]>
Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
23f78d4a IM |
2 | /* |
3 | * RT-Mutexes: simple blocking mutual exclusion locks with PI support | |
4 | * | |
5 | * started by Ingo Molnar and Thomas Gleixner. | |
6 | * | |
7 | * Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
8 | * Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com> | |
9 | * Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt | |
10 | * Copyright (C) 2006 Esben Nielsen | |
d07fe82c | 11 | * |
387b1468 | 12 | * See Documentation/locking/rt-mutex-design.rst for details. |
23f78d4a IM |
13 | */ |
14 | #include <linux/spinlock.h> | |
9984de1a | 15 | #include <linux/export.h> |
174cd4b1 | 16 | #include <linux/sched/signal.h> |
8bd75c77 | 17 | #include <linux/sched/rt.h> |
fb00aca4 | 18 | #include <linux/sched/deadline.h> |
84f001e1 | 19 | #include <linux/sched/wake_q.h> |
b17b0153 | 20 | #include <linux/sched/debug.h> |
23f78d4a IM |
21 | #include <linux/timer.h> |
22 | ||
23 | #include "rtmutex_common.h" | |
24 | ||
23f78d4a IM |
25 | /* |
26 | * lock->owner state tracking: | |
27 | * | |
8161239a LJ |
28 | * lock->owner holds the task_struct pointer of the owner. Bit 0 |
29 | * is used to keep track of the "lock has waiters" state. | |
23f78d4a | 30 | * |
8161239a LJ |
31 | * owner bit0 |
32 | * NULL 0 lock is free (fast acquire possible) | |
33 | * NULL 1 lock is free and has waiters and the top waiter | |
34 | * is going to take the lock* | |
35 | * taskpointer 0 lock is held (fast release possible) | |
36 | * taskpointer 1 lock is held and has waiters** | |
23f78d4a IM |
37 | * |
38 | * The fast atomic compare exchange based acquire and release is only | |
8161239a LJ |
39 | * possible when bit 0 of lock->owner is 0. |
40 | * | |
41 | * (*) It also can be a transitional state when grabbing the lock | |
42 | * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock, | |
43 | * we need to set the bit0 before looking at the lock, and the owner may be | |
44 | * NULL in this small time, hence this can be a transitional state. | |
23f78d4a | 45 | * |
8161239a LJ |
46 | * (**) There is a small time when bit 0 is set but there are no |
47 | * waiters. This can happen when grabbing the lock in the slow path. | |
48 | * To prevent a cmpxchg of the owner releasing the lock, we need to | |
49 | * set this bit before looking at the lock. | |
23f78d4a IM |
50 | */ |
51 | ||
bd197234 | 52 | static void |
8161239a | 53 | rt_mutex_set_owner(struct rt_mutex *lock, struct task_struct *owner) |
23f78d4a | 54 | { |
8161239a | 55 | unsigned long val = (unsigned long)owner; |
23f78d4a IM |
56 | |
57 | if (rt_mutex_has_waiters(lock)) | |
58 | val |= RT_MUTEX_HAS_WAITERS; | |
59 | ||
60 | lock->owner = (struct task_struct *)val; | |
61 | } | |
62 | ||
63 | static inline void clear_rt_mutex_waiters(struct rt_mutex *lock) | |
64 | { | |
65 | lock->owner = (struct task_struct *) | |
66 | ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS); | |
67 | } | |
68 | ||
69 | static void fixup_rt_mutex_waiters(struct rt_mutex *lock) | |
70 | { | |
dbb26055 TG |
71 | unsigned long owner, *p = (unsigned long *) &lock->owner; |
72 | ||
73 | if (rt_mutex_has_waiters(lock)) | |
74 | return; | |
75 | ||
76 | /* | |
77 | * The rbtree has no waiters enqueued, now make sure that the | |
78 | * lock->owner still has the waiters bit set, otherwise the | |
79 | * following can happen: | |
80 | * | |
81 | * CPU 0 CPU 1 CPU2 | |
82 | * l->owner=T1 | |
83 | * rt_mutex_lock(l) | |
84 | * lock(l->lock) | |
85 | * l->owner = T1 | HAS_WAITERS; | |
86 | * enqueue(T2) | |
87 | * boost() | |
88 | * unlock(l->lock) | |
89 | * block() | |
90 | * | |
91 | * rt_mutex_lock(l) | |
92 | * lock(l->lock) | |
93 | * l->owner = T1 | HAS_WAITERS; | |
94 | * enqueue(T3) | |
95 | * boost() | |
96 | * unlock(l->lock) | |
97 | * block() | |
98 | * signal(->T2) signal(->T3) | |
99 | * lock(l->lock) | |
100 | * dequeue(T2) | |
101 | * deboost() | |
102 | * unlock(l->lock) | |
103 | * lock(l->lock) | |
104 | * dequeue(T3) | |
105 | * ==> wait list is empty | |
106 | * deboost() | |
107 | * unlock(l->lock) | |
108 | * lock(l->lock) | |
109 | * fixup_rt_mutex_waiters() | |
110 | * if (wait_list_empty(l) { | |
111 | * l->owner = owner | |
112 | * owner = l->owner & ~HAS_WAITERS; | |
113 | * ==> l->owner = T1 | |
114 | * } | |
115 | * lock(l->lock) | |
116 | * rt_mutex_unlock(l) fixup_rt_mutex_waiters() | |
117 | * if (wait_list_empty(l) { | |
118 | * owner = l->owner & ~HAS_WAITERS; | |
119 | * cmpxchg(l->owner, T1, NULL) | |
120 | * ===> Success (l->owner = NULL) | |
121 | * | |
122 | * l->owner = owner | |
123 | * ==> l->owner = T1 | |
124 | * } | |
125 | * | |
126 | * With the check for the waiter bit in place T3 on CPU2 will not | |
127 | * overwrite. All tasks fiddling with the waiters bit are | |
128 | * serialized by l->lock, so nothing else can modify the waiters | |
129 | * bit. If the bit is set then nothing can change l->owner either | |
130 | * so the simple RMW is safe. The cmpxchg() will simply fail if it | |
131 | * happens in the middle of the RMW because the waiters bit is | |
132 | * still set. | |
133 | */ | |
134 | owner = READ_ONCE(*p); | |
135 | if (owner & RT_MUTEX_HAS_WAITERS) | |
136 | WRITE_ONCE(*p, owner & ~RT_MUTEX_HAS_WAITERS); | |
23f78d4a IM |
137 | } |
138 | ||
bd197234 | 139 | /* |
cede8841 SAS |
140 | * We can speed up the acquire/release, if there's no debugging state to be |
141 | * set up. | |
bd197234 | 142 | */ |
cede8841 | 143 | #ifndef CONFIG_DEBUG_RT_MUTEXES |
700318d1 DB |
144 | # define rt_mutex_cmpxchg_relaxed(l,c,n) (cmpxchg_relaxed(&l->owner, c, n) == c) |
145 | # define rt_mutex_cmpxchg_acquire(l,c,n) (cmpxchg_acquire(&l->owner, c, n) == c) | |
146 | # define rt_mutex_cmpxchg_release(l,c,n) (cmpxchg_release(&l->owner, c, n) == c) | |
147 | ||
148 | /* | |
149 | * Callers must hold the ->wait_lock -- which is the whole purpose as we force | |
150 | * all future threads that attempt to [Rmw] the lock to the slowpath. As such | |
151 | * relaxed semantics suffice. | |
152 | */ | |
bd197234 TG |
153 | static inline void mark_rt_mutex_waiters(struct rt_mutex *lock) |
154 | { | |
155 | unsigned long owner, *p = (unsigned long *) &lock->owner; | |
156 | ||
157 | do { | |
158 | owner = *p; | |
700318d1 DB |
159 | } while (cmpxchg_relaxed(p, owner, |
160 | owner | RT_MUTEX_HAS_WAITERS) != owner); | |
bd197234 | 161 | } |
27e35715 TG |
162 | |
163 | /* | |
164 | * Safe fastpath aware unlock: | |
165 | * 1) Clear the waiters bit | |
166 | * 2) Drop lock->wait_lock | |
167 | * 3) Try to unlock the lock with cmpxchg | |
168 | */ | |
b4abf910 TG |
169 | static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock, |
170 | unsigned long flags) | |
27e35715 TG |
171 | __releases(lock->wait_lock) |
172 | { | |
173 | struct task_struct *owner = rt_mutex_owner(lock); | |
174 | ||
175 | clear_rt_mutex_waiters(lock); | |
b4abf910 | 176 | raw_spin_unlock_irqrestore(&lock->wait_lock, flags); |
27e35715 TG |
177 | /* |
178 | * If a new waiter comes in between the unlock and the cmpxchg | |
179 | * we have two situations: | |
180 | * | |
181 | * unlock(wait_lock); | |
182 | * lock(wait_lock); | |
183 | * cmpxchg(p, owner, 0) == owner | |
184 | * mark_rt_mutex_waiters(lock); | |
185 | * acquire(lock); | |
186 | * or: | |
187 | * | |
188 | * unlock(wait_lock); | |
189 | * lock(wait_lock); | |
190 | * mark_rt_mutex_waiters(lock); | |
191 | * | |
192 | * cmpxchg(p, owner, 0) != owner | |
193 | * enqueue_waiter(); | |
194 | * unlock(wait_lock); | |
195 | * lock(wait_lock); | |
196 | * wake waiter(); | |
197 | * unlock(wait_lock); | |
198 | * lock(wait_lock); | |
199 | * acquire(lock); | |
200 | */ | |
700318d1 | 201 | return rt_mutex_cmpxchg_release(lock, owner, NULL); |
27e35715 TG |
202 | } |
203 | ||
bd197234 | 204 | #else |
700318d1 DB |
205 | # define rt_mutex_cmpxchg_relaxed(l,c,n) (0) |
206 | # define rt_mutex_cmpxchg_acquire(l,c,n) (0) | |
207 | # define rt_mutex_cmpxchg_release(l,c,n) (0) | |
208 | ||
bd197234 TG |
209 | static inline void mark_rt_mutex_waiters(struct rt_mutex *lock) |
210 | { | |
211 | lock->owner = (struct task_struct *) | |
212 | ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS); | |
213 | } | |
27e35715 TG |
214 | |
215 | /* | |
216 | * Simple slow path only version: lock->owner is protected by lock->wait_lock. | |
217 | */ | |
b4abf910 TG |
218 | static inline bool unlock_rt_mutex_safe(struct rt_mutex *lock, |
219 | unsigned long flags) | |
27e35715 TG |
220 | __releases(lock->wait_lock) |
221 | { | |
222 | lock->owner = NULL; | |
b4abf910 | 223 | raw_spin_unlock_irqrestore(&lock->wait_lock, flags); |
27e35715 TG |
224 | return true; |
225 | } | |
bd197234 TG |
226 | #endif |
227 | ||
19830e55 PZ |
228 | /* |
229 | * Only use with rt_mutex_waiter_{less,equal}() | |
230 | */ | |
231 | #define task_to_waiter(p) \ | |
232 | &(struct rt_mutex_waiter){ .prio = (p)->prio, .deadline = (p)->dl.deadline } | |
233 | ||
fb00aca4 PZ |
234 | static inline int |
235 | rt_mutex_waiter_less(struct rt_mutex_waiter *left, | |
236 | struct rt_mutex_waiter *right) | |
237 | { | |
2d3d891d | 238 | if (left->prio < right->prio) |
fb00aca4 PZ |
239 | return 1; |
240 | ||
241 | /* | |
2d3d891d DF |
242 | * If both waiters have dl_prio(), we check the deadlines of the |
243 | * associated tasks. | |
244 | * If left waiter has a dl_prio(), and we didn't return 1 above, | |
245 | * then right waiter has a dl_prio() too. | |
fb00aca4 | 246 | */ |
