Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * kernel/locking/mutex.c
4 : *
5 : * Mutexes: blocking mutual exclusion locks
6 : *
7 : * Started by Ingo Molnar:
8 : *
9 : * Copyright (C) 2004, 2005, 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
10 : *
11 : * Many thanks to Arjan van de Ven, Thomas Gleixner, Steven Rostedt and
12 : * David Howells for suggestions and improvements.
13 : *
14 : * - Adaptive spinning for mutexes by Peter Zijlstra. (Ported to mainline
15 : * from the -rt tree, where it was originally implemented for rtmutexes
16 : * by Steven Rostedt, based on work by Gregory Haskins, Peter Morreale
17 : * and Sven Dietrich.
18 : *
19 : * Also see Documentation/locking/mutex-design.rst.
20 : */
21 : #include <linux/mutex.h>
22 : #include <linux/ww_mutex.h>
23 : #include <linux/sched/signal.h>
24 : #include <linux/sched/rt.h>
25 : #include <linux/sched/wake_q.h>
26 : #include <linux/sched/debug.h>
27 : #include <linux/export.h>
28 : #include <linux/spinlock.h>
29 : #include <linux/interrupt.h>
30 : #include <linux/debug_locks.h>
31 : #include <linux/osq_lock.h>
32 :
33 : #ifndef CONFIG_PREEMPT_RT
34 : #include "mutex.h"
35 :
36 : #ifdef CONFIG_DEBUG_MUTEXES
37 : # define MUTEX_WARN_ON(cond) DEBUG_LOCKS_WARN_ON(cond)
38 : #else
39 : # define MUTEX_WARN_ON(cond)
40 : #endif
41 :
42 : void
43 996 : __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
44 : {
45 1992 : atomic_long_set(&lock->owner, 0);
46 : raw_spin_lock_init(&lock->wait_lock);
47 1992 : 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);
53 996 : }
54 : EXPORT_SYMBOL(__mutex_init);
55 :
56 : /*
57 : * @owner: contains: 'struct task_struct *' to the current lock owner,
58 : * NULL means not owned. Since task_struct pointers are aligned at
59 : * at least L1_CACHE_BYTES, we have low bits to store extra state.
60 : *
61 : * Bit0 indicates a non-empty waiter list; unlock must issue a wakeup.
62 : * Bit1 indicates unlock needs to hand the lock to the top-waiter
63 : * Bit2 indicates handoff has been done and we're waiting for pickup.
64 : */
65 : #define MUTEX_FLAG_WAITERS 0x01
66 : #define MUTEX_FLAG_HANDOFF 0x02
67 : #define MUTEX_FLAG_PICKUP 0x04
68 :
69 : #define MUTEX_FLAGS 0x07
70 :
71 : /*
72 : * Internal helper function; C doesn't allow us to hide it :/
73 : *
74 : * DO NOT USE (outside of mutex code).
75 : */
76 : static inline struct task_struct *__mutex_owner(struct mutex *lock)
77 : {
78 0 : return (struct task_struct *)(atomic_long_read(&lock->owner) & ~MUTEX_FLAGS);
79 : }
80 :
81 : static inline struct task_struct *__owner_task(unsigned long owner)
82 : {
83 0 : return (struct task_struct *)(owner & ~MUTEX_FLAGS);
84 : }
85 :
86 0 : bool mutex_is_locked(struct mutex *lock)
87 : {
88 0 : return __mutex_owner(lock) != NULL;
89 : }
90 : EXPORT_SYMBOL(mutex_is_locked);
91 :
92 : static inline unsigned long __owner_flags(unsigned long owner)
93 : {
94 0 : return owner & MUTEX_FLAGS;
95 : }
96 :
97 : /*
98 : * Returns: __mutex_owner(lock) on failure or NULL on success.
