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