]>
Commit | Line | Data |
---|---|---|
6053ee3b | 1 | /* |
67a6de49 | 2 | * kernel/locking/mutex.c |
6053ee3b IM |
3 | * |
4 | * Mutexes: blocking mutual exclusion locks | |
5 | * | |
6 | * Started by Ingo Molnar: | |
7 | * | |
8 | * Copyright (C) 2004, 2005, 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
9 | * | |
10 | * Many thanks to Arjan van de Ven, Thomas Gleixner, Steven Rostedt and | |
11 | * David Howells for suggestions and improvements. | |
12 | * | |
0d66bf6d PZ |
13 | * - Adaptive spinning for mutexes by Peter Zijlstra. (Ported to mainline |
14 | * from the -rt tree, where it was originally implemented for rtmutexes | |
15 | * by Steven Rostedt, based on work by Gregory Haskins, Peter Morreale | |
16 | * and Sven Dietrich. | |
17 | * | |
214e0aed | 18 | * Also see Documentation/locking/mutex-design.txt. |
6053ee3b IM |
19 | */ |
20 | #include <linux/mutex.h> | |
1b375dc3 | 21 | #include <linux/ww_mutex.h> |
174cd4b1 | 22 | #include <linux/sched/signal.h> |
8bd75c77 | 23 | #include <linux/sched/rt.h> |
84f001e1 | 24 | #include <linux/sched/wake_q.h> |
9984de1a | 25 | #include <linux/export.h> |
6053ee3b IM |
26 | #include <linux/spinlock.h> |
27 | #include <linux/interrupt.h> | |
9a11b49a | 28 | #include <linux/debug_locks.h> |
7a215f89 | 29 | #include <linux/osq_lock.h> |
6053ee3b | 30 | |
6053ee3b IM |
31 | #ifdef CONFIG_DEBUG_MUTEXES |
32 | # include "mutex-debug.h" | |
6053ee3b IM |
33 | #else |
34 | # include "mutex.h" | |
6053ee3b IM |
35 | #endif |
36 | ||
ef5d4707 IM |
37 | void |
38 | __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key) | |
6053ee3b | 39 | { |
3ca0ff57 | 40 | atomic_long_set(&lock->owner, 0); |
6053ee3b IM |
41 | spin_lock_init(&lock->wait_lock); |
42 | INIT_LIST_HEAD(&lock->wait_list); | |
2bd2c92c | 43 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
4d9d951e | 44 | osq_lock_init(&lock->osq); |
2bd2c92c | 45 | #endif |
6053ee3b | 46 | |
ef5d4707 | 47 | debug_mutex_init(lock, name, key); |
6053ee3b | 48 | } |
6053ee3b IM |
49 | EXPORT_SYMBOL(__mutex_init); |
50 | ||
3ca0ff57 PZ |
51 | /* |
52 | * @owner: contains: 'struct task_struct *' to the current lock owner, | |
53 | * NULL means not owned. Since task_struct pointers are aligned at | |
e274795e | 54 | * at least L1_CACHE_BYTES, we have low bits to store extra state. |
3ca0ff57 PZ |
55 | * |
56 | * Bit0 indicates a non-empty waiter list; unlock must issue a wakeup. | |
9d659ae1 | 57 | * Bit1 indicates unlock needs to hand the lock to the top-waiter |
e274795e | 58 | * Bit2 indicates handoff has been done and we're waiting for pickup. |
3ca0ff57 PZ |
59 | */ |
60 | #define MUTEX_FLAG_WAITERS 0x01 | |
9d659ae1 | 61 | #define MUTEX_FLAG_HANDOFF 0x02 |
e274795e | 62 | #define MUTEX_FLAG_PICKUP 0x04 |
3ca0ff57 | 63 | |
e274795e | 64 | #define MUTEX_FLAGS 0x07 |
3ca0ff57 PZ |
65 | |
66 | static inline struct task_struct *__owner_task(unsigned long owner) | |
67 | { | |
68 | return (struct task_struct *)(owner & ~MUTEX_FLAGS); | |
69 | } | |
70 | ||
71 | static inline unsigned long __owner_flags(unsigned long owner) | |
72 | { | |
73 | return owner & MUTEX_FLAGS; | |
74 | } | |
75 | ||
76 | /* | |
e274795e | 77 | * Trylock variant that retuns the owning task on failure. |
3ca0ff57 | 78 | */ |
e274795e | 79 | static inline struct task_struct *__mutex_trylock_or_owner(struct mutex *lock) |
3ca0ff57 PZ |
80 | { |
81 | unsigned long owner, curr = (unsigned long)current; | |
82 | ||
83 | owner = atomic_long_read(&lock->owner); | |
84 | for (;;) { /* must loop, can race against a flag */ | |
9d659ae1 | 85 | unsigned long old, flags = __owner_flags(owner); |
e274795e | 86 | unsigned long task = owner & ~MUTEX_FLAGS; |
9d659ae1 | 87 | |
e274795e PZ |
88 | if (task) { |
89 | if (likely(task != curr)) | |
90 | break; | |
3ca0ff57 | 91 | |
e274795e PZ |
92 | if (likely(!(flags & MUTEX_FLAG_PICKUP))) |
93 | break; | |
9d659ae1 | 94 | |
e274795e PZ |
95 | flags &= ~MUTEX_FLAG_PICKUP; |
96 | } else { | |
97 | #ifdef CONFIG_DEBUG_MUTEXES | |
98 | DEBUG_LOCKS_WARN_ON(flags & MUTEX_FLAG_PICKUP); | |
99 | #endif | |
9d659ae1 PZ |
100 | } |
101 | ||
102 | /* | |
103 | * We set the HANDOFF bit, we must make sure it doesn't live | |
104 | * past the point where we acquire it. This would be possible | |
105 | * if we (accidentally) set the bit on an unlocked mutex. | |
106 | */ | |
e274795e | 107 | flags &= ~MUTEX_FLAG_HANDOFF; |
3ca0ff57 | 108 | |
9d659ae1 | 109 | old = atomic_long_cmpxchg_acquire(&lock->owner, owner, curr | flags); |
3ca0ff57 | 110 | if (old == owner) |
e274795e | 111 | return NULL; |
3ca0ff57 PZ |
112 | |
113 | owner = old; | |
114 | } | |
e274795e PZ |
115 | |
116 | return __owner_task(owner); | |
117 | } | |
118 | ||
119 | /* | |
120 | * Actual trylock that will work on any unlocked state. | |
121 | */ | |
122 | static inline bool __mutex_trylock(struct mutex *lock) | |
123 | { | |
124 | return !__mutex_trylock_or_owner(lock); | |
3ca0ff57 PZ |
125 | } |
126 | ||
127 | #ifndef CONFIG_DEBUG_LOCK_ALLOC | |
128 | /* | |
129 | * Lockdep annotations are contained to the slow paths for simplicity. | |
130 | * There is nothing that would stop spreading the lockdep annotations outwards | |
131 | * except more code. | |
132 | */ | |
133 | ||
134 | /* | |
135 | * Optimistic trylock that only works in the uncontended case. Make sure to | |
136 | * follow with a __mutex_trylock() before failing. | |
137 | */ | |
138 | static __always_inline bool __mutex_trylock_fast(struct mutex *lock) | |
139 | { | |
140 | unsigned long curr = (unsigned long)current; | |
141 | ||
142 | if (!atomic_long_cmpxchg_acquire(&lock->owner, 0UL, curr)) | |
143 | return true; | |
144 | ||
145 | return false; | |
146 | } | |
147 | ||
148 | static __always_inline bool __mutex_unlock_fast(struct mutex *lock) | |
149 | { | |
150 | unsigned long curr = (unsigned long)current; | |
151 | ||
152 | if (atomic_long_cmpxchg_release(&lock->owner, curr, 0UL) == curr) | |
