]>
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> |
b17b0153 | 25 | #include <linux/sched/debug.h> |
9984de1a | 26 | #include <linux/export.h> |
6053ee3b IM |
27 | #include <linux/spinlock.h> |
28 | #include <linux/interrupt.h> | |
9a11b49a | 29 | #include <linux/debug_locks.h> |
7a215f89 | 30 | #include <linux/osq_lock.h> |
6053ee3b | 31 | |
6053ee3b IM |
32 | #ifdef CONFIG_DEBUG_MUTEXES |
33 | # include "mutex-debug.h" | |
6053ee3b IM |
34 | #else |
35 | # include "mutex.h" | |
6053ee3b IM |
36 | #endif |
37 | ||
ef5d4707 IM |
38 | void |
39 | __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key) | |
6053ee3b | 40 | { |
3ca0ff57 | 41 | atomic_long_set(&lock->owner, 0); |
6053ee3b IM |
42 | spin_lock_init(&lock->wait_lock); |
43 | INIT_LIST_HEAD(&lock->wait_list); | |
2bd2c92c | 44 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
4d9d951e | 45 | osq_lock_init(&lock->osq); |
2bd2c92c | 46 | #endif |
6053ee3b | 47 | |
ef5d4707 | 48 | debug_mutex_init(lock, name, key); |
6053ee3b | 49 | } |
6053ee3b IM |
50 | EXPORT_SYMBOL(__mutex_init); |
51 | ||
3ca0ff57 PZ |
52 | /* |
53 | * @owner: contains: 'struct task_struct *' to the current lock owner, | |
54 | * NULL means not owned. Since task_struct pointers are aligned at | |
e274795e | 55 | * at least L1_CACHE_BYTES, we have low bits to store extra state. |
3ca0ff57 PZ |
56 | * |
57 | * Bit0 indicates a non-empty waiter list; unlock must issue a wakeup. | |
9d659ae1 | 58 | * Bit1 indicates unlock needs to hand the lock to the top-waiter |
e274795e | 59 | * Bit2 indicates handoff has been done and we're waiting for pickup. |
3ca0ff57 PZ |
60 | */ |
61 | #define MUTEX_FLAG_WAITERS 0x01 | |
9d659ae1 | 62 | #define MUTEX_FLAG_HANDOFF 0x02 |
e274795e | 63 | #define MUTEX_FLAG_PICKUP 0x04 |
3ca0ff57 | 64 | |
e274795e | 65 | #define MUTEX_FLAGS 0x07 |
3ca0ff57 PZ |
66 | |
67 | static inline struct task_struct *__owner_task(unsigned long owner) | |
68 | { | |
69 | return (struct task_struct *)(owner & ~MUTEX_FLAGS); | |
70 | } | |
71 | ||
72 | static inline unsigned long __owner_flags(unsigned long owner) | |
73 | { | |
74 | return owner & MUTEX_FLAGS; | |
75 | } | |
76 | ||
77 | /* | |
e274795e | 78 | * Trylock variant that retuns the owning task on failure. |
3ca0ff57 | 79 | */ |
e274795e | 80 | static inline struct task_struct *__mutex_trylock_or_owner(struct mutex *lock) |
3ca0ff57 PZ |
81 | { |
82 | unsigned long owner, curr = (unsigned long)current; | |
83 | ||
84 | owner = atomic_long_read(&lock->owner); | |
85 | for (;;) { /* must loop, can race against a flag */ | |
9d659ae1 | 86 | unsigned long old, flags = __owner_flags(owner); |
e274795e | 87 | unsigned long task = owner & ~MUTEX_FLAGS; |
9d659ae1 | 88 | |
e274795e PZ |
89 | if (task) { |
90 | if (likely(task != curr)) | |
91 | break; | |
3ca0ff57 | 92 | |
e274795e PZ |
93 | if (likely(!(flags & MUTEX_FLAG_PICKUP))) |
94 | break; | |
9d659ae1 | 95 | |
e274795e PZ |
96 | flags &= ~MUTEX_FLAG_PICKUP; |
97 | } else { | |
98 | #ifdef CONFIG_DEBUG_MUTEXES | |
99 | DEBUG_LOCKS_WARN_ON(flags & MUTEX_FLAG_PICKUP); | |
100 | #endif | |
9d659ae1 PZ |
101 | } |
102 | ||
103 | /* | |
104 | * We set the HANDOFF bit, we must make sure it doesn't live | |
105 | * past the point where we acquire it. This would be possible | |
106 | * if we (accidentally) set the bit on an unlocked mutex. | |
107 | */ | |
e274795e | 108 | flags &= ~MUTEX_FLAG_HANDOFF; |
3ca0ff57 | 109 | |
9d659ae1 | 110 | old = atomic_long_cmpxchg_acquire(&lock->owner, owner, curr | flags); |
3ca0ff57 | 111 | if (old == owner) |
e274795e | 112 | return NULL; |
3ca0ff57 PZ |
113 | |
114 | owner = old; | |
115 | } | |
e274795e PZ |
116 | |
117 | return __owner_task(owner); | |
118 | } | |
119 | ||
120 | /* | |
121 | * Actual trylock that will work on any unlocked state. | |
122 | */ | |
123 | static inline bool __mutex_trylock(struct mutex *lock) | |
124 | { | |
125 | return !__mutex_trylock_or_owner(lock); | |
3ca0ff57 PZ |
126 | } |
127 | ||
128 | #ifndef CONFIG_DEBUG_LOCK_ALLOC | |
129 | /* | |
130 | * Lockdep annotations are contained to the slow paths for simplicity. | |
131 | * There is nothing that would stop spreading the lockdep annotations outwards | |
132 | * except more code. | |
133 | */ | |
134 | ||
135 | /* | |
136 | * Optimistic trylock that only works in the uncontended case. Make sure to | |
137 | * follow with a __mutex_trylock() before failing. | |
138 | */ | |
139 | static __always_inline bool __mutex_trylock_fast(struct mutex *lock) | |
140 | { | |
141 | unsigned long curr = (unsigned long)current; | |
c427f695 | 142 | unsigned long zero = 0UL; |
3ca0ff57 | 143 | |
c427f695 | 144 | if (atomic_long_try_cmpxchg_acquire(&lock->owner, &zero, curr)) |
3ca0ff57 PZ |
145 | return true; |
146 | ||
147 | return false; | |
148 | } | |
149 | ||
150 | static __always_inline bool __mutex_unlock_fast(struct mutex *lock) | |
151 | { | |
152 | unsigned long curr = (unsigned long)current; | |
153 | ||
154 | if (atomic_long_cmpxchg_release(&lock->owner, curr, 0UL) == curr) | |
155 | return true; | |
156 | ||
157 | return false; | |
158 | } | |
159 | #endif | |
160 | ||
161 | static inline void __mutex_set_flag(struct mutex *lock, unsigned long flag) | |
162 | { | |
163 | atomic_long_or(flag, &lock->owner); | |
164 | } | |
165 | ||
166 | static inline void __mutex_clear_flag(struct mutex *lock, unsigned long flag) | |
167 | { | |
168 | atomic_long_andnot(flag, &lock->owner); | |
169 | } | |
170 | ||
9d659ae1 PZ |
171 | static inline bool __mutex_waiter_is_first(struct mutex *lock, struct mutex_waiter *waiter) |
172 | { | |
173 | return list_first_entry(&lock->wait_list, struct mutex_waiter, list) == waiter; | |
174 | } | |
175 | ||
08295b3b TH |
176 | /* |
177 | * Add @waiter to a given location in the lock wait_list and set the | |
178 | * FLAG_WAITERS flag if it's the first waiter. | |
179 | */ | |
180 | static void __sched | |
181 | __mutex_add_waiter(struct mutex *lock, struct mutex_waiter *waiter, | |
182 | struct list_head *list) | |
183 | { | |
184 | debug_mutex_add_waiter(lock, waiter, current); | |
185 | ||
186 | list_add_tail(&waiter->list, list); | |
187 | if (__mutex_waiter_is_first(lock, waiter)) | |
188 | __mutex_set_flag(lock, MUTEX_FLAG_WAITERS); | |
189 | } | |
190 | ||
9d659ae1 PZ |
191 | /* |
192 | * Give up ownership to a specific task, when @task = NULL, this is equivalent | |
e274795e PZ |
193 | * to a regular unlock. Sets PICKUP on a handoff, clears HANDOF, preserves |
194 | * WAITERS. Provides RELEASE semantics like a regular unlock, the | |
195 | * __mutex_trylock() provides a matching ACQUIRE semantics for the handoff. | |
9d659ae1 PZ |
196 | */ |
197 | static void __mutex_handoff(struct mutex *lock, struct task_struct *task) | |
198 | { | |
199 | unsigned long owner = atomic_long_read(&lock->owner); | |
200 | ||
201 | for (;;) { | |