2d3d891d | 247 | if (dl_prio(left->prio)) |
e0aad5b4 | 248 | return dl_time_before(left->deadline, right->deadline); |
fb00aca4 PZ |
249 | |
250 | return 0; | |
251 | } | |
252 | ||
19830e55 PZ |
253 | static inline int |
254 | rt_mutex_waiter_equal(struct rt_mutex_waiter *left, | |
255 | struct rt_mutex_waiter *right) | |
256 | { | |
257 | if (left->prio != right->prio) | |
258 | return 0; | |
259 | ||
260 | /* | |
261 | * If both waiters have dl_prio(), we check the deadlines of the | |
262 | * associated tasks. | |
263 | * If left waiter has a dl_prio(), and we didn't return 0 above, | |
264 | * then right waiter has a dl_prio() too. | |
265 | */ | |
266 | if (dl_prio(left->prio)) | |
267 | return left->deadline == right->deadline; | |
268 | ||
269 | return 1; | |
270 | } | |
271 | ||
fb00aca4 PZ |
272 | static void |
273 | rt_mutex_enqueue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter) | |
274 | { | |
a23ba907 | 275 | struct rb_node **link = &lock->waiters.rb_root.rb_node; |
fb00aca4 PZ |
276 | struct rb_node *parent = NULL; |
277 | struct rt_mutex_waiter *entry; | |
a23ba907 | 278 | bool leftmost = true; |
fb00aca4 PZ |
279 | |
280 | while (*link) { | |
281 | parent = *link; | |
282 | entry = rb_entry(parent, struct rt_mutex_waiter, tree_entry); | |
283 | if (rt_mutex_waiter_less(waiter, entry)) { | |
284 | link = &parent->rb_left; | |
285 | } else { | |
286 | link = &parent->rb_right; | |
a23ba907 | 287 | leftmost = false; |
fb00aca4 PZ |
288 | } |
289 | } | |
290 | ||
fb00aca4 | 291 | rb_link_node(&waiter->tree_entry, parent, link); |
a23ba907 | 292 | rb_insert_color_cached(&waiter->tree_entry, &lock->waiters, leftmost); |
fb00aca4 PZ |
293 | } |
294 | ||
295 | static void | |
296 | rt_mutex_dequeue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter) | |
297 | { | |
298 | if (RB_EMPTY_NODE(&waiter->tree_entry)) | |
299 | return; | |
300 | ||
a23ba907 | 301 | rb_erase_cached(&waiter->tree_entry, &lock->waiters); |
fb00aca4 PZ |
302 | RB_CLEAR_NODE(&waiter->tree_entry); |
303 | } | |
304 | ||
305 | static void | |
306 | rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter) | |
307 | { | |
a23ba907 | 308 | struct rb_node **link = &task->pi_waiters.rb_root.rb_node; |
fb00aca4 PZ |
309 | struct rb_node *parent = NULL; |
310 | struct rt_mutex_waiter *entry; | |
a23ba907 | 311 | bool leftmost = true; |
fb00aca4 PZ |
312 | |
313 | while (*link) { | |
314 | parent = *link; | |
315 | entry = rb_entry(parent, struct rt_mutex_waiter, pi_tree_entry); | |
316 | if (rt_mutex_waiter_less(waiter, entry)) { | |
317 | link = &parent->rb_left; | |
318 | } else { | |
319 | link = &parent->rb_right; | |
a23ba907 | 320 | leftmost = false; |
fb00aca4 PZ |
321 | } |
322 | } | |
323 | ||
fb00aca4 | 324 | rb_link_node(&waiter->pi_tree_entry, parent, link); |
a23ba907 | 325 | rb_insert_color_cached(&waiter->pi_tree_entry, &task->pi_waiters, leftmost); |
fb00aca4 PZ |
326 | } |
327 | ||
328 | static void | |
329 | rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter) | |
330 | { | |
331 | if (RB_EMPTY_NODE(&waiter->pi_tree_entry)) | |
332 | return; | |
333 | ||
a23ba907 | 334 | rb_erase_cached(&waiter->pi_tree_entry, &task->pi_waiters); |
fb00aca4 PZ |
335 | RB_CLEAR_NODE(&waiter->pi_tree_entry); |
336 | } | |
337 | ||
acd58620 | 338 | static void rt_mutex_adjust_prio(struct task_struct *p) |
c365c292 | 339 | { |
acd58620 | 340 | struct task_struct *pi_task = NULL; |
e96a7705 | 341 | |
acd58620 | 342 | lockdep_assert_held(&p->pi_lock); |
c365c292 | 343 | |
acd58620 PZ |
344 | if (task_has_pi_waiters(p)) |
345 | pi_task = task_top_pi_waiter(p)->task; | |
c365c292 | 346 | |
acd58620 | 347 | rt_mutex_setprio(p, pi_task); |
23f78d4a IM |
348 | } |
349 | ||
8930ed80 TG |
350 | /* |
351 | * Deadlock detection is conditional: | |
352 | * | |
353 | * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted | |
354 | * if the detect argument is == RT_MUTEX_FULL_CHAINWALK. | |
355 | * | |
356 | * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always | |
357 | * conducted independent of the detect argument. | |
358 | * | |
359 | * If the waiter argument is NULL this indicates the deboost path and | |
360 | * deadlock detection is disabled independent of the detect argument | |
361 | * and the config settings. | |
362 | */ | |
363 | static bool rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter, | |
364 | enum rtmutex_chainwalk chwalk) | |
365 | { | |
366 | /* | |
367 | * This is just a wrapper function for the following call, | |
368 | * because debug_rt_mutex_detect_deadlock() smells like a magic | |
369 | * debug feature and I wanted to keep the cond function in the | |
370 | * main source file along with the comments instead of having | |
371 | * two of the same in the headers. | |
372 | */ | |
373 | return debug_rt_mutex_detect_deadlock(waiter, chwalk); | |
374 | } | |
375 | ||
23f78d4a IM |
376 | /* |
377 | * Max number of times we'll walk the boosting chain: | |
378 | */ | |
379 | int max_lock_depth = 1024; | |
380 | ||
82084984 TG |
381 | static inline struct rt_mutex *task_blocked_on_lock(struct task_struct *p) |
382 | { | |
383 | return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL; | |
384 | } | |
385 | ||
23f78d4a IM |
386 | /* |
387 | * Adjust the priority chain. Also used for deadlock detection. | |
388 | * Decreases task's usage by one - may thus free the task. | |
0c106173 | 389 | * |
82084984 TG |
390 | * @task: the task owning the mutex (owner) for which a chain walk is |
391 | * probably needed | |
e6beaa36 | 392 | * @chwalk: do we have to carry out deadlock detection? |
82084984 TG |
393 | * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck |
394 | * things for a task that has just got its priority adjusted, and | |
395 | * is waiting on a mutex) | |
396 | * @next_lock: the mutex on which the owner of @orig_lock was blocked before | |
397 | * we dropped its pi_lock. Is never dereferenced, only used for | |
398 | * comparison to detect lock chain changes. | |
0c106173 | 399 | * @orig_waiter: rt_mutex_waiter struct for the task that has just donated |
82084984 TG |
400 | * its priority to the mutex owner (can be NULL in the case |
401 | * depicted above or if the top waiter is gone away and we are | |
402 | * actually deboosting the owner) | |
403 | * @top_task: the current top waiter | |
0c106173 | 404 | * |
23f78d4a | 405 | * Returns 0 or -EDEADLK. |
3eb65aea TG |
406 | * |
407 | * Chain walk basics and protection scope | |
408 | * | |
409 | * [R] refcount on task | |
410 | * [P] task->pi_lock held | |
411 | * [L] rtmutex->wait_lock held | |
412 | * | |
413 | * Step Description Protected by | |
414 | * function arguments: | |
415 | * @task [R] | |
416 | * @orig_lock if != NULL @top_task is blocked on it | |
417 | * @next_lock Unprotected. Cannot be | |
418 | * dereferenced. Only used for | |
419 | * comparison. | |
420 | * @orig_waiter if != NULL @top_task is blocked on it | |
421 | * @top_task current, or in case of proxy | |
422 | * locking protected by calling | |
423 | * code | |
424 | * again: | |
425 | * loop_sanity_check(); | |
426 | * retry: | |
427 | * [1] lock(task->pi_lock); [R] acquire [P] | |
428 | * [2] waiter = task->pi_blocked_on; [P] | |
429 | * [3] check_exit_conditions_1(); [P] | |
430 | * [4] lock = waiter->lock; [P] | |
431 | * [5] if (!try_lock(lock->wait_lock)) { [P] try to acquire [L] | |
432 | * unlock(task->pi_lock); release [P] | |
433 | * goto retry; | |
434 | * } | |
435 | * [6] check_exit_conditions_2(); [P] + [L] | |
436 | * [7] requeue_lock_waiter(lock, waiter); [P] + [L] | |
437 | * [8] unlock(task->pi_lock); release [P] | |
438 | * put_task_struct(task); release [R] | |
439 | * [9] check_exit_conditions_3(); [L] | |
440 | * [10] task = owner(lock); [L] | |
441 | * get_task_struct(task); [L] acquire [R] | |
442 | * lock(task->pi_lock); [L] acquire [P] | |
443 | * [11] requeue_pi_waiter(tsk, waiters(lock));[P] + [L] | |
444 | * [12] check_exit_conditions_4(); [P] + [L] | |
445 | * [13] unlock(task->pi_lock); release [P] | |
446 | * unlock(lock->wait_lock); release [L] | |
447 | * goto again; | |
23f78d4a | 448 | */ |
bd197234 | 449 | static int rt_mutex_adjust_prio_chain(struct task_struct *task, |
8930ed80 | 450 | enum rtmutex_chainwalk chwalk, |
bd197234 | 451 | struct rt_mutex *orig_lock, |
82084984 | 452 | struct rt_mutex *next_lock, |
bd197234 TG |
453 | struct rt_mutex_waiter *orig_waiter, |
454 | struct task_struct *top_task) | |
23f78d4a | 455 | { |
23f78d4a | 456 | struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter; |
a57594a1 | 457 | struct rt_mutex_waiter *prerequeue_top_waiter; |
8930ed80 | 458 | int ret = 0, depth = 0; |
a57594a1 | 459 | struct rt_mutex *lock; |
8930ed80 | 460 | bool detect_deadlock; |
67792e2c | 461 | bool requeue = true; |
23f78d4a | 462 | |
8930ed80 | 463 | detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk); |
23f78d4a IM |
464 | |
465 | /* | |
466 | * The (de)boosting is a step by step approach with a lot of | |
467 | * pitfalls. We want this to be preemptible and we want hold a | |
468 | * maximum of two locks per step. So we have to check | |
469 | * carefully whether things change under us. | |
470 | */ | |
471 | again: | |
3eb65aea TG |
472 | /* |
473 | * We limit the lock chain length for each invocation. | |
474 | */ | |
23f78d4a IM |
475 | if (++depth > max_lock_depth) { |
476 | static int prev_max; | |
477 | ||
478 | /* | |
479 | * Print this only once. If the admin changes the limit, | |