99 : */
100 0 : static inline struct task_struct *__mutex_trylock_common(struct mutex *lock, bool handoff)
101 : {
102 0 : unsigned long owner, curr = (unsigned long)current;
103 :
104 0 : owner = atomic_long_read(&lock->owner);
105 : for (;;) { /* must loop, can race against a flag */
106 0 : unsigned long flags = __owner_flags(owner);
107 0 : unsigned long task = owner & ~MUTEX_FLAGS;
108 :
109 0 : if (task) {
110 0 : if (flags & MUTEX_FLAG_PICKUP) {
111 0 : if (task != curr)
112 : break;
113 0 : flags &= ~MUTEX_FLAG_PICKUP;
114 0 : } else if (handoff) {
115 0 : if (flags & MUTEX_FLAG_HANDOFF)
116 : break;
117 0 : 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 0 : if (atomic_long_try_cmpxchg_acquire(&lock->owner, &owner, task | flags)) {
127 0 : if (task == curr)
128 : return NULL;
129 : break;
130 : }
131 : }
132 :
133 0 : return __owner_task(owner);
134 : }
135 :
136 : /*
137 : * Trylock or set HANDOFF
138 : */
139 : static inline bool __mutex_trylock_or_handoff(struct mutex *lock, bool handoff)
140 : {
141 0 : return !__mutex_trylock_common(lock, handoff);
142 : }
143 :
144 : /*
145 : * Actual trylock that will work on any unlocked state.
146 : */
147 : static inline bool __mutex_trylock(struct mutex *lock)
148 : {
149 0 : 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 : */
163 : static __always_inline bool __mutex_trylock_fast(struct mutex *lock)
164 : {
165 3475 : unsigned long curr = (unsigned long)current;
166 3475 : unsigned long zero = 0UL;
167 :
168 6950 : if (atomic_long_try_cmpxchg_acquire(&lock->owner, &zero, curr))
169 : return true;
170 :
171 : return false;
172 : }
173 :
174 : static __always_inline bool __mutex_unlock_fast(struct mutex *lock)
175 : {
176 3475 : unsigned long curr = (unsigned long)current;
177 :
178 6950 : return atomic_long_try_cmpxchg_release(&lock->owner, &curr, 0UL);
179 : }
180 : #endif
181 :
182 : static inline void __mutex_set_flag(struct mutex *lock, unsigned long flag)
183 : {
184 0 : atomic_long_or(flag, &lock->owner);
185 : }
186 :
187 : static inline void __mutex_clear_flag(struct mutex *lock, unsigned long flag)
188 : {
189 0 : atomic_long_andnot(flag, &lock->owner);
190 : }
191 :
192 : static inline bool __mutex_waiter_is_first(struct mutex *lock, struct mutex_waiter *waiter)
193 : {
194 0 : 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 : */
201 : static 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 0 : list_add_tail(&waiter->list, list);
208 0 : if (__mutex_waiter_is_first(lock, waiter))
209 : __mutex_set_flag(lock, MUTEX_FLAG_WAITERS);
210 : }
211 :
212 : static void
213 0 : __mutex_remove_waiter(struct mutex *lock, struct mutex_waiter *waiter)
214 : {
215 0 : list_del(&waiter->list);
216 0 : if (likely(list_empty(&lock->wait_list)))
217 : __mutex_clear_flag(lock, MUTEX_FLAGS);
218 :
219 : debug_mutex_remove_waiter(lock, waiter, current);
220 0 : }
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 : */
228 : static void __mutex_handoff(struct mutex *lock, struct task_struct *task)
229 : {
230 0 : 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 0 : new = (owner & MUTEX_FLAG_WAITERS);
239 0 : new |= (unsigned long)task;
240 0 : if (task)
241 0 : new |= MUTEX_FLAG_PICKUP;
242 :
243 0 : 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 : */
255 : static 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 : */
278 3237 : void __sched mutex_lock(struct mutex *lock)
279 : {
280 : might_sleep();
281 :
282 3237 : if (!__mutex_trylock_fast(lock))
283 0 : __mutex_lock_slowpath(lock);
284 3237 : }
285 : EXPORT_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 : */
295 : static inline struct task_struct *__mutex_trylock_or_owner(struct mutex *lock)
296 : {
297 : return __mutex_trylock_common(lock, false);
298 : }
299 :
300 : static inline
301 : bool 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 : */
348 : static noinline
349 : bool 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(owner) || need_resched()) {
371 : ret = false;
372 : break;
373 : }
374 :
375 : if (ww_ctx && !ww_mutex_spin_on_owner(lock, ww_ctx, waiter)) {
376 : ret = false;
377 : break;
378 : }
379 :
380 : cpu_relax();
381 : }
382 :
383 : return ret;
384 : }
385 :
386 : /*
387 : * Initial check for entering the mutex spinning loop
388 : */
389 : static inline int mutex_can_spin_on_owner(struct mutex *lock)
390 : {
391 : struct task_struct *owner;
392 : int retval = 1;
393 :
394 : lockdep_assert_preemption_disabled();
395 :
396 : if (need_resched())
397 : return 0;
398 :
399 : /*
400 : * We already disabled preemption which is equal to the RCU read-side
401 : * crital section in optimistic spinning code. Thus the task_strcut
402 : * structure won't go away during the spinning period.