202 | unsigned long old, new; | |
203 | ||
204 | #ifdef CONFIG_DEBUG_MUTEXES | |
205 | DEBUG_LOCKS_WARN_ON(__owner_task(owner) != current); | |
e274795e | 206 | DEBUG_LOCKS_WARN_ON(owner & MUTEX_FLAG_PICKUP); |
9d659ae1 PZ |
207 | #endif |
208 | ||
209 | new = (owner & MUTEX_FLAG_WAITERS); | |
210 | new |= (unsigned long)task; | |
e274795e PZ |
211 | if (task) |
212 | new |= MUTEX_FLAG_PICKUP; | |
9d659ae1 PZ |
213 | |
214 | old = atomic_long_cmpxchg_release(&lock->owner, owner, new); | |
215 | if (old == owner) | |
216 | break; | |
217 | ||
218 | owner = old; | |
219 | } | |
220 | } | |
221 | ||
e4564f79 | 222 | #ifndef CONFIG_DEBUG_LOCK_ALLOC |
6053ee3b IM |
223 | /* |
224 | * We split the mutex lock/unlock logic into separate fastpath and | |
225 | * slowpath functions, to reduce the register pressure on the fastpath. | |
226 | * We also put the fastpath first in the kernel image, to make sure the | |
227 | * branch is predicted by the CPU as default-untaken. | |
228 | */ | |
3ca0ff57 | 229 | static void __sched __mutex_lock_slowpath(struct mutex *lock); |
6053ee3b | 230 | |
ef5dc121 | 231 | /** |
6053ee3b IM |
232 | * mutex_lock - acquire the mutex |
233 | * @lock: the mutex to be acquired | |
234 | * | |
235 | * Lock the mutex exclusively for this task. If the mutex is not | |
236 | * available right now, it will sleep until it can get it. | |
237 | * | |
238 | * The mutex must later on be released by the same task that | |
239 | * acquired it. Recursive locking is not allowed. The task | |
240 | * may not exit without first unlocking the mutex. Also, kernel | |
139b6fd2 | 241 | * memory where the mutex resides must not be freed with |
6053ee3b IM |
242 | * the mutex still locked. The mutex must first be initialized |
243 | * (or statically defined) before it can be locked. memset()-ing | |
244 | * the mutex to 0 is not allowed. | |
245 | * | |
7b4ff1ad MCC |
246 | * (The CONFIG_DEBUG_MUTEXES .config option turns on debugging |
247 | * checks that will enforce the restrictions and will also do | |
248 | * deadlock debugging) | |
6053ee3b IM |
249 | * |
250 | * This function is similar to (but not equivalent to) down(). | |
251 | */ | |
b09d2501 | 252 | void __sched mutex_lock(struct mutex *lock) |
6053ee3b | 253 | { |
c544bdb1 | 254 | might_sleep(); |
6053ee3b | 255 | |
3ca0ff57 PZ |
256 | if (!__mutex_trylock_fast(lock)) |
257 | __mutex_lock_slowpath(lock); | |
258 | } | |
6053ee3b | 259 | EXPORT_SYMBOL(mutex_lock); |
e4564f79 | 260 | #endif |
6053ee3b | 261 | |
55f036ca PZ |
262 | /* |
263 | * Wait-Die: | |
264 | * The newer transactions are killed when: | |
265 | * It (the new transaction) makes a request for a lock being held | |
266 | * by an older transaction. | |
08295b3b TH |
267 | * |
268 | * Wound-Wait: | |
269 | * The newer transactions are wounded when: | |
270 | * An older transaction makes a request for a lock being held by | |
271 | * the newer transaction. | |
55f036ca PZ |
272 | */ |
273 | ||
274 | /* | |
275 | * Associate the ww_mutex @ww with the context @ww_ctx under which we acquired | |
276 | * it. | |
277 | */ | |
427b1820 PZ |
278 | static __always_inline void |
279 | ww_mutex_lock_acquired(struct ww_mutex *ww, struct ww_acquire_ctx *ww_ctx) | |
76916515 DB |
280 | { |
281 | #ifdef CONFIG_DEBUG_MUTEXES | |
282 | /* | |
283 | * If this WARN_ON triggers, you used ww_mutex_lock to acquire, | |
284 | * but released with a normal mutex_unlock in this call. | |
285 | * | |
286 | * This should never happen, always use ww_mutex_unlock. | |
287 | */ | |
288 | DEBUG_LOCKS_WARN_ON(ww->ctx); | |
289 | ||
290 | /* | |
291 | * Not quite done after calling ww_acquire_done() ? | |
292 | */ | |
293 | DEBUG_LOCKS_WARN_ON(ww_ctx->done_acquire); | |
294 | ||
295 | if (ww_ctx->contending_lock) { | |
296 | /* | |
297 | * After -EDEADLK you tried to | |
298 | * acquire a different ww_mutex? Bad! | |
299 | */ | |
300 | DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock != ww); | |
301 | ||
302 | /* | |
303 | * You called ww_mutex_lock after receiving -EDEADLK, | |
304 | * but 'forgot' to unlock everything else first? | |
305 | */ | |
306 | DEBUG_LOCKS_WARN_ON(ww_ctx->acquired > 0); | |
307 | ww_ctx->contending_lock = NULL; | |
308 | } | |
309 | ||
310 | /* | |
311 | * Naughty, using a different class will lead to undefined behavior! | |
312 | */ | |
313 | DEBUG_LOCKS_WARN_ON(ww_ctx->ww_class != ww->ww_class); | |
314 | #endif | |
315 | ww_ctx->acquired++; | |
55f036ca | 316 | ww->ctx = ww_ctx; |
76916515 DB |
317 | } |
318 | ||
55f036ca PZ |
319 | /* |
320 | * Determine if context @a is 'after' context @b. IOW, @a is a younger | |
321 | * transaction than @b and depending on algorithm either needs to wait for | |
322 | * @b or die. | |
323 | */ | |
3822da3e NH |
324 | static inline bool __sched |
325 | __ww_ctx_stamp_after(struct ww_acquire_ctx *a, struct ww_acquire_ctx *b) | |
326 | { | |
55f036ca PZ |
327 | |
328 | return (signed long)(a->stamp - b->stamp) > 0; | |
329 | } | |
330 | ||
331 | /* | |
332 | * Wait-Die; wake a younger waiter context (when locks held) such that it can | |
333 | * die. | |
334 | * | |
335 | * Among waiters with context, only the first one can have other locks acquired | |
336 | * already (ctx->acquired > 0), because __ww_mutex_add_waiter() and | |
337 | * __ww_mutex_check_kill() wake any but the earliest context. | |
338 | */ | |
339 | static bool __sched | |
340 | __ww_mutex_die(struct mutex *lock, struct mutex_waiter *waiter, | |
341 | struct ww_acquire_ctx *ww_ctx) | |
342 | { | |
08295b3b TH |
343 | if (!ww_ctx->is_wait_die) |
344 | return false; | |
345 | ||
55f036ca PZ |
346 | if (waiter->ww_ctx->acquired > 0 && |
347 | __ww_ctx_stamp_after(waiter->ww_ctx, ww_ctx)) { | |
348 | debug_mutex_wake_waiter(lock, waiter); | |
349 | wake_up_process(waiter->task); | |
350 | } | |
351 | ||
352 | return true; | |
3822da3e NH |
353 | } |
354 | ||
08295b3b TH |
355 | /* |
356 | * Wound-Wait; wound a younger @hold_ctx if it holds the lock. | |
357 | * | |
358 | * Wound the lock holder if there are waiters with older transactions than | |
359 | * the lock holders. Even if multiple waiters may wound the lock holder, | |
360 | * it's sufficient that only one does. | |
361 | */ | |
362 | static bool __ww_mutex_wound(struct mutex *lock, | |
363 | struct ww_acquire_ctx *ww_ctx, | |
364 | struct ww_acquire_ctx *hold_ctx) | |
365 | { | |
366 | struct task_struct *owner = __mutex_owner(lock); | |
367 | ||
368 | lockdep_assert_held(&lock->wait_lock); | |
369 | ||
370 | /* | |
371 | * Possible through __ww_mutex_add_waiter() when we race with | |
372 | * ww_mutex_set_context_fastpath(). In that case we'll get here again | |
373 | * through __ww_mutex_check_waiters(). | |
374 | */ | |
375 | if (!hold_ctx) | |
376 | return false; | |
377 | ||
378 | /* | |
379 | * Can have !owner because of __mutex_unlock_slowpath(), but if owner, | |
