOL_GPL(mutex_lock_nested);
621
622void __sched
623_mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest)
624{
625	might_sleep();
626	__mutex_lock_common(lock, TASK_UNINTERRUPTIBLE,
627			    0, nest, _RET_IP_, NULL, 0);
628}
629
630EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock);
631
632int __sched
633mutex_lock_killable_nested(struct mutex *lock, unsigned int subclass)
634{
635	might_sleep();
636	return __mutex_lock_common(lock, TASK_KILLABLE,
637				   subclass, NULL, _RET_IP_, NULL, 0);
638}
639EXPORT_SYMBOL_GPL(mutex_lock_killable_nested);
640
641int __sched
642mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
643{
644	might_sleep();
645	return __mutex_lock_common(lock, TASK_INTERRUPTIBLE,
646				   subclass, NULL, _RET_IP_, NULL, 0);
647}
648
649EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
650
651static inline int
652ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
653{
654#ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH
655	unsigned tmp;
656
657	if (ctx->deadlock_inject_countdown-- == 0) {
658		tmp = ctx->deadlock_inject_interval;
659		if (tmp > UINT_MAX/4)
660			tmp = UINT_MAX;
661		else
662			tmp = tmp*2 + tmp + tmp/2;
663
664		ctx->deadlock_inject_interval = tmp;
665		ctx->deadlock_inject_countdown = tmp;
666		ctx->contending_lock = lock;
667
668		ww_mutex_unlock(lock);
669
670		return -EDEADLK;
671	}
672#endif
673
674	return 0;
675}
676
677int __sched
678__ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
679{
680	int ret;
681
682	might_sleep();
683	ret =  __mutex_lock_common(&lock->base, TASK_UNINTERRUPTIBLE,
684				   0, &ctx->dep_map, _RET_IP_, ctx, 1);
685	if (!ret && ctx->acquired > 1)
686		return ww_mutex_deadlock_injection(lock, ctx);
687
688	return ret;
689}
690EXPORT_SYMBOL_GPL(__ww_mutex_lock);
691
692int __sched
693__ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
694{
695	int ret;
696
697	might_sleep();
698	ret = __mutex_lock_common(&lock->base, TASK_INTERRUPTIBLE,
699				  0, &ctx->dep_map, _RET_IP_, ctx, 1);
700
701	if (!ret && ctx->acquired > 1)
702		return ww_mutex_deadlock_injection(lock, ctx);
703
704	return ret;
705}
706EXPORT_SYMBOL_GPL(__ww_mutex_lock_interruptible);
707
708#endif
709
710/*
711 * Release the lock, slowpath:
712 */
713static inline void
714__mutex_unlock_common_slowpath(struct mutex *lock, int nested)
715{
716	unsigned long flags;
717
718	/*
719	 * As a performance measurement, release the lock before doing other
720	 * wakeup related duties to follow. This allows other tasks to acquire
721	 * the lock sooner, while still handling cleanups in past unlock calls.
722	 * This can be done as we do not enforce strict equivalence between the
723	 * mutex counter and wait_list.
724	 *
725	 *
726	 * Some architectures leave the lock unlocked in the fastpath failure
727	 * case, others need to leave it locked. In the later case we have to
728	 * unlock it here - as the lock counter is currently 0 or negative.
729	 */
730	if (__mutex_slowpath_needs_to_unlock())
731		atomic_set(&lock->count, 1);
732
733	spin_lock_mutex(&lock->wait_lock, flags);
734	mutex_release(&lock->dep_map, nested, _RET_IP_);
735	debug_mutex_unlock(lock);
736
737	if (!list_empty(&lock->wait_list)) {
738		/* get the first entry from the wait-list: */
739		struct mutex_waiter *waiter =
740				list_entry(lock->wait_list.next,
741					   struct mutex_waiter, list);
742
743		debug_mutex_wake_waiter(lock, waiter);
744
745		wake_up_process(waiter->task);
746	}
747
748	spin_unlock_mutex(&lock->wait_lock, flags);
749}
750
751/*
752 * Release the lock, slowpath:
753 */
754__visible void
755__mutex_unlock_slowpath(atomic_t *lock_count)
756{
757	struct mutex *lock = container_of(lock_count, struct mutex, count);
758
759	__mutex_unlock_common_slowpath(lock, 1);
760}
761
762#ifndef CONFIG_DEBUG_LOCK_ALLOC
763/*
764 * Here come the less common (and hence less performance-critical) APIs:
765 * mutex_lock_interruptible() and mutex_trylock().
766 */
767static noinline int __sched
768__mutex_lock_killable_slowpath(struct mutex *lock);
769
770static noinline int __sched
771__mutex_lock_interruptible_slowpath(struct mutex *lock);
772
773/**
774 * mutex_lock_interruptible - acquire the mutex, interruptible
775 * @lock: the mutex to be acquired
776 *
777 * Lock the mutex like mutex_lock(), and return 0 if the mutex has
778 * been acquired or sleep until the mutex becomes available. If a
779 * signal arrives while waiting for the lock then this function
780 * returns -EINTR.
781 *
782 * This function is similar to (but not equivalent to) down_interruptible().
