xref: /openbmc/linux/kernel/locking/mutex.c (revision 214e0aed639ef40987bf6159fad303171a6de31e)
101768b42SPeter Zijlstra /*
267a6de49SPeter Zijlstra  * kernel/locking/mutex.c
301768b42SPeter Zijlstra  *
401768b42SPeter Zijlstra  * Mutexes: blocking mutual exclusion locks
501768b42SPeter Zijlstra  *
601768b42SPeter Zijlstra  * Started by Ingo Molnar:
701768b42SPeter Zijlstra  *
801768b42SPeter Zijlstra  *  Copyright (C) 2004, 2005, 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
901768b42SPeter Zijlstra  *
1001768b42SPeter Zijlstra  * Many thanks to Arjan van de Ven, Thomas Gleixner, Steven Rostedt and
1101768b42SPeter Zijlstra  * David Howells for suggestions and improvements.
1201768b42SPeter Zijlstra  *
1301768b42SPeter Zijlstra  *  - Adaptive spinning for mutexes by Peter Zijlstra. (Ported to mainline
1401768b42SPeter Zijlstra  *    from the -rt tree, where it was originally implemented for rtmutexes
1501768b42SPeter Zijlstra  *    by Steven Rostedt, based on work by Gregory Haskins, Peter Morreale
1601768b42SPeter Zijlstra  *    and Sven Dietrich.
1701768b42SPeter Zijlstra  *
18*214e0aedSDavidlohr Bueso  * Also see Documentation/locking/mutex-design.txt.
1901768b42SPeter Zijlstra  */
2001768b42SPeter Zijlstra #include <linux/mutex.h>
2101768b42SPeter Zijlstra #include <linux/ww_mutex.h>
2201768b42SPeter Zijlstra #include <linux/sched.h>
2301768b42SPeter Zijlstra #include <linux/sched/rt.h>
2401768b42SPeter Zijlstra #include <linux/export.h>
2501768b42SPeter Zijlstra #include <linux/spinlock.h>
2601768b42SPeter Zijlstra #include <linux/interrupt.h>
2701768b42SPeter Zijlstra #include <linux/debug_locks.h>
28c9122da1SPeter Zijlstra #include "mcs_spinlock.h"
2901768b42SPeter Zijlstra 
3001768b42SPeter Zijlstra /*
3101768b42SPeter Zijlstra  * In the DEBUG case we are using the "NULL fastpath" for mutexes,
3201768b42SPeter Zijlstra  * which forces all calls into the slowpath:
3301768b42SPeter Zijlstra  */
3401768b42SPeter Zijlstra #ifdef CONFIG_DEBUG_MUTEXES
3501768b42SPeter Zijlstra # include "mutex-debug.h"
3601768b42SPeter Zijlstra # include <asm-generic/mutex-null.h>
376f008e72SPeter Zijlstra /*
386f008e72SPeter Zijlstra  * Must be 0 for the debug case so we do not do the unlock outside of the
396f008e72SPeter Zijlstra  * wait_lock region. debug_mutex_unlock() will do the actual unlock in this
406f008e72SPeter Zijlstra  * case.
416f008e72SPeter Zijlstra  */
426f008e72SPeter Zijlstra # undef __mutex_slowpath_needs_to_unlock
436f008e72SPeter Zijlstra # define  __mutex_slowpath_needs_to_unlock()	0
4401768b42SPeter Zijlstra #else
4501768b42SPeter Zijlstra # include "mutex.h"
4601768b42SPeter Zijlstra # include <asm/mutex.h>
4701768b42SPeter Zijlstra #endif
4801768b42SPeter Zijlstra 
4901768b42SPeter Zijlstra void
5001768b42SPeter Zijlstra __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
5101768b42SPeter Zijlstra {
5201768b42SPeter Zijlstra 	atomic_set(&lock->count, 1);
5301768b42SPeter Zijlstra 	spin_lock_init(&lock->wait_lock);
5401768b42SPeter Zijlstra 	INIT_LIST_HEAD(&lock->wait_list);
5501768b42SPeter Zijlstra 	mutex_clear_owner(lock);
5601768b42SPeter Zijlstra #ifdef CONFIG_MUTEX_SPIN_ON_OWNER
574d9d951eSJason Low 	osq_lock_init(&lock->osq);
5801768b42SPeter Zijlstra #endif
5901768b42SPeter Zijlstra 
6001768b42SPeter Zijlstra 	debug_mutex_init(lock, name, key);
6101768b42SPeter Zijlstra }
6201768b42SPeter Zijlstra 
6301768b42SPeter Zijlstra EXPORT_SYMBOL(__mutex_init);
6401768b42SPeter Zijlstra 
6501768b42SPeter Zijlstra #ifndef CONFIG_DEBUG_LOCK_ALLOC
6601768b42SPeter Zijlstra /*
6701768b42SPeter Zijlstra  * We split the mutex lock/unlock logic into separate fastpath and
6801768b42SPeter Zijlstra  * slowpath functions, to reduce the register pressure on the fastpath.
6901768b42SPeter Zijlstra  * We also put the fastpath first in the kernel image, to make sure the
7001768b42SPeter Zijlstra  * branch is predicted by the CPU as default-untaken.
7101768b42SPeter Zijlstra  */
7222d9fd34SAndi Kleen __visible void __sched __mutex_lock_slowpath(atomic_t *lock_count);
7301768b42SPeter Zijlstra 
7401768b42SPeter Zijlstra /**
7501768b42SPeter Zijlstra  * mutex_lock - acquire the mutex
7601768b42SPeter Zijlstra  * @lock: the mutex to be acquired
7701768b42SPeter Zijlstra  *
7801768b42SPeter Zijlstra  * Lock the mutex exclusively for this task. If the mutex is not
7901768b42SPeter Zijlstra  * available right now, it will sleep until it can get it.
8001768b42SPeter Zijlstra  *
8101768b42SPeter Zijlstra  * The mutex must later on be released by the same task that
8201768b42SPeter Zijlstra  * acquired it. Recursive locking is not allowed. The task
8301768b42SPeter Zijlstra  * may not exit without first unlocking the mutex. Also, kernel
8401768b42SPeter Zijlstra  * memory where the mutex resides mutex must not be freed with
8501768b42SPeter Zijlstra  * the mutex still locked. The mutex must first be initialized
8601768b42SPeter Zijlstra  * (or statically defined) before it can be locked. memset()-ing
8701768b42SPeter Zijlstra  * the mutex to 0 is not allowed.
8801768b42SPeter Zijlstra  *
8901768b42SPeter Zijlstra  * ( The CONFIG_DEBUG_MUTEXES .config option turns on debugging
9001768b42SPeter Zijlstra  *   checks that will enforce the restrictions and will also do
9101768b42SPeter Zijlstra  *   deadlock debugging. )
9201768b42SPeter Zijlstra  *
9301768b42SPeter Zijlstra  * This function is similar to (but not equivalent to) down().
9401768b42SPeter Zijlstra  */
9501768b42SPeter Zijlstra void __sched mutex_lock(struct mutex *lock)
9601768b42SPeter Zijlstra {
9701768b42SPeter Zijlstra 	might_sleep();
9801768b42SPeter Zijlstra 	/*
9901768b42SPeter Zijlstra 	 * The locking fastpath is the 1->0 transition from
10001768b42SPeter Zijlstra 	 * 'unlocked' into 'locked' state.
10101768b42SPeter Zijlstra 	 */
10201768b42SPeter Zijlstra 	__mutex_fastpath_lock(&lock->count, __mutex_lock_slowpath);
10301768b42SPeter Zijlstra 	mutex_set_owner(lock);
10401768b42SPeter Zijlstra }
10501768b42SPeter Zijlstra 
10601768b42SPeter Zijlstra EXPORT_SYMBOL(mutex_lock);
10701768b42SPeter Zijlstra #endif
10801768b42SPeter Zijlstra 
10976916515SDavidlohr Bueso static __always_inline void ww_mutex_lock_acquired(struct ww_mutex *ww,
11076916515SDavidlohr Bueso 						   struct ww_acquire_ctx *ww_ctx)
11176916515SDavidlohr Bueso {
11276916515SDavidlohr Bueso #ifdef CONFIG_DEBUG_MUTEXES
11376916515SDavidlohr Bueso 	/*
11476916515SDavidlohr Bueso 	 * If this WARN_ON triggers, you used ww_mutex_lock to acquire,
11576916515SDavidlohr Bueso 	 * but released with a normal mutex_unlock in this call.
