projects/linux·blob v4.18 · refs

1267 lines · 31.7 KiB

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 <[email protected]>
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/signal.h>
23#include <linux/sched/rt.h>
24#include <linux/sched/wake_q.h>
25#include <linux/sched/debug.h>
26#include <linux/export.h>
27#include <linux/spinlock.h>
28#include <linux/interrupt.h>
29#include <linux/debug_locks.h>
30#include <linux/osq_lock.h>
31
32#ifdef CONFIG_DEBUG_MUTEXES
33# include "mutex-debug.h"
34#else
35# include "mutex.h"
36#endif
37
38void
39__mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
40{
41 atomic_long_set(&lock->owner, 0);
42 spin_lock_init(&lock->wait_lock);
43 INIT_LIST_HEAD(&lock->wait_list);
44#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
45 osq_lock_init(&lock->osq);
46#endif
47
48 debug_mutex_init(lock, name, key);
49}
50EXPORT_SYMBOL(__mutex_init);
51
52/*
53 * @owner: contains: 'struct task_struct *' to the current lock owner,
54 * NULL means not owned. Since task_struct pointers are aligned at
55 * at least L1_CACHE_BYTES, we have low bits to store extra state.
56 *
57 * Bit0 indicates a non-empty waiter list; unlock must issue a wakeup.
58 * Bit1 indicates unlock needs to hand the lock to the top-waiter
59 * Bit2 indicates handoff has been done and we're waiting for pickup.
60 */
61#define MUTEX_FLAG_WAITERS 0x01
62#define MUTEX_FLAG_HANDOFF 0x02
63#define MUTEX_FLAG_PICKUP 0x04
64
65#define MUTEX_FLAGS 0x07
66
67static inline struct task_struct *__owner_task(unsigned long owner)
68{
69 return (struct task_struct *)(owner & ~MUTEX_FLAGS);
70}
71
72static inline unsigned long __owner_flags(unsigned long owner)
73{
74 return owner & MUTEX_FLAGS;
75}
76
77/*
78 * Trylock variant that retuns the owning task on failure.
79 */
80static inline struct task_struct *__mutex_trylock_or_owner(struct mutex *lock)
81{
82 unsigned long owner, curr = (unsigned long)current;
83
84 owner = atomic_long_read(&lock->owner);
85 for (;;) { /* must loop, can race against a flag */
86 unsigned long old, flags = __owner_flags(owner);
87 unsigned long task = owner & ~MUTEX_FLAGS;
88
89 if (task) {
90 if (likely(task != curr))
91 break;
92
93 if (likely(!(flags & MUTEX_FLAG_PICKUP)))
94 break;
95
96 flags &= ~MUTEX_FLAG_PICKUP;
97 } else {
98#ifdef CONFIG_DEBUG_MUTEXES
99 DEBUG_LOCKS_WARN_ON(flags & MUTEX_FLAG_PICKUP);
100#endif
101 }
102
103 /*
104 * We set the HANDOFF bit, we must make sure it doesn't live
105 * past the point where we acquire it. This would be possible
106 * if we (accidentally) set the bit on an unlocked mutex.
107 */
108 flags &= ~MUTEX_FLAG_HANDOFF;
109
110 old = atomic_long_cmpxchg_acquire(&lock->owner, owner, curr | flags);
111 if (old == owner)
112 return NULL;
113
114 owner = old;
115 }
116
117 return __owner_task(owner);
118}
119
120/*
121 * Actual trylock that will work on any unlocked state.
122 */
123static inline bool __mutex_trylock(struct mutex *lock)
124{
125 return !__mutex_trylock_or_owner(lock);
126}
127
128#ifndef CONFIG_DEBUG_LOCK_ALLOC
129/*
130 * Lockdep annotations are contained to the slow paths for simplicity.
131 * There is nothing that would stop spreading the lockdep annotations outwards
132 * except more code.
133 */
134
135/*
136 * Optimistic trylock that only works in the uncontended case. Make sure to
137 * follow with a __mutex_trylock() before failing.
138 */
139static __always_inline bool __mutex_trylock_fast(struct mutex *lock)
140{
141 unsigned long curr = (unsigned long)current;
142 unsigned long zero = 0UL;
143
144 if (atomic_long_try_cmpxchg_acquire(&lock->owner, &zero, curr))
145 return true;
146
147 return false;
148}
149
150static __always_inline bool __mutex_unlock_fast(struct mutex *lock)
151{
152 unsigned long curr = (unsigned long)current;
153
154 if (atomic_long_cmpxchg_release(&lock->owner, curr, 0UL) == curr)
155 return true;
156
157 return false;
158}
159#endif
160
161static inline void __mutex_set_flag(struct mutex *lock, unsigned long flag)
162{
163 atomic_long_or(flag, &lock->owner);
164}
165
166static inline void __mutex_clear_flag(struct mutex *lock, unsigned long flag)
167{
168 atomic_long_andnot(flag, &lock->owner);
169}
170
171static inline bool __mutex_waiter_is_first(struct mutex *lock, struct mutex_waiter *waiter)
172{
173 return list_first_entry(&lock->wait_list, struct mutex_waiter, list) == waiter;
174}
175
176/*
177 * Give up ownership to a specific task, when @task = NULL, this is equivalent
178 * to a regular unlock. Sets PICKUP on a handoff, clears HANDOF, preserves
179 * WAITERS. Provides RELEASE semantics like a regular unlock, the
180 * __mutex_trylock() provides a matching ACQUIRE semantics for the handoff.