153 | return true; | |
154 | ||
155 | return false; | |
156 | } | |
157 | #endif | |
158 | ||
159 | static inline void __mutex_set_flag(struct mutex *lock, unsigned long flag) | |
160 | { | |
161 | atomic_long_or(flag, &lock->owner); | |
162 | } | |
163 | ||
164 | static inline void __mutex_clear_flag(struct mutex *lock, unsigned long flag) | |
165 | { | |
166 | atomic_long_andnot(flag, &lock->owner); | |
167 | } | |
168 | ||
9d659ae1 PZ |
169 | static inline bool __mutex_waiter_is_first(struct mutex *lock, struct mutex_waiter *waiter) |
170 | { | |
171 | return list_first_entry(&lock->wait_list, struct mutex_waiter, list) == waiter; | |
172 | } | |
173 | ||
174 | /* | |
175 | * Give up ownership to a specific task, when @task = NULL, this is equivalent | |
e274795e PZ |
176 | * to a regular unlock. Sets PICKUP on a handoff, clears HANDOF, preserves |
177 | * WAITERS. Provides RELEASE semantics like a regular unlock, the | |
178 | * __mutex_trylock() provides a matching ACQUIRE semantics for the handoff. | |
9d659ae1 PZ |
179 | */ |
180 | static void __mutex_handoff(struct mutex *lock, struct task_struct *task) | |
181 | { | |
182 | unsigned long owner = atomic_long_read(&lock->owner); | |
183 | ||
184 | for (;;) { | |
185 | unsigned long old, new; | |
186 | ||
187 | #ifdef CONFIG_DEBUG_MUTEXES | |
188 | DEBUG_LOCKS_WARN_ON(__owner_task(owner) != current); | |
e274795e | 189 | DEBUG_LOCKS_WARN_ON(owner & MUTEX_FLAG_PICKUP); |
9d659ae1 PZ |
190 | #endif |
191 | ||
192 | new = (owner & MUTEX_FLAG_WAITERS); | |
193 | new |= (unsigned long)task; | |
e274795e PZ |
194 | if (task) |
195 | new |= MUTEX_FLAG_PICKUP; | |
9d659ae1 PZ |
196 | |
197 | old = atomic_long_cmpxchg_release(&lock->owner, owner, new); | |
198 | if (old == owner) | |
199 | break; | |
200 | ||
201 | owner = old; | |
202 | } | |
203 | } | |
204 | ||
e4564f79 | 205 | #ifndef CONFIG_DEBUG_LOCK_ALLOC |
6053ee3b IM |
206 | /* |
207 | * We split the mutex lock/unlock logic into separate fastpath and | |
208 | * slowpath functions, to reduce the register pressure on the fastpath. | |
209 | * We also put the fastpath first in the kernel image, to make sure the | |
210 | * branch is predicted by the CPU as default-untaken. | |
211 | */ | |
3ca0ff57 | 212 | static void __sched __mutex_lock_slowpath(struct mutex *lock); |
6053ee3b | 213 | |
ef5dc121 | 214 | /** |
6053ee3b IM |
215 | * mutex_lock - acquire the mutex |
216 | * @lock: the mutex to be acquired | |
217 | * | |
218 | * Lock the mutex exclusively for this task. If the mutex is not | |
219 | * available right now, it will sleep until it can get it. | |
220 | * | |
221 | * The mutex must later on be released by the same task that | |
222 | * acquired it. Recursive locking is not allowed. The task | |
223 | * may not exit without first unlocking the mutex. Also, kernel | |
139b6fd2 | 224 | * memory where the mutex resides must not be freed with |
6053ee3b IM |
225 | * the mutex still locked. The mutex must first be initialized |
226 | * (or statically defined) before it can be locked. memset()-ing | |
227 | * the mutex to 0 is not allowed. | |
228 | * | |
229 | * ( The CONFIG_DEBUG_MUTEXES .config option turns on debugging | |
230 | * checks that will enforce the restrictions and will also do | |
231 | * deadlock debugging. ) | |
232 | * | |
233 | * This function is similar to (but not equivalent to) down(). | |
234 | */ | |
b09d2501 | 235 | void __sched mutex_lock(struct mutex *lock) |
6053ee3b | 236 | { |
c544bdb1 | 237 | might_sleep(); |
6053ee3b | 238 | |
3ca0ff57 PZ |
239 | if (!__mutex_trylock_fast(lock)) |
240 | __mutex_lock_slowpath(lock); | |
241 | } | |
6053ee3b | 242 | EXPORT_SYMBOL(mutex_lock); |
e4564f79 | 243 | #endif |
6053ee3b | 244 | |
427b1820 PZ |
245 | static __always_inline void |
246 | ww_mutex_lock_acquired(struct ww_mutex *ww, struct ww_acquire_ctx *ww_ctx) | |
76916515 DB |
247 | { |
248 | #ifdef CONFIG_DEBUG_MUTEXES | |
249 | /* | |
250 | * If this WARN_ON triggers, you used ww_mutex_lock to acquire, | |
251 | * but released with a normal mutex_unlock in this call. | |
252 | * | |
253 | * This should never happen, always use ww_mutex_unlock. | |
254 | */ | |
255 | DEBUG_LOCKS_WARN_ON(ww->ctx); | |
256 | ||
257 | /* | |
258 | * Not quite done after calling ww_acquire_done() ? | |
259 | */ | |
260 | DEBUG_LOCKS_WARN_ON(ww_ctx->done_acquire); | |
261 | ||
262 | if (ww_ctx->contending_lock) { | |
263 | /* | |
264 | * After -EDEADLK you tried to | |
265 | * acquire a different ww_mutex? Bad! | |
266 | */ | |
267 | DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock != ww); | |
268 | ||
269 | /* | |
270 | * You called ww_mutex_lock after receiving -EDEADLK, | |
271 | * but 'forgot' to unlock everything else first? | |
272 | */ | |
273 | DEBUG_LOCKS_WARN_ON(ww_ctx->acquired > 0); | |
274 | ww_ctx->contending_lock = NULL; | |
275 | } | |
276 | ||
277 | /* | |
278 | * Naughty, using a different class will lead to undefined behavior! | |
279 | */ | |
280 | DEBUG_LOCKS_WARN_ON(ww_ctx->ww_class != ww->ww_class); | |
281 | #endif | |
282 | ww_ctx->acquired++; | |
283 | } | |
284 | ||
3822da3e NH |
285 | static inline bool __sched |
286 | __ww_ctx_stamp_after(struct ww_acquire_ctx *a, struct ww_acquire_ctx *b) | |
287 | { | |
288 | return a->stamp - b->stamp <= LONG_MAX && | |
289 | (a->stamp != b->stamp || a > b); | |
290 | } | |
291 | ||
659cf9f5 NH |
292 | /* |
293 | * Wake up any waiters that may have to back off when the lock is held by the | |
294 | * given context. | |
295 | * | |
296 | * Due to the invariants on the wait list, this can only affect the first | |
297 | * waiter with a context. | |
298 | * | |
299 | * The current task must not be on the wait list. | |
300 | */ | |
301 | static void __sched | |
302 | __ww_mutex_wakeup_for_backoff(struct mutex *lock, struct ww_acquire_ctx *ww_ctx) | |
303 | { | |
304 | struct mutex_waiter *cur; | |
