xref: /openbmc/linux/kernel/locking/mutex.c (revision f3956ebb)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * kernel/locking/mutex.c
4  *
5  * Mutexes: blocking mutual exclusion locks
6  *
7  * Started by Ingo Molnar:
8  *
9  *  Copyright (C) 2004, 2005, 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
10  *
11  * Many thanks to Arjan van de Ven, Thomas Gleixner, Steven Rostedt and
12  * David Howells for suggestions and improvements.
13  *
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
17  *    and Sven Dietrich.
18  *
19  * Also see Documentation/locking/mutex-design.rst.
20  */
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>
32 
33 #ifndef CONFIG_PREEMPT_RT
34 #include "mutex.h"
35 
36 #ifdef CONFIG_DEBUG_MUTEXES
37 # define MUTEX_WARN_ON(cond) DEBUG_LOCKS_WARN_ON(cond)
38 #else
39 # define MUTEX_WARN_ON(cond)
40 #endif
41 
42 void
43 __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
44 {
45 	atomic_long_set(&lock->owner, 0);
46 	raw_spin_lock_init(&lock->wait_lock);
47 	INIT_LIST_HEAD(&lock->wait_list);
48 #ifdef CONFIG_MUTEX_SPIN_ON_OWNER
49 	osq_lock_init(&lock->osq);
50 #endif
51 
52 	debug_mutex_init(lock, name, key);
53 }
54 EXPORT_SYMBOL(__mutex_init);
55 
56 /*
57  * @owner: contains: 'struct task_struct *' to the current lock owner,
58  * NULL means not owned. Since task_struct pointers are aligned at
59  * at least L1_CACHE_BYTES, we have low bits to store extra state.
60  *
61  * Bit0 indicates a non-empty waiter list; unlock must issue a wakeup.
62  * Bit1 indicates unlock needs to hand the lock to the top-waiter
63  * Bit2 indicates handoff has been done and we're waiting for pickup.
64  */
65 #define MUTEX_FLAG_WAITERS	0x01
66 #define MUTEX_FLAG_HANDOFF	0x02
67 #define MUTEX_FLAG_PICKUP	0x04
68 
69 #define MUTEX_FLAGS		0x07
70 
71 /*
72  * Internal helper function; C doesn't allow us to hide it :/
73  *
74  * DO NOT USE (outside of mutex code).
75  */
76 static inline struct task_struct *__mutex_owner(struct mutex *lock)
77 {
78 	return (struct task_struct *)(atomic_long_read(&lock->owner) & ~MUTEX_FLAGS);
79 }
80 
81 static inline struct task_struct *__owner_task(unsigned long owner)
82 {
83 	return (struct task_struct *)(owner & ~MUTEX_FLAGS);
84 }
85 
86 bool mutex_is_locked(struct mutex *lock)
87 {
88 	return __mutex_owner(lock) != NULL;
89 }
90 EXPORT_SYMBOL(mutex_is_locked);
91 
92 static inline unsigned long __owner_flags(unsigned long owner)
93 {
94 	return owner & MUTEX_FLAGS;
95 }
96 
97 static inline struct task_struct *__mutex_trylock_common(struct mutex *lock, bool handoff)
98 {
99 	unsigned long owner, curr = (unsigned long)current;
100 
101 	owner = atomic_long_read(&lock->owner);
102 	for (;;) { /* must loop, can race against a flag */
103 		unsigned long flags = __owner_flags(owner);
104 		unsigned long task = owner & ~MUTEX_FLAGS;
105 
106 		if (task) {
107 			if (flags & MUTEX_FLAG_PICKUP) {
108 				if (task != curr)
109 					break;
110 				flags &= ~MUTEX_FLAG_PICKUP;
111 			} else if (handoff) {
112 				if (flags & MUTEX_FLAG_HANDOFF)
113 					break;
114 				flags |= MUTEX_FLAG_HANDOFF;
115 			} else {
116 				break;
117 			}
118 		} else {
119 			MUTEX_WARN_ON(flags & (MUTEX_FLAG_HANDOFF | MUTEX_FLAG_PICKUP));
120 			task = curr;
121 		}
122 
123 		if (atomic_long_try_cmpxchg_acquire(&lock->owner, &owner, task | flags)) {
124 			if (task == curr)
125 				return NULL;
126 			break;
127 		}
128 	}
129 
130 	return __owner_task(owner);
131 }
132 
133 /*
134  * Trylock or set HANDOFF
135  */
136 static inline bool __mutex_trylock_or_handoff(struct mutex *lock, bool handoff)
137 {
138 	return !__mutex_trylock_common(lock, handoff);
139 }
140 
141 /*
142  * Actual trylock that will work on any unlocked state.