181 */
182static void __mutex_handoff(struct mutex *lock, struct task_struct *task)
183{
184 unsigned long owner = atomic_long_read(&lock->owner);
185
186 for (;;) {
187 unsigned long old, new;
188
189#ifdef CONFIG_DEBUG_MUTEXES
190 DEBUG_LOCKS_WARN_ON(__owner_task(owner) != current);
191 DEBUG_LOCKS_WARN_ON(owner & MUTEX_FLAG_PICKUP);
192#endif
193
194 new = (owner & MUTEX_FLAG_WAITERS);
195 new |= (unsigned long)task;
196 if (task)
197 new |= MUTEX_FLAG_PICKUP;
198
199 old = atomic_long_cmpxchg_release(&lock->owner, owner, new);
200 if (old == owner)
201 break;
202
203 owner = old;
204 }
205}
206
207#ifndef CONFIG_DEBUG_LOCK_ALLOC
208/*
209 * We split the mutex lock/unlock logic into separate fastpath and
210 * slowpath functions, to reduce the register pressure on the fastpath.
211 * We also put the fastpath first in the kernel image, to make sure the
212 * branch is predicted by the CPU as default-untaken.
213 */
214static void __sched __mutex_lock_slowpath(struct mutex *lock);
215
216/**
217 * mutex_lock - acquire the mutex
218 * @lock: the mutex to be acquired
219 *
220 * Lock the mutex exclusively for this task. If the mutex is not
221 * available right now, it will sleep until it can get it.
222 *
223 * The mutex must later on be released by the same task that
224 * acquired it. Recursive locking is not allowed. The task
225 * may not exit without first unlocking the mutex. Also, kernel
226 * memory where the mutex resides must not be freed with
227 * the mutex still locked. The mutex must first be initialized
228 * (or statically defined) before it can be locked. memset()-ing
229 * the mutex to 0 is not allowed.
230 *
231 * (The CONFIG_DEBUG_MUTEXES .config option turns on debugging
232 * checks that will enforce the restrictions and will also do
233 * deadlock debugging)
234 *
235 * This function is similar to (but not equivalent to) down().
236 */
237void __sched mutex_lock(struct mutex *lock)
238{
239 might_sleep();
240
241 if (!__mutex_trylock_fast(lock))
242 __mutex_lock_slowpath(lock);
243}
244EXPORT_SYMBOL(mutex_lock);
245#endif
246
247static __always_inline void
248ww_mutex_lock_acquired(struct ww_mutex *ww, struct ww_acquire_ctx *ww_ctx)
249{
250#ifdef CONFIG_DEBUG_MUTEXES
251 /*
252 * If this WARN_ON triggers, you used ww_mutex_lock to acquire,
253 * but released with a normal mutex_unlock in this call.
254 *
255 * This should never happen, always use ww_mutex_unlock.
256 */
257 DEBUG_LOCKS_WARN_ON(ww->ctx);
258
259 /*
260 * Not quite done after calling ww_acquire_done() ?
261 */
262 DEBUG_LOCKS_WARN_ON(ww_ctx->done_acquire);
263
264 if (ww_ctx->contending_lock) {
265 /*
266 * After -EDEADLK you tried to
267 * acquire a different ww_mutex? Bad!
268 */
269 DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock != ww);
270
271 /*
272 * You called ww_mutex_lock after receiving -EDEADLK,
273 * but 'forgot' to unlock everything else first?
274 */
275 DEBUG_LOCKS_WARN_ON(ww_ctx->acquired > 0);
276 ww_ctx->contending_lock = NULL;
277 }
278
279 /*
280 * Naughty, using a different class will lead to undefined behavior!
281 */
282 DEBUG_LOCKS_WARN_ON(ww_ctx->ww_class != ww->ww_class);
283#endif
284 ww_ctx->acquired++;
285}
286
287static inline bool __sched
288__ww_ctx_stamp_after(struct ww_acquire_ctx *a, struct ww_acquire_ctx *b)
289{
290 return a->stamp - b->stamp <= LONG_MAX &&
291 (a->stamp != b->stamp || a > b);
292}
293
294/*
295 * Wake up any waiters that may have to back off when the lock is held by the
296 * given context.
297 *
298 * Due to the invariants on the wait list, this can only affect the first
299 * waiter with a context.
300 *
301 * The current task must not be on the wait list.