305 | ||
306 | lockdep_assert_held(&lock->wait_lock); | |
307 | ||
308 | list_for_each_entry(cur, &lock->wait_list, list) { | |
309 | if (!cur->ww_ctx) | |
310 | continue; | |
311 | ||
312 | if (cur->ww_ctx->acquired > 0 && | |
313 | __ww_ctx_stamp_after(cur->ww_ctx, ww_ctx)) { | |
314 | debug_mutex_wake_waiter(lock, cur); | |
315 | wake_up_process(cur->task); | |
316 | } | |
317 | ||
318 | break; | |
319 | } | |
320 | } | |
321 | ||
76916515 | 322 | /* |
4bd19084 | 323 | * After acquiring lock with fastpath or when we lost out in contested |
76916515 | 324 | * slowpath, set ctx and wake up any waiters so they can recheck. |
76916515 DB |
325 | */ |
326 | static __always_inline void | |
427b1820 | 327 | ww_mutex_set_context_fastpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
76916515 | 328 | { |
76916515 DB |
329 | ww_mutex_lock_acquired(lock, ctx); |
330 | ||
331 | lock->ctx = ctx; | |
332 | ||
333 | /* | |
334 | * The lock->ctx update should be visible on all cores before | |
335 | * the atomic read is done, otherwise contended waiters might be | |
336 | * missed. The contended waiters will either see ww_ctx == NULL | |
337 | * and keep spinning, or it will acquire wait_lock, add itself | |
338 | * to waiter list and sleep. | |
339 | */ | |
340 | smp_mb(); /* ^^^ */ | |
341 | ||
342 | /* | |
343 | * Check if lock is contended, if not there is nobody to wake up | |
344 | */ | |
3ca0ff57 | 345 | if (likely(!(atomic_long_read(&lock->base.owner) & MUTEX_FLAG_WAITERS))) |
76916515 DB |
346 | return; |
347 | ||
348 | /* | |
349 | * Uh oh, we raced in fastpath, wake up everyone in this case, | |
350 | * so they can see the new lock->ctx. | |
351 | */ | |
b9c16a0e | 352 | spin_lock(&lock->base.wait_lock); |
659cf9f5 | 353 | __ww_mutex_wakeup_for_backoff(&lock->base, ctx); |
b9c16a0e | 354 | spin_unlock(&lock->base.wait_lock); |
76916515 DB |
355 | } |
356 | ||
4bd19084 | 357 | /* |
659cf9f5 NH |
358 | * After acquiring lock in the slowpath set ctx. |
359 | * | |
360 | * Unlike for the fast path, the caller ensures that waiters are woken up where | |
361 | * necessary. | |
4bd19084 DB |
362 | * |
363 | * Callers must hold the mutex wait_lock. | |
364 | */ | |
365 | static __always_inline void | |
427b1820 | 366 | ww_mutex_set_context_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
4bd19084 | 367 | { |
4bd19084 DB |
368 | ww_mutex_lock_acquired(lock, ctx); |
369 | lock->ctx = ctx; | |
4bd19084 | 370 | } |
76916515 | 371 | |
41fcb9f2 | 372 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
c516df97 NH |
373 | |
374 | static inline | |
375 | bool ww_mutex_spin_on_owner(struct mutex *lock, struct ww_acquire_ctx *ww_ctx, | |
376 | struct mutex_waiter *waiter) | |
377 | { | |
378 | struct ww_mutex *ww; | |
379 | ||
380 | ww = container_of(lock, struct ww_mutex, base); | |
4bd19084 DB |
381 | |
382 | /* | |
c516df97 NH |
383 | * If ww->ctx is set the contents are undefined, only |
384 | * by acquiring wait_lock there is a guarantee that | |
385 | * they are not invalid when reading. | |
386 | * | |
387 | * As such, when deadlock detection needs to be | |
388 | * performed the optimistic spinning cannot be done. | |
389 | * | |
390 | * Check this in every inner iteration because we may | |
391 | * be racing against another thread's ww_mutex_lock. | |
4bd19084 | 392 | */ |
c516df97 NH |
393 | if (ww_ctx->acquired > 0 && READ_ONCE(ww->ctx)) |
394 | return false; | |
395 | ||
396 | /* | |
397 | * If we aren't on the wait list yet, cancel the spin | |
398 | * if there are waiters. We want to avoid stealing the | |
399 | * lock from a waiter with an earlier stamp, since the | |
400 | * other thread may already own a lock that we also | |
401 | * need. | |
402 | */ | |
403 | if (!waiter && (atomic_long_read(&lock->owner) & MUTEX_FLAG_WAITERS)) | |
404 | return false; | |
405 | ||
406 | /* | |
407 | * Similarly, stop spinning if we are no longer the | |
408 | * first waiter. | |
409 | */ | |
410 | if (waiter && !__mutex_waiter_is_first(lock, waiter)) | |
411 | return false; | |
412 | ||
413 | return true; | |
4bd19084 | 414 | } |
76916515 | 415 | |
41fcb9f2 | 416 | /* |
25f13b40 NH |
417 | * Look out! "owner" is an entirely speculative pointer access and not |
418 | * reliable. | |
419 | * | |
420 | * "noinline" so that this function shows up on perf profiles. | |
41fcb9f2 WL |
421 | */ |
422 | static noinline | |
25f13b40 | 423 | bool mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner, |
c516df97 | 424 | struct ww_acquire_ctx *ww_ctx, struct mutex_waiter *waiter) |
41fcb9f2 | 425 | { |
01ac33c1 | 426 | bool ret = true; |
be1f7bf2 | 427 | |
41fcb9f2 | 428 | rcu_read_lock(); |
3ca0ff57 | 429 | while (__mutex_owner(lock) == owner) { |
be1f7bf2 JL |
430 | /* |
431 | * Ensure we emit the owner->on_cpu, dereference _after_ | |
01ac33c1 JL |
432 | * checking lock->owner still matches owner. If that fails, |
433 | * owner might point to freed memory. If it still matches, | |
be1f7bf2 JL |
434 | * the rcu_read_lock() ensures the memory stays valid. |
435 | */ | |
436 | barrier(); | |
437 | ||
05ffc951 PX |
438 | /* |
439 | * Use vcpu_is_preempted to detect lock holder preemption issue. | |
440 | */ | |
441 | if (!owner->on_cpu || need_resched() || | |
442 | vcpu_is_preempted(task_cpu(owner))) { | |
be1f7bf2 JL |
443 | ret = false; |
444 | break; | |
445 | } | |
41fcb9f2 | 446 | |
c516df97 NH |
447 | if (ww_ctx && !ww_mutex_spin_on_owner(lock, ww_ctx, waiter)) { |
448 | ret = false; | |
449 | break; | |
25f13b40 NH |
450 | } |
451 | ||
f2f09a4c | 452 | cpu_relax(); |
41fcb9f2 WL |
453 | } |
454 | rcu_read_unlock(); | |
455 | ||
be1f7bf2 | 456 | return ret; |
41fcb9f2 | 457 | } |
2bd2c92c WL |
458 | |
459 | /* | |
460 | * Initial check for entering the mutex spinning loop | |
461 | */ | |
462 | static inline int mutex_can_spin_on_owner(struct mutex *lock) | |
463 | { | |
1e40c2ed | 464 | struct task_struct *owner; |