143  */
144 static inline bool __mutex_trylock(struct mutex *lock)
145 {
146 	return !__mutex_trylock_common(lock, false);
147 }
148 
149 #ifndef CONFIG_DEBUG_LOCK_ALLOC
150 /*
151  * Lockdep annotations are contained to the slow paths for simplicity.
152  * There is nothing that would stop spreading the lockdep annotations outwards
153  * except more code.
154  */
155 
156 /*
157  * Optimistic trylock that only works in the uncontended case. Make sure to
158  * follow with a __mutex_trylock() before failing.
159  */
160 static __always_inline bool __mutex_trylock_fast(struct mutex *lock)
161 {
162 	unsigned long curr = (unsigned long)current;
163 	unsigned long zero = 0UL;
164 
165 	if (atomic_long_try_cmpxchg_acquire(&lock->owner, &zero, curr))
166 		return true;
167 
168 	return false;
169 }
170 
171 static __always_inline bool __mutex_unlock_fast(struct mutex *lock)
172 {
173 	unsigned long curr = (unsigned long)current;
174 
175 	return atomic_long_try_cmpxchg_release(&lock->owner, &curr, 0UL);
176 }
177 #endif
178 
179 static inline void __mutex_set_flag(struct mutex *lock, unsigned long flag)
180 {
181 	atomic_long_or(flag, &lock->owner);
182 }
183 
184 static inline void __mutex_clear_flag(struct mutex *lock, unsigned long flag)
185 {
186 	atomic_long_andnot(flag, &lock->owner);
187 }
188 
189 static inline bool __mutex_waiter_is_first(struct mutex *lock, struct mutex_waiter *waiter)
190 {
191 	return list_first_entry(&lock->wait_list, struct mutex_waiter, list) == waiter;
192 }
193 
194 /*
195  * Add @waiter to a given location in the lock wait_list and set the
196  * FLAG_WAITERS flag if it's the first waiter.
197  */
198 static void
199 __mutex_add_waiter(struct mutex *lock, struct mutex_waiter *waiter,
200 		   struct list_head *list)
201 {
202 	debug_mutex_add_waiter(lock, waiter, current);
203 
204 	list_add_tail(&waiter->list, list);
205 	if (__mutex_waiter_is_first(lock, waiter))
206 		__mutex_set_flag(lock, MUTEX_FLAG_WAITERS);
207 }
208 
209 static void
210 __mutex_remove_waiter(struct mutex *lock, struct mutex_waiter *waiter)
211 {
212 	list_del(&waiter->list);
213 	if (likely(list_empty(&lock->wait_list)))
214 		__mutex_clear_flag(lock, MUTEX_FLAGS);
215 
216 	debug_mutex_remove_waiter(lock, waiter, current);
217 }
218 
219 /*
220  * Give up ownership to a specific task, when @task = NULL, this is equivalent
221  * to a regular unlock. Sets PICKUP on a handoff, clears HANDOFF, preserves
222  * WAITERS. Provides RELEASE semantics like a regular unlock, the
223  * __mutex_trylock() provides a matching ACQUIRE semantics for the handoff.
224  */
225 static void __mutex_handoff(struct mutex *lock, struct task_struct *task)
226 {
227 	unsigned long owner = atomic_long_read(&lock->owner);
228 
229 	for (;;) {
230 		unsigned long new;
231 
232 		MUTEX_WARN_ON(__owner_task(owner) != current);
233 		MUTEX_WARN_ON(owner & MUTEX_FLAG_PICKUP);
234 
235 		new = (owner & MUTEX_FLAG_WAITERS);
236 		new |= (unsigned long)task;
237 		if (task)
238 			new |= MUTEX_FLAG_PICKUP;
239 
240 		if (atomic_long_try_cmpxchg_release(&lock->owner, &owner, new))
241 			break;
242 	}
243 }
244 
245 #ifndef CONFIG_DEBUG_LOCK_ALLOC
246 /*
247  * We split the mutex lock/unlock logic into separate fastpath and
248  * slowpath functions, to reduce the register pressure on the fastpath.
249  * We also put the fastpath first in the kernel image, to make sure the
250  * branch is predicted by the CPU as default-untaken.
251  */
252 static void __sched __mutex_lock_slowpath(struct mutex *lock);
253 
254 /**
255  * mutex_lock - acquire the mutex
256  * @lock: the mutex to be acquired
257  *
258  * Lock the mutex exclusively for this task. If the mutex is not
259  * available right now, it will sleep until it can get it.
260  *
261  * The mutex must later on be released by the same task that
262  * acquired it. Recursive locking is not allowed. The task
263  * may not exit without first unlocking the mutex. Also, kernel
264  * memory where the mutex resides must not be freed with
265  * the mutex still locked. The mutex must first be initialized
266  * (or statically defined) before it can be locked. memset()-ing
267  * the mutex to 0 is not allowed.