403 : */
404 : owner = __mutex_owner(lock);
405 : if (owner)
406 : retval = owner_on_cpu(owner);
407 :
408 : /*
409 : * If lock->owner is not set, the mutex has been released. Return true
410 : * such that we'll trylock in the spin path, which is a faster option
411 : * than the blocking slow path.
412 : */
413 : return retval;
414 : }
415 :
416 : /*
417 : * Optimistic spinning.
418 : *
419 : * We try to spin for acquisition when we find that the lock owner
420 : * is currently running on a (different) CPU and while we don't
421 : * need to reschedule. The rationale is that if the lock owner is
422 : * running, it is likely to release the lock soon.
423 : *
424 : * The mutex spinners are queued up using MCS lock so that only one
425 : * spinner can compete for the mutex. However, if mutex spinning isn't
426 : * going to happen, there is no point in going through the lock/unlock
427 : * overhead.
428 : *
429 : * Returns true when the lock was taken, otherwise false, indicating
430 : * that we need to jump to the slowpath and sleep.
431 : *
432 : * The waiter flag is set to true if the spinner is a waiter in the wait
433 : * queue. The waiter-spinner will spin on the lock directly and concurrently
434 : * with the spinner at the head of the OSQ, if present, until the owner is
435 : * changed to itself.
436 : */
437 : static __always_inline bool
438 : mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
439 : struct mutex_waiter *waiter)
440 : {
441 : if (!waiter) {
442 : /*
443 : * The purpose of the mutex_can_spin_on_owner() function is
444 : * to eliminate the overhead of osq_lock() and osq_unlock()
445 : * in case spinning isn't possible. As a waiter-spinner
446 : * is not going to take OSQ lock anyway, there is no need
447 : * to call mutex_can_spin_on_owner().
448 : */
449 : if (!mutex_can_spin_on_owner(lock))
450 : goto fail;
451 :
452 : /*
453 : * In order to avoid a stampede of mutex spinners trying to
454 : * acquire the mutex all at once, the spinners need to take a
455 : * MCS (queued) lock first before spinning on the owner field.
456 : */
457 : if (!osq_lock(&lock->osq))
458 : goto fail;
459 : }
460 :
461 : for (;;) {
462 : struct task_struct *owner;
463 :
464 : /* Try to acquire the mutex... */
465 : owner = __mutex_trylock_or_owner(lock);
466 : if (!owner)
467 : break;
468 :
469 : /*
470 : * There's an owner, wait for it to either
471 : * release the lock or go to sleep.
472 : */
473 : if (!mutex_spin_on_owner(lock, owner, ww_ctx, waiter))
474 : goto fail_unlock;
475 :
476 : /*
477 : * The cpu_relax() call is a compiler barrier which forces
478 : * everything in this loop to be re-loaded. We don't need
479 : * memory barriers as we'll eventually observe the right
480 : * values at the cost of a few extra spins.
481 : */
482 : cpu_relax();
483 : }
484 :
485 : if (!waiter)
486 : osq_unlock(&lock->osq);
487 :
488 : return true;
489 :
490 :
491 : fail_unlock:
492 : if (!waiter)
493 : osq_unlock(&lock->osq);
494 :
495 : fail:
496 : /*
497 : * If we fell out of the spin path because of need_resched(),
498 : * reschedule now, before we try-lock the mutex. This avoids getting
499 : * scheduled out right after we obtained the mutex.
500 : */
501 : if (need_resched()) {
502 : /*
503 : * We _should_ have TASK_RUNNING here, but just in case
504 : * we do not, make it so, otherwise we might get stuck.
505 : */
506 : __set_current_state(TASK_RUNNING);
507 : schedule_preempt_disabled();
508 : }
509 :
510 : return false;
511 : }
512 : #else
513 : static __always_inline bool
514 : mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx,
515 : struct mutex_waiter *waiter)
516 : {
517 : return false;
518 : }
519 : #endif
520 :
521 : static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip);
522 :
523 : /**
524 : * mutex_unlock - release the mutex
525 : * @lock: the mutex to be released
526 : *
527 : * Unlock a mutex that has been locked by this task previously.
528 : *
529 : * This function must not be used in interrupt context. Unlocking
530 : * of a not locked mutex is not allowed.
531 : *
532 : * This function is similar to (but not equivalent to) up().