11676916515SDavidlohr Bueso 	 *
11776916515SDavidlohr Bueso 	 * This should never happen, always use ww_mutex_unlock.
11876916515SDavidlohr Bueso 	 */
11976916515SDavidlohr Bueso 	DEBUG_LOCKS_WARN_ON(ww->ctx);
12076916515SDavidlohr Bueso 
12176916515SDavidlohr Bueso 	/*
12276916515SDavidlohr Bueso 	 * Not quite done after calling ww_acquire_done() ?
12376916515SDavidlohr Bueso 	 */
12476916515SDavidlohr Bueso 	DEBUG_LOCKS_WARN_ON(ww_ctx->done_acquire);
12576916515SDavidlohr Bueso 
12676916515SDavidlohr Bueso 	if (ww_ctx->contending_lock) {
12776916515SDavidlohr Bueso 		/*
12876916515SDavidlohr Bueso 		 * After -EDEADLK you tried to
12976916515SDavidlohr Bueso 		 * acquire a different ww_mutex? Bad!
13076916515SDavidlohr Bueso 		 */
13176916515SDavidlohr Bueso 		DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock != ww);
13276916515SDavidlohr Bueso 
13376916515SDavidlohr Bueso 		/*
13476916515SDavidlohr Bueso 		 * You called ww_mutex_lock after receiving -EDEADLK,
13576916515SDavidlohr Bueso 		 * but 'forgot' to unlock everything else first?
13676916515SDavidlohr Bueso 		 */
13776916515SDavidlohr Bueso 		DEBUG_LOCKS_WARN_ON(ww_ctx->acquired > 0);
13876916515SDavidlohr Bueso 		ww_ctx->contending_lock = NULL;
13976916515SDavidlohr Bueso 	}
14076916515SDavidlohr Bueso 
14176916515SDavidlohr Bueso 	/*
14276916515SDavidlohr Bueso 	 * Naughty, using a different class will lead to undefined behavior!
14376916515SDavidlohr Bueso 	 */
14476916515SDavidlohr Bueso 	DEBUG_LOCKS_WARN_ON(ww_ctx->ww_class != ww->ww_class);
14576916515SDavidlohr Bueso #endif
14676916515SDavidlohr Bueso 	ww_ctx->acquired++;
14776916515SDavidlohr Bueso }
14876916515SDavidlohr Bueso 
14976916515SDavidlohr Bueso /*
15076916515SDavidlohr Bueso  * after acquiring lock with fastpath or when we lost out in contested
15176916515SDavidlohr Bueso  * slowpath, set ctx and wake up any waiters so they can recheck.
15276916515SDavidlohr Bueso  *
15376916515SDavidlohr Bueso  * This function is never called when CONFIG_DEBUG_LOCK_ALLOC is set,
15476916515SDavidlohr Bueso  * as the fastpath and opportunistic spinning are disabled in that case.
15576916515SDavidlohr Bueso  */
15676916515SDavidlohr Bueso static __always_inline void
15776916515SDavidlohr Bueso ww_mutex_set_context_fastpath(struct ww_mutex *lock,
15876916515SDavidlohr Bueso 			       struct ww_acquire_ctx *ctx)
15976916515SDavidlohr Bueso {
16076916515SDavidlohr Bueso 	unsigned long flags;
16176916515SDavidlohr Bueso 	struct mutex_waiter *cur;
16276916515SDavidlohr Bueso 
16376916515SDavidlohr Bueso 	ww_mutex_lock_acquired(lock, ctx);
16476916515SDavidlohr Bueso 
16576916515SDavidlohr Bueso 	lock->ctx = ctx;
16676916515SDavidlohr Bueso 
16776916515SDavidlohr Bueso 	/*
16876916515SDavidlohr Bueso 	 * The lock->ctx update should be visible on all cores before
16976916515SDavidlohr Bueso 	 * the atomic read is done, otherwise contended waiters might be
17076916515SDavidlohr Bueso 	 * missed. The contended waiters will either see ww_ctx == NULL
17176916515SDavidlohr Bueso 	 * and keep spinning, or it will acquire wait_lock, add itself
17276916515SDavidlohr Bueso 	 * to waiter list and sleep.
17376916515SDavidlohr Bueso 	 */
17476916515SDavidlohr Bueso 	smp_mb(); /* ^^^ */
17576916515SDavidlohr Bueso 
17676916515SDavidlohr Bueso 	/*
17776916515SDavidlohr Bueso 	 * Check if lock is contended, if not there is nobody to wake up
17876916515SDavidlohr Bueso 	 */
17976916515SDavidlohr Bueso 	if (likely(atomic_read(&lock->base.count) == 0))
18076916515SDavidlohr Bueso 		return;
18176916515SDavidlohr Bueso 
18276916515SDavidlohr Bueso 	/*
18376916515SDavidlohr Bueso 	 * Uh oh, we raced in fastpath, wake up everyone in this case,
18476916515SDavidlohr Bueso 	 * so they can see the new lock->ctx.
18576916515SDavidlohr Bueso 	 */
18676916515SDavidlohr Bueso 	spin_lock_mutex(&lock->base.wait_lock, flags);
18776916515SDavidlohr Bueso 	list_for_each_entry(cur, &lock->base.wait_list, list) {
18876916515SDavidlohr Bueso 		debug_mutex_wake_waiter(&lock->base, cur);
18976916515SDavidlohr Bueso 		wake_up_process(cur->task);
19076916515SDavidlohr Bueso 	}
19176916515SDavidlohr Bueso 	spin_unlock_mutex(&lock->base.wait_lock, flags);
19276916515SDavidlohr Bueso }
19376916515SDavidlohr Bueso 
19476916515SDavidlohr Bueso 
19501768b42SPeter Zijlstra #ifdef CONFIG_MUTEX_SPIN_ON_OWNER
19601768b42SPeter Zijlstra /*
19701768b42SPeter Zijlstra  * In order to avoid a stampede of mutex spinners from acquiring the mutex
19801768b42SPeter Zijlstra  * more or less simultaneously, the spinners need to acquire a MCS lock
19901768b42SPeter Zijlstra  * first before spinning on the owner field.
20001768b42SPeter Zijlstra  *
20101768b42SPeter Zijlstra  */
20201768b42SPeter Zijlstra 
20301768b42SPeter Zijlstra /*
20401768b42SPeter Zijlstra  * Mutex spinning code migrated from kernel/sched/core.c
20501768b42SPeter Zijlstra  */
20601768b42SPeter Zijlstra 
20701768b42SPeter Zijlstra static inline bool owner_running(struct mutex *lock, struct task_struct *owner)
20801768b42SPeter Zijlstra {
20901768b42SPeter Zijlstra 	if (lock->owner != owner)
21001768b42SPeter Zijlstra 		return false;
21101768b42SPeter Zijlstra 
21201768b42SPeter Zijlstra 	/*
21301768b42SPeter Zijlstra 	 * Ensure we emit the owner->on_cpu, dereference _after_ checking
21401768b42SPeter Zijlstra 	 * lock->owner still matches owner, if that fails, owner might
21501768b42SPeter Zijlstra 	 * point to free()d memory, if it still matches, the rcu_read_lock()
21601768b42SPeter Zijlstra 	 * ensures the memory stays valid.
21701768b42SPeter Zijlstra 	 */
21801768b42SPeter Zijlstra 	barrier();
21901768b42SPeter Zijlstra 
22001768b42SPeter Zijlstra 	return owner->on_cpu;
22101768b42SPeter Zijlstra }
22201768b42SPeter Zijlstra 
22301768b42SPeter Zijlstra /*
22401768b42SPeter Zijlstra  * Look out! "owner" is an entirely speculative pointer
22501768b42SPeter Zijlstra  * access and not reliable.
22601768b42SPeter Zijlstra  */
22701768b42SPeter Zijlstra static noinline
22801768b42SPeter Zijlstra int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner)
22901768b42SPeter Zijlstra {
23001768b42SPeter Zijlstra 	rcu_read_lock();
23101768b42SPeter Zijlstra 	while (owner_running(lock, owner)) {
23201768b42SPeter Zijlstra 		if (need_resched())
23301768b42SPeter Zijlstra 			break;
23401768b42SPeter Zijlstra 
2353a6bfbc9SDavidlohr Bueso 		cpu_relax_lowlatency();
23601768b42SPeter Zijlstra 	}
23701768b42SPeter Zijlstra 	rcu_read_unlock();
23801768b42SPeter Zijlstra 
23901768b42SPeter Zijlstra 	/*
24001768b42SPeter Zijlstra 	 * We break out the loop above on need_resched() and when the
24101768b42SPeter Zijlstra 	 * owner changed, which is a sign for heavy contention. Return
24201768b42SPeter Zijlstra 	 * success only when lock->owner is NULL.