2bd2c92c WL |
465 | int retval = 1; |
466 | ||
46af29e4 JL |
467 | if (need_resched()) |
468 | return 0; | |
469 | ||
2bd2c92c | 470 | rcu_read_lock(); |
3ca0ff57 | 471 | owner = __mutex_owner(lock); |
05ffc951 PX |
472 | |
473 | /* | |
474 | * As lock holder preemption issue, we both skip spinning if task is not | |
475 | * on cpu or its cpu is preempted | |
476 | */ | |
1e40c2ed | 477 | if (owner) |
05ffc951 | 478 | retval = owner->on_cpu && !vcpu_is_preempted(task_cpu(owner)); |
2bd2c92c | 479 | rcu_read_unlock(); |
3ca0ff57 | 480 | |
2bd2c92c | 481 | /* |
3ca0ff57 PZ |
482 | * If lock->owner is not set, the mutex has been released. Return true |
483 | * such that we'll trylock in the spin path, which is a faster option | |
484 | * than the blocking slow path. | |
2bd2c92c WL |
485 | */ |
486 | return retval; | |
487 | } | |
76916515 | 488 | |
76916515 DB |
489 | /* |
490 | * Optimistic spinning. | |
491 | * | |
492 | * We try to spin for acquisition when we find that the lock owner | |
493 | * is currently running on a (different) CPU and while we don't | |
494 | * need to reschedule. The rationale is that if the lock owner is | |
495 | * running, it is likely to release the lock soon. | |
496 | * | |
76916515 DB |
497 | * The mutex spinners are queued up using MCS lock so that only one |
498 | * spinner can compete for the mutex. However, if mutex spinning isn't | |
499 | * going to happen, there is no point in going through the lock/unlock | |
500 | * overhead. | |
501 | * | |
502 | * Returns true when the lock was taken, otherwise false, indicating | |
503 | * that we need to jump to the slowpath and sleep. | |
b341afb3 WL |
504 | * |
505 | * The waiter flag is set to true if the spinner is a waiter in the wait | |
506 | * queue. The waiter-spinner will spin on the lock directly and concurrently | |
507 | * with the spinner at the head of the OSQ, if present, until the owner is | |
508 | * changed to itself. | |
76916515 | 509 | */ |
427b1820 PZ |
510 | static __always_inline bool |
511 | mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx, | |
c516df97 | 512 | const bool use_ww_ctx, struct mutex_waiter *waiter) |
76916515 | 513 | { |
b341afb3 WL |
514 | if (!waiter) { |
515 | /* | |
516 | * The purpose of the mutex_can_spin_on_owner() function is | |
517 | * to eliminate the overhead of osq_lock() and osq_unlock() | |
518 | * in case spinning isn't possible. As a waiter-spinner | |
519 | * is not going to take OSQ lock anyway, there is no need | |
520 | * to call mutex_can_spin_on_owner(). | |
521 | */ | |
522 | if (!mutex_can_spin_on_owner(lock)) | |
523 | goto fail; | |
76916515 | 524 | |
b341afb3 WL |
525 | /* |
526 | * In order to avoid a stampede of mutex spinners trying to | |
527 | * acquire the mutex all at once, the spinners need to take a | |
528 | * MCS (queued) lock first before spinning on the owner field. | |
529 | */ | |
530 | if (!osq_lock(&lock->osq)) | |
531 | goto fail; | |
532 | } | |
76916515 | 533 | |
b341afb3 | 534 | for (;;) { |
76916515 DB |
535 | struct task_struct *owner; |
536 | ||
e274795e PZ |
537 | /* Try to acquire the mutex... */ |
538 | owner = __mutex_trylock_or_owner(lock); | |
539 | if (!owner) | |
540 | break; | |
76916515 DB |
541 | |
542 | /* | |
e274795e | 543 | * There's an owner, wait for it to either |
76916515 DB |
544 | * release the lock or go to sleep. |
545 | */ | |
c516df97 | 546 | if (!mutex_spin_on_owner(lock, owner, ww_ctx, waiter)) |
e274795e | 547 | goto fail_unlock; |
b341afb3 | 548 | |
76916515 DB |
549 | /* |
550 | * The cpu_relax() call is a compiler barrier which forces | |
551 | * everything in this loop to be re-loaded. We don't need | |
552 | * memory barriers as we'll eventually observe the right | |
553 | * values at the cost of a few extra spins. | |
554 | */ | |
f2f09a4c | 555 | cpu_relax(); |
76916515 DB |
556 | } |
557 | ||
b341afb3 WL |
558 | if (!waiter) |
559 | osq_unlock(&lock->osq); | |
560 | ||
561 | return true; | |
562 | ||
563 | ||
564 | fail_unlock: | |
565 | if (!waiter) | |
566 | osq_unlock(&lock->osq); | |
567 | ||
568 | fail: | |
76916515 DB |
569 | /* |
570 | * If we fell out of the spin path because of need_resched(), | |
571 | * reschedule now, before we try-lock the mutex. This avoids getting | |
572 | * scheduled out right after we obtained the mutex. | |
573 | */ | |
6f942a1f PZ |
574 | if (need_resched()) { |
575 | /* | |
576 | * We _should_ have TASK_RUNNING here, but just in case | |
577 | * we do not, make it so, otherwise we might get stuck. | |
578 | */ | |
579 | __set_current_state(TASK_RUNNING); | |
76916515 | 580 | schedule_preempt_disabled(); |
6f942a1f | 581 | } |
76916515 DB |
582 | |
583 | return false; | |
584 | } | |
585 | #else | |
427b1820 PZ |
586 | static __always_inline bool |
587 | mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx, | |
c516df97 | 588 | const bool use_ww_ctx, struct mutex_waiter *waiter) |
76916515 DB |
589 | { |
590 | return false; | |
591 | } | |
41fcb9f2 WL |
592 | #endif |
593 | ||
3ca0ff57 | 594 | static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip); |
6053ee3b | 595 | |
ef5dc121 | 596 | /** |
6053ee3b IM |
597 | * mutex_unlock - release the mutex |
598 | * @lock: the mutex to be released | |
599 | * | |
600 | * Unlock a mutex that has been locked by this task previously. | |
601 | * | |
602 | * This function must not be used in interrupt context. Unlocking | |
603 | * of a not locked mutex is not allowed. | |
604 | * | |
605 | * This function is similar to (but not equivalent to) up(). | |
606 | */ | |
7ad5b3a5 | 607 | void __sched mutex_unlock(struct mutex *lock) |
6053ee3b | 608 | { |
3ca0ff57 PZ |
609 | #ifndef CONFIG_DEBUG_LOCK_ALLOC |
610 | if (__mutex_unlock_fast(lock)) | |
611 | return; | |
0d66bf6d | 612 | #endif |
3ca0ff57 | 613 | __mutex_unlock_slowpath(lock, _RET_IP_); |
6053ee3b | 614 | } |
6053ee3b IM |