480 | * print a new message when reaching the limit again. | |
481 | */ | |
482 | if (prev_max != max_lock_depth) { | |
483 | prev_max = max_lock_depth; | |
484 | printk(KERN_WARNING "Maximum lock depth %d reached " | |
485 | "task: %s (%d)\n", max_lock_depth, | |
ba25f9dc | 486 | top_task->comm, task_pid_nr(top_task)); |
23f78d4a IM |
487 | } |
488 | put_task_struct(task); | |
489 | ||
3d5c9340 | 490 | return -EDEADLK; |
23f78d4a | 491 | } |
3eb65aea TG |
492 | |
493 | /* | |
494 | * We are fully preemptible here and only hold the refcount on | |
495 | * @task. So everything can have changed under us since the | |
496 | * caller or our own code below (goto retry/again) dropped all | |
497 | * locks. | |
498 | */ | |
23f78d4a IM |
499 | retry: |
500 | /* | |
3eb65aea | 501 | * [1] Task cannot go away as we did a get_task() before ! |
23f78d4a | 502 | */ |
b4abf910 | 503 | raw_spin_lock_irq(&task->pi_lock); |
23f78d4a | 504 | |
3eb65aea TG |
505 | /* |
506 | * [2] Get the waiter on which @task is blocked on. | |
507 | */ | |
23f78d4a | 508 | waiter = task->pi_blocked_on; |
3eb65aea TG |
509 | |
510 | /* | |
511 | * [3] check_exit_conditions_1() protected by task->pi_lock. | |
512 | */ | |
513 | ||
23f78d4a IM |
514 | /* |
515 | * Check whether the end of the boosting chain has been | |
516 | * reached or the state of the chain has changed while we | |
517 | * dropped the locks. | |
518 | */ | |
8161239a | 519 | if (!waiter) |
23f78d4a IM |
520 | goto out_unlock_pi; |
521 | ||
1a539a87 TG |
522 | /* |
523 | * Check the orig_waiter state. After we dropped the locks, | |
8161239a | 524 | * the previous owner of the lock might have released the lock. |
1a539a87 | 525 | */ |
8161239a | 526 | if (orig_waiter && !rt_mutex_owner(orig_lock)) |
1a539a87 TG |
527 | goto out_unlock_pi; |
528 | ||
82084984 TG |
529 | /* |
530 | * We dropped all locks after taking a refcount on @task, so | |
531 | * the task might have moved on in the lock chain or even left | |
532 | * the chain completely and blocks now on an unrelated lock or | |
533 | * on @orig_lock. | |
534 | * | |
535 | * We stored the lock on which @task was blocked in @next_lock, | |
536 | * so we can detect the chain change. | |
537 | */ | |
538 | if (next_lock != waiter->lock) | |
539 | goto out_unlock_pi; | |
540 | ||
1a539a87 TG |
541 | /* |
542 | * Drop out, when the task has no waiters. Note, | |
543 | * top_waiter can be NULL, when we are in the deboosting | |
544 | * mode! | |
545 | */ | |
397335f0 TG |
546 | if (top_waiter) { |
547 | if (!task_has_pi_waiters(task)) | |
548 | goto out_unlock_pi; | |
549 | /* | |
550 | * If deadlock detection is off, we stop here if we | |
67792e2c TG |
551 | * are not the top pi waiter of the task. If deadlock |
552 | * detection is enabled we continue, but stop the | |
553 | * requeueing in the chain walk. | |
397335f0 | 554 | */ |
67792e2c TG |
555 | if (top_waiter != task_top_pi_waiter(task)) { |
556 | if (!detect_deadlock) | |
557 | goto out_unlock_pi; | |
558 | else | |
559 | requeue = false; | |
560 | } | |
397335f0 | 561 | } |
23f78d4a IM |
562 | |
563 | /* | |
67792e2c TG |
564 | * If the waiter priority is the same as the task priority |
565 | * then there is no further priority adjustment necessary. If | |
566 | * deadlock detection is off, we stop the chain walk. If its | |
567 | * enabled we continue, but stop the requeueing in the chain | |
568 | * walk. | |
23f78d4a | 569 | */ |
19830e55 | 570 | if (rt_mutex_waiter_equal(waiter, task_to_waiter(task))) { |
67792e2c TG |
571 | if (!detect_deadlock) |
572 | goto out_unlock_pi; | |
573 | else | |
574 | requeue = false; | |
575 | } | |
23f78d4a | 576 | |
3eb65aea TG |
577 | /* |
578 | * [4] Get the next lock | |
579 | */ | |
23f78d4a | 580 | lock = waiter->lock; |
3eb65aea TG |
581 | /* |
582 | * [5] We need to trylock here as we are holding task->pi_lock, | |
583 | * which is the reverse lock order versus the other rtmutex | |
584 | * operations. | |
585 | */ | |
d209d74d | 586 | if (!raw_spin_trylock(&lock->wait_lock)) { |
b4abf910 | 587 | raw_spin_unlock_irq(&task->pi_lock); |
23f78d4a IM |
588 | cpu_relax(); |
589 | goto retry; | |
590 | } | |
591 | ||
397335f0 | 592 | /* |
3eb65aea TG |
593 | * [6] check_exit_conditions_2() protected by task->pi_lock and |
594 | * lock->wait_lock. | |
595 | * | |
397335f0 TG |
596 | * Deadlock detection. If the lock is the same as the original |
597 | * lock which caused us to walk the lock chain or if the | |
598 | * current lock is owned by the task which initiated the chain | |
599 | * walk, we detected a deadlock. | |
600 | */ | |
95e02ca9 | 601 | if (lock == orig_lock || rt_mutex_owner(lock) == top_task) { |
8930ed80 | 602 | debug_rt_mutex_deadlock(chwalk, orig_waiter, lock); |
d209d74d | 603 | raw_spin_unlock(&lock->wait_lock); |
3d5c9340 | 604 | ret = -EDEADLK; |
23f78d4a IM |
605 | goto out_unlock_pi; |
606 | } | |
607 | ||
67792e2c TG |
608 | /* |
609 | * If we just follow the lock chain for deadlock detection, no | |
610 | * need to do all the requeue operations. To avoid a truckload | |
611 | * of conditionals around the various places below, just do the | |
612 | * minimum chain walk checks. | |
613 | */ | |
614 | if (!requeue) { | |
615 | /* | |
616 | * No requeue[7] here. Just release @task [8] | |
617 | */ | |
b4abf910 | 618 | raw_spin_unlock(&task->pi_lock); |
67792e2c TG |
619 | put_task_struct(task); |
620 | ||
621 | /* | |
622 | * [9] check_exit_conditions_3 protected by lock->wait_lock. | |
623 | * If there is no owner of the lock, end of chain. | |
624 | */ | |
625 | if (!rt_mutex_owner(lock)) { | |
b4abf910 | 626 | raw_spin_unlock_irq(&lock->wait_lock); |
67792e2c TG |
627 | return 0; |
628 | } | |
629 | ||
630 | /* [10] Grab the next task, i.e. owner of @lock */ | |
7b3c92b8 | 631 | task = get_task_struct(rt_mutex_owner(lock)); |
b4abf910 | 632 | raw_spin_lock(&task->pi_lock); |
67792e2c TG |
633 | |
634 | /* | |
635 | * No requeue [11] here. We just do deadlock detection. | |
636 | * | |
637 | * [12] Store whether owner is blocked | |
638 | * itself. Decision is made after dropping the locks | |
639 | */ | |
640 | next_lock = task_blocked_on_lock(task); | |
641 | /* | |
642 | * Get the top waiter for the next iteration | |
643 | */ | |
644 | top_waiter = rt_mutex_top_waiter(lock); | |
645 | ||
646 | /* [13] Drop locks */ | |
b4abf910 TG |
647 | raw_spin_unlock(&task->pi_lock); |
648 | raw_spin_unlock_irq(&lock->wait_lock); | |
67792e2c TG |
649 | |
650 | /* If owner is not blocked, end of chain. */ | |
651 | if (!next_lock) | |
652 | goto out_put_task; | |
653 | goto again; | |
654 | } | |
655 | ||
a57594a1 TG |
656 | /* |
657 | * Store the current top waiter before doing the requeue | |
658 | * operation on @lock. We need it for the boost/deboost | |
659 | * decision below. | |
660 | */ | |
661 | prerequeue_top_waiter = rt_mutex_top_waiter(lock); | |
23f78d4a | 662 | |
9f40a51a | 663 | /* [7] Requeue the waiter in the lock waiter tree. */ |
fb00aca4 | 664 | rt_mutex_dequeue(lock, waiter); |
e0aad5b4 PZ |
665 | |
666 | /* | |
667 | * Update the waiter prio fields now that we're dequeued. | |
668 | * | |
669 | * These values can have changed through either: | |
670 | * | |
671 | * sys_sched_set_scheduler() / sys_sched_setattr() | |
672 | * | |
673 | * or | |
674 | * | |
675 | * DL CBS enforcement advancing the effective deadline. | |
676 | * | |
677 | * Even though pi_waiters also uses these fields, and that tree is only | |
678 | * updated in [11], we can do this here, since we hold [L], which | |
679 | * serializes all pi_waiters access and rb_erase() does not care about | |
680 | * the values of the node being removed. | |
681 | */ | |
2d3d891d | 682 | waiter->prio = task->prio; |
e0aad5b4 PZ |
683 | waiter->deadline = task->dl.deadline; |
684 | ||
fb00aca4 | 685 | rt_mutex_enqueue(lock, waiter); |
23f78d4a | 686 | |
3eb65aea | 687 | /* [8] Release the task */ |
b4abf910 | 688 | raw_spin_unlock(&task->pi_lock); |
2ffa5a5c TG |
689 | put_task_struct(task); |
690 | ||
a57594a1 | 691 | /* |
3eb65aea TG |
692 | * [9] check_exit_conditions_3 protected by lock->wait_lock. |
693 | * | |
a57594a1 TG |
694 | * We must abort the chain walk if there is no lock owner even |
695 | * in the dead lock detection case, as we have nothing to | |
696 | * follow here. This is the end of the chain we are walking. | |
697 | */ | |
8161239a LJ |
698 | if (!rt_mutex_owner(lock)) { |
699 | /* | |
3eb65aea TG |
700 | * If the requeue [7] above changed the top waiter, |
701 | * then we need to wake the new top waiter up to try | |
702 | * to get the lock. | |
8161239a | 703 | */ |
a57594a1 | 704 | if (prerequeue_top_waiter != rt_mutex_top_waiter(lock)) |
8161239a | 705 | wake_up_process(rt_mutex_top_waiter(lock)->task); |
b4abf910 | 706 | raw_spin_unlock_irq(&lock->wait_lock); |
2ffa5a5c | 707 | return 0; |
8161239a | 708 | } |
23f78d4a | 709 | |
3eb65aea | 710 | /* [10] Grab the next task, i.e. the owner of @lock */ |
7b3c92b8 | 711 | task = get_task_struct(rt_mutex_owner(lock)); |
b4abf910 | 712 | raw_spin_lock(&task->pi_lock); |
23f78d4a | 713 | |
3eb65aea | 714 | /* [11] requeue the pi waiters if necessary */ |
23f78d4a | 715 | if (waiter == rt_mutex_top_waiter(lock)) { |
a57594a1 TG |
716 | /* |
717 | * The waiter became the new top (highest priority) | |
718 | * waiter on the lock. Replace the previous top waiter | |
9f40a51a | 719 | * in the owner tasks pi waiters tree with this waiter |
a57594a1 TG |