533 : */
534 3475 : void __sched mutex_unlock(struct mutex *lock)
535 : {
536 : #ifndef CONFIG_DEBUG_LOCK_ALLOC
537 3475 : if (__mutex_unlock_fast(lock))
538 : return;
539 : #endif
540 0 : __mutex_unlock_slowpath(lock, _RET_IP_);
541 : }
542 : EXPORT_SYMBOL(mutex_unlock);
543 :
544 : /**
545 : * ww_mutex_unlock - release the w/w mutex
546 : * @lock: the mutex to be released
547 : *
548 : * Unlock a mutex that has been locked by this task previously with any of the
549 : * ww_mutex_lock* functions (with or without an acquire context). It is
550 : * forbidden to release the locks after releasing the acquire context.
551 : *
552 : * This function must not be used in interrupt context. Unlocking
553 : * of a unlocked mutex is not allowed.
554 : */
555 0 : void __sched ww_mutex_unlock(struct ww_mutex *lock)
556 : {
557 0 : __ww_mutex_unlock(lock);
558 0 : mutex_unlock(&lock->base);
559 0 : }
560 : EXPORT_SYMBOL(ww_mutex_unlock);
561 :
562 : /*
563 : * Lock a mutex (possibly interruptible), slowpath:
564 : */
565 : static __always_inline int __sched
566 : __mutex_lock_common(struct mutex *lock, unsigned int state, unsigned int subclass,
567 : struct lockdep_map *nest_lock, unsigned long ip,
568 : struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx)
569 : {
570 : struct mutex_waiter waiter;
571 : struct ww_mutex *ww;
572 : int ret;
573 :
574 : if (!use_ww_ctx)
575 0 : ww_ctx = NULL;
576 :
577 : might_sleep();
578 :
579 : MUTEX_WARN_ON(lock->magic != lock);
580 :
581 0 : ww = container_of(lock, struct ww_mutex, base);
582 0 : if (ww_ctx) {
583 0 : if (unlikely(ww_ctx == READ_ONCE(ww->ctx)))
584 : return -EALREADY;
585 :
586 : /*
587 : * Reset the wounded flag after a kill. No other process can
588 : * race and wound us here since they can't have a valid owner
589 : * pointer if we don't have any locks held.
590 : */
591 0 : if (ww_ctx->acquired == 0)
592 0 : ww_ctx->wounded = 0;
593 :
594 : #ifdef CONFIG_DEBUG_LOCK_ALLOC
595 : nest_lock = &ww_ctx->dep_map;
596 : #endif
597 : }
598 :
599 0 : preempt_disable();
600 : mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip);
601 :
602 0 : if (__mutex_trylock(lock) ||
603 0 : mutex_optimistic_spin(lock, ww_ctx, NULL)) {
604 : /* got the lock, yay! */
605 : lock_acquired(&lock->dep_map, ip);
606 0 : if (ww_ctx)
607 : ww_mutex_set_context_fastpath(ww, ww_ctx);
608 0 : preempt_enable();
609 : return 0;
610 : }
611 :
612 0 : raw_spin_lock(&lock->wait_lock);
613 : /*
614 : * After waiting to acquire the wait_lock, try again.
615 : */
616 0 : if (__mutex_trylock(lock)) {
617 0 : if (ww_ctx)
618 0 : __ww_mutex_check_waiters(lock, ww_ctx);
619 :
620 : goto skip_wait;
621 : }
622 :
623 : debug_mutex_lock_common(lock, &waiter);
624 0 : waiter.task = current;
625 : if (use_ww_ctx)
626 0 : waiter.ww_ctx = ww_ctx;
627 :
628 : lock_contended(&lock->dep_map, ip);
629 :
630 : if (!use_ww_ctx) {
631 : /* add waiting tasks to the end of the waitqueue (FIFO): */
632 0 : __mutex_add_waiter(lock, &waiter, &lock->wait_list);
633 : } else {
634 : /*
635 : * Add in stamp order, waking up waiters that must kill
636 : * themselves.
637 : */
638 0 : ret = __ww_mutex_add_waiter(&waiter, lock, ww_ctx);
639 0 : if (ret)
640 : goto err_early_kill;
641 : }
642 :
643 0 : set_current_state(state);
644 : for (;;) {
645 : bool first;
646 :
647 : /*
648 : * Once we hold wait_lock, we're serialized against
649 : * mutex_unlock() handing the lock off to us, do a trylock
650 : * before testing the error conditions to make sure we pick up
651 : * the handoff.