380 | * it cannot go away because we'll have FLAG_WAITERS set and hold | |
381 | * wait_lock. | |
382 | */ | |
383 | if (!owner) | |
384 | return false; | |
385 | ||
386 | if (ww_ctx->acquired > 0 && __ww_ctx_stamp_after(hold_ctx, ww_ctx)) { | |
387 | hold_ctx->wounded = 1; | |
388 | ||
389 | /* | |
390 | * wake_up_process() paired with set_current_state() | |
391 | * inserts sufficient barriers to make sure @owner either sees | |
e13e2366 | 392 | * it's wounded in __ww_mutex_check_kill() or has a |
08295b3b TH |
393 | * wakeup pending to re-read the wounded state. |
394 | */ | |
395 | if (owner != current) | |
396 | wake_up_process(owner); | |
397 | ||
398 | return true; | |
399 | } | |
400 | ||
401 | return false; | |
402 | } | |
403 | ||
659cf9f5 | 404 | /* |
55f036ca | 405 | * We just acquired @lock under @ww_ctx, if there are later contexts waiting |
08295b3b | 406 | * behind us on the wait-list, check if they need to die, or wound us. |
659cf9f5 | 407 | * |
55f036ca PZ |
408 | * See __ww_mutex_add_waiter() for the list-order construction; basically the |
409 | * list is ordered by stamp, smallest (oldest) first. | |
659cf9f5 | 410 | * |
08295b3b TH |
411 | * This relies on never mixing wait-die/wound-wait on the same wait-list; |
412 | * which is currently ensured by that being a ww_class property. | |
413 | * | |
659cf9f5 NH |
414 | * The current task must not be on the wait list. |
415 | */ | |
416 | static void __sched | |
55f036ca | 417 | __ww_mutex_check_waiters(struct mutex *lock, struct ww_acquire_ctx *ww_ctx) |
659cf9f5 NH |
418 | { |
419 | struct mutex_waiter *cur; | |
420 | ||
421 | lockdep_assert_held(&lock->wait_lock); | |
422 | ||
423 | list_for_each_entry(cur, &lock->wait_list, list) { | |
424 | if (!cur->ww_ctx) | |
425 | continue; | |
426 | ||
08295b3b TH |
427 | if (__ww_mutex_die(lock, cur, ww_ctx) || |
428 | __ww_mutex_wound(lock, cur->ww_ctx, ww_ctx)) | |
55f036ca | 429 | break; |
659cf9f5 NH |
430 | } |
431 | } | |
432 | ||
76916515 | 433 | /* |
55f036ca PZ |
434 | * After acquiring lock with fastpath, where we do not hold wait_lock, set ctx |
435 | * and wake up any waiters so they can recheck. | |
76916515 DB |
436 | */ |
437 | static __always_inline void | |
427b1820 | 438 | ww_mutex_set_context_fastpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
76916515 | 439 | { |
76916515 DB |
440 | ww_mutex_lock_acquired(lock, ctx); |
441 | ||
76916515 DB |
442 | /* |
443 | * The lock->ctx update should be visible on all cores before | |
55f036ca | 444 | * the WAITERS check is done, otherwise contended waiters might be |
76916515 DB |
445 | * missed. The contended waiters will either see ww_ctx == NULL |
446 | * and keep spinning, or it will acquire wait_lock, add itself | |
447 | * to waiter list and sleep. | |
448 | */ | |
08295b3b | 449 | smp_mb(); /* See comments above and below. */ |
76916515 DB |
450 | |
451 | /* | |
08295b3b TH |
452 | * [W] ww->ctx = ctx [W] MUTEX_FLAG_WAITERS |
453 | * MB MB | |
454 | * [R] MUTEX_FLAG_WAITERS [R] ww->ctx | |
455 | * | |
456 | * The memory barrier above pairs with the memory barrier in | |
457 | * __ww_mutex_add_waiter() and makes sure we either observe ww->ctx | |
458 | * and/or !empty list. | |
76916515 | 459 | */ |
3ca0ff57 | 460 | if (likely(!(atomic_long_read(&lock->base.owner) & MUTEX_FLAG_WAITERS))) |
76916515 DB |
461 | return; |
462 | ||
463 | /* | |
55f036ca | 464 | * Uh oh, we raced in fastpath, check if any of the waiters need to |
08295b3b | 465 | * die or wound us. |
76916515 | 466 | */ |
b9c16a0e | 467 | spin_lock(&lock->base.wait_lock); |
55f036ca | 468 | __ww_mutex_check_waiters(&lock->base, ctx); |
b9c16a0e | 469 | spin_unlock(&lock->base.wait_lock); |
76916515 DB |
470 | } |
471 | ||
41fcb9f2 | 472 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
c516df97 NH |
473 | |
474 | static inline | |
475 | bool ww_mutex_spin_on_owner(struct mutex *lock, struct ww_acquire_ctx *ww_ctx, | |
476 | struct mutex_waiter *waiter) | |
477 | { | |
478 | struct ww_mutex *ww; | |
479 | ||
480 | ww = container_of(lock, struct ww_mutex, base); | |
4bd19084 DB |
481 | |
482 | /* | |
c516df97 NH |
483 | * If ww->ctx is set the contents are undefined, only |
484 | * by acquiring wait_lock there is a guarantee that | |
485 | * they are not invalid when reading. | |
486 | * | |
487 | * As such, when deadlock detection needs to be | |
488 | * performed the optimistic spinning cannot be done. | |
489 | * | |
490 | * Check this in every inner iteration because we may | |
491 | * be racing against another thread's ww_mutex_lock. | |
4bd19084 | 492 | */ |
c516df97 NH |
493 | if (ww_ctx->acquired > 0 && READ_ONCE(ww->ctx)) |
494 | return false; | |
495 | ||
496 | /* | |
497 | * If we aren't on the wait list yet, cancel the spin | |
498 | * if there are waiters. We want to avoid stealing the | |
499 | * lock from a waiter with an earlier stamp, since the | |
500 | * other thread may already own a lock that we also | |
501 | * need. | |
502 | */ | |
503 | if (!waiter && (atomic_long_read(&lock->owner) & MUTEX_FLAG_WAITERS)) | |
504 | return false; | |
505 | ||
506 | /* | |
507 | * Similarly, stop spinning if we are no longer the | |
508 | * first waiter. | |
509 | */ | |
510 | if (waiter && !__mutex_waiter_is_first(lock, waiter)) | |
511 | return false; | |
512 | ||
513 | return true; | |
4bd19084 | 514 | } |
76916515 | 515 | |
41fcb9f2 | 516 | /* |
25f13b40 NH |
517 | * Look out! "owner" is an entirely speculative pointer access and not |
518 | * reliable. | |
519 | * | |
520 | * "noinline" so that this function shows up on perf profiles. | |
41fcb9f2 WL |
521 | */ |
522 | static noinline | |
25f13b40 | 523 | bool mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner, |
c516df97 | 524 | struct ww_acquire_ctx *ww_ctx, struct mutex_waiter *waiter) |
41fcb9f2 | 525 | { |
01ac33c1 | 526 | bool ret = true; |
be1f7bf2 | 527 | |
41fcb9f2 | 528 | rcu_read_lock(); |
3ca0ff57 | 529 | while (__mutex_owner(lock) == owner) { |
be1f7bf2 JL |
530 | /* |
531 | * Ensure we emit the owner->on_cpu, dereference _after_ | |
01ac33c1 JL |
532 | * checking lock->owner still matches owner. If that fails, |
533 | * owner might point to freed memory. If it still matches, | |
be1f7bf2 JL |
534 | * the rcu_read_lock() ensures the memory stays valid. |
535 | */ | |
536 | barrier(); | |
537 | ||
05ffc951 PX |
538 | /* |
539 | * Use vcpu_is_preempted to detect lock holder preemption issue. | |
540 | */ | |
541 | if (!owner->on_cpu || need_resched() || | |
542 | vcpu_is_preempted(task_cpu(owner))) { | |
be1f7bf2 JL |
543 | ret = false; |
544 | break; | |
545 | } | |
41fcb9f2 | 546 | |
c516df97 NH |
547 | if (ww_ctx && !ww_mutex_spin_on_owner(lock, ww_ctx, waiter)) { |
548 | ret = false; | |
549 | break; | |
25f13b40 NH |
550 | } |
551 | ||
f2f09a4c | 552 | cpu_relax(); |