720 | * and adjust the priority of the owner. |
721 | */ | |
722 | rt_mutex_dequeue_pi(task, prerequeue_top_waiter); | |
fb00aca4 | 723 | rt_mutex_enqueue_pi(task, waiter); |
acd58620 | 724 | rt_mutex_adjust_prio(task); |
23f78d4a | 725 | |
a57594a1 TG |
726 | } else if (prerequeue_top_waiter == waiter) { |
727 | /* | |
728 | * The waiter was the top waiter on the lock, but is | |
729 | * no longer the top prority waiter. Replace waiter in | |
9f40a51a | 730 | * the owner tasks pi waiters tree with the new top |
a57594a1 TG |
731 | * (highest priority) waiter and adjust the priority |
732 | * of the owner. | |
733 | * The new top waiter is stored in @waiter so that | |
734 | * @waiter == @top_waiter evaluates to true below and | |
735 | * we continue to deboost the rest of the chain. | |
736 | */ | |
fb00aca4 | 737 | rt_mutex_dequeue_pi(task, waiter); |
23f78d4a | 738 | waiter = rt_mutex_top_waiter(lock); |
fb00aca4 | 739 | rt_mutex_enqueue_pi(task, waiter); |
acd58620 | 740 | rt_mutex_adjust_prio(task); |
a57594a1 TG |
741 | } else { |
742 | /* | |
743 | * Nothing changed. No need to do any priority | |
744 | * adjustment. | |
745 | */ | |
23f78d4a IM |
746 | } |
747 | ||
82084984 | 748 | /* |
3eb65aea TG |
749 | * [12] check_exit_conditions_4() protected by task->pi_lock |
750 | * and lock->wait_lock. The actual decisions are made after we | |
751 | * dropped the locks. | |
752 | * | |
82084984 TG |
753 | * Check whether the task which owns the current lock is pi |
754 | * blocked itself. If yes we store a pointer to the lock for | |
755 | * the lock chain change detection above. After we dropped | |
756 | * task->pi_lock next_lock cannot be dereferenced anymore. | |
757 | */ | |
758 | next_lock = task_blocked_on_lock(task); | |
a57594a1 TG |
759 | /* |
760 | * Store the top waiter of @lock for the end of chain walk | |
761 | * decision below. | |
762 | */ | |
23f78d4a | 763 | top_waiter = rt_mutex_top_waiter(lock); |
3eb65aea TG |
764 | |
765 | /* [13] Drop the locks */ | |
b4abf910 TG |
766 | raw_spin_unlock(&task->pi_lock); |
767 | raw_spin_unlock_irq(&lock->wait_lock); | |
23f78d4a | 768 | |
82084984 | 769 | /* |
3eb65aea TG |
770 | * Make the actual exit decisions [12], based on the stored |
771 | * values. | |
772 | * | |
82084984 TG |
773 | * We reached the end of the lock chain. Stop right here. No |
774 | * point to go back just to figure that out. | |
775 | */ | |
776 | if (!next_lock) | |
777 | goto out_put_task; | |
778 | ||
a57594a1 TG |
779 | /* |
780 | * If the current waiter is not the top waiter on the lock, | |
781 | * then we can stop the chain walk here if we are not in full | |
782 | * deadlock detection mode. | |
783 | */ | |
23f78d4a IM |
784 | if (!detect_deadlock && waiter != top_waiter) |
785 | goto out_put_task; | |
786 | ||
787 | goto again; | |
788 | ||
789 | out_unlock_pi: | |
b4abf910 | 790 | raw_spin_unlock_irq(&task->pi_lock); |
23f78d4a IM |
791 | out_put_task: |
792 | put_task_struct(task); | |
36c8b586 | 793 | |
23f78d4a IM |
794 | return ret; |
795 | } | |
796 | ||
23f78d4a IM |
797 | /* |
798 | * Try to take an rt-mutex | |
799 | * | |
b4abf910 | 800 | * Must be called with lock->wait_lock held and interrupts disabled |
8161239a | 801 | * |
358c331f TG |
802 | * @lock: The lock to be acquired. |
803 | * @task: The task which wants to acquire the lock | |
9f40a51a | 804 | * @waiter: The waiter that is queued to the lock's wait tree if the |
358c331f | 805 | * callsite called task_blocked_on_lock(), otherwise NULL |
23f78d4a | 806 | */ |
8161239a | 807 | static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task, |
358c331f | 808 | struct rt_mutex_waiter *waiter) |
23f78d4a | 809 | { |
e0aad5b4 PZ |
810 | lockdep_assert_held(&lock->wait_lock); |
811 | ||
23f78d4a | 812 | /* |
358c331f TG |
813 | * Before testing whether we can acquire @lock, we set the |
814 | * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all | |
815 | * other tasks which try to modify @lock into the slow path | |
816 | * and they serialize on @lock->wait_lock. | |
23f78d4a | 817 | * |
358c331f TG |
818 | * The RT_MUTEX_HAS_WAITERS bit can have a transitional state |
819 | * as explained at the top of this file if and only if: | |
23f78d4a | 820 | * |
358c331f TG |
821 | * - There is a lock owner. The caller must fixup the |
822 | * transient state if it does a trylock or leaves the lock | |
823 | * function due to a signal or timeout. | |
824 | * | |
825 | * - @task acquires the lock and there are no other | |
826 | * waiters. This is undone in rt_mutex_set_owner(@task) at | |
827 | * the end of this function. | |
23f78d4a IM |
828 | */ |
829 | mark_rt_mutex_waiters(lock); | |
830 | ||
358c331f TG |
831 | /* |
832 | * If @lock has an owner, give up. | |
833 | */ | |
8161239a | 834 | if (rt_mutex_owner(lock)) |
23f78d4a IM |
835 | return 0; |
836 | ||
8161239a | 837 | /* |
358c331f | 838 | * If @waiter != NULL, @task has already enqueued the waiter |
9f40a51a | 839 | * into @lock waiter tree. If @waiter == NULL then this is a |
358c331f | 840 | * trylock attempt. |
8161239a | 841 | */ |
358c331f TG |
842 | if (waiter) { |
843 | /* | |
844 | * If waiter is not the highest priority waiter of | |
845 | * @lock, give up. | |
846 | */ | |
847 | if (waiter != rt_mutex_top_waiter(lock)) | |
848 | return 0; | |
8161239a | 849 | |
358c331f TG |
850 | /* |
851 | * We can acquire the lock. Remove the waiter from the | |
9f40a51a | 852 | * lock waiters tree. |
358c331f TG |
853 | */ |
854 | rt_mutex_dequeue(lock, waiter); | |
8161239a | 855 | |
358c331f | 856 | } else { |
8161239a | 857 | /* |
358c331f TG |
858 | * If the lock has waiters already we check whether @task is |
859 | * eligible to take over the lock. | |
860 | * | |
861 | * If there are no other waiters, @task can acquire | |
862 | * the lock. @task->pi_blocked_on is NULL, so it does | |
863 | * not need to be dequeued. | |
8161239a LJ |
864 | */ |
865 | if (rt_mutex_has_waiters(lock)) { | |
358c331f TG |
866 | /* |
867 | * If @task->prio is greater than or equal to | |
868 | * the top waiter priority (kernel view), | |
869 | * @task lost. | |
870 | */ | |
19830e55 PZ |
871 | if (!rt_mutex_waiter_less(task_to_waiter(task), |
872 | rt_mutex_top_waiter(lock))) | |
358c331f TG |
873 | return 0; |
874 | ||
875 | /* | |
876 | * The current top waiter stays enqueued. We | |
877 | * don't have to change anything in the lock | |
878 | * waiters order. | |
879 | */ | |
880 | } else { | |
881 | /* | |
882 | * No waiters. Take the lock without the | |
883 | * pi_lock dance.@task->pi_blocked_on is NULL | |
884 | * and we have no waiters to enqueue in @task | |
9f40a51a | 885 | * pi waiters tree. |
358c331f TG |
886 | */ |
887 | goto takeit; | |
8161239a | 888 | } |
8161239a LJ |
889 | } |
890 | ||
358c331f TG |
891 | /* |
892 | * Clear @task->pi_blocked_on. Requires protection by | |
893 | * @task->pi_lock. Redundant operation for the @waiter == NULL | |
894 | * case, but conditionals are more expensive than a redundant | |
895 | * store. | |
896 | */ | |
b4abf910 | 897 | raw_spin_lock(&task->pi_lock); |
358c331f TG |
898 | task->pi_blocked_on = NULL; |
899 | /* | |
900 | * Finish the lock acquisition. @task is the new owner. If | |
901 | * other waiters exist we have to insert the highest priority | |
9f40a51a | 902 | * waiter into @task->pi_waiters tree. |
358c331f TG |
903 | */ |
904 | if (rt_mutex_has_waiters(lock)) | |
905 | rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock)); | |
b4abf910 | 906 | raw_spin_unlock(&task->pi_lock); |
358c331f TG |
907 | |
908 | takeit: | |
23f78d4a | 909 | /* We got the lock. */ |
9a11b49a | 910 | debug_rt_mutex_lock(lock); |
23f78d4a | 911 | |
358c331f TG |
912 | /* |
913 | * This either preserves the RT_MUTEX_HAS_WAITERS bit if there | |
914 | * are still waiters or clears it. | |
915 | */ | |
8161239a | 916 | rt_mutex_set_owner(lock, task); |
23f78d4a | 917 | |
23f78d4a IM |
918 | return 1; |
919 | } | |
920 | ||
921 | /* | |
922 | * Task blocks on lock. | |
923 | * | |
924 | * Prepare waiter and propagate pi chain | |
925 | * | |
b4abf910 | 926 | * This must be called with lock->wait_lock held and interrupts disabled |
23f78d4a IM |
927 | */ |
928 | static int task_blocks_on_rt_mutex(struct rt_mutex *lock, | |
929 | struct rt_mutex_waiter *waiter, | |
8dac456a | 930 | struct task_struct *task, |
8930ed80 | 931 | enum rtmutex_chainwalk chwalk) |
23f78d4a | 932 | { |
36c8b586 | 933 | struct task_struct *owner = rt_mutex_owner(lock); |
23f78d4a | 934 | struct rt_mutex_waiter *top_waiter = waiter; |
82084984 | 935 | struct rt_mutex *next_lock; |
db630637 | 936 | int chain_walk = 0, res; |
23f78d4a | 937 | |
e0aad5b4 PZ |
938 | lockdep_assert_held(&lock->wait_lock); |
939 | ||
397335f0 TG |
940 | /* |
941 | * Early deadlock detection. We really don't want the task to | |
942 | * enqueue on itself just to untangle the mess later. It's not | |
943 | * only an optimization. We drop the locks, so another waiter | |
944 | * can come in before the chain walk detects the deadlock. So | |
945 | * the other will detect the deadlock and return -EDEADLOCK, | |
946 | * which is wrong, as the other waiter is not in a deadlock | |
947 | * situation. | |
948 | */ | |
3d5c9340 | 949 | if (owner == task) |
397335f0 TG |
950 | return -EDEADLK; |
951 | ||
b4abf910 | 952 | raw_spin_lock(&task->pi_lock); |
8dac456a | 953 | waiter->task = task; |
23f78d4a | 954 | waiter->lock = lock; |
2d3d891d | 955 | waiter->prio = task->prio; |