652 : */
653 0 : if (__mutex_trylock(lock))
654 : goto acquired;
655 :
656 : /*
657 : * Check for signals and kill conditions while holding
658 : * wait_lock. This ensures the lock cancellation is ordered
659 : * against mutex_unlock() and wake-ups do not go missing.
660 : */
661 0 : if (signal_pending_state(state, current)) {
662 : ret = -EINTR;
663 : goto err;
664 : }
665 :
666 0 : if (ww_ctx) {
667 0 : ret = __ww_mutex_check_kill(lock, &waiter, ww_ctx);
668 0 : if (ret)
669 : goto err;
670 : }
671 :
672 0 : raw_spin_unlock(&lock->wait_lock);
673 0 : schedule_preempt_disabled();
674 :
675 0 : first = __mutex_waiter_is_first(lock, &waiter);
676 :
677 0 : set_current_state(state);
678 : /*
679 : * Here we order against unlock; we must either see it change
680 : * state back to RUNNING and fall through the next schedule(),
681 : * or we must see its unlock and acquire.
682 : */
683 0 : if (__mutex_trylock_or_handoff(lock, first) ||
684 : (first && mutex_optimistic_spin(lock, ww_ctx, &waiter)))
685 : break;
686 :
687 0 : raw_spin_lock(&lock->wait_lock);
688 : }
689 0 : raw_spin_lock(&lock->wait_lock);
690 : acquired:
691 0 : __set_current_state(TASK_RUNNING);
692 :
693 0 : if (ww_ctx) {
694 : /*
695 : * Wound-Wait; we stole the lock (!first_waiter), check the
696 : * waiters as anyone might want to wound us.
697 : */
698 0 : if (!ww_ctx->is_wait_die &&
699 0 : !__mutex_waiter_is_first(lock, &waiter))
700 0 : __ww_mutex_check_waiters(lock, ww_ctx);
701 : }
702 :
703 0 : __mutex_remove_waiter(lock, &waiter);
704 :
705 : debug_mutex_free_waiter(&waiter);
706 :
707 : skip_wait:
708 : /* got the lock - cleanup and rejoice! */
709 : lock_acquired(&lock->dep_map, ip);
710 :
711 0 : if (ww_ctx)
712 : ww_mutex_lock_acquired(ww, ww_ctx);
713 :
714 0 : raw_spin_unlock(&lock->wait_lock);
715 0 : preempt_enable();
716 : return 0;
717 :
718 : err:
719 0 : __set_current_state(TASK_RUNNING);
720 0 : __mutex_remove_waiter(lock, &waiter);
721 : err_early_kill:
722 0 : raw_spin_unlock(&lock->wait_lock);
723 : debug_mutex_free_waiter(&waiter);
724 : mutex_release(&lock->dep_map, ip);
725 0 : preempt_enable();
726 : return ret;
727 : }
728 :
729 : static int __sched
730 0 : __mutex_lock(struct mutex *lock, unsigned int state, unsigned int subclass,
731 : struct lockdep_map *nest_lock, unsigned long ip)
732 : {
733 0 : return __mutex_lock_common(lock, state, subclass, nest_lock, ip, NULL, false);
734 : }
735 :
736 : static int __sched
737 0 : __ww_mutex_lock(struct mutex *lock, unsigned int state, unsigned int subclass,
738 : unsigned long ip, struct ww_acquire_ctx *ww_ctx)
739 : {
740 0 : return __mutex_lock_common(lock, state, subclass, NULL, ip, ww_ctx, true);
741 : }
742 :
743 : /**
744 : * ww_mutex_trylock - tries to acquire the w/w mutex with optional acquire context
745 : * @ww: mutex to lock
746 : * @ww_ctx: optional w/w acquire context
747 : *
748 : * Trylocks a mutex with the optional acquire context; no deadlock detection is
749 : * possible. Returns 1 if the mutex has been acquired successfully, 0 otherwise.
750 : *
751 : * Unlike ww_mutex_lock, no deadlock handling is performed. However, if a @ctx is
752 : * specified, -EALREADY handling may happen in calls to ww_mutex_trylock.
753 : *
754 : * A mutex acquired with this function must be released with ww_mutex_unlock.
755 : */
756 0 : int ww_mutex_trylock(struct ww_mutex *ww, struct ww_acquire_ctx *ww_ctx)
757 : {
758 0 : if (!ww_ctx)
759 0 : return mutex_trylock(&ww->base);
760 :
761 : MUTEX_WARN_ON(ww->base.magic != &ww->base);
762 :
763 : /*
764 : * Reset the wounded flag after a kill. No other process can
765 : * race and wound us here, since they can't have a valid owner
766 : * pointer if we don't have any locks held.