302 */
303static void __sched
304__ww_mutex_wakeup_for_backoff(struct mutex *lock, struct ww_acquire_ctx *ww_ctx)
305{
306 struct mutex_waiter *cur;
307
308 lockdep_assert_held(&lock->wait_lock);
309
310 list_for_each_entry(cur, &lock->wait_list, list) {
311 if (!cur->ww_ctx)
312 continue;
313
314 if (cur->ww_ctx->acquired > 0 &&
315 __ww_ctx_stamp_after(cur->ww_ctx, ww_ctx)) {
316 debug_mutex_wake_waiter(lock, cur);
317 wake_up_process(cur->task);
318 }
319
320 break;
321 }
322}
323
324/*
325 * After acquiring lock with fastpath or when we lost out in contested
326 * slowpath, set ctx and wake up any waiters so they can recheck.
327 */
328static __always_inline void
329ww_mutex_set_context_fastpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
330{
331 ww_mutex_lock_acquired(lock, ctx);
332
333 lock->ctx = ctx;
334
335 /*
336 * The lock->ctx update should be visible on all cores before
337 * the atomic read is done, otherwise contended waiters might be
338 * missed. The contended waiters will either see ww_ctx == NULL
339 * and keep spinning, or it will acquire wait_lock, add itself
340 * to waiter list and sleep.
341 */
342 smp_mb(); /* ^^^ */
343
344 /*
345 * Check if lock is contended, if not there is nobody to wake up
346 */
347 if (likely(!(atomic_long_read(&lock->base.owner) & MUTEX_FLAG_WAITERS)))
348 return;
349
350 /*
351 * Uh oh, we raced in fastpath, wake up everyone in this case,
352 * so they can see the new lock->ctx.
353 */
354 spin_lock(&lock->base.wait_lock);
355 __ww_mutex_wakeup_for_backoff(&lock->base, ctx);
356 spin_unlock(&lock->base.wait_lock);
357}
358
359/*
360 * After acquiring lock in the slowpath set ctx.
361 *
362 * Unlike for the fast path, the caller ensures that waiters are woken up where
363 * necessary.
364 *
365 * Callers must hold the mutex wait_lock.
366 */
367static __always_inline void
368ww_mutex_set_context_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
369{
370 ww_mutex_lock_acquired(lock, ctx);
371 lock->ctx = ctx;
372}
373
374#ifdef CONFIG_MUTEX_SPIN_ON_OWNER
375
376static inline
377bool ww_mutex_spin_on_owner(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
378 struct mutex_waiter *waiter)
379{
380 struct ww_mutex *ww;
381
382 ww = container_of(lock, struct ww_mutex, base);
383
384 /*
385 * If ww->ctx is set the contents are undefined, only
386 * by acquiring wait_lock there is a guarantee that
387 * they are not invalid when reading.
388 *
389 * As such, when deadlock detection needs to be
390 * performed the optimistic spinning cannot be done.
391 *
392 * Check this in every inner iteration because we may
393 * be racing against another thread's ww_mutex_lock.
394 */
395 if (ww_ctx->acquired > 0 && READ_ONCE(ww->ctx))
396 return false;
397
398 /*
399 * If we aren't on the wait list yet, cancel the spin
400 * if there are waiters. We want to avoid stealing the
401 * lock from a waiter with an earlier stamp, since the
402 * other thread may already own a lock that we also
403 * need.
404 */
405 if (!waiter && (atomic_long_read(&lock->owner) & MUTEX_FLAG_WAITERS))
406 return false;
407
408 /*
409 * Similarly, stop spinning if we are no longer the
410 * first waiter.
411 */
412 if (waiter && !__mutex_waiter_is_first(lock, waiter))
413 return false;
414
415 return true;
416}
417
418/*
419 * Look out! "owner" is an entirely speculative pointer access and not
420 * reliable.
421 *
422 * "noinline" so that this function shows up on perf profiles.
423 */
424static noinline
425bool mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner,
426 struct ww_acquire_ctx *ww_ctx, struct mutex_waiter *waiter)
427{
428 bool ret = true;
429
430 rcu_read_lock();
431 while (__mutex_owner(lock) == owner) {
432 /*
433 * Ensure we emit the owner->on_cpu, dereference _after_
434 * checking lock->owner still matches owner. If that fails,
435 * owner might point to freed memory. If it still matches,
436 * the rcu_read_lock() ensures the memory stays valid.
437 */
438 barrier();
439
440 /*
441 * Use vcpu_is_preempted to detect lock holder preemption issue.
442 */
443 if (!owner->on_cpu || need_resched() ||
444 vcpu_is_preempted(task_cpu(owner))) {
445 ret = false;
446 break;
447 }
448
449 if (ww_ctx && !ww_mutex_spin_on_owner(lock, ww_ctx, waiter)) {
450 ret = false;
451 break;
452 }
453
454 cpu_relax();
455 }
456 rcu_read_unlock();
457
458 return ret;
459}
460
461/*
462 * Initial check for entering the mutex spinning loop
463 */
464static inline int mutex_can_spin_on_owner(struct mutex *lock)
465{
466 struct task_struct *owner;
467 int retval = 1;
468
469 if (need_resched())
470 return 0;
471
472 rcu_read_lock();
473 owner = __mutex_owner(lock);
474
475 /*
476 * As lock holder preemption issue, we both skip spinning if task is not
477 * on cpu or its cpu is preempted
478 */
479 if (owner)
480 retval = owner->on_cpu && !vcpu_is_preempted(task_cpu(owner));
481 rcu_read_unlock();
482
483 /*
484 * If lock->owner is not set, the mutex has been released. Return true
485 * such that we'll trylock in the spin path, which is a faster option
486 * than the blocking slow path.