615 | EXPORT_SYMBOL(mutex_unlock); |
616 | ||
040a0a37 ML |
617 | /** |
618 | * ww_mutex_unlock - release the w/w mutex | |
619 | * @lock: the mutex to be released | |
620 | * | |
621 | * Unlock a mutex that has been locked by this task previously with any of the | |
622 | * ww_mutex_lock* functions (with or without an acquire context). It is | |
623 | * forbidden to release the locks after releasing the acquire context. | |
624 | * | |
625 | * This function must not be used in interrupt context. Unlocking | |
626 | * of a unlocked mutex is not allowed. | |
627 | */ | |
628 | void __sched ww_mutex_unlock(struct ww_mutex *lock) | |
629 | { | |
630 | /* | |
631 | * The unlocking fastpath is the 0->1 transition from 'locked' | |
632 | * into 'unlocked' state: | |
633 | */ | |
634 | if (lock->ctx) { | |
635 | #ifdef CONFIG_DEBUG_MUTEXES | |
636 | DEBUG_LOCKS_WARN_ON(!lock->ctx->acquired); | |
637 | #endif | |
638 | if (lock->ctx->acquired > 0) | |
639 | lock->ctx->acquired--; | |
640 | lock->ctx = NULL; | |
641 | } | |
642 | ||
3ca0ff57 | 643 | mutex_unlock(&lock->base); |
040a0a37 ML |
644 | } |
645 | EXPORT_SYMBOL(ww_mutex_unlock); | |
646 | ||
647 | static inline int __sched | |
200b1874 NH |
648 | __ww_mutex_lock_check_stamp(struct mutex *lock, struct mutex_waiter *waiter, |
649 | struct ww_acquire_ctx *ctx) | |
040a0a37 ML |
650 | { |
651 | struct ww_mutex *ww = container_of(lock, struct ww_mutex, base); | |
4d3199e4 | 652 | struct ww_acquire_ctx *hold_ctx = READ_ONCE(ww->ctx); |
200b1874 | 653 | struct mutex_waiter *cur; |
040a0a37 | 654 | |
200b1874 NH |
655 | if (hold_ctx && __ww_ctx_stamp_after(ctx, hold_ctx)) |
656 | goto deadlock; | |
040a0a37 | 657 | |
200b1874 NH |
658 | /* |
659 | * If there is a waiter in front of us that has a context, then its | |
660 | * stamp is earlier than ours and we must back off. | |
661 | */ | |
662 | cur = waiter; | |
663 | list_for_each_entry_continue_reverse(cur, &lock->wait_list, list) { | |
664 | if (cur->ww_ctx) | |
665 | goto deadlock; | |
040a0a37 ML |
666 | } |
667 | ||
668 | return 0; | |
200b1874 NH |
669 | |
670 | deadlock: | |
671 | #ifdef CONFIG_DEBUG_MUTEXES | |
672 | DEBUG_LOCKS_WARN_ON(ctx->contending_lock); | |
673 | ctx->contending_lock = ww; | |
674 | #endif | |
675 | return -EDEADLK; | |
040a0a37 ML |
676 | } |
677 | ||
6baa5c60 NH |
678 | static inline int __sched |
679 | __ww_mutex_add_waiter(struct mutex_waiter *waiter, | |
680 | struct mutex *lock, | |
681 | struct ww_acquire_ctx *ww_ctx) | |
682 | { | |
683 | struct mutex_waiter *cur; | |
684 | struct list_head *pos; | |
685 | ||
686 | if (!ww_ctx) { | |
687 | list_add_tail(&waiter->list, &lock->wait_list); | |
040a0a37 | 688 | return 0; |
6baa5c60 | 689 | } |
040a0a37 | 690 | |
6baa5c60 NH |
691 | /* |
692 | * Add the waiter before the first waiter with a higher stamp. | |
693 | * Waiters without a context are skipped to avoid starving | |
694 | * them. | |
695 | */ | |
696 | pos = &lock->wait_list; | |
697 | list_for_each_entry_reverse(cur, &lock->wait_list, list) { | |
698 | if (!cur->ww_ctx) | |
699 | continue; | |
700 | ||
701 | if (__ww_ctx_stamp_after(ww_ctx, cur->ww_ctx)) { | |
702 | /* Back off immediately if necessary. */ | |
703 | if (ww_ctx->acquired > 0) { | |
040a0a37 | 704 | #ifdef CONFIG_DEBUG_MUTEXES |
6baa5c60 NH |
705 | struct ww_mutex *ww; |
706 | ||
707 | ww = container_of(lock, struct ww_mutex, base); | |
708 | DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock); | |
709 | ww_ctx->contending_lock = ww; | |
040a0a37 | 710 | #endif |
6baa5c60 NH |
711 | return -EDEADLK; |
712 | } | |
713 | ||
714 | break; | |
715 | } | |
716 | ||
717 | pos = &cur->list; | |
200b1874 NH |
718 | |
719 | /* | |
720 | * Wake up the waiter so that it gets a chance to back | |
721 | * off. | |
722 | */ | |
723 | if (cur->ww_ctx->acquired > 0) { | |
724 | debug_mutex_wake_waiter(lock, cur); | |
725 | wake_up_process(cur->task); | |
726 | } | |
040a0a37 ML |
727 | } |
728 | ||
6baa5c60 | 729 | list_add_tail(&waiter->list, pos); |
040a0a37 ML |
730 | return 0; |
731 | } | |
732 | ||
6053ee3b IM |
733 | /* |
734 | * Lock a mutex (possibly interruptible), slowpath: | |
735 | */ | |
040a0a37 | 736 | static __always_inline int __sched |
e4564f79 | 737 | __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass, |
040a0a37 | 738 | struct lockdep_map *nest_lock, unsigned long ip, |
b0267507 | 739 | struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx) |
6053ee3b | 740 | { |
6053ee3b | 741 | struct mutex_waiter waiter; |
9d659ae1 | 742 | bool first = false; |
a40ca565 | 743 | struct ww_mutex *ww; |
040a0a37 | 744 | int ret; |
6053ee3b | 745 | |
427b1820 | 746 | might_sleep(); |
ea9e0fb8 | 747 | |
427b1820 | 748 | ww = container_of(lock, struct ww_mutex, base); |
ea9e0fb8 | 749 | if (use_ww_ctx && ww_ctx) { |
0422e83d CW |
750 | if (unlikely(ww_ctx == READ_ONCE(ww->ctx))) |
751 | return -EALREADY; | |
752 | } | |
753 | ||
41719b03 | 754 | preempt_disable(); |
e4c70a66 | 755 | mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip); |
c0226027 | 756 | |
e274795e | 757 | if (__mutex_trylock(lock) || |
c516df97 | 758 | mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, NULL)) { |
76916515 | 759 | /* got the lock, yay! */ |
3ca0ff57 | 760 | lock_acquired(&lock->dep_map, ip); |
ea9e0fb8 | 761 | if (use_ww_ctx && ww_ctx) |
3ca0ff57 | 762 | ww_mutex_set_context_fastpath(ww, ww_ctx); |
76916515 DB |
763 | preempt_enable(); |
764 | return 0; | |
0d66bf6d | 765 | } |
76916515 | 766 | |
b9c16a0e | 767 | spin_lock(&lock->wait_lock); |
1e820c96 | 768 | /* |
3ca0ff57 | 769 | * After waiting to acquire the wait_lock, try again. |
1e820c96 | 770 | */ |
659cf9f5 NH |
771 | if (__mutex_trylock(lock)) { |
772 | if (use_ww_ctx && ww_ctx) | |
773 | __ww_mutex_wakeup_for_backoff(lock, ww_ctx); | |
774 | ||
ec83f425 | 775 | goto skip_wait; |
659cf9f5 | 776 | } |
ec83f425 | 777 | |
9a11b49a | 778 | debug_mutex_lock_common(lock, &waiter); |