767 : */
768 0 : if (ww_ctx->acquired == 0)
769 0 : ww_ctx->wounded = 0;
770 :
771 0 : if (__mutex_trylock(&ww->base)) {
772 : ww_mutex_set_context_fastpath(ww, ww_ctx);
773 : mutex_acquire_nest(&ww->base.dep_map, 0, 1, &ww_ctx->dep_map, _RET_IP_);
774 : return 1;
775 : }
776 :
777 : return 0;
778 : }
779 : EXPORT_SYMBOL(ww_mutex_trylock);
780 :
781 : #ifdef CONFIG_DEBUG_LOCK_ALLOC
782 : void __sched
783 : mutex_lock_nested(struct mutex *lock, unsigned int subclass)
784 : {
785 : __mutex_lock(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_);
786 : }
787 :
788 : EXPORT_SYMBOL_GPL(mutex_lock_nested);
789 :
790 : void __sched
791 : _mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest)
792 : {
793 : __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, nest, _RET_IP_);
794 : }
795 : EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock);
796 :
797 : int __sched
798 : mutex_lock_killable_nested(struct mutex *lock, unsigned int subclass)
799 : {
800 : return __mutex_lock(lock, TASK_KILLABLE, subclass, NULL, _RET_IP_);
801 : }
802 : EXPORT_SYMBOL_GPL(mutex_lock_killable_nested);
803 :
804 : int __sched
805 : mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
806 : {
807 : return __mutex_lock(lock, TASK_INTERRUPTIBLE, subclass, NULL, _RET_IP_);
808 : }
809 : EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
810 :
811 : void __sched
812 : mutex_lock_io_nested(struct mutex *lock, unsigned int subclass)
813 : {
814 : int token;
815 :
816 : might_sleep();
817 :
818 : token = io_schedule_prepare();
819 : __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE,
820 : subclass, NULL, _RET_IP_, NULL, 0);
821 : io_schedule_finish(token);
822 : }
823 : EXPORT_SYMBOL_GPL(mutex_lock_io_nested);
824 :
825 : static inline int
826 : ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
827 : {
828 : #ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH
829 : unsigned tmp;
830 :
831 : if (ctx->deadlock_inject_countdown-- == 0) {
832 : tmp = ctx->deadlock_inject_interval;
833 : if (tmp > UINT_MAX/4)
834 : tmp = UINT_MAX;
835 : else
836 : tmp = tmp*2 + tmp + tmp/2;
837 :
838 : ctx->deadlock_inject_interval = tmp;
839 : ctx->deadlock_inject_countdown = tmp;
840 : ctx->contending_lock = lock;
841 :
842 : ww_mutex_unlock(lock);
843 :
844 : return -EDEADLK;
845 : }
846 : #endif
847 :
848 : return 0;
849 : }
850 :
851 : int __sched
852 : ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
853 : {
854 : int ret;
855 :
856 : might_sleep();
857 : ret = __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE,
858 : 0, _RET_IP_, ctx);
859 : if (!ret && ctx && ctx->acquired > 1)
860 : return ww_mutex_deadlock_injection(lock, ctx);
861 :
862 : return ret;
863 : }
864 : EXPORT_SYMBOL_GPL(ww_mutex_lock);
865 :
866 : int __sched
867 : ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
868 : {
869 : int ret;
870 :
871 : might_sleep();
872 : ret = __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE,
873 : 0, _RET_IP_, ctx);
874 :
875 : if (!ret && ctx && ctx->acquired > 1)
876 : return ww_mutex_deadlock_injection(lock, ctx);
877 :
878 : return ret;
879 : }
880 : EXPORT_SYMBOL_GPL(ww_mutex_lock_interruptible);
881 :
882 : #endif
883 :
884 : /*
885 : * Release the lock, slowpath:
886 : */
887 0 : static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip)
888 : {
889 0 : struct task_struct *next = NULL;
890 0 : DEFINE_WAKE_Q(wake_q);
891 : unsigned long owner;
892 :
893 : mutex_release(&lock->dep_map, ip);
894 :
895 : /*
896 : * Release the lock before (potentially) taking the spinlock such that
897 : * other contenders can get on with things ASAP.
898 : *
899 : * Except when HANDOFF, in that case we must not clear the owner field,
900 : * but instead set it to the top waiter.