487 */
488 return retval;
489}
490
491/*
492 * Optimistic spinning.
493 *
494 * We try to spin for acquisition when we find that the lock owner
495 * is currently running on a (different) CPU and while we don't
496 * need to reschedule. The rationale is that if the lock owner is
497 * running, it is likely to release the lock soon.
498 *
499 * The mutex spinners are queued up using MCS lock so that only one
500 * spinner can compete for the mutex. However, if mutex spinning isn't
501 * going to happen, there is no point in going through the lock/unlock
502 * overhead.
503 *
504 * Returns true when the lock was taken, otherwise false, indicating
505 * that we need to jump to the slowpath and sleep.
506 *
507 * The waiter flag is set to true if the spinner is a waiter in the wait
508 * queue. The waiter-spinner will spin on the lock directly and concurrently
509 * with the spinner at the head of the OSQ, if present, until the owner is
510 * changed to itself.
511 */
512static __always_inline bool
513mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
514 const bool use_ww_ctx, struct mutex_waiter *waiter)
515{
516 if (!waiter) {
517 /*
518 * The purpose of the mutex_can_spin_on_owner() function is
519 * to eliminate the overhead of osq_lock() and osq_unlock()
520 * in case spinning isn't possible. As a waiter-spinner
521 * is not going to take OSQ lock anyway, there is no need
522 * to call mutex_can_spin_on_owner().
523 */
524 if (!mutex_can_spin_on_owner(lock))
525 goto fail;
526
527 /*
528 * In order to avoid a stampede of mutex spinners trying to
529 * acquire the mutex all at once, the spinners need to take a
530 * MCS (queued) lock first before spinning on the owner field.
531 */
532 if (!osq_lock(&lock->osq))
533 goto fail;
534 }
535
536 for (;;) {
537 struct task_struct *owner;
538
539 /* Try to acquire the mutex... */
540 owner = __mutex_trylock_or_owner(lock);
541 if (!owner)
542 break;
543
544 /*
545 * There's an owner, wait for it to either
546 * release the lock or go to sleep.
547 */
548 if (!mutex_spin_on_owner(lock, owner, ww_ctx, waiter))
549 goto fail_unlock;
550
551 /*
552 * The cpu_relax() call is a compiler barrier which forces
553 * everything in this loop to be re-loaded. We don't need
554 * memory barriers as we'll eventually observe the right
555 * values at the cost of a few extra spins.
556 */
557 cpu_relax();
558 }
559
560 if (!waiter)
561 osq_unlock(&lock->osq);
562
563 return true;
564
565
566fail_unlock:
567 if (!waiter)
568 osq_unlock(&lock->osq);
569
570fail:
571 /*
572 * If we fell out of the spin path because of need_resched(),
573 * reschedule now, before we try-lock the mutex. This avoids getting
574 * scheduled out right after we obtained the mutex.
575 */
576 if (need_resched()) {
577 /*
578 * We _should_ have TASK_RUNNING here, but just in case
579 * we do not, make it so, otherwise we might get stuck.
580 */
581 __set_current_state(TASK_RUNNING);
582 schedule_preempt_disabled();
583 }
584
585 return false;
586}
587#else
588static __always_inline bool
589mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
590 const bool use_ww_ctx, struct mutex_waiter *waiter)
591{
592 return false;
593}
594#endif
595
596static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip);
597
598/**
599 * mutex_unlock - release the mutex
600 * @lock: the mutex to be released
601 *
602 * Unlock a mutex that has been locked by this task previously.
603 *
604 * This function must not be used in interrupt context. Unlocking
605 * of a not locked mutex is not allowed.
606 *
607 * This function is similar to (but not equivalent to) up().
608 */
609void __sched mutex_unlock(struct mutex *lock)
610{
611#ifndef CONFIG_DEBUG_LOCK_ALLOC
612 if (__mutex_unlock_fast(lock))
613 return;
614#endif
615 __mutex_unlock_slowpath(lock, _RET_IP_);
616}
617EXPORT_SYMBOL(mutex_unlock);
618
619/**
620 * ww_mutex_unlock - release the w/w mutex
621 * @lock: the mutex to be released
622 *
623 * Unlock a mutex that has been locked by this task previously with any of the
624 * ww_mutex_lock* functions (with or without an acquire context). It is
625 * forbidden to release the locks after releasing the acquire context.
626 *
627 * This function must not be used in interrupt context. Unlocking
628 * of a unlocked mutex is not allowed.