268  *
269  * (The CONFIG_DEBUG_MUTEXES .config option turns on debugging
270  * checks that will enforce the restrictions and will also do
271  * deadlock debugging)
272  *
273  * This function is similar to (but not equivalent to) down().
274  */
275 void __sched mutex_lock(struct mutex *lock)
276 {
277 	might_sleep();
278 
279 	if (!__mutex_trylock_fast(lock))
280 		__mutex_lock_slowpath(lock);
281 }
282 EXPORT_SYMBOL(mutex_lock);
283 #endif
284 
285 #include "ww_mutex.h"
286 
287 #ifdef CONFIG_MUTEX_SPIN_ON_OWNER
288 
289 /*
290  * Trylock variant that returns the owning task on failure.
291  */
292 static inline struct task_struct *__mutex_trylock_or_owner(struct mutex *lock)
293 {
294 	return __mutex_trylock_common(lock, false);
295 }
296 
297 static inline
298 bool ww_mutex_spin_on_owner(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
299 			    struct mutex_waiter *waiter)
300 {
301 	struct ww_mutex *ww;
302 
303 	ww = container_of(lock, struct ww_mutex, base);
304 
305 	/*
306 	 * If ww->ctx is set the contents are undefined, only
307 	 * by acquiring wait_lock there is a guarantee that
308 	 * they are not invalid when reading.
309 	 *
310 	 * As such, when deadlock detection needs to be
311 	 * performed the optimistic spinning cannot be done.
312 	 *
313 	 * Check this in every inner iteration because we may
314 	 * be racing against another thread's ww_mutex_lock.
315 	 */
316 	if (ww_ctx->acquired > 0 && READ_ONCE(ww->ctx))
317 		return false;
318 
319 	/*
320 	 * If we aren't on the wait list yet, cancel the spin
321 	 * if there are waiters. We want  to avoid stealing the
322 	 * lock from a waiter with an earlier stamp, since the
323 	 * other thread may already own a lock that we also
324 	 * need.
325 	 */
326 	if (!waiter && (atomic_long_read(&lock->owner) & MUTEX_FLAG_WAITERS))
327 		return false;
328 
329 	/*
330 	 * Similarly, stop spinning if we are no longer the
331 	 * first waiter.
332 	 */
333 	if (waiter && !__mutex_waiter_is_first(lock, waiter))
334 		return false;
335 
336 	return true;
337 }
338 
339 /*
340  * Look out! "owner" is an entirely speculative pointer access and not
341  * reliable.
342  *
343  * "noinline" so that this function shows up on perf profiles.
344  */
345 static noinline
346 bool mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner,
347 			 struct ww_acquire_ctx *ww_ctx, struct mutex_waiter *waiter)
348 {
349 	bool ret = true;
350 
351 	rcu_read_lock();
352 	while (__mutex_owner(lock) == owner) {
353 		/*
354 		 * Ensure we emit the owner->on_cpu, dereference _after_
355 		 * checking lock->owner still matches owner. If that fails,
356 		 * owner might point to freed memory. If it still matches,
357 		 * the rcu_read_lock() ensures the memory stays valid.
358 		 */
359 		barrier();
360 
361 		/*
362 		 * Use vcpu_is_preempted to detect lock holder preemption issue.
363 		 */
364 		if (!owner->on_cpu || need_resched() ||
365 				vcpu_is_preempted(task_cpu(owner))) {
366 			ret = false;
367 			break;
368 		}
369 
370 		if (ww_ctx && !ww_mutex_spin_on_owner(lock, ww_ctx, waiter)) {
371 			ret = false;
372 			break;
373 		}
374 
375 		cpu_relax();
376 	}
377 	rcu_read_unlock();
378 
379 	return ret;
380 }
381 
382 /*
383  * Initial check for entering the mutex spinning loop
384  */
385 static inline int mutex_can_spin_on_owner(struct mutex *lock)
386 {
387 	struct task_struct *owner;
388 	int retval = 1;
389 
390 	if (need_resched())
391 		return 0;
392 
393 	rcu_read_lock();
394 	owner = __mutex_owner(lock);
395 
396 	/*
397 	 * As lock holder preemption issue, we both skip spinning if task is not
398 	 * on cpu or its cpu is preempted
399 	 */
400 	if (owner)
401 		retval = owner->on_cpu && !vcpu_is_preempted(task_cpu(owner));
402 	rcu_read_unlock();
403 
404 	/*
405 	 * If lock->owner is not set, the mutex has been released. Return true
406 	 * such that we'll trylock in the spin path, which is a faster option
407 	 * than the blocking slow path.