901 : */
902 0 : owner = atomic_long_read(&lock->owner);
903 : for (;;) {
904 : MUTEX_WARN_ON(__owner_task(owner) != current);
905 : MUTEX_WARN_ON(owner & MUTEX_FLAG_PICKUP);
906 :
907 0 : if (owner & MUTEX_FLAG_HANDOFF)
908 : break;
909 :
910 0 : if (atomic_long_try_cmpxchg_release(&lock->owner, &owner, __owner_flags(owner))) {
911 0 : if (owner & MUTEX_FLAG_WAITERS)
912 : break;
913 :
914 0 : return;
915 : }
916 : }
917 :
918 0 : raw_spin_lock(&lock->wait_lock);
919 : debug_mutex_unlock(lock);
920 0 : if (!list_empty(&lock->wait_list)) {
921 : /* get the first entry from the wait-list: */
922 0 : struct mutex_waiter *waiter =
923 0 : list_first_entry(&lock->wait_list,
924 : struct mutex_waiter, list);
925 :
926 0 : next = waiter->task;
927 :
928 : debug_mutex_wake_waiter(lock, waiter);
929 0 : wake_q_add(&wake_q, next);
930 : }
931 :
932 0 : if (owner & MUTEX_FLAG_HANDOFF)
933 : __mutex_handoff(lock, next);
934 :
935 0 : raw_spin_unlock(&lock->wait_lock);
936 :
937 0 : wake_up_q(&wake_q);
938 : }
939 :
940 : #ifndef CONFIG_DEBUG_LOCK_ALLOC
941 : /*
942 : * Here come the less common (and hence less performance-critical) APIs:
943 : * mutex_lock_interruptible() and mutex_trylock().
944 : */
945 : static noinline int __sched
946 : __mutex_lock_killable_slowpath(struct mutex *lock);
947 :
948 : static noinline int __sched
949 : __mutex_lock_interruptible_slowpath(struct mutex *lock);
950 :
951 : /**
952 : * mutex_lock_interruptible() - Acquire the mutex, interruptible by signals.
953 : * @lock: The mutex to be acquired.
954 : *
955 : * Lock the mutex like mutex_lock(). If a signal is delivered while the
956 : * process is sleeping, this function will return without acquiring the
957 : * mutex.
958 : *
959 : * Context: Process context.
960 : * Return: 0 if the lock was successfully acquired or %-EINTR if a
961 : * signal arrived.
962 : */
963 0 : int __sched mutex_lock_interruptible(struct mutex *lock)
964 : {
965 : might_sleep();
966 :
967 0 : if (__mutex_trylock_fast(lock))
968 : return 0;
969 :
970 0 : return __mutex_lock_interruptible_slowpath(lock);
971 : }
972 :
973 : EXPORT_SYMBOL(mutex_lock_interruptible);
974 :
975 : /**
976 : * mutex_lock_killable() - Acquire the mutex, interruptible by fatal signals.
977 : * @lock: The mutex to be acquired.
978 : *
979 : * Lock the mutex like mutex_lock(). If a signal which will be fatal to
980 : * the current process is delivered while the process is sleeping, this
981 : * function will return without acquiring the mutex.
982 : *
983 : * Context: Process context.
984 : * Return: 0 if the lock was successfully acquired or %-EINTR if a
985 : * fatal signal arrived.
986 : */
987 238 : int __sched mutex_lock_killable(struct mutex *lock)
988 : {
989 : might_sleep();
990 :
991 238 : if (__mutex_trylock_fast(lock))
992 : return 0;
993 :
994 0 : return __mutex_lock_killable_slowpath(lock);
995 : }
996 : EXPORT_SYMBOL(mutex_lock_killable);
997 :
998 : /**
999 : * mutex_lock_io() - Acquire the mutex and mark the process as waiting for I/O
1000 : * @lock: The mutex to be acquired.
1001 : *
1002 : * Lock the mutex like mutex_lock(). While the task is waiting for this
1003 : * mutex, it will be accounted as being in the IO wait state by the
1004 : * scheduler.
1005 : *
1006 : * Context: Process context.