41fcb9f2 WL |
553 | } |
554 | rcu_read_unlock(); | |
555 | ||
be1f7bf2 | 556 | return ret; |
41fcb9f2 | 557 | } |
2bd2c92c WL |
558 | |
559 | /* | |
560 | * Initial check for entering the mutex spinning loop | |
561 | */ | |
562 | static inline int mutex_can_spin_on_owner(struct mutex *lock) | |
563 | { | |
1e40c2ed | 564 | struct task_struct *owner; |
2bd2c92c WL |
565 | int retval = 1; |
566 | ||
46af29e4 JL |
567 | if (need_resched()) |
568 | return 0; | |
569 | ||
2bd2c92c | 570 | rcu_read_lock(); |
3ca0ff57 | 571 | owner = __mutex_owner(lock); |
05ffc951 PX |
572 | |
573 | /* | |
574 | * As lock holder preemption issue, we both skip spinning if task is not | |
575 | * on cpu or its cpu is preempted | |
576 | */ | |
1e40c2ed | 577 | if (owner) |
05ffc951 | 578 | retval = owner->on_cpu && !vcpu_is_preempted(task_cpu(owner)); |
2bd2c92c | 579 | rcu_read_unlock(); |
3ca0ff57 | 580 | |
2bd2c92c | 581 | /* |
3ca0ff57 PZ |
582 | * If lock->owner is not set, the mutex has been released. Return true |
583 | * such that we'll trylock in the spin path, which is a faster option | |
584 | * than the blocking slow path. | |
2bd2c92c WL |
585 | */ |
586 | return retval; | |
587 | } | |
76916515 | 588 | |
76916515 DB |
589 | /* |
590 | * Optimistic spinning. | |
591 | * | |
592 | * We try to spin for acquisition when we find that the lock owner | |
593 | * is currently running on a (different) CPU and while we don't | |
594 | * need to reschedule. The rationale is that if the lock owner is | |
595 | * running, it is likely to release the lock soon. | |
596 | * | |
76916515 DB |
597 | * The mutex spinners are queued up using MCS lock so that only one |
598 | * spinner can compete for the mutex. However, if mutex spinning isn't | |
599 | * going to happen, there is no point in going through the lock/unlock | |
600 | * overhead. | |
601 | * | |
602 | * Returns true when the lock was taken, otherwise false, indicating | |
603 | * that we need to jump to the slowpath and sleep. | |
b341afb3 WL |
604 | * |
605 | * The waiter flag is set to true if the spinner is a waiter in the wait | |
606 | * queue. The waiter-spinner will spin on the lock directly and concurrently | |
607 | * with the spinner at the head of the OSQ, if present, until the owner is | |
608 | * changed to itself. | |
76916515 | 609 | */ |
427b1820 PZ |
610 | static __always_inline bool |
611 | mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx, | |
c516df97 | 612 | const bool use_ww_ctx, struct mutex_waiter *waiter) |
76916515 | 613 | { |
b341afb3 WL |
614 | if (!waiter) { |
615 | /* | |
616 | * The purpose of the mutex_can_spin_on_owner() function is | |
617 | * to eliminate the overhead of osq_lock() and osq_unlock() | |
618 | * in case spinning isn't possible. As a waiter-spinner | |
619 | * is not going to take OSQ lock anyway, there is no need | |
620 | * to call mutex_can_spin_on_owner(). | |
621 | */ | |
622 | if (!mutex_can_spin_on_owner(lock)) | |
623 | goto fail; | |
76916515 | 624 | |
b341afb3 WL |
625 | /* |
626 | * In order to avoid a stampede of mutex spinners trying to | |
627 | * acquire the mutex all at once, the spinners need to take a | |
628 | * MCS (queued) lock first before spinning on the owner field. | |
629 | */ | |
630 | if (!osq_lock(&lock->osq)) | |
631 | goto fail; | |
632 | } | |
76916515 | 633 | |
b341afb3 | 634 | for (;;) { |
76916515 DB |
635 | struct task_struct *owner; |
636 | ||
e274795e PZ |
637 | /* Try to acquire the mutex... */ |
638 | owner = __mutex_trylock_or_owner(lock); | |
639 | if (!owner) | |
640 | break; | |
76916515 DB |
641 | |
642 | /* | |
e274795e | 643 | * There's an owner, wait for it to either |
76916515 DB |
644 | * release the lock or go to sleep. |
645 | */ | |
c516df97 | 646 | if (!mutex_spin_on_owner(lock, owner, ww_ctx, waiter)) |
e274795e | 647 | goto fail_unlock; |
b341afb3 | 648 | |
76916515 DB |
649 | /* |
650 | * The cpu_relax() call is a compiler barrier which forces | |
651 | * everything in this loop to be re-loaded. We don't need | |
652 | * memory barriers as we'll eventually observe the right | |
653 | * values at the cost of a few extra spins. | |
654 | */ | |
f2f09a4c | 655 | cpu_relax(); |
76916515 DB |
656 | } |
657 | ||
b341afb3 WL |
658 | if (!waiter) |
659 | osq_unlock(&lock->osq); | |
660 | ||
661 | return true; | |
662 | ||
663 | ||
664 | fail_unlock: | |
665 | if (!waiter) | |
666 | osq_unlock(&lock->osq); | |
667 | ||
668 | fail: | |
76916515 DB |
669 | /* |
670 | * If we fell out of the spin path because of need_resched(), | |
671 | * reschedule now, before we try-lock the mutex. This avoids getting | |
672 | * scheduled out right after we obtained the mutex. | |
673 | */ | |
6f942a1f PZ |
674 | if (need_resched()) { |
675 | /* | |
676 | * We _should_ have TASK_RUNNING here, but just in case | |
677 | * we do not, make it so, otherwise we might get stuck. | |
678 | */ | |
679 | __set_current_state(TASK_RUNNING); | |
76916515 | 680 | schedule_preempt_disabled(); |
6f942a1f | 681 | } |
76916515 DB |
682 | |
683 | return false; | |
684 | } | |
685 | #else | |
427b1820 PZ |
686 | static __always_inline bool |
687 | mutex_optimistic_spin(struct mutex *lock, struct ww_acquire_ctx *ww_ctx, | |
c516df97 | 688 | const bool use_ww_ctx, struct mutex_waiter *waiter) |
76916515 DB |
689 | { |
690 | return false; | |
691 | } | |
41fcb9f2 WL |
692 | #endif |
693 | ||
3ca0ff57 | 694 | static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip); |
6053ee3b | 695 | |
ef5dc121 | 696 | /** |
6053ee3b IM |
697 | * mutex_unlock - release the mutex |
698 | * @lock: the mutex to be released | |
699 | * | |
700 | * Unlock a mutex that has been locked by this task previously. | |
701 | * | |
702 | * This function must not be used in interrupt context. Unlocking | |
703 | * of a not locked mutex is not allowed. | |
704 | * | |
705 | * This function is similar to (but not equivalent to) up(). | |
706 | */ | |
7ad5b3a5 | 707 | void __sched mutex_unlock(struct mutex *lock) |
6053ee3b | 708 | { |
3ca0ff57 PZ |
709 | #ifndef CONFIG_DEBUG_LOCK_ALLOC |
710 | if (__mutex_unlock_fast(lock)) | |
711 | return; | |
0d66bf6d | 712 | #endif |
3ca0ff57 | 713 | __mutex_unlock_slowpath(lock, _RET_IP_); |
6053ee3b | 714 | } |
6053ee3b IM |
715 | EXPORT_SYMBOL(mutex_unlock); |
716 | ||
040a0a37 ML |
717 | /** |
718 | * ww_mutex_unlock - release the w/w mutex | |
719 | * @lock: the mutex to be released | |
720 | * | |
721 | * Unlock a mutex that has been locked by this task previously with any of the | |
722 | * ww_mutex_lock* functions (with or without an acquire context). It is | |
723 | * forbidden to release the locks after releasing the acquire context. | |
724 | * | |
725 | * This function must not be used in interrupt context. Unlocking | |
726 | * of a unlocked mutex is not allowed. | |
727 | */ | |