e0aad5b4 | 956 | waiter->deadline = task->dl.deadline; |
23f78d4a IM |
957 | |
958 | /* Get the top priority waiter on the lock */ | |
959 | if (rt_mutex_has_waiters(lock)) | |
960 | top_waiter = rt_mutex_top_waiter(lock); | |
fb00aca4 | 961 | rt_mutex_enqueue(lock, waiter); |
23f78d4a | 962 | |
8dac456a | 963 | task->pi_blocked_on = waiter; |
23f78d4a | 964 | |
b4abf910 | 965 | raw_spin_unlock(&task->pi_lock); |
23f78d4a | 966 | |
8161239a LJ |
967 | if (!owner) |
968 | return 0; | |
969 | ||
b4abf910 | 970 | raw_spin_lock(&owner->pi_lock); |
23f78d4a | 971 | if (waiter == rt_mutex_top_waiter(lock)) { |
fb00aca4 PZ |
972 | rt_mutex_dequeue_pi(owner, top_waiter); |
973 | rt_mutex_enqueue_pi(owner, waiter); | |
23f78d4a | 974 | |
acd58620 | 975 | rt_mutex_adjust_prio(owner); |
db630637 SR |
976 | if (owner->pi_blocked_on) |
977 | chain_walk = 1; | |
8930ed80 | 978 | } else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) { |
db630637 | 979 | chain_walk = 1; |
82084984 | 980 | } |
db630637 | 981 | |
82084984 TG |
982 | /* Store the lock on which owner is blocked or NULL */ |
983 | next_lock = task_blocked_on_lock(owner); | |
984 | ||
b4abf910 | 985 | raw_spin_unlock(&owner->pi_lock); |
82084984 TG |
986 | /* |
987 | * Even if full deadlock detection is on, if the owner is not | |
988 | * blocked itself, we can avoid finding this out in the chain | |
989 | * walk. | |
990 | */ | |
991 | if (!chain_walk || !next_lock) | |
23f78d4a IM |
992 | return 0; |
993 | ||
db630637 SR |
994 | /* |
995 | * The owner can't disappear while holding a lock, | |
996 | * so the owner struct is protected by wait_lock. | |
997 | * Gets dropped in rt_mutex_adjust_prio_chain()! | |
998 | */ | |
999 | get_task_struct(owner); | |
1000 | ||
b4abf910 | 1001 | raw_spin_unlock_irq(&lock->wait_lock); |
23f78d4a | 1002 | |
8930ed80 | 1003 | res = rt_mutex_adjust_prio_chain(owner, chwalk, lock, |
82084984 | 1004 | next_lock, waiter, task); |
23f78d4a | 1005 | |
b4abf910 | 1006 | raw_spin_lock_irq(&lock->wait_lock); |
23f78d4a IM |
1007 | |
1008 | return res; | |
1009 | } | |
1010 | ||
1011 | /* | |
9f40a51a | 1012 | * Remove the top waiter from the current tasks pi waiter tree and |
45ab4eff | 1013 | * queue it up. |
23f78d4a | 1014 | * |
b4abf910 | 1015 | * Called with lock->wait_lock held and interrupts disabled. |
23f78d4a | 1016 | */ |
45ab4eff DB |
1017 | static void mark_wakeup_next_waiter(struct wake_q_head *wake_q, |
1018 | struct rt_mutex *lock) | |
23f78d4a IM |
1019 | { |
1020 | struct rt_mutex_waiter *waiter; | |
23f78d4a | 1021 | |
b4abf910 | 1022 | raw_spin_lock(¤t->pi_lock); |
23f78d4a IM |
1023 | |
1024 | waiter = rt_mutex_top_waiter(lock); | |
23f78d4a IM |
1025 | |
1026 | /* | |
acd58620 PZ |
1027 | * Remove it from current->pi_waiters and deboost. |
1028 | * | |
1029 | * We must in fact deboost here in order to ensure we call | |
1030 | * rt_mutex_setprio() to update p->pi_top_task before the | |
1031 | * task unblocks. | |
23f78d4a | 1032 | */ |
fb00aca4 | 1033 | rt_mutex_dequeue_pi(current, waiter); |
acd58620 | 1034 | rt_mutex_adjust_prio(current); |
23f78d4a | 1035 | |
27e35715 TG |
1036 | /* |
1037 | * As we are waking up the top waiter, and the waiter stays | |
1038 | * queued on the lock until it gets the lock, this lock | |
1039 | * obviously has waiters. Just set the bit here and this has | |
1040 | * the added benefit of forcing all new tasks into the | |
1041 | * slow path making sure no task of lower priority than | |
1042 | * the top waiter can steal this lock. | |
1043 | */ | |
1044 | lock->owner = (void *) RT_MUTEX_HAS_WAITERS; | |
23f78d4a | 1045 | |
acd58620 PZ |
1046 | /* |
1047 | * We deboosted before waking the top waiter task such that we don't | |
1048 | * run two tasks with the 'same' priority (and ensure the | |
1049 | * p->pi_top_task pointer points to a blocked task). This however can | |
1050 | * lead to priority inversion if we would get preempted after the | |
1051 | * deboost but before waking our donor task, hence the preempt_disable() | |
1052 | * before unlock. | |
1053 | * | |
1054 | * Pairs with preempt_enable() in rt_mutex_postunlock(); | |
1055 | */ | |
1056 | preempt_disable(); | |
45ab4eff | 1057 | wake_q_add(wake_q, waiter->task); |
acd58620 | 1058 | raw_spin_unlock(¤t->pi_lock); |
23f78d4a IM |
1059 | } |
1060 | ||
1061 | /* | |
8161239a | 1062 | * Remove a waiter from a lock and give up |
23f78d4a | 1063 | * |
b4abf910 | 1064 | * Must be called with lock->wait_lock held and interrupts disabled. I must |
8161239a | 1065 | * have just failed to try_to_take_rt_mutex(). |
23f78d4a | 1066 | */ |
bd197234 TG |
1067 | static void remove_waiter(struct rt_mutex *lock, |
1068 | struct rt_mutex_waiter *waiter) | |
23f78d4a | 1069 | { |
1ca7b860 | 1070 | bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock)); |
36c8b586 | 1071 | struct task_struct *owner = rt_mutex_owner(lock); |
1ca7b860 | 1072 | struct rt_mutex *next_lock; |
23f78d4a | 1073 | |
e0aad5b4 PZ |
1074 | lockdep_assert_held(&lock->wait_lock); |
1075 | ||
b4abf910 | 1076 | raw_spin_lock(¤t->pi_lock); |
fb00aca4 | 1077 | rt_mutex_dequeue(lock, waiter); |
23f78d4a | 1078 | current->pi_blocked_on = NULL; |
b4abf910 | 1079 | raw_spin_unlock(¤t->pi_lock); |
23f78d4a | 1080 | |
1ca7b860 TG |
1081 | /* |
1082 | * Only update priority if the waiter was the highest priority | |
1083 | * waiter of the lock and there is an owner to update. | |
1084 | */ | |
1085 | if (!owner || !is_top_waiter) | |
8161239a LJ |
1086 | return; |
1087 | ||
b4abf910 | 1088 | raw_spin_lock(&owner->pi_lock); |
23f78d4a | 1089 | |
1ca7b860 | 1090 | rt_mutex_dequeue_pi(owner, waiter); |
23f78d4a | 1091 | |
1ca7b860 TG |
1092 | if (rt_mutex_has_waiters(lock)) |
1093 | rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock)); | |
23f78d4a | 1094 | |
acd58620 | 1095 | rt_mutex_adjust_prio(owner); |
23f78d4a | 1096 | |
1ca7b860 TG |
1097 | /* Store the lock on which owner is blocked or NULL */ |
1098 | next_lock = task_blocked_on_lock(owner); | |
db630637 | 1099 | |
b4abf910 | 1100 | raw_spin_unlock(&owner->pi_lock); |
23f78d4a | 1101 | |
1ca7b860 TG |
1102 | /* |
1103 | * Don't walk the chain, if the owner task is not blocked | |
1104 | * itself. | |
1105 | */ | |
82084984 | 1106 | if (!next_lock) |
23f78d4a IM |
1107 | return; |
1108 | ||
db630637 SR |
1109 | /* gets dropped in rt_mutex_adjust_prio_chain()! */ |
1110 | get_task_struct(owner); | |
1111 | ||
b4abf910 | 1112 | raw_spin_unlock_irq(&lock->wait_lock); |
23f78d4a | 1113 | |
8930ed80 TG |
1114 | rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock, |
1115 | next_lock, NULL, current); | |
23f78d4a | 1116 | |
b4abf910 | 1117 | raw_spin_lock_irq(&lock->wait_lock); |
23f78d4a IM |
1118 | } |
1119 | ||
95e02ca9 TG |
1120 | /* |
1121 | * Recheck the pi chain, in case we got a priority setting | |
1122 | * | |
1123 | * Called from sched_setscheduler | |
1124 | */ | |
1125 | void rt_mutex_adjust_pi(struct task_struct *task) | |
1126 | { | |
1127 | struct rt_mutex_waiter *waiter; | |
82084984 | 1128 | struct rt_mutex *next_lock; |
95e02ca9 TG |
1129 | unsigned long flags; |
1130 | ||
1d615482 | 1131 | raw_spin_lock_irqsave(&task->pi_lock, flags); |
95e02ca9 TG |
1132 | |
1133 | waiter = task->pi_blocked_on; | |
19830e55 | 1134 | if (!waiter || rt_mutex_waiter_equal(waiter, task_to_waiter(task))) { |
1d615482 | 1135 | raw_spin_unlock_irqrestore(&task->pi_lock, flags); |
95e02ca9 TG |
1136 | return; |
1137 | } | |
82084984 | 1138 | next_lock = waiter->lock; |
1d615482 | 1139 | raw_spin_unlock_irqrestore(&task->pi_lock, flags); |
95e02ca9 | 1140 | |
db630637 SR |
1141 | /* gets dropped in rt_mutex_adjust_prio_chain()! */ |
1142 | get_task_struct(task); | |
82084984 | 1143 | |
8930ed80 TG |
1144 | rt_mutex_adjust_prio_chain(task, RT_MUTEX_MIN_CHAINWALK, NULL, |
1145 | next_lock, NULL, task); | |
95e02ca9 TG |
1146 | } |
1147 | ||
50809358 PZ |
1148 | void rt_mutex_init_waiter(struct rt_mutex_waiter *waiter) |
1149 | { | |
1150 | debug_rt_mutex_init_waiter(waiter); | |
1151 | RB_CLEAR_NODE(&waiter->pi_tree_entry); | |
1152 | RB_CLEAR_NODE(&waiter->tree_entry); | |
1153 | waiter->task = NULL; | |
1154 | } | |
1155 | ||
8dac456a DH |
1156 | /** |
1157 | * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop | |
1158 | * @lock: the rt_mutex to take | |
1159 | * @state: the state the task should block in (TASK_INTERRUPTIBLE | |
b4abf910 | 1160 | * or TASK_UNINTERRUPTIBLE) |
8dac456a DH |
1161 | * @timeout: the pre-initialized and started timer, or NULL for none |
1162 | * @waiter: the pre-initialized rt_mutex_waiter | |
8dac456a | 1163 | * |
b4abf910 | 1164 | * Must be called with lock->wait_lock held and interrupts disabled |
23f78d4a IM |
1165 | */ |
1166 | static int __sched | |
8dac456a DH |
1167 | __rt_mutex_slowlock(struct rt_mutex *lock, int state, |
1168 | struct hrtimer_sleeper *timeout, | |
8161239a | 1169 | struct rt_mutex_waiter *waiter) |
23f78d4a | 1170 | { |
23f78d4a IM |
1171 | int ret = 0; |
1172 | ||
23f78d4a IM |
1173 | for (;;) { |
1174 | /* Try to acquire the lock: */ | |
8161239a | 1175 | if (try_to_take_rt_mutex(lock, current, waiter)) |
23f78d4a IM |
1176 | break; |
1177 | ||
1178 | /* | |
1179 | * TASK_INTERRUPTIBLE checks for signals and | |
1180 | * timeout. Ignored otherwise. | |
1181 | */ | |
4009f4b3 | 1182 | if (likely(state == TASK_INTERRUPTIBLE)) { |
23f78d4a IM |
1183 | /* Signal pending? */ |
1184 | if (signal_pending(current)) | |
1185 | ret = -EINTR; | |
1186 | if (timeout && !timeout->task) | |