1007 : */
1008 0 : void __sched mutex_lock_io(struct mutex *lock)
1009 : {
1010 : int token;
1011 :
1012 0 : token = io_schedule_prepare();
1013 0 : mutex_lock(lock);
1014 0 : io_schedule_finish(token);
1015 0 : }
1016 : EXPORT_SYMBOL_GPL(mutex_lock_io);
1017 :
1018 : static noinline void __sched
1019 0 : __mutex_lock_slowpath(struct mutex *lock)
1020 : {
1021 0 : __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_);
1022 0 : }
1023 :
1024 : static noinline int __sched
1025 0 : __mutex_lock_killable_slowpath(struct mutex *lock)
1026 : {
1027 0 : return __mutex_lock(lock, TASK_KILLABLE, 0, NULL, _RET_IP_);
1028 : }
1029 :
1030 : static noinline int __sched
1031 0 : __mutex_lock_interruptible_slowpath(struct mutex *lock)
1032 : {
1033 0 : return __mutex_lock(lock, TASK_INTERRUPTIBLE, 0, NULL, _RET_IP_);
1034 : }
1035 :
1036 : static noinline int __sched
1037 0 : __ww_mutex_lock_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1038 : {
1039 0 : return __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE, 0,
1040 0 : _RET_IP_, ctx);
1041 : }
1042 :
1043 : static noinline int __sched
1044 0 : __ww_mutex_lock_interruptible_slowpath(struct ww_mutex *lock,
1045 : struct ww_acquire_ctx *ctx)
1046 : {
1047 0 : return __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE, 0,
1048 0 : _RET_IP_, ctx);
1049 : }
1050 :
1051 : #endif
1052 :
1053 : /**
1054 : * mutex_trylock - try to acquire the mutex, without waiting
1055 : * @lock: the mutex to be acquired
1056 : *
1057 : * Try to acquire the mutex atomically. Returns 1 if the mutex
1058 : * has been acquired successfully, and 0 on contention.
1059 : *
1060 : * NOTE: this function follows the spin_trylock() convention, so
1061 : * it is negated from the down_trylock() return values! Be careful
1062 : * about this when converting semaphore users to mutexes.
1063 : *
1064 : * This function must not be used in interrupt context. The
1065 : * mutex must be released by the same task that acquired it.
1066 : */
1067 0 : int __sched mutex_trylock(struct mutex *lock)
1068 : {
1069 : bool locked;
1070 :
1071 : MUTEX_WARN_ON(lock->magic != lock);
1072 :
1073 0 : locked = __mutex_trylock(lock);
1074 : if (locked)
1075 : mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
1076 :
1077 0 : return locked;
1078 : }
1079 : EXPORT_SYMBOL(mutex_trylock);
1080 :
1081 : #ifndef CONFIG_DEBUG_LOCK_ALLOC
1082 : int __sched
1083 0 : ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1084 : {
1085 : might_sleep();
1086 :
1087 0 : if (__mutex_trylock_fast(&lock->base)) {
1088 0 : if (ctx)
1089 : ww_mutex_set_context_fastpath(lock, ctx);
1090 : return 0;
1091 : }
1092 :
1093 0 : return __ww_mutex_lock_slowpath(lock, ctx);
1094 : }
1095 : EXPORT_SYMBOL(ww_mutex_lock);
1096 :
1097 : int __sched
1098 0 : ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx)
1099 : {
1100 : might_sleep();
1101 :
1102 0 : if (__mutex_trylock_fast(&lock->base)) {
1103 0 : if (ctx)
1104 : ww_mutex_set_context_fastpath(lock, ctx);
1105 : return 0;
1106 : }
1107 :
1108 0 : return __ww_mutex_lock_interruptible_slowpath(lock, ctx);
1109 : }
1110 : EXPORT_SYMBOL(ww_mutex_lock_interruptible);
1111 :
1112 : #endif /* !CONFIG_DEBUG_LOCK_ALLOC */
1113 : #endif /* !CONFIG_PREEMPT_RT */
1114 :
1115 : /**
1116 : * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0
1117 : * @cnt: the atomic which we are to dec
1118 : * @lock: the mutex to return holding if we dec to 0
1119 : *
1120 : * return true and hold lock if we dec to 0, return false otherwise
1121 : */
1122 0 : int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock)
1123 : {
1124 : /* dec if we can't possibly hit 0 */
1125 0 : if (atomic_add_unless(cnt, -1, 1))
1126 : return 0;
1127 : /* we might hit 0, so take the lock */
1128 0 : mutex_lock(lock);
1129 0 : if (!atomic_dec_and_test(cnt)) {
1130 : /* when we actually did the dec, we didn't hit 0 */
1131 : mutex_unlock(lock);
1132 : return 0;
1133 : }
1134 : /* we hit 0, and we hold the lock */
1135 : return 1;
1136 : }
1137 : EXPORT_SYMBOL(atomic_dec_and_mutex_lock);
|