24301768b42SPeter Zijlstra 	 */
24401768b42SPeter Zijlstra 	return lock->owner == NULL;
24501768b42SPeter Zijlstra }
24601768b42SPeter Zijlstra 
24701768b42SPeter Zijlstra /*
24801768b42SPeter Zijlstra  * Initial check for entering the mutex spinning loop
24901768b42SPeter Zijlstra  */
25001768b42SPeter Zijlstra static inline int mutex_can_spin_on_owner(struct mutex *lock)
25101768b42SPeter Zijlstra {
25201768b42SPeter Zijlstra 	struct task_struct *owner;
25301768b42SPeter Zijlstra 	int retval = 1;
25401768b42SPeter Zijlstra 
25546af29e4SJason Low 	if (need_resched())
25646af29e4SJason Low 		return 0;
25746af29e4SJason Low 
25801768b42SPeter Zijlstra 	rcu_read_lock();
25901768b42SPeter Zijlstra 	owner = ACCESS_ONCE(lock->owner);
26001768b42SPeter Zijlstra 	if (owner)
26101768b42SPeter Zijlstra 		retval = owner->on_cpu;
26201768b42SPeter Zijlstra 	rcu_read_unlock();
26301768b42SPeter Zijlstra 	/*
26401768b42SPeter Zijlstra 	 * if lock->owner is not set, the mutex owner may have just acquired
26501768b42SPeter Zijlstra 	 * it and not set the owner yet or the mutex has been released.
26601768b42SPeter Zijlstra 	 */
26701768b42SPeter Zijlstra 	return retval;
26801768b42SPeter Zijlstra }
26976916515SDavidlohr Bueso 
27076916515SDavidlohr Bueso /*
27176916515SDavidlohr Bueso  * Atomically try to take the lock when it is available
27276916515SDavidlohr Bueso  */
27376916515SDavidlohr Bueso static inline bool mutex_try_to_acquire(struct mutex *lock)
27476916515SDavidlohr Bueso {
27576916515SDavidlohr Bueso 	return !mutex_is_locked(lock) &&
27676916515SDavidlohr Bueso 		(atomic_cmpxchg(&lock->count, 1, 0) == 1);
27776916515SDavidlohr Bueso }
27876916515SDavidlohr Bueso 
27976916515SDavidlohr Bueso /*
28076916515SDavidlohr Bueso  * Optimistic spinning.
28176916515SDavidlohr Bueso  *
28276916515SDavidlohr Bueso  * We try to spin for acquisition when we find that the lock owner
28376916515SDavidlohr Bueso  * is currently running on a (different) CPU and while we don't
28476916515SDavidlohr Bueso  * need to reschedule. The rationale is that if the lock owner is
28576916515SDavidlohr Bueso  * running, it is likely to release the lock soon.
28676916515SDavidlohr Bueso  *
28776916515SDavidlohr Bueso  * Since this needs the lock owner, and this mutex implementation
28876916515SDavidlohr Bueso  * doesn't track the owner atomically in the lock field, we need to
28976916515SDavidlohr Bueso  * track it non-atomically.
29076916515SDavidlohr Bueso  *
29176916515SDavidlohr Bueso  * We can't do this for DEBUG_MUTEXES because that relies on wait_lock
29276916515SDavidlohr Bueso  * to serialize everything.
29376916515SDavidlohr Bueso  *
29476916515SDavidlohr Bueso  * The mutex spinners are queued up using MCS lock so that only one
29576916515SDavidlohr Bueso  * spinner can compete for the mutex. However, if mutex spinning isn't
29676916515SDavidlohr Bueso  * going to happen, there is no point in going through the lock/unlock
29776916515SDavidlohr Bueso  * overhead.
29876916515SDavidlohr Bueso  *
29976916515SDavidlohr Bueso  * Returns true when the lock was taken, otherwise false, indicating
30076916515SDavidlohr Bueso  * that we need to jump to the slowpath and sleep.
30176916515SDavidlohr Bueso  */
30276916515SDavidlohr Bueso static bool mutex_optimistic_spin(struct mutex *lock,
30376916515SDavidlohr Bueso 				  struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
30476916515SDavidlohr Bueso {
30576916515SDavidlohr Bueso 	struct task_struct *task = current;
30676916515SDavidlohr Bueso 
30776916515SDavidlohr Bueso 	if (!mutex_can_spin_on_owner(lock))
30876916515SDavidlohr Bueso 		goto done;
30976916515SDavidlohr Bueso 
31076916515SDavidlohr Bueso 	if (!osq_lock(&lock->osq))
31176916515SDavidlohr Bueso 		goto done;
31276916515SDavidlohr Bueso 
31376916515SDavidlohr Bueso 	while (true) {
31476916515SDavidlohr Bueso 		struct task_struct *owner;
31576916515SDavidlohr Bueso 
31676916515SDavidlohr Bueso 		if (use_ww_ctx && ww_ctx->acquired > 0) {
31776916515SDavidlohr Bueso 			struct ww_mutex *ww;
31876916515SDavidlohr Bueso 
31976916515SDavidlohr Bueso 			ww = container_of(lock, struct ww_mutex, base);
32076916515SDavidlohr Bueso 			/*
32176916515SDavidlohr Bueso 			 * If ww->ctx is set the contents are undefined, only
32276916515SDavidlohr Bueso 			 * by acquiring wait_lock there is a guarantee that
32376916515SDavidlohr Bueso 			 * they are not invalid when reading.
32476916515SDavidlohr Bueso 			 *
32576916515SDavidlohr Bueso 			 * As such, when deadlock detection needs to be
32676916515SDavidlohr Bueso 			 * performed the optimistic spinning cannot be done.
32776916515SDavidlohr Bueso 			 */
32876916515SDavidlohr Bueso 			if (ACCESS_ONCE(ww->ctx))
32976916515SDavidlohr Bueso 				break;
33076916515SDavidlohr Bueso 		}
33176916515SDavidlohr Bueso 
33276916515SDavidlohr Bueso 		/*
33376916515SDavidlohr Bueso 		 * If there's an owner, wait for it to either
33476916515SDavidlohr Bueso 		 * release the lock or go to sleep.
33576916515SDavidlohr Bueso 		 */
33676916515SDavidlohr Bueso 		owner = ACCESS_ONCE(lock->owner);
33776916515SDavidlohr Bueso 		if (owner && !mutex_spin_on_owner(lock, owner))
33876916515SDavidlohr Bueso 			break;
33976916515SDavidlohr Bueso 
34076916515SDavidlohr Bueso 		/* Try to acquire the mutex if it is unlocked. */
34176916515SDavidlohr Bueso 		if (mutex_try_to_acquire(lock)) {
34276916515SDavidlohr Bueso 			lock_acquired(&lock->dep_map, ip);
34376916515SDavidlohr Bueso 
34476916515SDavidlohr Bueso 			if (use_ww_ctx) {
34576916515SDavidlohr Bueso 				struct ww_mutex *ww;
34676916515SDavidlohr Bueso 				ww = container_of(lock, struct ww_mutex, base);
34776916515SDavidlohr Bueso 
34876916515SDavidlohr Bueso 				ww_mutex_set_context_fastpath(ww, ww_ctx);
34976916515SDavidlohr Bueso 			}
35076916515SDavidlohr Bueso 
35176916515SDavidlohr Bueso 			mutex_set_owner(lock);
35276916515SDavidlohr Bueso 			osq_unlock(&lock->osq);
35376916515SDavidlohr Bueso 			return true;
35476916515SDavidlohr Bueso 		}
35576916515SDavidlohr Bueso 
35676916515SDavidlohr Bueso 		/*
35776916515SDavidlohr Bueso 		 * When there's no owner, we might have preempted between the
35876916515SDavidlohr Bueso 		 * owner acquiring the lock and setting the owner field. If
35976916515SDavidlohr Bueso 		 * we're an RT task that will live-lock because we won't let
36076916515SDavidlohr Bueso 		 * the owner complete.