1187 | ret = -ETIMEDOUT; | |
1188 | if (ret) | |
1189 | break; | |
1190 | } | |
1191 | ||
b4abf910 | 1192 | raw_spin_unlock_irq(&lock->wait_lock); |
23f78d4a | 1193 | |
8dac456a | 1194 | debug_rt_mutex_print_deadlock(waiter); |
23f78d4a | 1195 | |
1b0b7c17 | 1196 | schedule(); |
23f78d4a | 1197 | |
b4abf910 | 1198 | raw_spin_lock_irq(&lock->wait_lock); |
23f78d4a IM |
1199 | set_current_state(state); |
1200 | } | |
1201 | ||
afffc6c1 | 1202 | __set_current_state(TASK_RUNNING); |
8dac456a DH |
1203 | return ret; |
1204 | } | |
1205 | ||
3d5c9340 TG |
1206 | static void rt_mutex_handle_deadlock(int res, int detect_deadlock, |
1207 | struct rt_mutex_waiter *w) | |
1208 | { | |
1209 | /* | |
1210 | * If the result is not -EDEADLOCK or the caller requested | |
1211 | * deadlock detection, nothing to do here. | |
1212 | */ | |
1213 | if (res != -EDEADLOCK || detect_deadlock) | |
1214 | return; | |
1215 | ||
1216 | /* | |
1217 | * Yell lowdly and stop the task right here. | |
1218 | */ | |
1219 | rt_mutex_print_deadlock(w); | |
1220 | while (1) { | |
1221 | set_current_state(TASK_INTERRUPTIBLE); | |
1222 | schedule(); | |
1223 | } | |
1224 | } | |
1225 | ||
8dac456a DH |
1226 | /* |
1227 | * Slow path lock function: | |
1228 | */ | |
1229 | static int __sched | |
1230 | rt_mutex_slowlock(struct rt_mutex *lock, int state, | |
1231 | struct hrtimer_sleeper *timeout, | |
8930ed80 | 1232 | enum rtmutex_chainwalk chwalk) |
8dac456a DH |
1233 | { |
1234 | struct rt_mutex_waiter waiter; | |
b4abf910 | 1235 | unsigned long flags; |
8dac456a DH |
1236 | int ret = 0; |
1237 | ||
50809358 | 1238 | rt_mutex_init_waiter(&waiter); |
8dac456a | 1239 | |
b4abf910 TG |
1240 | /* |
1241 | * Technically we could use raw_spin_[un]lock_irq() here, but this can | |
1242 | * be called in early boot if the cmpxchg() fast path is disabled | |
1243 | * (debug, no architecture support). In this case we will acquire the | |
1244 | * rtmutex with lock->wait_lock held. But we cannot unconditionally | |
1245 | * enable interrupts in that early boot case. So we need to use the | |
1246 | * irqsave/restore variants. | |
1247 | */ | |
1248 | raw_spin_lock_irqsave(&lock->wait_lock, flags); | |
8dac456a DH |
1249 | |
1250 | /* Try to acquire the lock again: */ | |
8161239a | 1251 | if (try_to_take_rt_mutex(lock, current, NULL)) { |
b4abf910 | 1252 | raw_spin_unlock_irqrestore(&lock->wait_lock, flags); |
8dac456a DH |
1253 | return 0; |
1254 | } | |
1255 | ||
1256 | set_current_state(state); | |
1257 | ||
1258 | /* Setup the timer, when timeout != NULL */ | |
ccdd92c1 | 1259 | if (unlikely(timeout)) |
8dac456a | 1260 | hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS); |
8dac456a | 1261 | |
8930ed80 | 1262 | ret = task_blocks_on_rt_mutex(lock, &waiter, current, chwalk); |
8161239a LJ |
1263 | |
1264 | if (likely(!ret)) | |
afffc6c1 | 1265 | /* sleep on the mutex */ |
8161239a | 1266 | ret = __rt_mutex_slowlock(lock, state, timeout, &waiter); |
8dac456a | 1267 | |
3d5c9340 | 1268 | if (unlikely(ret)) { |
9d3e2d02 | 1269 | __set_current_state(TASK_RUNNING); |
c28d62cf | 1270 | remove_waiter(lock, &waiter); |
8930ed80 | 1271 | rt_mutex_handle_deadlock(ret, chwalk, &waiter); |
3d5c9340 | 1272 | } |
23f78d4a IM |
1273 | |
1274 | /* | |
1275 | * try_to_take_rt_mutex() sets the waiter bit | |
1276 | * unconditionally. We might have to fix that up. | |
1277 | */ | |
1278 | fixup_rt_mutex_waiters(lock); | |
1279 | ||
b4abf910 | 1280 | raw_spin_unlock_irqrestore(&lock->wait_lock, flags); |
23f78d4a IM |
1281 | |
1282 | /* Remove pending timer: */ | |
1283 | if (unlikely(timeout)) | |
1284 | hrtimer_cancel(&timeout->timer); | |
1285 | ||
23f78d4a IM |
1286 | debug_rt_mutex_free_waiter(&waiter); |
1287 | ||
1288 | return ret; | |
1289 | } | |
1290 | ||
c1e2f0ea PZ |
1291 | static inline int __rt_mutex_slowtrylock(struct rt_mutex *lock) |
1292 | { | |
1293 | int ret = try_to_take_rt_mutex(lock, current, NULL); | |
1294 | ||
1295 | /* | |
1296 | * try_to_take_rt_mutex() sets the lock waiters bit | |
1297 | * unconditionally. Clean this up. | |
1298 | */ | |
1299 | fixup_rt_mutex_waiters(lock); | |
1300 | ||
1301 | return ret; | |
1302 | } | |
1303 | ||
23f78d4a IM |
1304 | /* |
1305 | * Slow path try-lock function: | |
1306 | */ | |
88f2b4c1 | 1307 | static inline int rt_mutex_slowtrylock(struct rt_mutex *lock) |
23f78d4a | 1308 | { |
b4abf910 | 1309 | unsigned long flags; |
88f2b4c1 TG |
1310 | int ret; |
1311 | ||
1312 | /* | |
1313 | * If the lock already has an owner we fail to get the lock. | |
1314 | * This can be done without taking the @lock->wait_lock as | |
1315 | * it is only being read, and this is a trylock anyway. | |
1316 | */ | |
1317 | if (rt_mutex_owner(lock)) | |
1318 | return 0; | |
23f78d4a | 1319 | |
88f2b4c1 | 1320 | /* |
b4abf910 TG |
1321 | * The mutex has currently no owner. Lock the wait lock and try to |
1322 | * acquire the lock. We use irqsave here to support early boot calls. | |
88f2b4c1 | 1323 | */ |
b4abf910 | 1324 | raw_spin_lock_irqsave(&lock->wait_lock, flags); |
23f78d4a | 1325 | |
c1e2f0ea | 1326 | ret = __rt_mutex_slowtrylock(lock); |
23f78d4a | 1327 | |
b4abf910 | 1328 | raw_spin_unlock_irqrestore(&lock->wait_lock, flags); |
23f78d4a IM |
1329 | |
1330 | return ret; | |
1331 | } | |
1332 | ||
1333 | /* | |
802ab58d | 1334 | * Slow path to release a rt-mutex. |
aa2bfe55 PZ |
1335 | * |
1336 | * Return whether the current task needs to call rt_mutex_postunlock(). | |
23f78d4a | 1337 | */ |
802ab58d SAS |
1338 | static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock, |
1339 | struct wake_q_head *wake_q) | |
23f78d4a | 1340 | { |
b4abf910 TG |
1341 | unsigned long flags; |
1342 | ||
1343 | /* irqsave required to support early boot calls */ | |
1344 | raw_spin_lock_irqsave(&lock->wait_lock, flags); | |
23f78d4a IM |
1345 | |
1346 | debug_rt_mutex_unlock(lock); | |
1347 | ||
27e35715 TG |
1348 | /* |
1349 | * We must be careful here if the fast path is enabled. If we | |
1350 | * have no waiters queued we cannot set owner to NULL here | |
1351 | * because of: | |
1352 | * | |
1353 | * foo->lock->owner = NULL; | |
1354 | * rtmutex_lock(foo->lock); <- fast path | |
1355 | * free = atomic_dec_and_test(foo->refcnt); | |
1356 | * rtmutex_unlock(foo->lock); <- fast path | |
1357 | * if (free) | |
1358 | * kfree(foo); | |
1359 | * raw_spin_unlock(foo->lock->wait_lock); | |
1360 | * | |
1361 | * So for the fastpath enabled kernel: | |
1362 | * | |
1363 | * Nothing can set the waiters bit as long as we hold | |
1364 | * lock->wait_lock. So we do the following sequence: | |
1365 | * | |
1366 | * owner = rt_mutex_owner(lock); | |
1367 | * clear_rt_mutex_waiters(lock); | |
1368 | * raw_spin_unlock(&lock->wait_lock); | |
1369 | * if (cmpxchg(&lock->owner, owner, 0) == owner) | |
1370 | * return; | |
1371 | * goto retry; | |
1372 | * | |
1373 | * The fastpath disabled variant is simple as all access to | |
1374 | * lock->owner is serialized by lock->wait_lock: | |
1375 | * | |
1376 | * lock->owner = NULL; | |
1377 | * raw_spin_unlock(&lock->wait_lock); | |
1378 | */ | |
1379 | while (!rt_mutex_has_waiters(lock)) { | |
1380 | /* Drops lock->wait_lock ! */ | |
b4abf910 | 1381 | if (unlock_rt_mutex_safe(lock, flags) == true) |
802ab58d | 1382 | return false; |
27e35715 | 1383 | /* Relock the rtmutex and try again */ |
b4abf910 | 1384 | raw_spin_lock_irqsave(&lock->wait_lock, flags); |
23f78d4a IM |
1385 | } |
1386 | ||
27e35715 TG |
1387 | /* |
1388 | * The wakeup next waiter path does not suffer from the above | |
1389 | * race. See the comments there. | |
45ab4eff DB |
1390 | * |
1391 | * Queue the next waiter for wakeup once we release the wait_lock. | |
27e35715 | 1392 | */ |
802ab58d | 1393 | mark_wakeup_next_waiter(wake_q, lock); |
b4abf910 | 1394 | raw_spin_unlock_irqrestore(&lock->wait_lock, flags); |
23f78d4a | 1395 | |
aa2bfe55 | 1396 | return true; /* call rt_mutex_postunlock() */ |
23f78d4a IM |
1397 | } |
1398 | ||
1399 | /* | |
1400 | * debug aware fast / slowpath lock,trylock,unlock | |
1401 | * | |
1402 | * The atomic acquire/release ops are compiled away, when either the | |
1403 | * architecture does not support cmpxchg or when debugging is enabled. | |
1404 | */ | |
1405 | static inline int | |
1406 | rt_mutex_fastlock(struct rt_mutex *lock, int state, | |
23f78d4a IM |
1407 | int (*slowfn)(struct rt_mutex *lock, int state, |
1408 | struct hrtimer_sleeper *timeout, | |
8930ed80 | 1409 | enum rtmutex_chainwalk chwalk)) |
23f78d4a | 1410 | { |
fffa954f | 1411 | if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current))) |
23f78d4a | 1412 | return 0; |
fffa954f PZ |
1413 | |
1414 | return slowfn(lock, state, NULL, RT_MUTEX_MIN_CHAINWALK); | |
23f78d4a IM |
1415 | } |
1416 | ||
1417 | static inline int | |
1418 | rt_mutex_timed_fastlock(struct rt_mutex *lock, int state, | |
8930ed80 TG |
1419 | struct hrtimer_sleeper *timeout, |
1420 | enum rtmutex_chainwalk chwalk, | |
23f78d4a IM |
1421 | int (*slowfn)(struct rt_mutex *lock, int state, |
1422 | struct hrtimer_sleeper *timeout, | |
8930ed80 | 1423 | enum rtmutex_chainwalk chwalk)) |
23f78d4a | 1424 | { |
8930ed80 | 1425 | if (chwalk == RT_MUTEX_MIN_CHAINWALK && |
fffa954f | 1426 | likely(rt_mutex_cmpxchg_acquire(lock, NULL, current))) |
23f78d4a | 1427 | return 0; |
fffa954f PZ |
1428 | |
1429 | return slowfn(lock, state, timeout, chwalk); | |
23f78d4a IM |
1430 | } |
1431 | ||
1432 | static inline int | |
1433 | rt_mutex_fasttrylock(struct rt_mutex *lock, | |