783 */
784int __sched mutex_lock_interruptible(struct mutex *lock)
785{
786	int ret;
787
788	might_sleep();
789	ret =  __mutex_fastpath_lock_retval(&lock->count);
790	if (likely(!ret)) {
791		mutex_set_owner(lock);
792		return 0;
793	} else
794		return __mutex_lock_interruptible_slowpath(lock);
795}
796
797EXPORT_SYMBOL(mutex_lock_interruptible);
798
799int __sched mutex_lock_killable(struct mutex *lock)
800{
801	int ret;
802
803	might_sleep();
804	ret = __mutex_fastpath_lock_retval(&lock->count);
805	if (likely(!ret)) {
806		mutex_set_owner(lock);
807		return 0;
808	} else
809		return __mutex_lock_killable_slowpath(lock);
810}
811EXPORT_SYMBOL(mutex_lock_killable);
812
813__visible void __sched
814__mutex_lock_slowpath(atomic_t *lock_count)
815{
816	struct mutex *lock = container_of(lock_count, struct mutex, count);
817
818	__mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, 0,
819			    NULL, _RET_IP_, NULL, 0);
820}
821
822static noinline int __sched
823__mutex_lock_killable_slowpath(struct mutex *lock)
824{
825	return __mutex_lock_common(lock, TASK_KILLABLE, 0,
826				   NULL, _RET_IP_, NULL, 0);
827}
828
829static noinline int __sched
830__mutex_lock_interruptible_slowpath(struct mutex *lock)
831{
832	return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, 0,
833				   NULL, _RET_IP_, NULL, 0);
834}
835
836static noinline int __sched
837__ww_mutex_lock_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
838{
839	return __mutex_lock_common(&lock->base, TASK_UNINTERRUPTIBLE, 0,
840				   NULL, _RET_IP_, ctx, 1);
841}
842
843static noinline int __sched
844__ww_mutex_lock_interruptible_slowpath(struct ww_mutex *lock,
845					    struct ww_acquire_ctx *ctx)
846{
847	return __mutex_lock_common(&lock->base, TASK_INTERRUPTIBLE, 0,
848				   NULL, _RET_IP_, ctx, 1);
849}
850
851#endif
852
853/*
854 * Spinlock based trylock, we take the spinlock and check whether we
855 * can get the lock:
856 */
857static inline int __mutex_trylock_slowpath(atomic_t *lock_count)
858{
859	struct mutex *lock = container_of(lock_count, struct mutex, count);
860	unsigned long flags;
861	int prev;
862
863	/* No need to trylock if the mutex is locked. */
864	if (mutex_is_locked(lock))
865		return 0;
866
867	spin_lock_mutex(&lock->wait_lock, flags);
868
869	prev = atomic_xchg(&lock->count, -1);
870	if (likely(prev == 1)) {
871		mutex_set_owner(lock);
872		mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
873	}
874
875	/* Set it back to 0 if there are no waiters: */
876	if (likely(list_empty(&lock->wait_list)))
877		atomic_set(&lock->count, 0);
878
879	spin_unlock_mutex(&lock->wait_lock, flags);
880
881	return prev == 1;
882}
883
884/**
885 * mutex_trylock - try to acquire the mutex, without waiting
886 * @lock: the mutex to be acquired
887 *
888 * Try to acquire the mutex atomically. Returns 1 if the mutex
889 * has been acquired successfully, and 0 on contention.
890 *
891 * NOTE: this function follows the spin_trylock() convention, so
892 * it is negated from the down_trylock() return values! Be careful
893 * about this when converting semaphore users to mutexes.
894 *
895 * This function must not be used in interrupt context. The
896 * mutex must be released by the same task that acquired it.
897 */
898int __sched mutex_trylock(struct mutex *lock)
899{
900	int ret;
901
902	ret = __mutex_fastpath_trylock(&lock->count, __mutex_trylock_slowpath);
903	if (ret)
904		mutex_set_owner(lock);
905
906	return ret;
907}
908EXPORT_SYMBOL(mutex_trylock);
909
910#ifndef CONFIG_DEBUG_LOCK_ALLOC
911int __sched
912__ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
913{
914	int ret;
915
916	might_sleep();
917
918	ret = __mutex_fastpath_lock_retval(&lock->base.count);
919
920	if (likely(!ret)) {
921		ww_mutex_set_context_fastpath(lock, ctx);
922		mutex_set_owner(&lock->base);
923	} else
924		ret = __ww_mutex_lock_slowpath(lock, ctx);
925	return ret;
926}
927EXPORT_SYMBOL(__ww_mutex_lock);
928
929int __sched
930__ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
931{
932	int ret;
933
934	might_sleep();
935
936	ret = __mutex_fastpath_lock_retval(&lock->base.count);
937
938	if (likely(!ret)) {
939		ww_mutex_set_context_fastpath(lock, ctx);
940		mutex_set_owner(&lock->base);
941	} else
942		ret = __ww_mutex_lock_interruptible_slowpath(lock, ctx);
943	return ret;
944}
945EXPORT_SYMBOL(__ww_mutex_lock_interruptible);
946
947#endif
948
949/**
950 * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0
951 * @cnt: the atomic which we are to dec
952 * @lock: the mutex to return holding if we dec to 0
953 *
954 * return true and hold lock if we dec to 0, return false otherwise
955 */
956int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock)
957{
958	/* dec if we can't possibly hit 0 */
959	if (atomic_add_unless(cnt, -1, 1))
960		return 0;
961	/* we might hit 0, so take the lock */
962	mutex_lock(lock);
963	if (!atomic_dec_and_test(cnt)) {
964		/* when we actually did the dec, we didn't hit 0 */
965		mutex_unlock(lock);
966		return 0;
967	}
968	/* we hit 0, and we hold the lock */
969	return 1;
970}
971EXPORT_SYMBOL(atomic_dec_and_mutex_lock);
972