408 	 */
409 	return retval;
410 }
411 
412 /*
413  * Optimistic spinning.
414  *
415  * We try to spin for acquisition when we find that the lock owner
416  * is currently running on a (different) CPU and while we don't
417  * need to reschedule. The rationale is that if the lock owner is
418  * running, it is likely to release the lock soon.
419  *
420  * The mutex spinners are queued up using MCS lock so that only one
421  * spinner can compete for the mutex. However, if mutex spinning isn't
422  * going to happen, there is no point in going through the lock/unlock
423  * overhead.
424  *
425  * Returns true when the lock was taken, otherwise false, indicating
426  * that we need to jump to the slowpath and sleep.
427  *
428  * The waiter flag is set to true if the spinner is a waiter in the wait
429  * queue. The waiter-spinner will spin on the lock directly and concurrently
430  * with the spinner at the head of the OSQ, if present, until the owner is
431  * changed to itself.
432  */
433 static __always_inline bool
434 mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
435 		      struct mutex_waiter *waiter)
436 {
437 	if (!waiter) {
438 		/*
439 		 * The purpose of the mutex_can_spin_on_owner() function is
440 		 * to eliminate the overhead of osq_lock() and osq_unlock()
441 		 * in case spinning isn't possible. As a waiter-spinner
442 		 * is not going to take OSQ lock anyway, there is no need
443 		 * to call mutex_can_spin_on_owner().
444 		 */
445 		if (!mutex_can_spin_on_owner(lock))
446 			goto fail;
447 
448 		/*
449 		 * In order to avoid a stampede of mutex spinners trying to
450 		 * acquire the mutex all at once, the spinners need to take a
451 		 * MCS (queued) lock first before spinning on the owner field.
452 		 */
453 		if (!osq_lock(&lock->osq))
454 			goto fail;
455 	}
456 
457 	for (;;) {
458 		struct task_struct *owner;
459 
460 		/* Try to acquire the mutex... */
461 		owner = __mutex_trylock_or_owner(lock);
462 		if (!owner)
463 			break;
464 
465 		/*
466 		 * There's an owner, wait for it to either
467 		 * release the lock or go to sleep.
468 		 */
469 		if (!mutex_spin_on_owner(lock, owner, ww_ctx, waiter))
470 			goto fail_unlock;
471 
472 		/*
473 		 * The cpu_relax() call is a compiler barrier which forces
474 		 * everything in this loop to be re-loaded. We don't need
475 		 * memory barriers as we'll eventually observe the right
476 		 * values at the cost of a few extra spins.
477 		 */
478 		cpu_relax();
479 	}
480 
481 	if (!waiter)
482 		osq_unlock(&lock->osq);
483 
484 	return true;
485 
486 
487 fail_unlock:
488 	if (!waiter)
489 		osq_unlock(&lock->osq);
490 
491 fail:
492 	/*
493 	 * If we fell out of the spin path because of need_resched(),
494 	 * reschedule now, before we try-lock the mutex. This avoids getting
495 	 * scheduled out right after we obtained the mutex.
496 	 */
497 	if (need_resched()) {
498 		/*
499 		 * We _should_ have TASK_RUNNING here, but just in case
500 		 * we do not, make it so, otherwise we might get stuck.
501 		 */
502 		__set_current_state(TASK_RUNNING);
503 		schedule_preempt_disabled();
504 	}
505 
506 	return false;
507 }
508 #else
509 static __always_inline bool
510 mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
511 		      struct mutex_waiter *waiter)
512 {
513 	return false;
514 }
515 #endif
516 
517 static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip);
518 
519 /**
520  * mutex_unlock - release the mutex
521  * @lock: the mutex to be released
522  *
523  * Unlock a mutex that has been locked by this task previously.
524  *
525  * This function must not be used in interrupt context. Unlocking
526  * of a not locked mutex is not allowed.
527  *
528  * This function is similar to (but not equivalent to) up().
529  */
530 void __sched mutex_unlock(struct mutex *lock)
531 {
532 #ifndef CONFIG_DEBUG_LOCK_ALLOC
533 	if (__mutex_unlock_fast(lock))
534 		return;
535 #endif
536 	__mutex_unlock_slowpath(lock, _RET_IP_);
537 }
538 EXPORT_SYMBOL(mutex_unlock);
539 
540 /**
541  * ww_mutex_unlock - release the w/w mutex
542  * @lock: the mutex to be released
543  *
544  * Unlock a mutex that has been locked by this task previously with any of the
545  * ww_mutex_lock* functions (with or without an acquire context). It is
546  * forbidden to release the locks after releasing the acquire context.