9a11b49a | 1434 | int (*slowfn)(struct rt_mutex *lock)) |
23f78d4a | 1435 | { |
fffa954f | 1436 | if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current))) |
23f78d4a | 1437 | return 1; |
fffa954f | 1438 | |
9a11b49a | 1439 | return slowfn(lock); |
23f78d4a IM |
1440 | } |
1441 | ||
2a1c6029 | 1442 | /* |
aa2bfe55 | 1443 | * Performs the wakeup of the the top-waiter and re-enables preemption. |
2a1c6029 | 1444 | */ |
aa2bfe55 | 1445 | void rt_mutex_postunlock(struct wake_q_head *wake_q) |
2a1c6029 XP |
1446 | { |
1447 | wake_up_q(wake_q); | |
1448 | ||
1449 | /* Pairs with preempt_disable() in rt_mutex_slowunlock() */ | |
aa2bfe55 | 1450 | preempt_enable(); |
2a1c6029 XP |
1451 | } |
1452 | ||
23f78d4a IM |
1453 | static inline void |
1454 | rt_mutex_fastunlock(struct rt_mutex *lock, | |
802ab58d SAS |
1455 | bool (*slowfn)(struct rt_mutex *lock, |
1456 | struct wake_q_head *wqh)) | |
23f78d4a | 1457 | { |
194a6b5b | 1458 | DEFINE_WAKE_Q(wake_q); |
802ab58d | 1459 | |
fffa954f PZ |
1460 | if (likely(rt_mutex_cmpxchg_release(lock, current, NULL))) |
1461 | return; | |
802ab58d | 1462 | |
aa2bfe55 PZ |
1463 | if (slowfn(lock, &wake_q)) |
1464 | rt_mutex_postunlock(&wake_q); | |
23f78d4a IM |
1465 | } |
1466 | ||
62cedf3e PR |
1467 | static inline void __rt_mutex_lock(struct rt_mutex *lock, unsigned int subclass) |
1468 | { | |
1469 | might_sleep(); | |
1470 | ||
1471 | mutex_acquire(&lock->dep_map, subclass, 0, _RET_IP_); | |
1472 | rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, rt_mutex_slowlock); | |
1473 | } | |
1474 | ||
1475 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
1476 | /** | |
1477 | * rt_mutex_lock_nested - lock a rt_mutex | |
1478 | * | |
1479 | * @lock: the rt_mutex to be locked | |
1480 | * @subclass: the lockdep subclass | |
1481 | */ | |
1482 | void __sched rt_mutex_lock_nested(struct rt_mutex *lock, unsigned int subclass) | |
1483 | { | |
1484 | __rt_mutex_lock(lock, subclass); | |
1485 | } | |
1486 | EXPORT_SYMBOL_GPL(rt_mutex_lock_nested); | |
62cedf3e | 1487 | |
84818af2 SRV |
1488 | #else /* !CONFIG_DEBUG_LOCK_ALLOC */ |
1489 | ||
23f78d4a IM |
1490 | /** |
1491 | * rt_mutex_lock - lock a rt_mutex | |
1492 | * | |
1493 | * @lock: the rt_mutex to be locked | |
1494 | */ | |
1495 | void __sched rt_mutex_lock(struct rt_mutex *lock) | |
1496 | { | |
62cedf3e | 1497 | __rt_mutex_lock(lock, 0); |
23f78d4a IM |
1498 | } |
1499 | EXPORT_SYMBOL_GPL(rt_mutex_lock); | |
62cedf3e | 1500 | #endif |
23f78d4a IM |
1501 | |
1502 | /** | |
1503 | * rt_mutex_lock_interruptible - lock a rt_mutex interruptible | |
1504 | * | |
c051b21f | 1505 | * @lock: the rt_mutex to be locked |
23f78d4a IM |
1506 | * |
1507 | * Returns: | |
c051b21f TG |
1508 | * 0 on success |
1509 | * -EINTR when interrupted by a signal | |
23f78d4a | 1510 | */ |
c051b21f | 1511 | int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock) |
23f78d4a | 1512 | { |
f5694788 PZ |
1513 | int ret; |
1514 | ||
23f78d4a IM |
1515 | might_sleep(); |
1516 | ||
f5694788 PZ |
1517 | mutex_acquire(&lock->dep_map, 0, 0, _RET_IP_); |
1518 | ret = rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE, rt_mutex_slowlock); | |
1519 | if (ret) | |
5facae4f | 1520 | mutex_release(&lock->dep_map, _RET_IP_); |
f5694788 PZ |
1521 | |
1522 | return ret; | |
23f78d4a IM |
1523 | } |
1524 | EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible); | |
1525 | ||
5293c2ef PZ |
1526 | /* |
1527 | * Futex variant, must not use fastpath. | |
1528 | */ | |
1529 | int __sched rt_mutex_futex_trylock(struct rt_mutex *lock) | |
1530 | { | |
1531 | return rt_mutex_slowtrylock(lock); | |
c051b21f TG |
1532 | } |
1533 | ||
c1e2f0ea PZ |
1534 | int __sched __rt_mutex_futex_trylock(struct rt_mutex *lock) |
1535 | { | |
1536 | return __rt_mutex_slowtrylock(lock); | |
1537 | } | |
1538 | ||
23f78d4a | 1539 | /** |
23b94b96 LH |
1540 | * rt_mutex_timed_lock - lock a rt_mutex interruptible |
1541 | * the timeout structure is provided | |
1542 | * by the caller | |
23f78d4a | 1543 | * |
c051b21f | 1544 | * @lock: the rt_mutex to be locked |
23f78d4a | 1545 | * @timeout: timeout structure or NULL (no timeout) |
23f78d4a IM |
1546 | * |
1547 | * Returns: | |
c051b21f TG |
1548 | * 0 on success |
1549 | * -EINTR when interrupted by a signal | |
3ac49a1c | 1550 | * -ETIMEDOUT when the timeout expired |
23f78d4a IM |
1551 | */ |
1552 | int | |
c051b21f | 1553 | rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout) |
23f78d4a | 1554 | { |
f5694788 PZ |
1555 | int ret; |
1556 | ||
23f78d4a IM |
1557 | might_sleep(); |
1558 | ||
f5694788 PZ |
1559 | mutex_acquire(&lock->dep_map, 0, 0, _RET_IP_); |
1560 | ret = rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout, | |
8930ed80 | 1561 | RT_MUTEX_MIN_CHAINWALK, |
c051b21f | 1562 | rt_mutex_slowlock); |
f5694788 | 1563 | if (ret) |
5facae4f | 1564 | mutex_release(&lock->dep_map, _RET_IP_); |
f5694788 PZ |
1565 | |
1566 | return ret; | |
23f78d4a IM |
1567 | } |
1568 | EXPORT_SYMBOL_GPL(rt_mutex_timed_lock); | |
1569 | ||
1570 | /** | |
1571 | * rt_mutex_trylock - try to lock a rt_mutex | |
1572 | * | |
1573 | * @lock: the rt_mutex to be locked | |
1574 | * | |
6ce47fd9 TG |
1575 | * This function can only be called in thread context. It's safe to |
1576 | * call it from atomic regions, but not from hard interrupt or soft | |
1577 | * interrupt context. | |
1578 | * | |
23f78d4a IM |
1579 | * Returns 1 on success and 0 on contention |
1580 | */ | |
1581 | int __sched rt_mutex_trylock(struct rt_mutex *lock) | |
1582 | { | |
f5694788 PZ |
1583 | int ret; |
1584 | ||
a461d587 | 1585 | if (WARN_ON_ONCE(in_irq() || in_nmi() || in_serving_softirq())) |
6ce47fd9 TG |
1586 | return 0; |
1587 | ||
f5694788 PZ |
1588 | ret = rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock); |
1589 | if (ret) | |
1590 | mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_); | |
1591 | ||
1592 | return ret; | |
23f78d4a IM |
1593 | } |
1594 | EXPORT_SYMBOL_GPL(rt_mutex_trylock); | |
1595 | ||
1596 | /** | |
1597 | * rt_mutex_unlock - unlock a rt_mutex | |
1598 | * | |
1599 | * @lock: the rt_mutex to be unlocked | |
1600 | */ | |
1601 | void __sched rt_mutex_unlock(struct rt_mutex *lock) | |
1602 | { | |
5facae4f | 1603 | mutex_release(&lock->dep_map, _RET_IP_); |
23f78d4a IM |
1604 | rt_mutex_fastunlock(lock, rt_mutex_slowunlock); |
1605 | } | |
1606 | EXPORT_SYMBOL_GPL(rt_mutex_unlock); | |
1607 | ||
802ab58d | 1608 | /** |
5293c2ef PZ |
1609 | * Futex variant, that since futex variants do not use the fast-path, can be |
1610 | * simple and will not need to retry. | |
802ab58d | 1611 | */ |
5293c2ef PZ |
1612 | bool __sched __rt_mutex_futex_unlock(struct rt_mutex *lock, |
1613 | struct wake_q_head *wake_q) | |
802ab58d | 1614 | { |
5293c2ef PZ |
1615 | lockdep_assert_held(&lock->wait_lock); |
1616 | ||
1617 | debug_rt_mutex_unlock(lock); | |
1618 | ||
1619 | if (!rt_mutex_has_waiters(lock)) { | |
1620 | lock->owner = NULL; | |
1621 | return false; /* done */ | |
1622 | } | |
1623 | ||
2a1c6029 | 1624 | /* |
def34eaa MG |
1625 | * We've already deboosted, mark_wakeup_next_waiter() will |
1626 | * retain preempt_disabled when we drop the wait_lock, to | |
1627 | * avoid inversion prior to the wakeup. preempt_disable() | |
1628 | * therein pairs with rt_mutex_postunlock(). | |
2a1c6029 | 1629 | */ |
def34eaa | 1630 | mark_wakeup_next_waiter(wake_q, lock); |
2a1c6029 | 1631 | |
aa2bfe55 | 1632 | return true; /* call postunlock() */ |
5293c2ef | 1633 | } |
fffa954f | 1634 | |
5293c2ef PZ |
1635 | void __sched rt_mutex_futex_unlock(struct rt_mutex *lock) |
1636 | { | |
1637 | DEFINE_WAKE_Q(wake_q); | |
6b0ef92f | 1638 | unsigned long flags; |
aa2bfe55 | 1639 | bool postunlock; |
5293c2ef | 1640 | |
6b0ef92f | 1641 | raw_spin_lock_irqsave(&lock->wait_lock, flags); |
aa2bfe55 | 1642 | postunlock = __rt_mutex_futex_unlock(lock, &wake_q); |
6b0ef92f | 1643 | raw_spin_unlock_irqrestore(&lock->wait_lock, flags); |
5293c2ef | 1644 | |
aa2bfe55 PZ |
1645 | if (postunlock) |
1646 | rt_mutex_postunlock(&wake_q); | |
802ab58d SAS |
1647 | } |
1648 | ||
23b94b96 | 1649 | /** |
23f78d4a IM |
1650 | * rt_mutex_destroy - mark a mutex unusable |
1651 | * @lock: the mutex to be destroyed | |
1652 | * | |
1653 | * This function marks the mutex uninitialized, and any subsequent | |
1654 | * use of the mutex is forbidden. The mutex must not be locked when | |
1655 | * this function is called. | |
1656 | */ | |
1657 | void rt_mutex_destroy(struct rt_mutex *lock) | |
1658 | { | |
1659 | WARN_ON(rt_mutex_is_locked(lock)); | |
1660 | #ifdef CONFIG_DEBUG_RT_MUTEXES | |
1661 | lock->magic = NULL; | |
1662 | #endif | |
1663 | } | |
23f78d4a IM |
1664 | EXPORT_SYMBOL_GPL(rt_mutex_destroy); |
1665 | ||
1666 | /** | |
1667 | * __rt_mutex_init - initialize the rt lock | |
1668 | * | |
1669 | * @lock: the rt lock to be initialized | |
1670 | * | |
1671 | * Initialize the rt lock to unlocked state. | |
1672 | * | |
1673 | * Initializing of a locked rt lock is not allowed | |
1674 | */ | |
f5694788 PZ |
1675 | void __rt_mutex_init(struct rt_mutex *lock, const char *name, |
1676 | struct lock_class_key *key) | |
23f78d4a IM |
1677 | { |
1678 | lock->owner = NULL; | |
d209d74d | 1679 | raw_spin_lock_init(&lock->wait_lock); |
a23ba907 | 1680 | lock->waiters = RB_ROOT_CACHED; |
23f78d4a | 1681 | |
cde50a67 LASL |
1682 | if (name && key) |
1683 | debug_rt_mutex_init(lock, name, key); | |
23f78d4a IM |
1684 | } |