728 | void __sched ww_mutex_unlock(struct ww_mutex *lock) | |
729 | { | |
730 | /* | |
731 | * The unlocking fastpath is the 0->1 transition from 'locked' | |
732 | * into 'unlocked' state: | |
733 | */ | |
734 | if (lock->ctx) { | |
735 | #ifdef CONFIG_DEBUG_MUTEXES | |
736 | DEBUG_LOCKS_WARN_ON(!lock->ctx->acquired); | |
737 | #endif | |
738 | if (lock->ctx->acquired > 0) | |
739 | lock->ctx->acquired--; | |
740 | lock->ctx = NULL; | |
741 | } | |
742 | ||
3ca0ff57 | 743 | mutex_unlock(&lock->base); |
040a0a37 ML |
744 | } |
745 | EXPORT_SYMBOL(ww_mutex_unlock); | |
746 | ||
55f036ca PZ |
747 | |
748 | static __always_inline int __sched | |
749 | __ww_mutex_kill(struct mutex *lock, struct ww_acquire_ctx *ww_ctx) | |
750 | { | |
751 | if (ww_ctx->acquired > 0) { | |
752 | #ifdef CONFIG_DEBUG_MUTEXES | |
753 | struct ww_mutex *ww; | |
754 | ||
755 | ww = container_of(lock, struct ww_mutex, base); | |
756 | DEBUG_LOCKS_WARN_ON(ww_ctx->contending_lock); | |
757 | ww_ctx->contending_lock = ww; | |
758 | #endif | |
759 | return -EDEADLK; | |
760 | } | |
761 | ||
762 | return 0; | |
763 | } | |
764 | ||
765 | ||
766 | /* | |
08295b3b TH |
767 | * Check the wound condition for the current lock acquire. |
768 | * | |
769 | * Wound-Wait: If we're wounded, kill ourself. | |
55f036ca PZ |
770 | * |
771 | * Wait-Die: If we're trying to acquire a lock already held by an older | |
772 | * context, kill ourselves. | |
773 | * | |
774 | * Since __ww_mutex_add_waiter() orders the wait-list on stamp, we only have to | |
775 | * look at waiters before us in the wait-list. | |
776 | */ | |
040a0a37 | 777 | static inline int __sched |
55f036ca PZ |
778 | __ww_mutex_check_kill(struct mutex *lock, struct mutex_waiter *waiter, |
779 | struct ww_acquire_ctx *ctx) | |
040a0a37 ML |
780 | { |
781 | struct ww_mutex *ww = container_of(lock, struct ww_mutex, base); | |
4d3199e4 | 782 | struct ww_acquire_ctx *hold_ctx = READ_ONCE(ww->ctx); |
200b1874 | 783 | struct mutex_waiter *cur; |
040a0a37 | 784 | |
55f036ca PZ |
785 | if (ctx->acquired == 0) |
786 | return 0; | |
787 | ||
08295b3b TH |
788 | if (!ctx->is_wait_die) { |
789 | if (ctx->wounded) | |
790 | return __ww_mutex_kill(lock, ctx); | |
791 | ||
792 | return 0; | |
793 | } | |
794 | ||
200b1874 | 795 | if (hold_ctx && __ww_ctx_stamp_after(ctx, hold_ctx)) |
55f036ca | 796 | return __ww_mutex_kill(lock, ctx); |
040a0a37 | 797 | |
200b1874 NH |
798 | /* |
799 | * If there is a waiter in front of us that has a context, then its | |
55f036ca | 800 | * stamp is earlier than ours and we must kill ourself. |
200b1874 NH |
801 | */ |
802 | cur = waiter; | |
803 | list_for_each_entry_continue_reverse(cur, &lock->wait_list, list) { | |
55f036ca PZ |
804 | if (!cur->ww_ctx) |
805 | continue; | |
806 | ||
807 | return __ww_mutex_kill(lock, ctx); | |
040a0a37 ML |
808 | } |
809 | ||
810 | return 0; | |
811 | } | |
812 | ||
55f036ca PZ |
813 | /* |
814 | * Add @waiter to the wait-list, keep the wait-list ordered by stamp, smallest | |
815 | * first. Such that older contexts are preferred to acquire the lock over | |
816 | * younger contexts. | |
817 | * | |
818 | * Waiters without context are interspersed in FIFO order. | |
819 | * | |
820 | * Furthermore, for Wait-Die kill ourself immediately when possible (there are | |
08295b3b TH |
821 | * older contexts already waiting) to avoid unnecessary waiting and for |
822 | * Wound-Wait ensure we wound the owning context when it is younger. | |
55f036ca | 823 | */ |
6baa5c60 NH |
824 | static inline int __sched |
825 | __ww_mutex_add_waiter(struct mutex_waiter *waiter, | |
826 | struct mutex *lock, | |
827 | struct ww_acquire_ctx *ww_ctx) | |
828 | { | |
829 | struct mutex_waiter *cur; | |
830 | struct list_head *pos; | |
08295b3b | 831 | bool is_wait_die; |
6baa5c60 NH |
832 | |
833 | if (!ww_ctx) { | |
08295b3b | 834 | __mutex_add_waiter(lock, waiter, &lock->wait_list); |
040a0a37 | 835 | return 0; |
6baa5c60 | 836 | } |
040a0a37 | 837 | |
08295b3b TH |
838 | is_wait_die = ww_ctx->is_wait_die; |
839 | ||
6baa5c60 NH |
840 | /* |
841 | * Add the waiter before the first waiter with a higher stamp. | |
842 | * Waiters without a context are skipped to avoid starving | |
08295b3b TH |
843 | * them. Wait-Die waiters may die here. Wound-Wait waiters |
844 | * never die here, but they are sorted in stamp order and | |
845 | * may wound the lock holder. | |
6baa5c60 NH |
846 | */ |
847 | pos = &lock->wait_list; | |
848 | list_for_each_entry_reverse(cur, &lock->wait_list, list) { | |
849 | if (!cur->ww_ctx) | |
850 | continue; | |
851 | ||
852 | if (__ww_ctx_stamp_after(ww_ctx, cur->ww_ctx)) { | |
55f036ca PZ |
853 | /* |
854 | * Wait-Die: if we find an older context waiting, there | |
855 | * is no point in queueing behind it, as we'd have to | |
856 | * die the moment it would acquire the lock. | |
857 | */ | |
08295b3b TH |
858 | if (is_wait_die) { |
859 | int ret = __ww_mutex_kill(lock, ww_ctx); | |
6baa5c60 | 860 | |
08295b3b TH |
861 | if (ret) |
862 | return ret; | |
863 | } | |
6baa5c60 NH |
864 | |
865 | break; | |
866 | } | |
867 | ||
868 | pos = &cur->list; | |
200b1874 | 869 | |
55f036ca PZ |
870 | /* Wait-Die: ensure younger waiters die. */ |
871 | __ww_mutex_die(lock, cur, ww_ctx); | |
040a0a37 ML |
872 | } |
873 | ||
08295b3b TH |
874 | __mutex_add_waiter(lock, waiter, pos); |
875 | ||
876 | /* | |
877 | * Wound-Wait: if we're blocking on a mutex owned by a younger context, | |
878 | * wound that such that we might proceed. | |
879 | */ | |
880 | if (!is_wait_die) { | |
881 | struct ww_mutex *ww = container_of(lock, struct ww_mutex, base); | |
882 | ||
883 | /* | |
884 | * See ww_mutex_set_context_fastpath(). Orders setting | |
885 | * MUTEX_FLAG_WAITERS vs the ww->ctx load, | |
886 | * such that either we or the fastpath will wound @ww->ctx. | |
887 | */ | |
888 | smp_mb(); | |
889 | __ww_mutex_wound(lock, ww_ctx, ww->ctx); | |
890 | } | |
55f036ca | 891 | |
040a0a37 ML |
892 | return 0; |
893 | } | |
894 | ||
6053ee3b IM |
895 | /* |
896 | * Lock a mutex (possibly interruptible), slowpath: | |
897 | */ | |
040a0a37 | 898 | static __always_inline int __sched |
e4564f79 | 899 | __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass, |
040a0a37 | 900 | struct lockdep_map *nest_lock, unsigned long ip, |
b0267507 | 901 | struct ww_acquire_ctx *ww_ctx, const bool use_ww_ctx) |
6053ee3b | 902 | { |
6053ee3b | 903 | struct mutex_waiter waiter; |
9d659ae1 | 904 | bool first = false; |
a40ca565 | 905 | struct ww_mutex *ww; |
040a0a37 | 906 | int ret; |
6053ee3b | 907 | |
427b1820 | 908 | might_sleep(); |
ea9e0fb8 | 909 | |
427b1820 | 910 | ww = container_of(lock, struct ww_mutex, base); |
ea9e0fb8 | 911 | if (use_ww_ctx && ww_ctx) { |
0422e83d CW |
912 | if (unlikely(ww_ctx == READ_ONCE(ww->ctx))) |
913 | return -EALREADY; | |
08295b3b TH |
914 | |
915 | /* | |