36176916515SDavidlohr Bueso 		 */
36276916515SDavidlohr Bueso 		if (!owner && (need_resched() || rt_task(task)))
36376916515SDavidlohr Bueso 			break;
36476916515SDavidlohr Bueso 
36576916515SDavidlohr Bueso 		/*
36676916515SDavidlohr Bueso 		 * The cpu_relax() call is a compiler barrier which forces
36776916515SDavidlohr Bueso 		 * everything in this loop to be re-loaded. We don't need
36876916515SDavidlohr Bueso 		 * memory barriers as we'll eventually observe the right
36976916515SDavidlohr Bueso 		 * values at the cost of a few extra spins.
37076916515SDavidlohr Bueso 		 */
37176916515SDavidlohr Bueso 		cpu_relax_lowlatency();
37276916515SDavidlohr Bueso 	}
37376916515SDavidlohr Bueso 
37476916515SDavidlohr Bueso 	osq_unlock(&lock->osq);
37576916515SDavidlohr Bueso done:
37676916515SDavidlohr Bueso 	/*
37776916515SDavidlohr Bueso 	 * If we fell out of the spin path because of need_resched(),
37876916515SDavidlohr Bueso 	 * reschedule now, before we try-lock the mutex. This avoids getting
37976916515SDavidlohr Bueso 	 * scheduled out right after we obtained the mutex.
38076916515SDavidlohr Bueso 	 */
38176916515SDavidlohr Bueso 	if (need_resched())
38276916515SDavidlohr Bueso 		schedule_preempt_disabled();
38376916515SDavidlohr Bueso 
38476916515SDavidlohr Bueso 	return false;
38576916515SDavidlohr Bueso }
38676916515SDavidlohr Bueso #else
38776916515SDavidlohr Bueso static bool mutex_optimistic_spin(struct mutex *lock,
38876916515SDavidlohr Bueso 				  struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
38976916515SDavidlohr Bueso {
39076916515SDavidlohr Bueso 	return false;
39176916515SDavidlohr Bueso }
39201768b42SPeter Zijlstra #endif
39301768b42SPeter Zijlstra 
39422d9fd34SAndi Kleen __visible __used noinline
39522d9fd34SAndi Kleen void __sched __mutex_unlock_slowpath(atomic_t *lock_count);
39601768b42SPeter Zijlstra 
39701768b42SPeter Zijlstra /**
39801768b42SPeter Zijlstra  * mutex_unlock - release the mutex
39901768b42SPeter Zijlstra  * @lock: the mutex to be released
40001768b42SPeter Zijlstra  *
40101768b42SPeter Zijlstra  * Unlock a mutex that has been locked by this task previously.
40201768b42SPeter Zijlstra  *
40301768b42SPeter Zijlstra  * This function must not be used in interrupt context. Unlocking
40401768b42SPeter Zijlstra  * of a not locked mutex is not allowed.
40501768b42SPeter Zijlstra  *
40601768b42SPeter Zijlstra  * This function is similar to (but not equivalent to) up().
40701768b42SPeter Zijlstra  */
40801768b42SPeter Zijlstra void __sched mutex_unlock(struct mutex *lock)
40901768b42SPeter Zijlstra {
41001768b42SPeter Zijlstra 	/*
41101768b42SPeter Zijlstra 	 * The unlocking fastpath is the 0->1 transition from 'locked'
41201768b42SPeter Zijlstra 	 * into 'unlocked' state:
41301768b42SPeter Zijlstra 	 */
41401768b42SPeter Zijlstra #ifndef CONFIG_DEBUG_MUTEXES
41501768b42SPeter Zijlstra 	/*
41601768b42SPeter Zijlstra 	 * When debugging is enabled we must not clear the owner before time,
41701768b42SPeter Zijlstra 	 * the slow path will always be taken, and that clears the owner field
41801768b42SPeter Zijlstra 	 * after verifying that it was indeed current.
41901768b42SPeter Zijlstra 	 */
42001768b42SPeter Zijlstra 	mutex_clear_owner(lock);
42101768b42SPeter Zijlstra #endif
42201768b42SPeter Zijlstra 	__mutex_fastpath_unlock(&lock->count, __mutex_unlock_slowpath);
42301768b42SPeter Zijlstra }
42401768b42SPeter Zijlstra 
42501768b42SPeter Zijlstra EXPORT_SYMBOL(mutex_unlock);
42601768b42SPeter Zijlstra 
42701768b42SPeter Zijlstra /**
42801768b42SPeter Zijlstra  * ww_mutex_unlock - release the w/w mutex
42901768b42SPeter Zijlstra  * @lock: the mutex to be released
43001768b42SPeter Zijlstra  *
43101768b42SPeter Zijlstra  * Unlock a mutex that has been locked by this task previously with any of the
43201768b42SPeter Zijlstra  * ww_mutex_lock* functions (with or without an acquire context). It is
43301768b42SPeter Zijlstra  * forbidden to release the locks after releasing the acquire context.
43401768b42SPeter Zijlstra  *
43501768b42SPeter Zijlstra  * This function must not be used in interrupt context. Unlocking
43601768b42SPeter Zijlstra  * of a unlocked mutex is not allowed.
43701768b42SPeter Zijlstra  */
43801768b42SPeter Zijlstra void __sched ww_mutex_unlock(struct ww_mutex *lock)
43901768b42SPeter Zijlstra {
44001768b42SPeter Zijlstra 	/*
44101768b42SPeter Zijlstra 	 * The unlocking fastpath is the 0->1 transition from 'locked'
44201768b42SPeter Zijlstra 	 * into 'unlocked' state:
44301768b42SPeter Zijlstra 	 */
44401768b42SPeter Zijlstra 	if (lock->ctx) {
44501768b42SPeter Zijlstra #ifdef CONFIG_DEBUG_MUTEXES
44601768b42SPeter Zijlstra 		DEBUG_LOCKS_WARN_ON(!lock->ctx->acquired);
44701768b42SPeter Zijlstra #endif
44801768b42SPeter Zijlstra 		if (lock->ctx->acquired > 0)
44901768b42SPeter Zijlstra 			lock->ctx->acquired--;
45001768b42SPeter Zijlstra 		lock->ctx = NULL;
45101768b42SPeter Zijlstra 	}
45201768b42SPeter Zijlstra 
45301768b42SPeter Zijlstra #ifndef CONFIG_DEBUG_MUTEXES
45401768b42SPeter Zijlstra 	/*
45501768b42SPeter Zijlstra 	 * When debugging is enabled we must not clear the owner before time,
45601768b42SPeter Zijlstra 	 * the slow path will always be taken, and that clears the owner field
45701768b42SPeter Zijlstra 	 * after verifying that it was indeed current.
