projects/linux·blob v5.16 · refs

1144 lines · 29.0 KiB

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