916 | * Reset the wounded flag after a kill. No other process can | |
917 | * race and wound us here since they can't have a valid owner | |
918 | * pointer if we don't have any locks held. | |
919 | */ | |
920 | if (ww_ctx->acquired == 0) | |
921 | ww_ctx->wounded = 0; | |
0422e83d CW |
922 | } |
923 | ||
41719b03 | 924 | preempt_disable(); |
e4c70a66 | 925 | mutex_acquire_nest(&lock->dep_map, subclass, 0, nest_lock, ip); |
c0226027 | 926 | |
e274795e | 927 | if (__mutex_trylock(lock) || |
c516df97 | 928 | mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, NULL)) { |
76916515 | 929 | /* got the lock, yay! */ |
3ca0ff57 | 930 | lock_acquired(&lock->dep_map, ip); |
ea9e0fb8 | 931 | if (use_ww_ctx && ww_ctx) |
3ca0ff57 | 932 | ww_mutex_set_context_fastpath(ww, ww_ctx); |
76916515 DB |
933 | preempt_enable(); |
934 | return 0; | |
0d66bf6d | 935 | } |
76916515 | 936 | |
b9c16a0e | 937 | spin_lock(&lock->wait_lock); |
1e820c96 | 938 | /* |
3ca0ff57 | 939 | * After waiting to acquire the wait_lock, try again. |
1e820c96 | 940 | */ |
659cf9f5 NH |
941 | if (__mutex_trylock(lock)) { |
942 | if (use_ww_ctx && ww_ctx) | |
55f036ca | 943 | __ww_mutex_check_waiters(lock, ww_ctx); |
659cf9f5 | 944 | |
ec83f425 | 945 | goto skip_wait; |
659cf9f5 | 946 | } |
ec83f425 | 947 | |
9a11b49a | 948 | debug_mutex_lock_common(lock, &waiter); |
6053ee3b | 949 | |
6baa5c60 NH |
950 | lock_contended(&lock->dep_map, ip); |
951 | ||
952 | if (!use_ww_ctx) { | |
953 | /* add waiting tasks to the end of the waitqueue (FIFO): */ | |
08295b3b TH |
954 | __mutex_add_waiter(lock, &waiter, &lock->wait_list); |
955 | ||
977625a6 NH |
956 | |
957 | #ifdef CONFIG_DEBUG_MUTEXES | |
958 | waiter.ww_ctx = MUTEX_POISON_WW_CTX; | |
959 | #endif | |
6baa5c60 | 960 | } else { |
55f036ca PZ |
961 | /* |
962 | * Add in stamp order, waking up waiters that must kill | |
963 | * themselves. | |
964 | */ | |
6baa5c60 NH |
965 | ret = __ww_mutex_add_waiter(&waiter, lock, ww_ctx); |
966 | if (ret) | |
55f036ca | 967 | goto err_early_kill; |
6baa5c60 NH |
968 | |
969 | waiter.ww_ctx = ww_ctx; | |
970 | } | |
971 | ||
d269a8b8 | 972 | waiter.task = current; |
6053ee3b | 973 | |
642fa448 | 974 | set_current_state(state); |
6053ee3b | 975 | for (;;) { |
5bbd7e64 PZ |
976 | /* |
977 | * Once we hold wait_lock, we're serialized against | |
978 | * mutex_unlock() handing the lock off to us, do a trylock | |
979 | * before testing the error conditions to make sure we pick up | |
980 | * the handoff. | |
981 | */ | |
e274795e | 982 | if (__mutex_trylock(lock)) |
5bbd7e64 | 983 | goto acquired; |
6053ee3b IM |
984 | |
985 | /* | |
55f036ca | 986 | * Check for signals and kill conditions while holding |
5bbd7e64 PZ |
987 | * wait_lock. This ensures the lock cancellation is ordered |
988 | * against mutex_unlock() and wake-ups do not go missing. | |
6053ee3b | 989 | */ |
3bb5f4ac | 990 | if (signal_pending_state(state, current)) { |
040a0a37 ML |
991 | ret = -EINTR; |
992 | goto err; | |
993 | } | |
6053ee3b | 994 | |
55f036ca PZ |
995 | if (use_ww_ctx && ww_ctx) { |
996 | ret = __ww_mutex_check_kill(lock, &waiter, ww_ctx); | |
040a0a37 ML |
997 | if (ret) |
998 | goto err; | |
6053ee3b | 999 | } |
040a0a37 | 1000 | |
b9c16a0e | 1001 | spin_unlock(&lock->wait_lock); |
bd2f5536 | 1002 | schedule_preempt_disabled(); |
9d659ae1 | 1003 | |
6baa5c60 NH |
1004 | /* |
1005 | * ww_mutex needs to always recheck its position since its waiter | |
1006 | * list is not FIFO ordered. | |
1007 | */ | |
1008 | if ((use_ww_ctx && ww_ctx) || !first) { | |
1009 | first = __mutex_waiter_is_first(lock, &waiter); | |
1010 | if (first) | |
1011 | __mutex_set_flag(lock, MUTEX_FLAG_HANDOFF); | |
9d659ae1 | 1012 | } |
5bbd7e64 | 1013 | |
642fa448 | 1014 | set_current_state(state); |
5bbd7e64 PZ |
1015 | /* |
1016 | * Here we order against unlock; we must either see it change | |
1017 | * state back to RUNNING and fall through the next schedule(), | |
1018 | * or we must see its unlock and acquire. | |
1019 | */ | |
e274795e | 1020 | if (__mutex_trylock(lock) || |
c516df97 | 1021 | (first && mutex_optimistic_spin(lock, ww_ctx, use_ww_ctx, &waiter))) |
5bbd7e64 PZ |
1022 | break; |
1023 | ||
b9c16a0e | 1024 | spin_lock(&lock->wait_lock); |
6053ee3b | 1025 | } |
b9c16a0e | 1026 | spin_lock(&lock->wait_lock); |
5bbd7e64 | 1027 | acquired: |
642fa448 | 1028 | __set_current_state(TASK_RUNNING); |
51587bcf | 1029 | |
08295b3b TH |
1030 | if (use_ww_ctx && ww_ctx) { |
1031 | /* | |
1032 | * Wound-Wait; we stole the lock (!first_waiter), check the | |
1033 | * waiters as anyone might want to wound us. | |
1034 | */ | |
1035 | if (!ww_ctx->is_wait_die && | |
1036 | !__mutex_waiter_is_first(lock, &waiter)) | |
1037 | __ww_mutex_check_waiters(lock, ww_ctx); | |
1038 | } | |
1039 | ||
d269a8b8 | 1040 | mutex_remove_waiter(lock, &waiter, current); |
ec83f425 | 1041 | if (likely(list_empty(&lock->wait_list))) |
9d659ae1 | 1042 | __mutex_clear_flag(lock, MUTEX_FLAGS); |
3ca0ff57 | 1043 | |
ec83f425 | 1044 | debug_mutex_free_waiter(&waiter); |
6053ee3b | 1045 | |
ec83f425 DB |
1046 | skip_wait: |
1047 | /* got the lock - cleanup and rejoice! */ | |
c7e78cff | 1048 | lock_acquired(&lock->dep_map, ip); |
6053ee3b | 1049 | |
ea9e0fb8 | 1050 | if (use_ww_ctx && ww_ctx) |
55f036ca | 1051 | ww_mutex_lock_acquired(ww, ww_ctx); |
040a0a37 | 1052 | |
b9c16a0e | 1053 | spin_unlock(&lock->wait_lock); |
41719b03 | 1054 | preempt_enable(); |
6053ee3b | 1055 | return 0; |
040a0a37 ML |
1056 | |
1057 | err: | |
642fa448 | 1058 | __set_current_state(TASK_RUNNING); |
d269a8b8 | 1059 | mutex_remove_waiter(lock, &waiter, current); |
55f036ca | 1060 | err_early_kill: |
b9c16a0e | 1061 | spin_unlock(&lock->wait_lock); |
040a0a37 ML |
1062 | debug_mutex_free_waiter(&waiter); |
1063 | mutex_release(&lock->dep_map, 1, ip); | |
1064 | preempt_enable(); | |
1065 | return ret; | |
6053ee3b IM |
1066 | } |
1067 | ||
427b1820 PZ |
1068 | static int __sched |
1069 | __mutex_lock(struct mutex *lock, long state, unsigned int subclass, | |
1070 | struct lockdep_map *nest_lock, unsigned long ip) | |
1071 | { | |
1072 | return __mutex_lock_common(lock, state, subclass, nest_lock, ip, NULL, false); | |
1073 | } | |
1074 | ||
1075 | static int __sched | |
1076 | __ww_mutex_lock(struct mutex *lock, long state, unsigned int subclass, | |
1077 | struct lockdep_map *nest_lock, unsigned long ip, | |
1078 | struct ww_acquire_ctx *ww_ctx) | |
1079 | { | |
1080 | return __mutex_lock_common(lock, state, subclass, nest_lock, ip, ww_ctx, true); | |
1081 | } | |
1082 | ||
ef5d4707 IM |
1083 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
1084 | void __sched | |
1085 | mutex_lock_nested(struct mutex *lock, unsigned int subclass) | |
1086 | { | |
427b1820 | 1087 | __mutex_lock(lock, TASK_UNINTERRUPTIBLE, subclass, NULL, _RET_IP_); |
ef5d4707 IM |
1088 | } |
1089 | ||