45801768b42SPeter Zijlstra 	 */
45901768b42SPeter Zijlstra 	mutex_clear_owner(&lock->base);
46001768b42SPeter Zijlstra #endif
46101768b42SPeter Zijlstra 	__mutex_fastpath_unlock(&lock->base.count, __mutex_unlock_slowpath);
46201768b42SPeter Zijlstra }
46301768b42SPeter Zijlstra EXPORT_SYMBOL(ww_mutex_unlock);
46401768b42SPeter Zijlstra 
46501768b42SPeter Zijlstra static inline int __sched
46601768b42SPeter Zijlstra __mutex_lock_check_stamp(struct mutex *lock, struct ww_acquire_ctx *ctx)
46701768b42SPeter Zijlstra {
46801768b42SPeter Zijlstra 	struct ww_mutex *ww = container_of(lock, struct ww_mutex, base);
46901768b42SPeter Zijlstra 	struct ww_acquire_ctx *hold_ctx = ACCESS_ONCE(ww->ctx);
47001768b42SPeter Zijlstra 
47101768b42SPeter Zijlstra 	if (!hold_ctx)
47201768b42SPeter Zijlstra 		return 0;
47301768b42SPeter Zijlstra 
47401768b42SPeter Zijlstra 	if (unlikely(ctx == hold_ctx))
47501768b42SPeter Zijlstra 		return -EALREADY;
47601768b42SPeter Zijlstra 
47701768b42SPeter Zijlstra 	if (ctx->stamp - hold_ctx->stamp <= LONG_MAX &&
47801768b42SPeter Zijlstra 	    (ctx->stamp != hold_ctx->stamp || ctx > hold_ctx)) {
47901768b42SPeter Zijlstra #ifdef CONFIG_DEBUG_MUTEXES
48001768b42SPeter Zijlstra 		DEBUG_LOCKS_WARN_ON(ctx->contending_lock);
48101768b42SPeter Zijlstra 		ctx->contending_lock = ww;
48201768b42SPeter Zijlstra #endif
48301768b42SPeter Zijlstra 		return -EDEADLK;
48401768b42SPeter Zijlstra 	}
48501768b42SPeter Zijlstra 
48601768b42SPeter Zijlstra 	return 0;
48701768b42SPeter Zijlstra }
48801768b42SPeter Zijlstra 
48901768b42SPeter Zijlstra /*
49001768b42SPeter Zijlstra  * Lock a mutex (possibly interruptible), slowpath:
49101768b42SPeter Zijlstra  */
49201768b42SPeter Zijlstra static __always_inline int __sched
49301768b42SPeter Zijlstra __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
49401768b42SPeter Zijlstra 		    struct lockdep_map *nest_lock, unsigned long ip,
49501768b42SPeter Zijlstra 		    struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
49601768b42SPeter Zijlstra {
49701768b42SPeter Zijlstra 	struct task_struct *task = current;
49801768b42SPeter Zijlstra 	struct mutex_waiter waiter;
49901768b42SPeter Zijlstra 	unsigned long flags;
50001768b42SPeter Zijlstra 	int ret;
50101768b42SPeter Zijlstra 
50201768b42SPeter Zijlstra 	preempt_disable();
50301768b42SPeter Zijlstra 	mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);
50401768b42SPeter Zijlstra 
50576916515SDavidlohr Bueso 	if (mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx)) {
50676916515SDavidlohr Bueso 		/* got the lock, yay! */
50701768b42SPeter Zijlstra 		preempt_enable();
50801768b42SPeter Zijlstra 		return 0;
50901768b42SPeter Zijlstra 	}
51001768b42SPeter Zijlstra 
51101768b42SPeter Zijlstra 	spin_lock_mutex(&lock->wait_lock, flags);
51201768b42SPeter Zijlstra 
5131e820c96SJason Low 	/*
5141e820c96SJason Low 	 * Once more, try to acquire the lock. Only try-lock the mutex if
5150d968dd8SJason Low 	 * it is unlocked to reduce unnecessary xchg() operations.
5161e820c96SJason Low 	 */
5170d968dd8SJason Low 	if (!mutex_is_locked(lock) && (atomic_xchg(&lock->count, 0) == 1))
51801768b42SPeter Zijlstra 		goto skip_wait;
51901768b42SPeter Zijlstra 
52001768b42SPeter Zijlstra 	debug_mutex_lock_common(lock, &waiter);
52101768b42SPeter Zijlstra 	debug_mutex_add_waiter(lock, &waiter, task_thread_info(task));
52201768b42SPeter Zijlstra 
52301768b42SPeter Zijlstra 	/* add waiting tasks to the end of the waitqueue (FIFO): */
52401768b42SPeter Zijlstra 	list_add_tail(&waiter.list, &lock->wait_list);
52501768b42SPeter Zijlstra 	waiter.task = task;
52601768b42SPeter Zijlstra 
52701768b42SPeter Zijlstra 	lock_contended(&lock->dep_map, ip);
52801768b42SPeter Zijlstra 
52901768b42SPeter Zijlstra 	for (;;) {
53001768b42SPeter Zijlstra 		/*
53101768b42SPeter Zijlstra 		 * Lets try to take the lock again - this is needed even if
53201768b42SPeter Zijlstra 		 * we get here for the first time (shortly after failing to
53301768b42SPeter Zijlstra 		 * acquire the lock), to make sure that we get a wakeup once
53401768b42SPeter Zijlstra 		 * it's unlocked. Later on, if we sleep, this is the
53501768b42SPeter Zijlstra 		 * operation that gives us the lock. We xchg it to -1, so
53601768b42SPeter Zijlstra 		 * that when we release the lock, we properly wake up the
5371e820c96SJason Low 		 * other waiters. We only attempt the xchg if the count is
5381e820c96SJason Low 		 * non-negative in order to avoid unnecessary xchg operations:
53901768b42SPeter Zijlstra 		 */
5401e820c96SJason Low 		if (atomic_read(&lock->count) >= 0 &&
54101768b42SPeter Zijlstra 		    (atomic_xchg(&lock->count, -1) == 1))
54201768b42SPeter Zijlstra 			break;
54301768b42SPeter Zijlstra 
54401768b42SPeter Zijlstra 		/*
54501768b42SPeter Zijlstra 		 * got a signal? (This code gets eliminated in the
54601768b42SPeter Zijlstra 		 * TASK_UNINTERRUPTIBLE case.)
54701768b42SPeter Zijlstra 		 */
54801768b42SPeter Zijlstra 		if (unlikely(signal_pending_state(state, task))) {
54901768b42SPeter Zijlstra 			ret = -EINTR;
55001768b42SPeter Zijlstra 			goto err;
55101768b42SPeter Zijlstra 		}
55201768b42SPeter Zijlstra 
55301768b42SPeter Zijlstra 		if (use_ww_ctx && ww_ctx->acquired > 0) {
55401768b42SPeter Zijlstra 			ret = __mutex_lock_check_stamp(lock, ww_ctx);
55501768b42SPeter Zijlstra 			if (ret)
55601768b42SPeter Zijlstra 				goto err;
55701768b42SPeter Zijlstra 		}
55801768b42SPeter Zijlstra 
55901768b42SPeter Zijlstra 		__set_task_state(task, state);
56001768b42SPeter Zijlstra 
56101768b42SPeter Zijlstra 		/* didn't get the lock, go to sleep: */
56201768b42SPeter Zijlstra 		spin_unlock_mutex(&lock->wait_lock, flags);
56301768b42SPeter Zijlstra 		schedule_preempt_disabled();
56401768b42SPeter Zijlstra 		spin_lock_mutex(&lock->wait_lock, flags);
56501768b42SPeter Zijlstra 	}
56601768b42SPeter Zijlstra 	mutex_remove_waiter(lock, &waiter, current_thread_info());
56701768b42SPeter Zijlstra 	/* set it to 0 if there are no waiters left: */
56801768b42SPeter Zijlstra 	if (likely(list_empty(&lock->wait_list)))
56901768b42SPeter Zijlstra 		atomic_set(&lock->count, 0);
57001768b42SPeter Zijlstra 	debug_mutex_free_waiter(&waiter);
57101768b42SPeter Zijlstra 
57201768b42SPeter Zijlstra skip_wait:
57301768b42SPeter Zijlstra 	/* got the lock - cleanup and rejoice! */
57401768b42SPeter Zijlstra 	lock_acquired(&lock->dep_map, ip);
57501768b42SPeter Zijlstra 	mutex_set_owner(lock);
57601768b42SPeter Zijlstra 
57701768b42SPeter Zijlstra 	if (use_ww_ctx) {
57801768b42SPeter Zijlstra 		struct ww_mutex *ww = container_of(lock, struct ww_mutex, base);
57901768b42SPeter Zijlstra 		struct mutex_waiter *cur;
58001768b42SPeter Zijlstra 
58101768b42SPeter Zijlstra 		/*
58201768b42SPeter Zijlstra 		 * This branch gets optimized out for the common case,
58301768b42SPeter Zijlstra 		 * and is only important for ww_mutex_lock.
58401768b42SPeter Zijlstra 		 */
58501768b42SPeter Zijlstra 		ww_mutex_lock_acquired(ww, ww_ctx);
58601768b42SPeter Zijlstra 		ww->ctx = ww_ctx;
58701768b42SPeter Zijlstra 
58801768b42SPeter Zijlstra 		/*
58901768b42SPeter Zijlstra 		 * Give any possible sleeping processes the chance to wake up,
59001768b42SPeter Zijlstra 		 * so they can recheck if they have to back off.