629 */
630void __sched ww_mutex_unlock(struct ww_mutex *lock)
631{
632 /*
633 * The unlocking fastpath is the 0->1 transition from 'locked'
634 * into 'unlocked' state:
635 */
636 if (lock->ctx) {
637#ifdef CONFIG_DEBUG_MUTEXES
638 DEBUG_LOCKS_WARN_ON(!lock->ctx->acquired);
639#endif
640 if (lock->ctx->acquired > 0)
641 lock->ctx->acquired--;
642 lock->ctx = NULL;
643 }
644
645 mutex_unlock(&lock->base);
646}
647EXPORT_SYMBOL(ww_mutex_unlock);
648
649static inline int __sched
650__ww_mutex_lock_check_stamp(struct mutex *lock, struct mutex_waiter *waiter,
651 struct ww_acquire_ctx *ctx)
652{
653 struct ww_mutex *ww = container_of(lock, struct ww_mutex, base);
654 struct ww_acquire_ctx *hold_ctx = READ_ONCE(ww->ctx);
655 struct mutex_waiter *cur;
656
657 if (hold_ctx && __ww_ctx_stamp_after(ctx, hold_ctx))
658 goto deadlock;
659
660 /*
661 * If there is a waiter in front of us that has a context, then its
662 * stamp is earlier than ours and we must back off.
663 */
664 cur = waiter;
665 list_for_each_entry_continue_reverse(cur, &lock->wait_list, list) {
666 if (cur->ww_ctx)
667 goto deadlock;
668 }
669
670 return 0;
671
672deadlock:
673#ifdef CONFIG_DEBUG_MUTEXES
674 DEBUG_LOCKS_WARN_ON(ctx->contending_lock);
675 ctx->contending_lock = ww;
676#endif
677 return -EDEADLK;
678}
679
680static inline int __sched
681__ww_mutex_add_waiter(struct mutex_waiter *waiter,
682 struct mutex *lock,
683 struct ww_acquire_ctx *ww_ctx)
684{
685 struct mutex_waiter *cur;
686 struct list_head *pos;
687
688 if (!ww_ctx) {
689 list_add_tail(&waiter->list, &lock->wait_list);
690 return 0;
691 }
692
693 /*
694 * Add the waiter before the first waiter with a higher stamp.
695 * Waiters without a context are skipped to avoid starving
696 * them.
697 */
698 pos = &lock->wait_list;
699 list_for_each_entry_reverse(cur, &lock->wait_list, list) {
700 if (!cur->ww_ctx)
701 continue;
702
703 if (__ww_ctx_stamp_after(ww_ctx, cur->ww_ctx)) {
704 /* Back off immediately if necessary. */
705 if (ww_ctx->acquired > 0) {
706#ifdef CONFIG_DEBUG_MUTEXES
707 struct ww_mutex *ww;
708
709 ww = container_of(lock, struct ww_mutex, base);
710 DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock);
711 ww_ctx->contending_lock = ww;
712#endif
713 return -EDEADLK;
714 }
715
716 break;
717 }
718
719 pos = &cur->list;
720
721 /*
722 * Wake up the waiter so that it gets a chance to back
723 * off.
724 */
725 if (cur->ww_ctx->acquired > 0) {
726 debug_mutex_wake_waiter(lock, cur);
727 wake_up_process(cur->task);
728 }
729 }
730
731 list_add_tail(&waiter->list, pos);
732 return 0;
733}
734
735/*
736 * Lock a mutex (possibly interruptible), slowpath:
737 */
738static __always_inline int __sched
739__mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
740 struct lockdep_map *nest_lock, unsigned long ip,
741 struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
742{
743 struct mutex_waiter waiter;
744 bool first = false;
745 struct ww_mutex *ww;
746 int ret;
747
748 might_sleep();
749
750 ww = container_of(lock, struct ww_mutex, base);
751 if (use_ww_ctx && ww_ctx) {
752 if (unlikely(ww_ctx == READ_ONCE(ww->ctx)))
753 return -EALREADY;
754 }
755
756 preempt_disable();
757 mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);
758
759 if (__mutex_trylock(lock) ||
760 mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, NULL)) {
761 /* got the lock, yay! */
762 lock_acquired(&lock->dep_map, ip);
763 if (use_ww_ctx && ww_ctx)
764 ww_mutex_set_context_fastpath(ww, ww_ctx);
765 preempt_enable();
766 return 0;
767 }
768
769 spin_lock(&lock->wait_lock);
770 /*
771 * After waiting to acquire the wait_lock, try again.
772 */
773 if (__mutex_trylock(lock)) {
774 if (use_ww_ctx && ww_ctx)
775 __ww_mutex_wakeup_for_backoff(lock, ww_ctx);
776
777 goto skip_wait;
778 }
779
780 debug_mutex_lock_common(lock, &waiter);
781 debug_mutex_add_waiter(lock, &waiter, current);
782
783 lock_contended(&lock->dep_map, ip);
784
785 if (!use_ww_ctx) {
786 /* add waiting tasks to the end of the waitqueue (FIFO): */
787 list_add_tail(&waiter.list, &lock->wait_list);
788
789#ifdef CONFIG_DEBUG_MUTEXES
790 waiter.ww_ctx = MUTEX_POISON_WW_CTX;
791#endif
792 } else {
793 /* Add in stamp order, waking up waiters that must back off. */
794 ret = __ww_mutex_add_waiter(&waiter, lock, ww_ctx);
795 if (ret)
796 goto err_early_backoff;
797
798 waiter.ww_ctx = ww_ctx;
799 }
800
801 waiter.task = current;
802
803 if (__mutex_waiter_is_first(lock, &waiter))
804 __mutex_set_flag(lock, MUTEX_FLAG_WAITERS);
805
806 set_current_state(state);
807 for (;;) {
808 /*
809 * Once we hold wait_lock, we're serialized against
810 * mutex_unlock() handing the lock off to us, do a trylock
811 * before testing the error conditions to make sure we pick up
812 * the handoff.