d269a8b8 | 779 | debug_mutex_add_waiter(lock, &waiter, current); |
6053ee3b | 780 | |
6baa5c60 NH |
781 | lock_contended(&lock->dep_map, ip); |
782 | ||
783 | if (!use_ww_ctx) { | |
784 | /* add waiting tasks to the end of the waitqueue (FIFO): */ | |
785 | list_add_tail(&waiter.list, &lock->wait_list); | |
977625a6 NH |
786 | |
787 | #ifdef CONFIG_DEBUG_MUTEXES | |
788 | waiter.ww_ctx = MUTEX_POISON_WW_CTX; | |
789 | #endif | |
6baa5c60 NH |
790 | } else { |
791 | /* Add in stamp order, waking up waiters that must back off. */ | |
792 | ret = __ww_mutex_add_waiter(&waiter, lock, ww_ctx); | |
793 | if (ret) | |
794 | goto err_early_backoff; | |
795 | ||
796 | waiter.ww_ctx = ww_ctx; | |
797 | } | |
798 | ||
d269a8b8 | 799 | waiter.task = current; |
6053ee3b | 800 | |
9d659ae1 | 801 | if (__mutex_waiter_is_first(lock, &waiter)) |
3ca0ff57 PZ |
802 | __mutex_set_flag(lock, MUTEX_FLAG_WAITERS); |
803 | ||
642fa448 | 804 | set_current_state(state); |
6053ee3b | 805 | for (;;) { |
5bbd7e64 PZ |
806 | /* |
807 | * Once we hold wait_lock, we're serialized against | |
808 | * mutex_unlock() handing the lock off to us, do a trylock | |
809 | * before testing the error conditions to make sure we pick up | |
810 | * the handoff. | |
811 | */ | |
e274795e | 812 | if (__mutex_trylock(lock)) |
5bbd7e64 | 813 | goto acquired; |
6053ee3b IM |
814 | |
815 | /* | |
5bbd7e64 PZ |
816 | * Check for signals and wound conditions while holding |
817 | * wait_lock. This ensures the lock cancellation is ordered | |
818 | * against mutex_unlock() and wake-ups do not go missing. | |
6053ee3b | 819 | */ |
d269a8b8 | 820 | if (unlikely(signal_pending_state(state, current))) { |
040a0a37 ML |
821 | ret = -EINTR; |
822 | goto err; | |
823 | } | |
6053ee3b | 824 | |
ea9e0fb8 | 825 | if (use_ww_ctx && ww_ctx && ww_ctx->acquired > 0) { |
200b1874 | 826 | ret = __ww_mutex_lock_check_stamp(lock, &waiter, ww_ctx); |
040a0a37 ML |
827 | if (ret) |
828 | goto err; | |
6053ee3b | 829 | } |
040a0a37 | 830 | |
b9c16a0e | 831 | spin_unlock(&lock->wait_lock); |
bd2f5536 | 832 | schedule_preempt_disabled(); |
9d659ae1 | 833 | |
6baa5c60 NH |
834 | /* |
835 | * ww_mutex needs to always recheck its position since its waiter | |
836 | * list is not FIFO ordered. | |
837 | */ | |
838 | if ((use_ww_ctx && ww_ctx) || !first) { | |
839 | first = __mutex_waiter_is_first(lock, &waiter); | |
840 | if (first) | |
841 | __mutex_set_flag(lock, MUTEX_FLAG_HANDOFF); | |
9d659ae1 | 842 | } |
5bbd7e64 | 843 | |
642fa448 | 844 | set_current_state(state); |
5bbd7e64 PZ |
845 | /* |
846 | * Here we order against unlock; we must either see it change | |
847 | * state back to RUNNING and fall through the next schedule(), | |
848 | * or we must see its unlock and acquire. | |
849 | */ | |
e274795e | 850 | if (__mutex_trylock(lock) || |
c516df97 | 851 | (first && mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, &waiter))) |
5bbd7e64 PZ |
852 | break; |
853 | ||
b9c16a0e | 854 | spin_lock(&lock->wait_lock); |
6053ee3b | 855 | } |
b9c16a0e | 856 | spin_lock(&lock->wait_lock); |
5bbd7e64 | 857 | acquired: |
642fa448 | 858 | __set_current_state(TASK_RUNNING); |
51587bcf | 859 | |
d269a8b8 | 860 | mutex_remove_waiter(lock, &waiter, current); |
ec83f425 | 861 | if (likely(list_empty(&lock->wait_list))) |
9d659ae1 | 862 | __mutex_clear_flag(lock, MUTEX_FLAGS); |
3ca0ff57 | 863 | |
ec83f425 | 864 | debug_mutex_free_waiter(&waiter); |
6053ee3b | 865 | |
ec83f425 DB |
866 | skip_wait: |
867 | /* got the lock - cleanup and rejoice! */ | |
c7e78cff | 868 | lock_acquired(&lock->dep_map, ip); |
6053ee3b | 869 | |
ea9e0fb8 | 870 | if (use_ww_ctx && ww_ctx) |
4bd19084 | 871 | ww_mutex_set_context_slowpath(ww, ww_ctx); |
040a0a37 | 872 | |
b9c16a0e | 873 | spin_unlock(&lock->wait_lock); |
41719b03 | 874 | preempt_enable(); |
6053ee3b | 875 | return 0; |
040a0a37 ML |
876 | |
877 | err: | |
642fa448 | 878 | __set_current_state(TASK_RUNNING); |
d269a8b8 | 879 | mutex_remove_waiter(lock, &waiter, current); |
6baa5c60 | 880 | err_early_backoff: |
b9c16a0e | 881 | spin_unlock(&lock->wait_lock); |
040a0a37 ML |
882 | debug_mutex_free_waiter(&waiter); |
883 | mutex_release(&lock->dep_map, 1, ip); | |
884 | preempt_enable(); | |
885 | return ret; | |
6053ee3b IM |
886 | } |
887 | ||
427b1820 PZ |
888 | static int __sched |
889 | __mutex_lock(struct mutex *lock, long state, unsigned int subclass, | |
890 | struct lockdep_map *nest_lock, unsigned long ip) | |
891 | { | |
892 | return __mutex_lock_common(lock, state, subclass, nest_lock, ip, NULL, false); | |
893 | } | |
894 | ||
895 | static int __sched | |
896 | __ww_mutex_lock(struct mutex *lock, long state, unsigned int subclass, | |
897 | struct lockdep_map *nest_lock, unsigned long ip, | |
898 | struct ww_acquire_ctx *ww_ctx) | |
899 | { | |
900 | return __mutex_lock_common(lock, state, subclass, nest_lock, ip, ww_ctx, true); | |
901 | } | |
902 | ||
ef5d4707 IM |
903 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
904 | void __sched | |
905 | mutex_lock_nested(struct mutex *lock, unsigned int subclass) | |
906 | { | |
427b1820 | 907 | __mutex_lock(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_); |
ef5d4707 IM |
908 | } |
909 | ||
910 | EXPORT_SYMBOL_GPL(mutex_lock_nested); | |
d63a5a74 | 911 | |
e4c70a66 PZ |
912 | void __sched |
913 | _mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest) | |
914 | { | |
427b1820 | 915 | __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, nest, _RET_IP_); |
e4c70a66 | 916 | } |
e4c70a66 PZ |
917 | EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock); |
918 | ||
ad776537 LH |
919 | int __sched |
920 | mutex_lock_killable_nested(struct mutex *lock, unsigned int subclass) | |
921 | { | |
427b1820 | 922 | return __mutex_lock(lock, TASK_KILLABLE, subclass, NULL, _RET_IP_); |
ad776537 LH |
923 | } |
924 | EXPORT_SYMBOL_GPL(mutex_lock_killable_nested); | |