59101768b42SPeter Zijlstra 		 */
59201768b42SPeter Zijlstra 		list_for_each_entry(cur, &lock->wait_list, list) {
59301768b42SPeter Zijlstra 			debug_mutex_wake_waiter(lock, cur);
59401768b42SPeter Zijlstra 			wake_up_process(cur->task);
59501768b42SPeter Zijlstra 		}
59601768b42SPeter Zijlstra 	}
59701768b42SPeter Zijlstra 
59801768b42SPeter Zijlstra 	spin_unlock_mutex(&lock->wait_lock, flags);
59901768b42SPeter Zijlstra 	preempt_enable();
60001768b42SPeter Zijlstra 	return 0;
60101768b42SPeter Zijlstra 
60201768b42SPeter Zijlstra err:
60301768b42SPeter Zijlstra 	mutex_remove_waiter(lock, &waiter, task_thread_info(task));
60401768b42SPeter Zijlstra 	spin_unlock_mutex(&lock->wait_lock, flags);
60501768b42SPeter Zijlstra 	debug_mutex_free_waiter(&waiter);
60601768b42SPeter Zijlstra 	mutex_release(&lock->dep_map, 1, ip);
60701768b42SPeter Zijlstra 	preempt_enable();
60801768b42SPeter Zijlstra 	return ret;
60901768b42SPeter Zijlstra }
61001768b42SPeter Zijlstra 
61101768b42SPeter Zijlstra #ifdef CONFIG_DEBUG_LOCK_ALLOC
61201768b42SPeter Zijlstra void __sched
61301768b42SPeter Zijlstra mutex_lock_nested(struct mutex *lock, unsigned int subclass)
61401768b42SPeter Zijlstra {
61501768b42SPeter Zijlstra 	might_sleep();
61601768b42SPeter Zijlstra 	__mutex_lock_common(lock, TASK_UNINTERRUPTIBLE,
61701768b42SPeter Zijlstra 			    subclass, NULL, _RET_IP_, NULL, 0);
61801768b42SPeter Zijlstra }
61901768b42SPeter Zijlstra 
62001768b42SPeter Zijlstra EXPORT_SYMBOL_GPL(mutex_lock_nested);
62101768b42SPeter Zijlstra 
62201768b42SPeter Zijlstra void __sched
62301768b42SPeter Zijlstra _mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest)
62401768b42SPeter Zijlstra {
62501768b42SPeter Zijlstra 	might_sleep();
62601768b42SPeter Zijlstra 	__mutex_lock_common(lock, TASK_UNINTERRUPTIBLE,
62701768b42SPeter Zijlstra 			    0, nest, _RET_IP_, NULL, 0);
62801768b42SPeter Zijlstra }
62901768b42SPeter Zijlstra 
63001768b42SPeter Zijlstra EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock);
63101768b42SPeter Zijlstra 
63201768b42SPeter Zijlstra int __sched
63301768b42SPeter Zijlstra mutex_lock_killable_nested(struct mutex *lock, unsigned int subclass)
63401768b42SPeter Zijlstra {
63501768b42SPeter Zijlstra 	might_sleep();
63601768b42SPeter Zijlstra 	return __mutex_lock_common(lock, TASK_KILLABLE,
63701768b42SPeter Zijlstra 				   subclass, NULL, _RET_IP_, NULL, 0);
63801768b42SPeter Zijlstra }
63901768b42SPeter Zijlstra EXPORT_SYMBOL_GPL(mutex_lock_killable_nested);
64001768b42SPeter Zijlstra 
64101768b42SPeter Zijlstra int __sched
64201768b42SPeter Zijlstra mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
64301768b42SPeter Zijlstra {
64401768b42SPeter Zijlstra 	might_sleep();
64501768b42SPeter Zijlstra 	return __mutex_lock_common(lock, TASK_INTERRUPTIBLE,
64601768b42SPeter Zijlstra 				   subclass, NULL, _RET_IP_, NULL, 0);
64701768b42SPeter Zijlstra }
64801768b42SPeter Zijlstra 
64901768b42SPeter Zijlstra EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
65001768b42SPeter Zijlstra 
65101768b42SPeter Zijlstra static inline int
65201768b42SPeter Zijlstra ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
65301768b42SPeter Zijlstra {
65401768b42SPeter Zijlstra #ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH
65501768b42SPeter Zijlstra 	unsigned tmp;
65601768b42SPeter Zijlstra 
65701768b42SPeter Zijlstra 	if (ctx->deadlock_inject_countdown-- == 0) {
65801768b42SPeter Zijlstra 		tmp = ctx->deadlock_inject_interval;
65901768b42SPeter Zijlstra 		if (tmp > UINT_MAX/4)
66001768b42SPeter Zijlstra 			tmp = UINT_MAX;
66101768b42SPeter Zijlstra 		else
66201768b42SPeter Zijlstra 			tmp = tmp*2 + tmp + tmp/2;
66301768b42SPeter Zijlstra 
66401768b42SPeter Zijlstra 		ctx->deadlock_inject_interval = tmp;
66501768b42SPeter Zijlstra 		ctx->deadlock_inject_countdown = tmp;
66601768b42SPeter Zijlstra 		ctx->contending_lock = lock;
66701768b42SPeter Zijlstra 
66801768b42SPeter Zijlstra 		ww_mutex_unlock(lock);
66901768b42SPeter Zijlstra 
67001768b42SPeter Zijlstra 		return -EDEADLK;
67101768b42SPeter Zijlstra 	}
67201768b42SPeter Zijlstra #endif
67301768b42SPeter Zijlstra 
67401768b42SPeter Zijlstra 	return 0;
67501768b42SPeter Zijlstra }
67601768b42SPeter Zijlstra 
67701768b42SPeter Zijlstra int __sched
67801768b42SPeter Zijlstra __ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
67901768b42SPeter Zijlstra {
68001768b42SPeter Zijlstra 	int ret;
68101768b42SPeter Zijlstra 
68201768b42SPeter Zijlstra 	might_sleep();
68301768b42SPeter Zijlstra 	ret =  __mutex_lock_common(&lock->base, TASK_UNINTERRUPTIBLE,
68401768b42SPeter Zijlstra 				   0, &ctx->dep_map, _RET_IP_, ctx, 1);
68501768b42SPeter Zijlstra 	if (!ret && ctx->acquired > 1)
68601768b42SPeter Zijlstra 		return ww_mutex_deadlock_injection(lock, ctx);
68701768b42SPeter Zijlstra 
68801768b42SPeter Zijlstra 	return ret;
68901768b42SPeter Zijlstra }
69001768b42SPeter Zijlstra EXPORT_SYMBOL_GPL(__ww_mutex_lock);
69101768b42SPeter Zijlstra 
69201768b42SPeter Zijlstra int __sched
69301768b42SPeter Zijlstra __ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
69401768b42SPeter Zijlstra {
69501768b42SPeter Zijlstra 	int ret;
69601768b42SPeter Zijlstra 
69701768b42SPeter Zijlstra 	might_sleep();
69801768b42SPeter Zijlstra 	ret = __mutex_lock_common(&lock->base, TASK_INTERRUPTIBLE,
69901768b42SPeter Zijlstra 				  0, &ctx->dep_map, _RET_IP_, ctx, 1);
70001768b42SPeter Zijlstra 
70101768b42SPeter Zijlstra 	if (!ret && ctx->acquired > 1)
70201768b42SPeter Zijlstra 		return ww_mutex_deadlock_injection(lock, ctx);
70301768b42SPeter Zijlstra 
70401768b42SPeter Zijlstra 	return ret;
70501768b42SPeter Zijlstra }
70601768b42SPeter Zijlstra EXPORT_SYMBOL_GPL(__ww_mutex_lock_interruptible);
70701768b42SPeter Zijlstra 
70801768b42SPeter Zijlstra #endif
70901768b42SPeter Zijlstra 
71001768b42SPeter Zijlstra /*
71101768b42SPeter Zijlstra  * Release the lock, slowpath:
71201768b42SPeter Zijlstra  */
71301768b42SPeter Zijlstra static inline void
714242489cfSDavidlohr Bueso __mutex_unlock_common_slowpath(struct mutex *lock, int nested)
71501768b42SPeter Zijlstra {
71601768b42SPeter Zijlstra 	unsigned long flags;
71701768b42SPeter Zijlstra 
71801768b42SPeter Zijlstra 	/*
71942fa566bSDavidlohr Bueso 	 * As a performance measurement, release the lock before doing other
72042fa566bSDavidlohr Bueso 	 * wakeup related duties to follow. This allows other tasks to acquire
72142fa566bSDavidlohr Bueso 	 * the lock sooner, while still handling cleanups in past unlock calls.
72242fa566bSDavidlohr Bueso 	 * This can be done as we do not enforce strict equivalence between the
72342fa566bSDavidlohr Bueso 	 * mutex counter and wait_list.