813 */
814 if (__mutex_trylock(lock))
815 goto acquired;
816
817 /*
818 * Check for signals and wound conditions while holding
819 * wait_lock. This ensures the lock cancellation is ordered
820 * against mutex_unlock() and wake-ups do not go missing.
821 */
822 if (unlikely(signal_pending_state(state, current))) {
823 ret = -EINTR;
824 goto err;
825 }
826
827 if (use_ww_ctx && ww_ctx && ww_ctx->acquired > 0) {
828 ret = __ww_mutex_lock_check_stamp(lock, &waiter, ww_ctx);
829 if (ret)
830 goto err;
831 }
832
833 spin_unlock(&lock->wait_lock);
834 schedule_preempt_disabled();
835
836 /*
837 * ww_mutex needs to always recheck its position since its waiter
838 * list is not FIFO ordered.
839 */
840 if ((use_ww_ctx && ww_ctx) || !first) {
841 first = __mutex_waiter_is_first(lock, &waiter);
842 if (first)
843 __mutex_set_flag(lock, MUTEX_FLAG_HANDOFF);
844 }
845
846 set_current_state(state);
847 /*
848 * Here we order against unlock; we must either see it change
849 * state back to RUNNING and fall through the next schedule(),
850 * or we must see its unlock and acquire.
851 */
852 if (__mutex_trylock(lock) ||
853 (first && mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, &waiter)))
854 break;
855
856 spin_lock(&lock->wait_lock);
857 }
858 spin_lock(&lock->wait_lock);
859acquired:
860 __set_current_state(TASK_RUNNING);
861
862 mutex_remove_waiter(lock, &waiter, current);
863 if (likely(list_empty(&lock->wait_list)))
864 __mutex_clear_flag(lock, MUTEX_FLAGS);
865
866 debug_mutex_free_waiter(&waiter);
867
868skip_wait:
869 /* got the lock - cleanup and rejoice! */
870 lock_acquired(&lock->dep_map, ip);
871
872 if (use_ww_ctx && ww_ctx)
873 ww_mutex_set_context_slowpath(ww, ww_ctx);
874
875 spin_unlock(&lock->wait_lock);
876 preempt_enable();
877 return 0;
878
879err:
880 __set_current_state(TASK_RUNNING);
881 mutex_remove_waiter(lock, &waiter, current);
882err_early_backoff:
883 spin_unlock(&lock->wait_lock);
884 debug_mutex_free_waiter(&waiter);
885 mutex_release(&lock->dep_map, 1, ip);
886 preempt_enable();
887 return ret;
888}
889
890static int __sched
891__mutex_lock(struct mutex *lock, long state, unsigned int subclass,
892 struct lockdep_map *nest_lock, unsigned long ip)
893{
894 return __mutex_lock_common(lock, state, subclass, nest_lock, ip, NULL, false);
895}
896
897static int __sched
898__ww_mutex_lock(struct mutex *lock, long state, unsigned int subclass,
899 struct lockdep_map *nest_lock, unsigned long ip,
900 struct ww_acquire_ctx *ww_ctx)
901{
902 return __mutex_lock_common(lock, state, subclass, nest_lock, ip, ww_ctx, true);
903}
904
905#ifdef CONFIG_DEBUG_LOCK_ALLOC
906void __sched
907mutex_lock_nested(struct mutex *lock, unsigned int subclass)
908{
909 __mutex_lock(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_);
910}
911
912EXPORT_SYMBOL_GPL(mutex_lock_nested);
913
914void __sched
915_mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest)
916{
917 __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, nest, _RET_IP_);
918}
919EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock);
920
921int __sched
922mutex_lock_killable_nested(struct mutex *lock, unsigned int subclass)
923{
924 return __mutex_lock(lock, TASK_KILLABLE, subclass, NULL, _RET_IP_);
925}
926EXPORT_SYMBOL_GPL(mutex_lock_killable_nested);
927
928int __sched
929mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
930{
931 return __mutex_lock(lock, TASK_INTERRUPTIBLE, subclass, NULL, _RET_IP_);
932}
933EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
934
935void __sched
936mutex_lock_io_nested(struct mutex *lock, unsigned int subclass)
937{
938 int token;
939
940 might_sleep();
941
942 token = io_schedule_prepare();
943 __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE,
944 subclass, NULL, _RET_IP_, NULL, 0);
945 io_schedule_finish(token);
946}
947EXPORT_SYMBOL_GPL(mutex_lock_io_nested);
948
949static inline int
950ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
951{
952#ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH
953 unsigned tmp;
954
