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