72442fa566bSDavidlohr Bueso 	 *
72542fa566bSDavidlohr Bueso 	 *
72642fa566bSDavidlohr Bueso 	 * Some architectures leave the lock unlocked in the fastpath failure
72701768b42SPeter Zijlstra 	 * case, others need to leave it locked. In the later case we have to
72842fa566bSDavidlohr Bueso 	 * unlock it here - as the lock counter is currently 0 or negative.
72901768b42SPeter Zijlstra 	 */
73001768b42SPeter Zijlstra 	if (__mutex_slowpath_needs_to_unlock())
73101768b42SPeter Zijlstra 		atomic_set(&lock->count, 1);
73201768b42SPeter Zijlstra 
7331d8fe7dcSJason Low 	spin_lock_mutex(&lock->wait_lock, flags);
7341d8fe7dcSJason Low 	mutex_release(&lock->dep_map, nested, _RET_IP_);
7351d8fe7dcSJason Low 	debug_mutex_unlock(lock);
7361d8fe7dcSJason Low 
73701768b42SPeter Zijlstra 	if (!list_empty(&lock->wait_list)) {
73801768b42SPeter Zijlstra 		/* get the first entry from the wait-list: */
73901768b42SPeter Zijlstra 		struct mutex_waiter *waiter =
74001768b42SPeter Zijlstra 				list_entry(lock->wait_list.next,
74101768b42SPeter Zijlstra 					   struct mutex_waiter, list);
74201768b42SPeter Zijlstra 
74301768b42SPeter Zijlstra 		debug_mutex_wake_waiter(lock, waiter);
74401768b42SPeter Zijlstra 
74501768b42SPeter Zijlstra 		wake_up_process(waiter->task);
74601768b42SPeter Zijlstra 	}
74701768b42SPeter Zijlstra 
74801768b42SPeter Zijlstra 	spin_unlock_mutex(&lock->wait_lock, flags);
74901768b42SPeter Zijlstra }
75001768b42SPeter Zijlstra 
75101768b42SPeter Zijlstra /*
75201768b42SPeter Zijlstra  * Release the lock, slowpath:
75301768b42SPeter Zijlstra  */
75422d9fd34SAndi Kleen __visible void
75501768b42SPeter Zijlstra __mutex_unlock_slowpath(atomic_t *lock_count)
75601768b42SPeter Zijlstra {
757242489cfSDavidlohr Bueso 	struct mutex *lock = container_of(lock_count, struct mutex, count);
758242489cfSDavidlohr Bueso 
759242489cfSDavidlohr Bueso 	__mutex_unlock_common_slowpath(lock, 1);
76001768b42SPeter Zijlstra }
76101768b42SPeter Zijlstra 
76201768b42SPeter Zijlstra #ifndef CONFIG_DEBUG_LOCK_ALLOC
76301768b42SPeter Zijlstra /*
76401768b42SPeter Zijlstra  * Here come the less common (and hence less performance-critical) APIs:
76501768b42SPeter Zijlstra  * mutex_lock_interruptible() and mutex_trylock().
76601768b42SPeter Zijlstra  */
76701768b42SPeter Zijlstra static noinline int __sched
76801768b42SPeter Zijlstra __mutex_lock_killable_slowpath(struct mutex *lock);
76901768b42SPeter Zijlstra 
77001768b42SPeter Zijlstra static noinline int __sched
77101768b42SPeter Zijlstra __mutex_lock_interruptible_slowpath(struct mutex *lock);
77201768b42SPeter Zijlstra 
77301768b42SPeter Zijlstra /**
77401768b42SPeter Zijlstra  * mutex_lock_interruptible - acquire the mutex, interruptible
77501768b42SPeter Zijlstra  * @lock: the mutex to be acquired
77601768b42SPeter Zijlstra  *
77701768b42SPeter Zijlstra  * Lock the mutex like mutex_lock(), and return 0 if the mutex has
77801768b42SPeter Zijlstra  * been acquired or sleep until the mutex becomes available. If a
77901768b42SPeter Zijlstra  * signal arrives while waiting for the lock then this function
78001768b42SPeter Zijlstra  * returns -EINTR.
78101768b42SPeter Zijlstra  *
78201768b42SPeter Zijlstra  * This function is similar to (but not equivalent to) down_interruptible().
78301768b42SPeter Zijlstra  */
78401768b42SPeter Zijlstra int __sched mutex_lock_interruptible(struct mutex *lock)
78501768b42SPeter Zijlstra {
78601768b42SPeter Zijlstra 	int ret;
78701768b42SPeter Zijlstra 
78801768b42SPeter Zijlstra 	might_sleep();
78901768b42SPeter Zijlstra 	ret =  __mutex_fastpath_lock_retval(&lock->count);
79001768b42SPeter Zijlstra 	if (likely(!ret)) {
79101768b42SPeter Zijlstra 		mutex_set_owner(lock);
79201768b42SPeter Zijlstra 		return 0;
79301768b42SPeter Zijlstra 	} else
79401768b42SPeter Zijlstra 		return __mutex_lock_interruptible_slowpath(lock);
79501768b42SPeter Zijlstra }
79601768b42SPeter Zijlstra 
79701768b42SPeter Zijlstra EXPORT_SYMBOL(mutex_lock_interruptible);
79801768b42SPeter Zijlstra 
79901768b42SPeter Zijlstra int __sched mutex_lock_killable(struct mutex *lock)
80001768b42SPeter Zijlstra {
80101768b42SPeter Zijlstra 	int ret;
80201768b42SPeter Zijlstra 
80301768b42SPeter Zijlstra 	might_sleep();
80401768b42SPeter Zijlstra 	ret = __mutex_fastpath_lock_retval(&lock->count);
80501768b42SPeter Zijlstra 	if (likely(!ret)) {
80601768b42SPeter Zijlstra 		mutex_set_owner(lock);
80701768b42SPeter Zijlstra 		return 0;
80801768b42SPeter Zijlstra 	} else
80901768b42SPeter Zijlstra 		return __mutex_lock_killable_slowpath(lock);
81001768b42SPeter Zijlstra }
81101768b42SPeter Zijlstra EXPORT_SYMBOL(mutex_lock_killable);
81201768b42SPeter Zijlstra 
81322d9fd34SAndi Kleen __visible void __sched
81401768b42SPeter Zijlstra __mutex_lock_slowpath(atomic_t *lock_count)
81501768b42SPeter Zijlstra {
81601768b42SPeter Zijlstra 	struct mutex *lock = container_of(lock_count, struct mutex, count);
81701768b42SPeter Zijlstra 
81801768b42SPeter Zijlstra 	__mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, 0,
81901768b42SPeter Zijlstra 			    NULL, _RET_IP_, NULL, 0);
82001768b42SPeter Zijlstra }
82101768b42SPeter Zijlstra 
82201768b42SPeter Zijlstra static noinline int __sched
82301768b42SPeter Zijlstra __mutex_lock_killable_slowpath(struct mutex *lock)
82401768b42SPeter Zijlstra {
82501768b42SPeter Zijlstra 	return __mutex_lock_common(lock, TASK_KILLABLE, 0,
82601768b42SPeter Zijlstra 				   NULL, _RET_IP_, NULL, 0);
82701768b42SPeter Zijlstra }
82801768b42SPeter Zijlstra 
82901768b42SPeter Zijlstra static noinline int __sched
83001768b42SPeter Zijlstra __mutex_lock_interruptible_slowpath(struct mutex *lock)
83101768b42SPeter Zijlstra {
83201768b42SPeter Zijlstra 	return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, 0,
83301768b42SPeter Zijlstra 				   NULL, _RET_IP_, NULL, 0);
83401768b42SPeter Zijlstra }
83501768b42SPeter Zijlstra 
83601768b42SPeter Zijlstra static noinline int __sched
83701768b42SPeter Zijlstra __ww_mutex_lock_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
83801768b42SPeter Zijlstra {
83901768b42SPeter Zijlstra 	return __mutex_lock_common(&lock->base, TASK_UNINTERRUPTIBLE, 0,
84001768b42SPeter Zijlstra 				   NULL, _RET_IP_, ctx, 1);
84101768b42SPeter Zijlstra }
84201768b42SPeter Zijlstra 
84301768b42SPeter Zijlstra static noinline int __sched
84401768b42SPeter Zijlstra __ww_mutex_lock_interruptible_slowpath(struct ww_mutex *lock,
84501768b42SPeter Zijlstra 					    struct ww_acquire_ctx *ctx)