955 if (ctx->deadlock_inject_countdown-- == 0) {
956 tmp = ctx->deadlock_inject_interval;
957 if (tmp > UINT_MAX/4)
958 tmp = UINT_MAX;
959 else
960 tmp = tmp*2 + tmp + tmp/2;
961
962 ctx->deadlock_inject_interval = tmp;
963 ctx->deadlock_inject_countdown = tmp;
964 ctx->contending_lock = lock;
965
966 ww_mutex_unlock(lock);
967
968 return -EDEADLK;
969 }
970#endif
971
972 return 0;
973}
974
975int __sched
976ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
977{
978 int ret;
979
980 might_sleep();
981 ret = __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE,
982 0, ctx ? &ctx->dep_map : NULL, _RET_IP_,
983 ctx);
984 if (!ret && ctx && ctx->acquired > 1)
985 return ww_mutex_deadlock_injection(lock, ctx);
986
987 return ret;
988}
989EXPORT_SYMBOL_GPL(ww_mutex_lock);
990
991int __sched
992ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
993{
994 int ret;
995
996 might_sleep();
997 ret = __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE,
998 0, ctx ? &ctx->dep_map : NULL, _RET_IP_,
999 ctx);
1000
1001 if (!ret && ctx && ctx->acquired > 1)
1002 return ww_mutex_deadlock_injection(lock, ctx);
1003
1004 return ret;
1005}
1006EXPORT_SYMBOL_GPL(ww_mutex_lock_interruptible);
1007
1008#endif
1009
1010/*
1011 * Release the lock, slowpath:
1012 */
1013static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip)
1014{
1015 struct task_struct *next = NULL;
1016 DEFINE_WAKE_Q(wake_q);
1017 unsigned long owner;
1018
1019 mutex_release(&lock->dep_map, 1, ip);
1020
1021 /*
1022 * Release the lock before (potentially) taking the spinlock such that
1023 * other contenders can get on with things ASAP.
1024 *
1025 * Except when HANDOFF, in that case we must not clear the owner field,
1026 * but instead set it to the top waiter.
1027 */
1028 owner = atomic_long_read(&lock->owner);
1029 for (;;) {
1030 unsigned long old;
1031
1032#ifdef CONFIG_DEBUG_MUTEXES
1033 DEBUG_LOCKS_WARN_ON(__owner_task(owner) != current);
1034 DEBUG_LOCKS_WARN_ON(owner & MUTEX_FLAG_PICKUP);
1035#endif
1036
1037 if (owner & MUTEX_FLAG_HANDOFF)
1038 break;
1039
1040 old = atomic_long_cmpxchg_release(&lock->owner, owner,
1041 __owner_flags(owner));
1042 if (old == owner) {
1043 if (owner & MUTEX_FLAG_WAITERS)
1044 break;
1045
1046 return;
1047 }
1048
1049 owner = old;
1050 }
1051
1052 spin_lock(&lock->wait_lock);
1053 debug_mutex_unlock(lock);
1054 if (!list_empty(&lock->wait_list)) {
1055 /* get the first entry from the wait-list: */
1056 struct mutex_waiter *waiter =
1057 list_first_entry(&lock->wait_list,
1058 struct mutex_waiter, list);
1059
1060 next = waiter->task;
1061
1062 debug_mutex_wake_waiter(lock, waiter);
1063 wake_q_add(&wake_q, next);
1064 }
1065
1066 if (owner & MUTEX_FLAG_HANDOFF)
1067 __mutex_handoff(lock, next);
1068
1069 spin_unlock(&lock->wait_lock);
1070
1071 wake_up_q(&wake_q);
1072}
1073
1074#ifndef CONFIG_DEBUG_LOCK_ALLOC
1075/*
1076 * Here come the less common (and hence less performance-critical) APIs:
1077 * mutex_lock_interruptible() and mutex_trylock().
1078 */
1079static noinline int __sched
1080__mutex_lock_killable_slowpath(struct mutex *lock);
1081
1082static noinline int __sched
1083__mutex_lock_interruptible_slowpath(struct mutex *lock);
1084
1085/**
1086 * mutex_lock_interruptible() - Acquire the mutex, interruptible by signals.
1087 * @lock: The mutex to be acquired.
1088 *
1089 * Lock the mutex like mutex_lock(). If a signal is delivered while the
1090 * process is sleeping, this function will return without acquiring the
1091 * mutex.
1092 *
1093 * Context: Process context.
1094 * Return: 0 if the lock was successfully acquired or %-EINTR if a
1095 * signal arrived.
1096 */
1097int __sched mutex_lock_interruptible(struct mutex *lock)
1098{
1099 might_sleep();
1100
1101 if (__mutex_trylock_fast(lock))
1102 return 0;
1103
1104 return __mutex_lock_interruptible_slowpath(lock);
1105}
1106
1107EXPORT_SYMBOL(mutex_lock_interruptible);
1108
1109/**
1110 * mutex_lock_killable() - Acquire the mutex, interruptible by fatal signals.
1111 * @lock: The mutex to be acquired.