1090 | EXPORT_SYMBOL_GPL(mutex_lock_nested); | |
d63a5a74 | 1091 | |
e4c70a66 PZ |
1092 | void __sched |
1093 | _mutex_lock_nest_lock(struct mutex *lock, struct lockdep_map *nest) | |
1094 | { | |
427b1820 | 1095 | __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, nest, _RET_IP_); |
e4c70a66 | 1096 | } |
e4c70a66 PZ |
1097 | EXPORT_SYMBOL_GPL(_mutex_lock_nest_lock); |
1098 | ||
ad776537 LH |
1099 | int __sched |
1100 | mutex_lock_killable_nested(struct mutex *lock, unsigned int subclass) | |
1101 | { | |
427b1820 | 1102 | return __mutex_lock(lock, TASK_KILLABLE, subclass, NULL, _RET_IP_); |
ad776537 LH |
1103 | } |
1104 | EXPORT_SYMBOL_GPL(mutex_lock_killable_nested); | |
1105 | ||
d63a5a74 N |
1106 | int __sched |
1107 | mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass) | |
1108 | { | |
427b1820 | 1109 | return __mutex_lock(lock, TASK_INTERRUPTIBLE, subclass, NULL, _RET_IP_); |
d63a5a74 | 1110 | } |
d63a5a74 | 1111 | EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested); |
040a0a37 | 1112 | |
1460cb65 TH |
1113 | void __sched |
1114 | mutex_lock_io_nested(struct mutex *lock, unsigned int subclass) | |
1115 | { | |
1116 | int token; | |
1117 | ||
1118 | might_sleep(); | |
1119 | ||
1120 | token = io_schedule_prepare(); | |
1121 | __mutex_lock_common(lock, TASK_UNINTERRUPTIBLE, | |
1122 | subclass, NULL, _RET_IP_, NULL, 0); | |
1123 | io_schedule_finish(token); | |
1124 | } | |
1125 | EXPORT_SYMBOL_GPL(mutex_lock_io_nested); | |
1126 | ||
23010027 DV |
1127 | static inline int |
1128 | ww_mutex_deadlock_injection(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) | |
1129 | { | |
1130 | #ifdef CONFIG_DEBUG_WW_MUTEX_SLOWPATH | |
1131 | unsigned tmp; | |
1132 | ||
1133 | if (ctx->deadlock_inject_countdown-- == 0) { | |
1134 | tmp = ctx->deadlock_inject_interval; | |
1135 | if (tmp > UINT_MAX/4) | |
1136 | tmp = UINT_MAX; | |
1137 | else | |
1138 | tmp = tmp*2 + tmp + tmp/2; | |
1139 | ||
1140 | ctx->deadlock_inject_interval = tmp; | |
1141 | ctx->deadlock_inject_countdown = tmp; | |
1142 | ctx->contending_lock = lock; | |
1143 | ||
1144 | ww_mutex_unlock(lock); | |
1145 | ||
1146 | return -EDEADLK; | |
1147 | } | |
1148 | #endif | |
1149 | ||
1150 | return 0; | |
1151 | } | |
040a0a37 ML |
1152 | |
1153 | int __sched | |
c5470b22 | 1154 | ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
040a0a37 | 1155 | { |
23010027 DV |
1156 | int ret; |
1157 | ||
040a0a37 | 1158 | might_sleep(); |
427b1820 PZ |
1159 | ret = __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE, |
1160 | 0, ctx ? &ctx->dep_map : NULL, _RET_IP_, | |
1161 | ctx); | |
ea9e0fb8 | 1162 | if (!ret && ctx && ctx->acquired > 1) |
23010027 DV |
1163 | return ww_mutex_deadlock_injection(lock, ctx); |
1164 | ||
1165 | return ret; | |
040a0a37 | 1166 | } |
c5470b22 | 1167 | EXPORT_SYMBOL_GPL(ww_mutex_lock); |
040a0a37 ML |
1168 | |
1169 | int __sched | |
c5470b22 | 1170 | ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
040a0a37 | 1171 | { |
23010027 DV |
1172 | int ret; |
1173 | ||
040a0a37 | 1174 | might_sleep(); |
427b1820 PZ |
1175 | ret = __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE, |
1176 | 0, ctx ? &ctx->dep_map : NULL, _RET_IP_, | |
1177 | ctx); | |
23010027 | 1178 | |
ea9e0fb8 | 1179 | if (!ret && ctx && ctx->acquired > 1) |
23010027 DV |
1180 | return ww_mutex_deadlock_injection(lock, ctx); |
1181 | ||
1182 | return ret; | |
040a0a37 | 1183 | } |
c5470b22 | 1184 | EXPORT_SYMBOL_GPL(ww_mutex_lock_interruptible); |
040a0a37 | 1185 | |
ef5d4707 IM |
1186 | #endif |
1187 | ||
6053ee3b IM |
1188 | /* |
1189 | * Release the lock, slowpath: | |
1190 | */ | |
3ca0ff57 | 1191 | static noinline void __sched __mutex_unlock_slowpath(struct mutex *lock, unsigned long ip) |
6053ee3b | 1192 | { |
9d659ae1 | 1193 | struct task_struct *next = NULL; |
194a6b5b | 1194 | DEFINE_WAKE_Q(wake_q); |
b9c16a0e | 1195 | unsigned long owner; |
6053ee3b | 1196 | |
3ca0ff57 PZ |
1197 | mutex_release(&lock->dep_map, 1, ip); |
1198 | ||
6053ee3b | 1199 | /* |
9d659ae1 PZ |
1200 | * Release the lock before (potentially) taking the spinlock such that |
1201 | * other contenders can get on with things ASAP. | |
1202 | * | |
1203 | * Except when HANDOFF, in that case we must not clear the owner field, | |
1204 | * but instead set it to the top waiter. | |
6053ee3b | 1205 | */ |
9d659ae1 PZ |
1206 | owner = atomic_long_read(&lock->owner); |
1207 | for (;;) { | |
1208 | unsigned long old; | |
1209 | ||
1210 | #ifdef CONFIG_DEBUG_MUTEXES | |
1211 | DEBUG_LOCKS_WARN_ON(__owner_task(owner) != current); | |
e274795e | 1212 | DEBUG_LOCKS_WARN_ON(owner & MUTEX_FLAG_PICKUP); |
9d659ae1 PZ |
1213 | #endif |
1214 | ||
1215 | if (owner & MUTEX_FLAG_HANDOFF) | |
1216 | break; | |
1217 | ||
1218 | old = atomic_long_cmpxchg_release(&lock->owner, owner, | |
1219 | __owner_flags(owner)); | |
1220 | if (old == owner) { | |
1221 | if (owner & MUTEX_FLAG_WAITERS) | |
1222 | break; | |
1223 | ||
1224 | return; | |
1225 | } | |
1226 | ||
1227 | owner = old; | |
1228 | } | |
6053ee3b | 1229 | |
b9c16a0e | 1230 | spin_lock(&lock->wait_lock); |
1d8fe7dc | 1231 | debug_mutex_unlock(lock); |
6053ee3b IM |
1232 | if (!list_empty(&lock->wait_list)) { |
1233 | /* get the first entry from the wait-list: */ | |
1234 | struct mutex_waiter *waiter = | |
9d659ae1 PZ |
1235 | list_first_entry(&lock->wait_list, |
1236 | struct mutex_waiter, list); | |
1237 | ||
1238 | next = waiter->task; | |
6053ee3b IM |
1239 | |
1240 | debug_mutex_wake_waiter(lock, waiter); | |
9d659ae1 | 1241 | wake_q_add(&wake_q, next); |
6053ee3b IM |
1242 | } |
1243 | ||
9d659ae1 PZ |
1244 | if (owner & MUTEX_FLAG_HANDOFF) |
1245 | __mutex_handoff(lock, next); | |
1246 | ||
b9c16a0e | 1247 | spin_unlock(&lock->wait_lock); |
9d659ae1 | 1248 | |
1329ce6f | 1249 | wake_up_q(&wake_q); |
6053ee3b IM |
1250 | } |
1251 | ||
e4564f79 | 1252 | #ifndef CONFIG_DEBUG_LOCK_ALLOC |
6053ee3b IM |
1253 | /* |
1254 | * Here come the less common (and hence less performance-critical) APIs: | |
1255 | * mutex_lock_interruptible() and mutex_trylock(). | |
1256 | */ | |
7ad5b3a5 | 1257 | static noinline int __sched |
a41b56ef | 1258 | __mutex_lock_killable_slowpath(struct mutex *lock); |
ad776537 | 1259 | |
7ad5b3a5 | 1260 | static noinline int __sched |
a41b56ef | 1261 | __mutex_lock_interruptible_slowpath(struct mutex *lock); |
6053ee3b | 1262 | |
ef5dc121 | 1263 | /** |
45dbac0e MW |
1264 | * mutex_lock_interruptible() - Acquire the mutex, interruptible by signals. |
1265 | * @lock: The mutex to be acquired. | |
6053ee3b | 1266 | * |
45dbac0e MW |
1267 | * Lock the mutex like mutex_lock(). If a signal is delivered while the |
1268 | * process is sleeping, this function will return without acquiring the | |
1269 | * mutex. | |
6053ee3b | 1270 | * |
45dbac0e MW |
1271 | * Context: Process context. |