925 | ||
d63a5a74 N |
926 | int __sched |
927 | mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass) | |
928 | { | |
427b1820 | 929 | return __mutex_lock(lock, TASK_INTERRUPTIBLE, subclass, NULL, _RET_IP_); |
d63a5a74 | 930 | } |
d63a5a74 | 931 | EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested); |
040a0a37 | 932 | |
1460cb65 TH |
933 | void __sched |
934 | mutex_lock_io_nested(struct mutex *lock, unsigned int subclass) | |
935 | { | |
936 | int token; | |
937 | ||
938 | might_sleep(); | |
939 | ||
940 | token = io_schedule_prepare(); | |
941 | __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, | |
942 | subclass, NULL, _RET_IP_, NULL, 0); | |
943 | io_schedule_finish(token); | |
944 | } | |
945 | EXPORT_SYMBOL_GPL(mutex_lock_io_nested); | |
946 | ||
23010027 DV |
947 | static inline int |
948 | ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) | |
949 | { | |
950 | #ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH | |
951 | unsigned tmp; | |
952 | ||
953 | if (ctx->deadlock_inject_countdown-- == 0) { | |
954 | tmp = ctx->deadlock_inject_interval; | |
955 | if (tmp > UINT_MAX/4) | |
956 | tmp = UINT_MAX; | |
957 | else | |
958 | tmp = tmp*2 + tmp + tmp/2; | |
959 | ||
960 | ctx->deadlock_inject_interval = tmp; | |
961 | ctx->deadlock_inject_countdown = tmp; | |
962 | ctx->contending_lock = lock; | |
963 | ||
964 | ww_mutex_unlock(lock); | |
965 | ||
966 | return -EDEADLK; | |
967 | } | |
968 | #endif | |
969 | ||
970 | return 0; | |
971 | } | |
040a0a37 ML |
972 | |
973 | int __sched | |
c5470b22 | 974 | ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
040a0a37 | 975 | { |
23010027 DV |
976 | int ret; |
977 | ||
040a0a37 | 978 | might_sleep(); |
427b1820 PZ |
979 | ret = __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE, |
980 | 0, ctx ? &ctx->dep_map : NULL, _RET_IP_, | |
981 | ctx); | |
ea9e0fb8 | 982 | if (!ret && ctx && ctx->acquired > 1) |
23010027 DV |
983 | return ww_mutex_deadlock_injection(lock, ctx); |
984 | ||
985 | return ret; | |
040a0a37 | 986 | } |
c5470b22 | 987 | EXPORT_SYMBOL_GPL(ww_mutex_lock); |
040a0a37 ML |
988 | |
989 | int __sched | |
c5470b22 | 990 | ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
040a0a37 | 991 | { |
23010027 DV |
992 | int ret; |
993 | ||
040a0a37 | 994 | might_sleep(); |
427b1820 PZ |
995 | ret = __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE, |
996 | 0, ctx ? &ctx->dep_map : NULL, _RET_IP_, | |
997 | ctx); | |
23010027 | 998 | |
ea9e0fb8 | 999 | if (!ret && ctx && ctx->acquired > 1) |
23010027 DV |
1000 | return ww_mutex_deadlock_injection(lock, ctx); |
1001 | ||
1002 | return ret; | |
040a0a37 | 1003 | } |
c5470b22 | 1004 | EXPORT_SYMBOL_GPL(ww_mutex_lock_interruptible); |
040a0a37 | 1005 | |
ef5d4707 IM |
1006 | #endif |
1007 | ||
6053ee3b IM |
1008 | /* |
1009 | * Release the lock, slowpath: | |
1010 | */ | |
3ca0ff57 | 1011 | static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip) |
6053ee3b | 1012 | { |
9d659ae1 | 1013 | struct task_struct *next = NULL; |
194a6b5b | 1014 | DEFINE_WAKE_Q(wake_q); |
b9c16a0e | 1015 | unsigned long owner; |
6053ee3b | 1016 | |
3ca0ff57 PZ |
1017 | mutex_release(&lock->dep_map, 1, ip); |
1018 | ||
6053ee3b | 1019 | /* |
9d659ae1 PZ |
1020 | * Release the lock before (potentially) taking the spinlock such that |
1021 | * other contenders can get on with things ASAP. | |
1022 | * | |
1023 | * Except when HANDOFF, in that case we must not clear the owner field, | |
1024 | * but instead set it to the top waiter. | |
6053ee3b | 1025 | */ |
9d659ae1 PZ |
1026 | owner = atomic_long_read(&lock->owner); |
1027 | for (;;) { | |
1028 | unsigned long old; | |
1029 | ||
1030 | #ifdef CONFIG_DEBUG_MUTEXES | |
1031 | DEBUG_LOCKS_WARN_ON(__owner_task(owner) != current); | |
e274795e | 1032 | DEBUG_LOCKS_WARN_ON(owner & MUTEX_FLAG_PICKUP); |
9d659ae1 PZ |
1033 | #endif |
1034 | ||
1035 | if (owner & MUTEX_FLAG_HANDOFF) | |
1036 | break; | |
1037 | ||
1038 | old = atomic_long_cmpxchg_release(&lock->owner, owner, | |
1039 | __owner_flags(owner)); | |
1040 | if (old == owner) { | |
1041 | if (owner & MUTEX_FLAG_WAITERS) | |
1042 | break; | |
1043 | ||
1044 | return; | |
1045 | } | |
1046 | ||
1047 | owner = old; | |
1048 | } | |
6053ee3b | 1049 | |
b9c16a0e | 1050 | spin_lock(&lock->wait_lock); |
1d8fe7dc | 1051 | debug_mutex_unlock(lock); |
6053ee3b IM |
1052 | if (!list_empty(&lock->wait_list)) { |
1053 | /* get the first entry from the wait-list: */ | |
1054 | struct mutex_waiter *waiter = | |
9d659ae1 PZ |
1055 | list_first_entry(&lock->wait_list, |
1056 | struct mutex_waiter, list); | |
1057 | ||
1058 | next = waiter->task; | |
6053ee3b IM |
1059 | |
1060 | debug_mutex_wake_waiter(lock, waiter); | |
9d659ae1 | 1061 | wake_q_add(&wake_q, next); |
6053ee3b IM |
1062 | } |
1063 | ||
9d659ae1 PZ |
1064 | if (owner & MUTEX_FLAG_HANDOFF) |
1065 | __mutex_handoff(lock, next); | |
1066 | ||
b9c16a0e | 1067 | spin_unlock(&lock->wait_lock); |
9d659ae1 | 1068 | |
1329ce6f | 1069 | wake_up_q(&wake_q); |
6053ee3b IM |
1070 | } |
1071 | ||
e4564f79 | 1072 | #ifndef CONFIG_DEBUG_LOCK_ALLOC |
6053ee3b IM |
1073 | /* |
1074 | * Here come the less common (and hence less performance-critical) APIs: | |
1075 | * mutex_lock_interruptible() and mutex_trylock(). | |
1076 | */ | |
7ad5b3a5 | 1077 | static noinline int __sched |
a41b56ef | 1078 | __mutex_lock_killable_slowpath(struct mutex *lock); |
ad776537 | 1079 | |
7ad5b3a5 | 1080 | static noinline int __sched |
a41b56ef | 1081 | __mutex_lock_interruptible_slowpath(struct mutex *lock); |
6053ee3b | 1082 | |
ef5dc121 RD |
1083 | /** |
1084 | * mutex_lock_interruptible - acquire the mutex, interruptible | |
6053ee3b IM |
1085 | * @lock: the mutex to be acquired |
1086 | * | |
1087 | * Lock the mutex like mutex_lock(), and return 0 if the mutex has | |
1088 | * been acquired or sleep until the mutex becomes available. If a | |