84601768b42SPeter Zijlstra {
84701768b42SPeter Zijlstra 	return __mutex_lock_common(&lock->base, TASK_INTERRUPTIBLE, 0,
84801768b42SPeter Zijlstra 				   NULL, _RET_IP_, ctx, 1);
84901768b42SPeter Zijlstra }
85001768b42SPeter Zijlstra 
85101768b42SPeter Zijlstra #endif
85201768b42SPeter Zijlstra 
85301768b42SPeter Zijlstra /*
85401768b42SPeter Zijlstra  * Spinlock based trylock, we take the spinlock and check whether we
85501768b42SPeter Zijlstra  * can get the lock:
85601768b42SPeter Zijlstra  */
85701768b42SPeter Zijlstra static inline int __mutex_trylock_slowpath(atomic_t *lock_count)
85801768b42SPeter Zijlstra {
85901768b42SPeter Zijlstra 	struct mutex *lock = container_of(lock_count, struct mutex, count);
86001768b42SPeter Zijlstra 	unsigned long flags;
86101768b42SPeter Zijlstra 	int prev;
86201768b42SPeter Zijlstra 
86372d5305dSJason Low 	/* No need to trylock if the mutex is locked. */
86472d5305dSJason Low 	if (mutex_is_locked(lock))
86572d5305dSJason Low 		return 0;
86672d5305dSJason Low 
86701768b42SPeter Zijlstra 	spin_lock_mutex(&lock->wait_lock, flags);
86801768b42SPeter Zijlstra 
86901768b42SPeter Zijlstra 	prev = atomic_xchg(&lock->count, -1);
87001768b42SPeter Zijlstra 	if (likely(prev == 1)) {
87101768b42SPeter Zijlstra 		mutex_set_owner(lock);
87201768b42SPeter Zijlstra 		mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
87301768b42SPeter Zijlstra 	}
87401768b42SPeter Zijlstra 
87501768b42SPeter Zijlstra 	/* Set it back to 0 if there are no waiters: */
87601768b42SPeter Zijlstra 	if (likely(list_empty(&lock->wait_list)))
87701768b42SPeter Zijlstra 		atomic_set(&lock->count, 0);
87801768b42SPeter Zijlstra 
87901768b42SPeter Zijlstra 	spin_unlock_mutex(&lock->wait_lock, flags);
88001768b42SPeter Zijlstra 
88101768b42SPeter Zijlstra 	return prev == 1;
88201768b42SPeter Zijlstra }
88301768b42SPeter Zijlstra 
88401768b42SPeter Zijlstra /**
88501768b42SPeter Zijlstra  * mutex_trylock - try to acquire the mutex, without waiting
88601768b42SPeter Zijlstra  * @lock: the mutex to be acquired
88701768b42SPeter Zijlstra  *
88801768b42SPeter Zijlstra  * Try to acquire the mutex atomically. Returns 1 if the mutex
88901768b42SPeter Zijlstra  * has been acquired successfully, and 0 on contention.
89001768b42SPeter Zijlstra  *
89101768b42SPeter Zijlstra  * NOTE: this function follows the spin_trylock() convention, so
89201768b42SPeter Zijlstra  * it is negated from the down_trylock() return values! Be careful
89301768b42SPeter Zijlstra  * about this when converting semaphore users to mutexes.
89401768b42SPeter Zijlstra  *
89501768b42SPeter Zijlstra  * This function must not be used in interrupt context. The
89601768b42SPeter Zijlstra  * mutex must be released by the same task that acquired it.
89701768b42SPeter Zijlstra  */
89801768b42SPeter Zijlstra int __sched mutex_trylock(struct mutex *lock)
89901768b42SPeter Zijlstra {
90001768b42SPeter Zijlstra 	int ret;
90101768b42SPeter Zijlstra 
90201768b42SPeter Zijlstra 	ret = __mutex_fastpath_trylock(&lock->count, __mutex_trylock_slowpath);
90301768b42SPeter Zijlstra 	if (ret)
90401768b42SPeter Zijlstra 		mutex_set_owner(lock);
90501768b42SPeter Zijlstra 
90601768b42SPeter Zijlstra 	return ret;
90701768b42SPeter Zijlstra }
90801768b42SPeter Zijlstra EXPORT_SYMBOL(mutex_trylock);
90901768b42SPeter Zijlstra 
91001768b42SPeter Zijlstra #ifndef CONFIG_DEBUG_LOCK_ALLOC
91101768b42SPeter Zijlstra int __sched
91201768b42SPeter Zijlstra __ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
91301768b42SPeter Zijlstra {
91401768b42SPeter Zijlstra 	int ret;
91501768b42SPeter Zijlstra 
91601768b42SPeter Zijlstra 	might_sleep();
91701768b42SPeter Zijlstra 
91801768b42SPeter Zijlstra 	ret = __mutex_fastpath_lock_retval(&lock->base.count);
91901768b42SPeter Zijlstra 
92001768b42SPeter Zijlstra 	if (likely(!ret)) {
92101768b42SPeter Zijlstra 		ww_mutex_set_context_fastpath(lock, ctx);
92201768b42SPeter Zijlstra 		mutex_set_owner(&lock->base);
92301768b42SPeter Zijlstra 	} else
92401768b42SPeter Zijlstra 		ret = __ww_mutex_lock_slowpath(lock, ctx);
92501768b42SPeter Zijlstra 	return ret;
92601768b42SPeter Zijlstra }
92701768b42SPeter Zijlstra EXPORT_SYMBOL(__ww_mutex_lock);
92801768b42SPeter Zijlstra 
92901768b42SPeter Zijlstra int __sched
93001768b42SPeter Zijlstra __ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
93101768b42SPeter Zijlstra {
93201768b42SPeter Zijlstra 	int ret;
93301768b42SPeter Zijlstra 
93401768b42SPeter Zijlstra 	might_sleep();
93501768b42SPeter Zijlstra 
93601768b42SPeter Zijlstra 	ret = __mutex_fastpath_lock_retval(&lock->base.count);
93701768b42SPeter Zijlstra 
93801768b42SPeter Zijlstra 	if (likely(!ret)) {
93901768b42SPeter Zijlstra 		ww_mutex_set_context_fastpath(lock, ctx);
94001768b42SPeter Zijlstra 		mutex_set_owner(&lock->base);
94101768b42SPeter Zijlstra 	} else
94201768b42SPeter Zijlstra 		ret = __ww_mutex_lock_interruptible_slowpath(lock, ctx);
94301768b42SPeter Zijlstra 	return ret;
94401768b42SPeter Zijlstra }
94501768b42SPeter Zijlstra EXPORT_SYMBOL(__ww_mutex_lock_interruptible);
94601768b42SPeter Zijlstra 
94701768b42SPeter Zijlstra #endif
94801768b42SPeter Zijlstra 
94901768b42SPeter Zijlstra /**
95001768b42SPeter Zijlstra  * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0
95101768b42SPeter Zijlstra  * @cnt: the atomic which we are to dec
95201768b42SPeter Zijlstra  * @lock: the mutex to return holding if we dec to 0
95301768b42SPeter Zijlstra  *
95401768b42SPeter Zijlstra  * return true and hold lock if we dec to 0, return false otherwise
95501768b42SPeter Zijlstra  */
95601768b42SPeter Zijlstra int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock)
95701768b42SPeter Zijlstra {
95801768b42SPeter Zijlstra 	/* dec if we can't possibly hit 0 */
95901768b42SPeter Zijlstra 	if (atomic_add_unless(cnt, -1, 1))
96001768b42SPeter Zijlstra 		return 0;
96101768b42SPeter Zijlstra 	/* we might hit 0, so take the lock */
96201768b42SPeter Zijlstra 	mutex_lock(lock);
96301768b42SPeter Zijlstra 	if (!atomic_dec_and_test(cnt)) {
96401768b42SPeter Zijlstra 		/* when we actually did the dec, we didn't hit 0 */
96501768b42SPeter Zijlstra 		mutex_unlock(lock);
96601768b42SPeter Zijlstra 		return 0;
96701768b42SPeter Zijlstra 	}
96801768b42SPeter Zijlstra 	/* we hit 0, and we hold the lock */
96901768b42SPeter Zijlstra 	return 1;
97001768b42SPeter Zijlstra }
97101768b42SPeter Zijlstra EXPORT_SYMBOL(atomic_dec_and_mutex_lock);
972