547  *
548  * This function must not be used in interrupt context. Unlocking
549  * of a unlocked mutex is not allowed.
550  */
551 void __sched ww_mutex_unlock(struct ww_mutex *lock)
552 {
553 	__ww_mutex_unlock(lock);
554 	mutex_unlock(&lock->base);
555 }
556 EXPORT_SYMBOL(ww_mutex_unlock);
557 
558 /*
559  * Lock a mutex (possibly interruptible), slowpath:
560  */
561 static __always_inline int __sched
562 __mutex_lock_common(struct mutex *lock, unsigned int state, unsigned int subclass,
563 		    struct lockdep_map *nest_lock, unsigned long ip,
564 		    struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
565 {
566 	struct mutex_waiter waiter;
567 	struct ww_mutex *ww;
568 	int ret;
569 
570 	if (!use_ww_ctx)
571 		ww_ctx = NULL;
572 
573 	might_sleep();
574 
575 	MUTEX_WARN_ON(lock->magic != lock);
576 
577 	ww = container_of(lock, struct ww_mutex, base);
578 	if (ww_ctx) {
579 		if (unlikely(ww_ctx == READ_ONCE(ww->ctx)))
580 			return -EALREADY;
581 
582 		/*
583 		 * Reset the wounded flag after a kill. No other process can
584 		 * race and wound us here since they can't have a valid owner
585 		 * pointer if we don't have any locks held.
586 		 */
587 		if (ww_ctx->acquired == 0)
588 			ww_ctx->wounded = 0;
589 
590 #ifdef CONFIG_DEBUG_LOCK_ALLOC
591 		nest_lock = &ww_ctx->dep_map;
592 #endif
593 	}
594 
595 	preempt_disable();
596 	mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);
597 
598 	if (__mutex_trylock(lock) ||
599 	    mutex_optimistic_spin(lock, ww_ctx, NULL)) {
600 		/* got the lock, yay! */
601 		lock_acquired(&lock->dep_map, ip);
602 		if (ww_ctx)
603 			ww_mutex_set_context_fastpath(ww, ww_ctx);
604 		preempt_enable();
605 		return 0;
606 	}
607 
608 	raw_spin_lock(&lock->wait_lock);
609 	/*
610 	 * After waiting to acquire the wait_lock, try again.
611 	 */
612 	if (__mutex_trylock(lock)) {
613 		if (ww_ctx)
614 			__ww_mutex_check_waiters(lock, ww_ctx);
615 
616 		goto skip_wait;
617 	}
618 
619 	debug_mutex_lock_common(lock, &waiter);
620 	waiter.task = current;
621 	if (use_ww_ctx)
622 		waiter.ww_ctx = ww_ctx;
623 
624 	lock_contended(&lock->dep_map, ip);
625 
626 	if (!use_ww_ctx) {
627 		/* add waiting tasks to the end of the waitqueue (FIFO): */
628 		__mutex_add_waiter(lock, &waiter, &lock->wait_list);
629 	} else {
630 		/*
631 		 * Add in stamp order, waking up waiters that must kill
632 		 * themselves.
633 		 */
634 		ret = __ww_mutex_add_waiter(&waiter, lock, ww_ctx);
635 		if (ret)
636 			goto err_early_kill;
637 	}
638 
639 	set_current_state(state);
640 	for (;;) {
641 		bool first;
642 
643 		/*
644 		 * Once we hold wait_lock, we're serialized against
645 		 * mutex_unlock() handing the lock off to us, do a trylock
646 		 * before testing the error conditions to make sure we pick up
647 		 * the handoff.
648 		 */
649 		if (__mutex_trylock(lock))
650 			goto acquired;
651 
652 		/*
653 		 * Check for signals and kill conditions while holding
654 		 * wait_lock. This ensures the lock cancellation is ordered
655 		 * against mutex_unlock() and wake-ups do not go missing.
656 		 */
657 		if (signal_pending_state(state, current)) {
658 			ret = -EINTR;
659 			goto err;
660 		}
661 
662 		if (ww_ctx) {
663 			ret = __ww_mutex_check_kill(lock, &waiter, ww_ctx);
664 			if (ret)
665 				goto err;
666 		}
667 
668 		raw_spin_unlock(&lock->wait_lock);
669 		schedule_preempt_disabled();
670 
671 		first = __mutex_waiter_is_first(lock, &waiter);
672 
673 		set_current_state(state);
674 		/*
675 		 * Here we order against unlock; we must either see it change
676 		 * state back to RUNNING and fall through the next schedule(),
677 		 * or we must see its unlock and acquire.