1089 | * signal arrives while waiting for the lock then this function | |
1090 | * returns -EINTR. | |
1091 | * | |
1092 | * This function is similar to (but not equivalent to) down_interruptible(). | |
1093 | */ | |
7ad5b3a5 | 1094 | int __sched mutex_lock_interruptible(struct mutex *lock) |
6053ee3b | 1095 | { |
c544bdb1 | 1096 | might_sleep(); |
3ca0ff57 PZ |
1097 | |
1098 | if (__mutex_trylock_fast(lock)) | |
a41b56ef | 1099 | return 0; |
3ca0ff57 PZ |
1100 | |
1101 | return __mutex_lock_interruptible_slowpath(lock); | |
6053ee3b IM |
1102 | } |
1103 | ||
1104 | EXPORT_SYMBOL(mutex_lock_interruptible); | |
1105 | ||
7ad5b3a5 | 1106 | int __sched mutex_lock_killable(struct mutex *lock) |
ad776537 LH |
1107 | { |
1108 | might_sleep(); | |
3ca0ff57 PZ |
1109 | |
1110 | if (__mutex_trylock_fast(lock)) | |
a41b56ef | 1111 | return 0; |
3ca0ff57 PZ |
1112 | |
1113 | return __mutex_lock_killable_slowpath(lock); | |
ad776537 LH |
1114 | } |
1115 | EXPORT_SYMBOL(mutex_lock_killable); | |
1116 | ||
1460cb65 TH |
1117 | void __sched mutex_lock_io(struct mutex *lock) |
1118 | { | |
1119 | int token; | |
1120 | ||
1121 | token = io_schedule_prepare(); | |
1122 | mutex_lock(lock); | |
1123 | io_schedule_finish(token); | |
1124 | } | |
1125 | EXPORT_SYMBOL_GPL(mutex_lock_io); | |
1126 | ||
3ca0ff57 PZ |
1127 | static noinline void __sched |
1128 | __mutex_lock_slowpath(struct mutex *lock) | |
e4564f79 | 1129 | { |
427b1820 | 1130 | __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_); |
e4564f79 PZ |
1131 | } |
1132 | ||
7ad5b3a5 | 1133 | static noinline int __sched |
a41b56ef | 1134 | __mutex_lock_killable_slowpath(struct mutex *lock) |
ad776537 | 1135 | { |
427b1820 | 1136 | return __mutex_lock(lock, TASK_KILLABLE, 0, NULL, _RET_IP_); |
ad776537 LH |
1137 | } |
1138 | ||
7ad5b3a5 | 1139 | static noinline int __sched |
a41b56ef | 1140 | __mutex_lock_interruptible_slowpath(struct mutex *lock) |
6053ee3b | 1141 | { |
427b1820 | 1142 | return __mutex_lock(lock, TASK_INTERRUPTIBLE, 0, NULL, _RET_IP_); |
040a0a37 ML |
1143 | } |
1144 | ||
1145 | static noinline int __sched | |
1146 | __ww_mutex_lock_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) | |
1147 | { | |
427b1820 PZ |
1148 | return __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE, 0, NULL, |
1149 | _RET_IP_, ctx); | |
6053ee3b | 1150 | } |
040a0a37 ML |
1151 | |
1152 | static noinline int __sched | |
1153 | __ww_mutex_lock_interruptible_slowpath(struct ww_mutex *lock, | |
1154 | struct ww_acquire_ctx *ctx) | |
1155 | { | |
427b1820 PZ |
1156 | return __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE, 0, NULL, |
1157 | _RET_IP_, ctx); | |
040a0a37 ML |
1158 | } |
1159 | ||
e4564f79 | 1160 | #endif |
6053ee3b | 1161 | |
ef5dc121 RD |
1162 | /** |
1163 | * mutex_trylock - try to acquire the mutex, without waiting | |
6053ee3b IM |
1164 | * @lock: the mutex to be acquired |
1165 | * | |
1166 | * Try to acquire the mutex atomically. Returns 1 if the mutex | |
1167 | * has been acquired successfully, and 0 on contention. | |
1168 | * | |
1169 | * NOTE: this function follows the spin_trylock() convention, so | |
ef5dc121 | 1170 | * it is negated from the down_trylock() return values! Be careful |
6053ee3b IM |
1171 | * about this when converting semaphore users to mutexes. |
1172 | * | |
1173 | * This function must not be used in interrupt context. The | |
1174 | * mutex must be released by the same task that acquired it. | |
1175 | */ | |
7ad5b3a5 | 1176 | int __sched mutex_trylock(struct mutex *lock) |
6053ee3b | 1177 | { |
e274795e | 1178 | bool locked = __mutex_trylock(lock); |
0d66bf6d | 1179 | |
3ca0ff57 PZ |
1180 | if (locked) |
1181 | mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_); | |
0d66bf6d | 1182 | |
3ca0ff57 | 1183 | return locked; |
6053ee3b | 1184 | } |
6053ee3b | 1185 | EXPORT_SYMBOL(mutex_trylock); |
a511e3f9 | 1186 | |
040a0a37 ML |
1187 | #ifndef CONFIG_DEBUG_LOCK_ALLOC |
1188 | int __sched | |
c5470b22 | 1189 | ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
040a0a37 | 1190 | { |
040a0a37 ML |
1191 | might_sleep(); |
1192 | ||
3ca0ff57 | 1193 | if (__mutex_trylock_fast(&lock->base)) { |
ea9e0fb8 NH |
1194 | if (ctx) |
1195 | ww_mutex_set_context_fastpath(lock, ctx); | |
3ca0ff57 PZ |
1196 | return 0; |
1197 | } | |
1198 | ||
1199 | return __ww_mutex_lock_slowpath(lock, ctx); | |
040a0a37 | 1200 | } |
c5470b22 | 1201 | EXPORT_SYMBOL(ww_mutex_lock); |
040a0a37 ML |
1202 | |
1203 | int __sched | |
c5470b22 | 1204 | ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
040a0a37 | 1205 | { |
040a0a37 ML |
1206 | might_sleep(); |
1207 | ||
3ca0ff57 | 1208 | if (__mutex_trylock_fast(&lock->base)) { |
ea9e0fb8 NH |
1209 | if (ctx) |
1210 | ww_mutex_set_context_fastpath(lock, ctx); | |
3ca0ff57 PZ |
1211 | return 0; |
1212 | } | |
1213 | ||
1214 | return __ww_mutex_lock_interruptible_slowpath(lock, ctx); | |
040a0a37 | 1215 | } |
c5470b22 | 1216 | EXPORT_SYMBOL(ww_mutex_lock_interruptible); |
040a0a37 ML |
1217 | |
1218 | #endif | |
1219 | ||
a511e3f9 AM |
1220 | /** |
1221 | * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0 | |
1222 | * @cnt: the atomic which we are to dec | |
1223 | * @lock: the mutex to return holding if we dec to 0 | |
1224 | * | |
1225 | * return true and hold lock if we dec to 0, return false otherwise | |
1226 | */ | |
1227 | int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock) | |
1228 | { | |
1229 | /* dec if we can't possibly hit 0 */ | |
1230 | if (atomic_add_unless(cnt, -1, 1)) | |
1231 | return 0; | |
1232 | /* we might hit 0, so take the lock */ | |
1233 | mutex_lock(lock); | |
1234 | if (!atomic_dec_and_test(cnt)) { | |
1235 | /* when we actually did the dec, we didn't hit 0 */ | |
1236 | mutex_unlock(lock); | |
1237 | return 0; | |
1238 | } | |
1239 | /* we hit 0, and we hold the lock */ | |
1240 | return 1; | |
1241 | } | |
1242 | EXPORT_SYMBOL(atomic_dec_and_mutex_lock); |