1112 *
1113 * Lock the mutex like mutex_lock(). If a signal which will be fatal to
1114 * the current process is delivered while the process is sleeping, this
1115 * function will return without acquiring the mutex.
1116 *
1117 * Context: Process context.
1118 * Return: 0 if the lock was successfully acquired or %-EINTR if a
1119 * fatal signal arrived.
1120 */
1121int __sched mutex_lock_killable(struct mutex *lock)
1122{
1123 might_sleep();
1124
1125 if (__mutex_trylock_fast(lock))
1126 return 0;
1127
1128 return __mutex_lock_killable_slowpath(lock);
1129}
1130EXPORT_SYMBOL(mutex_lock_killable);
1131
1132/**
1133 * mutex_lock_io() - Acquire the mutex and mark the process as waiting for I/O
1134 * @lock: The mutex to be acquired.
1135 *
1136 * Lock the mutex like mutex_lock(). While the task is waiting for this
1137 * mutex, it will be accounted as being in the IO wait state by the
1138 * scheduler.
1139 *
1140 * Context: Process context.
1141 */
1142void __sched mutex_lock_io(struct mutex *lock)
1143{
1144 int token;
1145
1146 token = io_schedule_prepare();
1147 mutex_lock(lock);
1148 io_schedule_finish(token);
1149}
1150EXPORT_SYMBOL_GPL(mutex_lock_io);
1151
1152static noinline void __sched
1153__mutex_lock_slowpath(struct mutex *lock)
1154{
1155 __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_);
1156}
1157
1158static noinline int __sched
1159__mutex_lock_killable_slowpath(struct mutex *lock)
1160{
1161 return __mutex_lock(lock, TASK_KILLABLE, 0, NULL, _RET_IP_);
1162}
1163
1164static noinline int __sched
1165__mutex_lock_interruptible_slowpath(struct mutex *lock)
1166{
1167 return __mutex_lock(lock, TASK_INTERRUPTIBLE, 0, NULL, _RET_IP_);
1168}
1169
1170static noinline int __sched
1171__ww_mutex_lock_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1172{
1173 return __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE, 0, NULL,
1174 _RET_IP_, ctx);
1175}
1176
1177static noinline int __sched
1178__ww_mutex_lock_interruptible_slowpath(struct ww_mutex *lock,
1179 struct ww_acquire_ctx *ctx)
1180{
1181 return __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE, 0, NULL,
1182 _RET_IP_, ctx);
1183}
1184
1185#endif
1186
1187/**
1188 * mutex_trylock - try to acquire the mutex, without waiting
1189 * @lock: the mutex to be acquired
1190 *
1191 * Try to acquire the mutex atomically. Returns 1 if the mutex
1192 * has been acquired successfully, and 0 on contention.
1193 *
1194 * NOTE: this function follows the spin_trylock() convention, so
1195 * it is negated from the down_trylock() return values! Be careful
1196 * about this when converting semaphore users to mutexes.
1197 *
1198 * This function must not be used in interrupt context. The
1199 * mutex must be released by the same task that acquired it.
1200 */
1201int __sched mutex_trylock(struct mutex *lock)
1202{
1203 bool locked = __mutex_trylock(lock);
1204
1205 if (locked)
1206 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
1207
1208 return locked;
1209}
1210EXPORT_SYMBOL(mutex_trylock);
1211
1212#ifndef CONFIG_DEBUG_LOCK_ALLOC
1213int __sched
1214ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1215{
1216 might_sleep();
1217
1218 if (__mutex_trylock_fast(&lock->base)) {
1219 if (ctx)
1220 ww_mutex_set_context_fastpath(lock, ctx);
1221 return 0;
1222 }
1223
1224 return __ww_mutex_lock_slowpath(lock, ctx);
1225}
1226EXPORT_SYMBOL(ww_mutex_lock);
1227
1228int __sched
1229ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1230{
1231 might_sleep();
1232
1233 if (__mutex_trylock_fast(&lock->base)) {
1234 if (ctx)
1235 ww_mutex_set_context_fastpath(lock, ctx);
1236 return 0;
1237 }
1238
1239 return __ww_mutex_lock_interruptible_slowpath(lock, ctx);
1240}
1241EXPORT_SYMBOL(ww_mutex_lock_interruptible);
1242
1243#endif
1244
1245/**
1246 * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0
1247 * @cnt: the atomic which we are to dec
1248 * @lock: the mutex to return holding if we dec to 0
1249 *
1250 * return true and hold lock if we dec to 0, return false otherwise
1251 */
1252int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock)
1253{
1254 /* dec if we can't possibly hit 0 */
1255 if (atomic_add_unless(cnt, -1, 1))
1256 return 0;
1257 /* we might hit 0, so take the lock */
1258 mutex_lock(lock);
1259 if (!atomic_dec_and_test(cnt)) {
1260 /* when we actually did the dec, we didn't hit 0 */
1261 mutex_unlock(lock);
1262 return 0;
1263 }
1264 /* we hit 0, and we hold the lock */
1265 return 1;
1266}
1267EXPORT_SYMBOL(atomic_dec_and_mutex_lock);