678 		 */
679 		if (__mutex_trylock_or_handoff(lock, first) ||
680 		    (first && mutex_optimistic_spin(lock, ww_ctx, &waiter)))
681 			break;
682 
683 		raw_spin_lock(&lock->wait_lock);
684 	}
685 	raw_spin_lock(&lock->wait_lock);
686 acquired:
687 	__set_current_state(TASK_RUNNING);
688 
689 	if (ww_ctx) {
690 		/*
691 		 * Wound-Wait; we stole the lock (!first_waiter), check the
692 		 * waiters as anyone might want to wound us.
693 		 */
694 		if (!ww_ctx->is_wait_die &&
695 		    !__mutex_waiter_is_first(lock, &waiter))
696 			__ww_mutex_check_waiters(lock, ww_ctx);
697 	}
698 
699 	__mutex_remove_waiter(lock, &waiter);
700 
701 	debug_mutex_free_waiter(&waiter);
702 
703 skip_wait:
704 	/* got the lock - cleanup and rejoice! */
705 	lock_acquired(&lock->dep_map, ip);
706 
707 	if (ww_ctx)
708 		ww_mutex_lock_acquired(ww, ww_ctx);
709 
710 	raw_spin_unlock(&lock->wait_lock);
711 	preempt_enable();
712 	return 0;
713 
714 err:
715 	__set_current_state(TASK_RUNNING);
716 	__mutex_remove_waiter(lock, &waiter);
717 err_early_kill:
718 	raw_spin_unlock(&lock->wait_lock);
719 	debug_mutex_free_waiter(&waiter);
720 	mutex_release(&lock->dep_map, ip);
721 	preempt_enable();
722 	return ret;
723 }
724 
725 static int __sched
726 __mutex_lock(struct mutex *lock, unsigned int state, unsigned int subclass,
727 	     struct lockdep_map *nest_lock, unsigned long ip)
728 {
729 	return __mutex_lock_common(lock, state, subclass, nest_lock, ip, NULL, false);
730 }
731 
732 static int __sched
733 __ww_mutex_lock(struct mutex *lock, unsigned int state, unsigned int subclass,
734 		unsigned long ip, struct ww_acquire_ctx *ww_ctx)
735 {
736 	return __mutex_lock_common(lock, state, subclass, NULL, ip, ww_ctx, true);
737 }
738 
739 #ifdef CONFIG_DEBUG_LOCK_ALLOC
740 void __sched
741 mutex_lock_nested(struct mutex *lock, unsigned int subclass)
742 {
743 	__mutex_lock(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_);
744 }
745 
746 EXPORT_SYMBOL_GPL(mutex_lock_nested);
747 
748 void __sched
749 _mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest)
750 {
751 	__mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, nest, _RET_IP_);
752 }
753 EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock);
754 
755 int __sched
756 mutex_lock_killable_nested(struct mutex *lock, unsigned int subclass)
757 {
758 	return __mutex_lock(lock, TASK_KILLABLE, subclass, NULL, _RET_IP_);
759 }
760 EXPORT_SYMBOL_GPL(mutex_lock_killable_nested);
761 
762 int __sched
763 mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
764 {
765 	return __mutex_lock(lock, TASK_INTERRUPTIBLE, subclass, NULL, _RET_IP_);
766 }
767 EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
768 
769 void __sched
770 mutex_lock_io_nested(struct mutex *lock, unsigned int subclass)
771 {
772 	int token;
773 
774 	might_sleep();
775 
776 	token = io_schedule_prepare();
777 	__mutex_lock_common(lock, TASK_UNINTERRUPTIBLE,
778 			    subclass, NULL, _RET_IP_, NULL, 0);
779 	io_schedule_finish(token);
780 }
781 EXPORT_SYMBOL_GPL(mutex_lock_io_nested);
782 
783 static inline int
784 ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
785 {
786 #ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH
787 	unsigned tmp;
788 
789 	if (ctx->deadlock_inject_countdown-- == 0) {
790 		tmp = ctx->deadlock_inject_interval;
791 		if (tmp > UINT_MAX/4)
792 			tmp = UINT_MAX;
793 		else
794 			tmp = tmp*2 + tmp + tmp/2;
795 
796 		ctx->deadlock_inject_interval = tmp;
797 		ctx->deadlock_inject_countdown = tmp;
798 		ctx->contending_lock = lock;
799 
800 		ww_mutex_unlock(lock);
801 
802 		return -EDEADLK;
803 	}
804 #endif
805 
806 	return 0;
807 }
808 
809 int __sched
810 ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
811 {
812 	int ret;
813 
814 	might_sleep();
815 	ret =  __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE,
816 			       0, _RET_IP_, ctx);
817 	if (!ret && ctx && ctx->acquired > 1)
818 		return ww_mutex_deadlock_injection(lock, ctx);
819 
820 	return ret;
821 }
822 EXPORT_SYMBOL_GPL(ww_mutex_lock);
823 
824 int __sched
825 ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
826 {
827 	int ret;
828 
829 	might_sleep();
830 	ret = __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE,
831 			      0, _RET_IP_, ctx);
832 
833 	if (!ret && ctx && ctx->acquired > 1)
834 		return ww_mutex_deadlock_injection(lock, ctx);
835 
836 	return ret;
837 }
838 EXPORT_SYMBOL_GPL(ww_mutex_lock_interruptible);
839 
840 #endif
841 
842 /*
843  * Release the lock, slowpath:
844  */
845 static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip)
846 {
847 	struct task_struct *next = NULL;
848 	DEFINE_WAKE_Q(wake_q);
849 	unsigned long owner;
850 
851 	mutex_release(&lock->dep_map, ip);
852 
853 	/*
854 	 * Release the lock before (potentially) taking the spinlock such that
855 	 * other contenders can get on with things ASAP.