1272 | * Return: 0 if the lock was successfully acquired or %-EINTR if a | |
1273 | * signal arrived. | |
6053ee3b | 1274 | */ |
7ad5b3a5 | 1275 | int __sched mutex_lock_interruptible(struct mutex *lock) |
6053ee3b | 1276 | { |
c544bdb1 | 1277 | might_sleep(); |
3ca0ff57 PZ |
1278 | |
1279 | if (__mutex_trylock_fast(lock)) | |
a41b56ef | 1280 | return 0; |
3ca0ff57 PZ |
1281 | |
1282 | return __mutex_lock_interruptible_slowpath(lock); | |
6053ee3b IM |
1283 | } |
1284 | ||
1285 | EXPORT_SYMBOL(mutex_lock_interruptible); | |
1286 | ||
45dbac0e MW |
1287 | /** |
1288 | * mutex_lock_killable() - Acquire the mutex, interruptible by fatal signals. | |
1289 | * @lock: The mutex to be acquired. | |
1290 | * | |
1291 | * Lock the mutex like mutex_lock(). If a signal which will be fatal to | |
1292 | * the current process is delivered while the process is sleeping, this | |
1293 | * function will return without acquiring the mutex. | |
1294 | * | |
1295 | * Context: Process context. | |
1296 | * Return: 0 if the lock was successfully acquired or %-EINTR if a | |
1297 | * fatal signal arrived. | |
1298 | */ | |
7ad5b3a5 | 1299 | int __sched mutex_lock_killable(struct mutex *lock) |
ad776537 LH |
1300 | { |
1301 | might_sleep(); | |
3ca0ff57 PZ |
1302 | |
1303 | if (__mutex_trylock_fast(lock)) | |
a41b56ef | 1304 | return 0; |
3ca0ff57 PZ |
1305 | |
1306 | return __mutex_lock_killable_slowpath(lock); | |
ad776537 LH |
1307 | } |
1308 | EXPORT_SYMBOL(mutex_lock_killable); | |
1309 | ||
45dbac0e MW |
1310 | /** |
1311 | * mutex_lock_io() - Acquire the mutex and mark the process as waiting for I/O | |
1312 | * @lock: The mutex to be acquired. | |
1313 | * | |
1314 | * Lock the mutex like mutex_lock(). While the task is waiting for this | |
1315 | * mutex, it will be accounted as being in the IO wait state by the | |
1316 | * scheduler. | |
1317 | * | |
1318 | * Context: Process context. | |
1319 | */ | |
1460cb65 TH |
1320 | void __sched mutex_lock_io(struct mutex *lock) |
1321 | { | |
1322 | int token; | |
1323 | ||
1324 | token = io_schedule_prepare(); | |
1325 | mutex_lock(lock); | |
1326 | io_schedule_finish(token); | |
1327 | } | |
1328 | EXPORT_SYMBOL_GPL(mutex_lock_io); | |
1329 | ||
3ca0ff57 PZ |
1330 | static noinline void __sched |
1331 | __mutex_lock_slowpath(struct mutex *lock) | |
e4564f79 | 1332 | { |
427b1820 | 1333 | __mutex_lock(lock, TASK_UNINTERRUPTIBLE, 0, NULL, _RET_IP_); |
e4564f79 PZ |
1334 | } |
1335 | ||
7ad5b3a5 | 1336 | static noinline int __sched |
a41b56ef | 1337 | __mutex_lock_killable_slowpath(struct mutex *lock) |
ad776537 | 1338 | { |
427b1820 | 1339 | return __mutex_lock(lock, TASK_KILLABLE, 0, NULL, _RET_IP_); |
ad776537 LH |
1340 | } |
1341 | ||
7ad5b3a5 | 1342 | static noinline int __sched |
a41b56ef | 1343 | __mutex_lock_interruptible_slowpath(struct mutex *lock) |
6053ee3b | 1344 | { |
427b1820 | 1345 | return __mutex_lock(lock, TASK_INTERRUPTIBLE, 0, NULL, _RET_IP_); |
040a0a37 ML |
1346 | } |
1347 | ||
1348 | static noinline int __sched | |
1349 | __ww_mutex_lock_slowpath(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) | |
1350 | { | |
427b1820 PZ |
1351 | return __ww_mutex_lock(&lock->base, TASK_UNINTERRUPTIBLE, 0, NULL, |
1352 | _RET_IP_, ctx); | |
6053ee3b | 1353 | } |
040a0a37 ML |
1354 | |
1355 | static noinline int __sched | |
1356 | __ww_mutex_lock_interruptible_slowpath(struct ww_mutex *lock, | |
1357 | struct ww_acquire_ctx *ctx) | |
1358 | { | |
427b1820 PZ |
1359 | return __ww_mutex_lock(&lock->base, TASK_INTERRUPTIBLE, 0, NULL, |
1360 | _RET_IP_, ctx); | |
040a0a37 ML |
1361 | } |
1362 | ||
e4564f79 | 1363 | #endif |
6053ee3b | 1364 | |
ef5dc121 RD |
1365 | /** |
1366 | * mutex_trylock - try to acquire the mutex, without waiting | |
6053ee3b IM |
1367 | * @lock: the mutex to be acquired |
1368 | * | |
1369 | * Try to acquire the mutex atomically. Returns 1 if the mutex | |
1370 | * has been acquired successfully, and 0 on contention. | |
1371 | * | |
1372 | * NOTE: this function follows the spin_trylock() convention, so | |
ef5dc121 | 1373 | * it is negated from the down_trylock() return values! Be careful |
6053ee3b IM |
1374 | * about this when converting semaphore users to mutexes. |
1375 | * | |
1376 | * This function must not be used in interrupt context. The | |
1377 | * mutex must be released by the same task that acquired it. | |
1378 | */ | |
7ad5b3a5 | 1379 | int __sched mutex_trylock(struct mutex *lock) |
6053ee3b | 1380 | { |
e274795e | 1381 | bool locked = __mutex_trylock(lock); |
0d66bf6d | 1382 | |
3ca0ff57 PZ |
1383 | if (locked) |
1384 | mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_); | |
0d66bf6d | 1385 | |
3ca0ff57 | 1386 | return locked; |
6053ee3b | 1387 | } |
6053ee3b | 1388 | EXPORT_SYMBOL(mutex_trylock); |
a511e3f9 | 1389 | |
040a0a37 ML |
1390 | #ifndef CONFIG_DEBUG_LOCK_ALLOC |
1391 | int __sched | |
c5470b22 | 1392 | ww_mutex_lock(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
040a0a37 | 1393 | { |
040a0a37 ML |
1394 | might_sleep(); |
1395 | ||
3ca0ff57 | 1396 | if (__mutex_trylock_fast(&lock->base)) { |
ea9e0fb8 NH |
1397 | if (ctx) |
1398 | ww_mutex_set_context_fastpath(lock, ctx); | |
3ca0ff57 PZ |
1399 | return 0; |
1400 | } | |
1401 | ||
1402 | return __ww_mutex_lock_slowpath(lock, ctx); | |
040a0a37 | 1403 | } |
c5470b22 | 1404 | EXPORT_SYMBOL(ww_mutex_lock); |
040a0a37 ML |
1405 | |
1406 | int __sched | |
c5470b22 | 1407 | ww_mutex_lock_interruptible(struct ww_mutex *lock, struct ww_acquire_ctx *ctx) |
040a0a37 | 1408 | { |
040a0a37 ML |
1409 | might_sleep(); |
1410 | ||
3ca0ff57 | 1411 | if (__mutex_trylock_fast(&lock->base)) { |
ea9e0fb8 NH |
1412 | if (ctx) |
1413 | ww_mutex_set_context_fastpath(lock, ctx); | |
3ca0ff57 PZ |
1414 | return 0; |
1415 | } | |
1416 | ||
1417 | return __ww_mutex_lock_interruptible_slowpath(lock, ctx); | |
040a0a37 | 1418 | } |
c5470b22 | 1419 | EXPORT_SYMBOL(ww_mutex_lock_interruptible); |
040a0a37 ML |
1420 | |
1421 | #endif | |
1422 | ||
a511e3f9 AM |
1423 | /** |
1424 | * atomic_dec_and_mutex_lock - return holding mutex if we dec to 0 | |
1425 | * @cnt: the atomic which we are to dec | |
1426 | * @lock: the mutex to return holding if we dec to 0 | |
1427 | * | |
1428 | * return true and hold lock if we dec to 0, return false otherwise | |
1429 | */ | |
1430 | int atomic_dec_and_mutex_lock(atomic_t *cnt, struct mutex *lock) | |
1431 | { | |
1432 | /* dec if we can't possibly hit 0 */ | |
1433 | if (atomic_add_unless(cnt, -1, 1)) | |
1434 | return 0; | |
1435 | /* we might hit 0, so take the lock */ | |
1436 | mutex_lock(lock); | |
1437 | if (!atomic_dec_and_test(cnt)) { | |
1438 | /* when we actually did the dec, we didn't hit 0 */ | |
1439 | mutex_unlock(lock); | |
1440 | return 0; | |
1441 | } | |
1442 | /* we hit 0, and we hold the lock */ | |
1443 | return 1; | |
1444 | } | |
1445 | EXPORT_SYMBOL(atomic_dec_and_mutex_lock); |