856 	 *
857 	 * Except when HANDOFF, in that case we must not clear the owner field,
858 	 * but instead set it to the top waiter.
859 	 */
860 	owner = atomic_long_read(&lock->owner);
861 	for (;;) {
862 		MUTEX_WARN_ON(__owner_task(owner) != current);
863 		MUTEX_WARN_ON(owner & MUTEX_FLAG_PICKUP);
864 
865 		if (owner & MUTEX_FLAG_HANDOFF)
866 			break;
867 
868 		if (atomic_long_try_cmpxchg_release(&lock->owner, &owner, __owner_flags(owner))) {
869 			if (owner & MUTEX_FLAG_WAITERS)
870 				break;
871 
872 			return;
873 		}
874 	}
875 
876 	raw_spin_lock(&lock->wait_lock);
877 	debug_mutex_unlock(lock);
878 	if (!list_empty(&lock->wait_list)) {
879 		/* get the first entry from the wait-list: */
880 		struct mutex_waiter *waiter =
881 			list_first_entry(&lock->wait_list,
882 					 struct mutex_waiter, list);
883 
884 		next = waiter->task;
885 
886 		debug_mutex_wake_waiter(lock, waiter);
887 		wake_q_add(&wake_q, next);
888 	}
889 
890 	if (owner & MUTEX_FLAG_HANDOFF)
891 		__mutex_handoff(lock, next);
892 
893 	raw_spin_unlock(&lock->wait_lock);
894 
895 	wake_up_q(&wake_q);
896 }
897 
898 #ifndef CONFIG_DEBUG_LOCK_ALLOC
899 /*
900  * Here come the less common (and hence less performance-critical) APIs:
901  * mutex_lock_interruptible() and mutex_trylock().
902  */
903 static noinline int __sched
904 __mutex_lock_killable_slowpath(struct mutex *lock);
905 
906 static noinline int __sched
907 __mutex_lock_interruptible_slowpath(struct mutex *lock);
908 
909 /**
910  * mutex_lock_interruptible() - Acquire the mutex, interruptible by signals.
911  * @lock: The mutex to be acquired.
912  *
913  * Lock the mutex like mutex_lock().  If a signal is delivered while the
914  * process is sleeping, this function will return without acquiring the
915  * mutex.
916  *
917  * Context: Process context.
918  * Return: 0 if the lock was successfully acquired or %-EINTR if a
919  * signal arrived.
920  */
921 int __sched mutex_lock_interruptible(struct mutex *lock)
922 {
923 	might_sleep();
924 
925 	if (__mutex_trylock_fast(lock))
926 		return 0;
927 
928 	return __mutex_lock_interruptible_slowpath(lock);
929 }
930 
931 EXPORT_SYMBOL(mutex_lock_interruptible);
932 
933 /**
934  * mutex_lock_killable() - Acquire the mutex, interruptible by fatal signals.
935  * @lock: The mutex to be acquired.
936  *
937  * Lock the mutex like mutex_lock().  If a signal which will be fatal to
938  * the current process is delivered while the process is sleeping, this
939  * function will return without acquiring the mutex.
940  *
941  * Context: Process context.
942  * Return: 0 if the lock was successfully acquired or %-EINTR if a
943  * fatal signal arrived.
944  */
945 int __sched mutex_lock_killable(struct mutex *lock)
946 {
947 	might_sleep();
948 
949 	if (__mutex_trylock_fast(lock))
950 		return 0;
951 
952 	return __mutex_lock_killable_slowpath(lock);
953 }
954 EXPORT_SYMBOL(mutex_lock_killable);
955 
956 /**
957  * mutex_lock_io() - Acquire the mutex and mark the process as waiting for I/O
958  * @lock: The mutex to be acquired.
959  *
960  * Lock the mutex like mutex_lock().  While the task is waiting for this
961  * mutex, it will be accounted as being in the IO wait state by the
962  * scheduler.
963  *
964  * Context: Process context.
965  */
966 void __sched mutex_lock_io(struct mutex *lock)
967 {
968 	int token;
969 
970 	token = io_schedule_prepare();
971 	mutex_lock(lock);
972 	io_schedule_finish(token);
973 }
974 EXPORT_SYMBOL_GPL(mutex_lock_io);
975 
976 static noinline void __sched
977 __mutex_lock_slowpath(struct mutex *lock)
978 {
979 	__mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_);
980 }
981 
982 static noinline int __sched
983 __mutex_lock_killable_slowpath(struct mutex *lock)
984 {
985 	return __mutex_lock(lock, TASK_KILLABLE, 0, NULL, _RET_IP_);
986 }
987 
988 static noinline int __sched
989 __mutex_lock_interruptible_slowpath(struct mutex *lock)
990 {
991 	return __mutex_lock(lock, TASK_INTERRUPTIBLE, 0, NULL, _RET_IP_);
992 }
993 
994 static noinline int __sched
995 __ww_mutex_lock_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
996 {
997 	return __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE, 0,
998 			       _RET_IP_, ctx);
999 }
1000 
1001 static noinline int __sched
1002 __ww_mutex_lock_interruptible_slowpath(struct ww_mutex *lock,
1003 					    struct ww_acquire_ctx *ctx)
1004 {
1005 	return __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE, 0,
1006 			       _RET_IP_, ctx);
1007 }
1008 
1009 #endif
1010 
1011 /**
1012  * mutex_trylock - try to acquire the mutex, without waiting
1013  * @lock: the mutex to be acquired
1014  *
1015  * Try to acquire the mutex atomically. Returns 1 if the mutex
1016  * has been acquired successfully, and 0 on contention.
1017  *
1018  * NOTE: this function follows the spin_trylock() convention, so
1019  * it is negated from the down_trylock() return values! Be careful
1020  * about this when converting semaphore users to mutexes.
1021  *
1022  * This function must not be used in interrupt context. The
1023  * mutex must be released by the same task that acquired it.
1024  */
1025 int __sched mutex_trylock(struct mutex *lock)
1026 {
1027 	bool locked;
1028 
1029 	MUTEX_WARN_ON(lock->magic != lock);
1030 
1031 	locked = __mutex_trylock(lock);
1032 	if (locked)
1033 		mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
1034 
1035 	return locked;
1036 }
1037 EXPORT_SYMBOL(mutex_trylock);
1038 
1039 #ifndef CONFIG_DEBUG_LOCK_ALLOC
1040 int __sched
1041 ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1042 {
1043 	might_sleep();
1044 
1045 	if (__mutex_trylock_fast(&lock->base)) {
1046 		if (ctx)
1047 			ww_mutex_set_context_fastpath(lock, ctx);
1048 		return 0;
1049 	}
1050 
1051 	return __ww_mutex_lock_slowpath(lock, ctx);
1052 }
1053 EXPORT_SYMBOL(ww_mutex_lock);
1054 
1055 int __sched
1056 ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1057 {
1058 	might_sleep();
1059 
1060 	if (__mutex_trylock_fast(&lock->base)) {
1061 		if (ctx)
1062 			ww_mutex_set_context_fastpath(lock, ctx);
1063 		return 0;
1064 	}
1065 
1066 	return __ww_mutex_lock_interruptible_slowpath(lock, ctx);
1067 }
1068 EXPORT_SYMBOL(ww_mutex_lock_interruptible);
1069 
1070 #endif /* !CONFIG_DEBUG_LOCK_ALLOC */
1071 #endif /* !CONFIG_PREEMPT_RT */
1072 
1073 /**
1074  * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0
1075  * @cnt: the atomic which we are to dec
1076  * @lock: the mutex to return holding if we dec to 0
1077  *
1078  * return true and hold lock if we dec to 0, return false otherwise
1079  */
1080 int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock)
1081 {
1082 	/* dec if we can't possibly hit 0 */
1083 	if (atomic_add_unless(cnt, -1, 1))
1084 		return 0;
1085 	/* we might hit 0, so take the lock */
1086 	mutex_lock(lock);
1087 	if (!atomic_dec_and_test(cnt)) {
1088 		/* when we actually did the dec, we didn't hit 0 */
1089 		mutex_unlock(lock);
1090 		return 0;
1091 	}
1092 	/* we hit 0, and we hold the lock */
1093 	return 1;
1094 }
1095 EXPORT_SYMBOL(atomic_dec_and_mutex_lock);
1096