1685 | EXPORT_SYMBOL_GPL(__rt_mutex_init); | |
0cdbee99 IM |
1686 | |
1687 | /** | |
1688 | * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a | |
1689 | * proxy owner | |
1690 | * | |
84d82ec5 | 1691 | * @lock: the rt_mutex to be locked |
0cdbee99 IM |
1692 | * @proxy_owner:the task to set as owner |
1693 | * | |
1694 | * No locking. Caller has to do serializing itself | |
84d82ec5 TG |
1695 | * |
1696 | * Special API call for PI-futex support. This initializes the rtmutex and | |
1697 | * assigns it to @proxy_owner. Concurrent operations on the rtmutex are not | |
1698 | * possible at this point because the pi_state which contains the rtmutex | |
1699 | * is not yet visible to other tasks. | |
0cdbee99 IM |
1700 | */ |
1701 | void rt_mutex_init_proxy_locked(struct rt_mutex *lock, | |
1702 | struct task_struct *proxy_owner) | |
1703 | { | |
f5694788 | 1704 | __rt_mutex_init(lock, NULL, NULL); |
9a11b49a | 1705 | debug_rt_mutex_proxy_lock(lock, proxy_owner); |
8161239a | 1706 | rt_mutex_set_owner(lock, proxy_owner); |
0cdbee99 IM |
1707 | } |
1708 | ||
1709 | /** | |
1710 | * rt_mutex_proxy_unlock - release a lock on behalf of owner | |
1711 | * | |
84d82ec5 | 1712 | * @lock: the rt_mutex to be locked |
0cdbee99 IM |
1713 | * |
1714 | * No locking. Caller has to do serializing itself | |
84d82ec5 TG |
1715 | * |
1716 | * Special API call for PI-futex support. This merrily cleans up the rtmutex | |
1717 | * (debugging) state. Concurrent operations on this rt_mutex are not | |
1718 | * possible because it belongs to the pi_state which is about to be freed | |
1719 | * and it is not longer visible to other tasks. | |
0cdbee99 IM |
1720 | */ |
1721 | void rt_mutex_proxy_unlock(struct rt_mutex *lock, | |
1722 | struct task_struct *proxy_owner) | |
1723 | { | |
1724 | debug_rt_mutex_proxy_unlock(lock); | |
8161239a | 1725 | rt_mutex_set_owner(lock, NULL); |
0cdbee99 IM |
1726 | } |
1727 | ||
1a1fb985 TG |
1728 | /** |
1729 | * __rt_mutex_start_proxy_lock() - Start lock acquisition for another task | |
1730 | * @lock: the rt_mutex to take | |
1731 | * @waiter: the pre-initialized rt_mutex_waiter | |
1732 | * @task: the task to prepare | |
1733 | * | |
1734 | * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock | |
1735 | * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that. | |
1736 | * | |
1737 | * NOTE: does _NOT_ remove the @waiter on failure; must either call | |
1738 | * rt_mutex_wait_proxy_lock() or rt_mutex_cleanup_proxy_lock() after this. | |
1739 | * | |
1740 | * Returns: | |
1741 | * 0 - task blocked on lock | |
1742 | * 1 - acquired the lock for task, caller should wake it up | |
1743 | * <0 - error | |
1744 | * | |
1745 | * Special API call for PI-futex support. | |
1746 | */ | |
56222b21 | 1747 | int __rt_mutex_start_proxy_lock(struct rt_mutex *lock, |
8dac456a | 1748 | struct rt_mutex_waiter *waiter, |
c051b21f | 1749 | struct task_struct *task) |
8dac456a DH |
1750 | { |
1751 | int ret; | |
1752 | ||
1a1fb985 TG |
1753 | lockdep_assert_held(&lock->wait_lock); |
1754 | ||
56222b21 | 1755 | if (try_to_take_rt_mutex(lock, task, NULL)) |
8dac456a | 1756 | return 1; |
8dac456a | 1757 | |
3d5c9340 | 1758 | /* We enforce deadlock detection for futexes */ |
8930ed80 TG |
1759 | ret = task_blocks_on_rt_mutex(lock, waiter, task, |
1760 | RT_MUTEX_FULL_CHAINWALK); | |
8dac456a | 1761 | |
8161239a | 1762 | if (ret && !rt_mutex_owner(lock)) { |
8dac456a DH |
1763 | /* |
1764 | * Reset the return value. We might have | |
1765 | * returned with -EDEADLK and the owner | |
1766 | * released the lock while we were walking the | |
1767 | * pi chain. Let the waiter sort it out. | |
1768 | */ | |
1769 | ret = 0; | |
1770 | } | |
8161239a | 1771 | |
8dac456a DH |
1772 | debug_rt_mutex_print_deadlock(waiter); |
1773 | ||
1774 | return ret; | |
1775 | } | |
1776 | ||
56222b21 PZ |
1777 | /** |
1778 | * rt_mutex_start_proxy_lock() - Start lock acquisition for another task | |
1779 | * @lock: the rt_mutex to take | |
1780 | * @waiter: the pre-initialized rt_mutex_waiter | |
1781 | * @task: the task to prepare | |
1782 | * | |
1a1fb985 TG |
1783 | * Starts the rt_mutex acquire; it enqueues the @waiter and does deadlock |
1784 | * detection. It does not wait, see rt_mutex_wait_proxy_lock() for that. | |
1785 | * | |
1786 | * NOTE: unlike __rt_mutex_start_proxy_lock this _DOES_ remove the @waiter | |
1787 | * on failure. | |
1788 | * | |
56222b21 PZ |
1789 | * Returns: |
1790 | * 0 - task blocked on lock | |
1791 | * 1 - acquired the lock for task, caller should wake it up | |
1792 | * <0 - error | |
1793 | * | |
1a1fb985 | 1794 | * Special API call for PI-futex support. |
56222b21 PZ |
1795 | */ |
1796 | int rt_mutex_start_proxy_lock(struct rt_mutex *lock, | |
1797 | struct rt_mutex_waiter *waiter, | |
1798 | struct task_struct *task) | |
1799 | { | |
1800 | int ret; | |
1801 | ||
1802 | raw_spin_lock_irq(&lock->wait_lock); | |
1803 | ret = __rt_mutex_start_proxy_lock(lock, waiter, task); | |
1a1fb985 TG |
1804 | if (unlikely(ret)) |
1805 | remove_waiter(lock, waiter); | |
56222b21 PZ |
1806 | raw_spin_unlock_irq(&lock->wait_lock); |
1807 | ||
1808 | return ret; | |
1809 | } | |
1810 | ||
0cdbee99 IM |
1811 | /** |
1812 | * rt_mutex_next_owner - return the next owner of the lock | |
1813 | * | |
1814 | * @lock: the rt lock query | |
1815 | * | |
1816 | * Returns the next owner of the lock or NULL | |
1817 | * | |
1818 | * Caller has to serialize against other accessors to the lock | |
1819 | * itself. | |
1820 | * | |
1821 | * Special API call for PI-futex support | |
1822 | */ | |
1823 | struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock) | |
1824 | { | |
1825 | if (!rt_mutex_has_waiters(lock)) | |
1826 | return NULL; | |
1827 | ||
1828 | return rt_mutex_top_waiter(lock)->task; | |
1829 | } | |
8dac456a DH |
1830 | |
1831 | /** | |
38d589f2 | 1832 | * rt_mutex_wait_proxy_lock() - Wait for lock acquisition |
8dac456a DH |
1833 | * @lock: the rt_mutex we were woken on |
1834 | * @to: the timeout, null if none. hrtimer should already have | |
c051b21f | 1835 | * been started. |
8dac456a | 1836 | * @waiter: the pre-initialized rt_mutex_waiter |
8dac456a | 1837 | * |
38d589f2 PZ |
1838 | * Wait for the the lock acquisition started on our behalf by |
1839 | * rt_mutex_start_proxy_lock(). Upon failure, the caller must call | |
1840 | * rt_mutex_cleanup_proxy_lock(). | |
8dac456a DH |
1841 | * |
1842 | * Returns: | |
1843 | * 0 - success | |
c051b21f | 1844 | * <0 - error, one of -EINTR, -ETIMEDOUT |
8dac456a | 1845 | * |
38d589f2 | 1846 | * Special API call for PI-futex support |
8dac456a | 1847 | */ |
38d589f2 | 1848 | int rt_mutex_wait_proxy_lock(struct rt_mutex *lock, |
8dac456a | 1849 | struct hrtimer_sleeper *to, |
c051b21f | 1850 | struct rt_mutex_waiter *waiter) |
8dac456a DH |
1851 | { |
1852 | int ret; | |
1853 | ||
b4abf910 | 1854 | raw_spin_lock_irq(&lock->wait_lock); |
afffc6c1 | 1855 | /* sleep on the mutex */ |
04dc1b2f | 1856 | set_current_state(TASK_INTERRUPTIBLE); |
8161239a | 1857 | ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter); |
04dc1b2f PZ |
1858 | /* |
1859 | * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might | |
1860 | * have to fix that up. | |
1861 | */ | |
1862 | fixup_rt_mutex_waiters(lock); | |
b4abf910 | 1863 | raw_spin_unlock_irq(&lock->wait_lock); |
8dac456a | 1864 | |
8dac456a DH |
1865 | return ret; |
1866 | } | |
38d589f2 PZ |
1867 | |
1868 | /** | |
1869 | * rt_mutex_cleanup_proxy_lock() - Cleanup failed lock acquisition | |
1870 | * @lock: the rt_mutex we were woken on | |
1871 | * @waiter: the pre-initialized rt_mutex_waiter | |
1872 | * | |
1a1fb985 TG |
1873 | * Attempt to clean up after a failed __rt_mutex_start_proxy_lock() or |
1874 | * rt_mutex_wait_proxy_lock(). | |
38d589f2 PZ |
1875 | * |
1876 | * Unless we acquired the lock; we're still enqueued on the wait-list and can | |
1877 | * in fact still be granted ownership until we're removed. Therefore we can | |
1878 | * find we are in fact the owner and must disregard the | |
1879 | * rt_mutex_wait_proxy_lock() failure. | |
1880 | * | |
1881 | * Returns: | |
1882 | * true - did the cleanup, we done. | |
1883 | * false - we acquired the lock after rt_mutex_wait_proxy_lock() returned, | |
1884 | * caller should disregards its return value. | |
1885 | * | |
1886 | * Special API call for PI-futex support | |
1887 | */ | |
1888 | bool rt_mutex_cleanup_proxy_lock(struct rt_mutex *lock, | |
1889 | struct rt_mutex_waiter *waiter) | |
1890 | { | |
1891 | bool cleanup = false; | |
1892 | ||
1893 | raw_spin_lock_irq(&lock->wait_lock); | |
04dc1b2f PZ |
1894 | /* |
1895 | * Do an unconditional try-lock, this deals with the lock stealing | |
1896 | * state where __rt_mutex_futex_unlock() -> mark_wakeup_next_waiter() | |
1897 | * sets a NULL owner. | |
1898 | * | |
1899 | * We're not interested in the return value, because the subsequent | |
1900 | * test on rt_mutex_owner() will infer that. If the trylock succeeded, | |
1901 | * we will own the lock and it will have removed the waiter. If we | |
1902 | * failed the trylock, we're still not owner and we need to remove | |
1903 | * ourselves. | |
1904 | */ | |
1905 | try_to_take_rt_mutex(lock, current, waiter); | |
38d589f2 PZ |
1906 | /* |
1907 | * Unless we're the owner; we're still enqueued on the wait_list. | |
1908 | * So check if we became owner, if not, take us off the wait_list. | |
1909 | */ | |
1910 | if (rt_mutex_owner(lock) != current) { | |
1911 | remove_waiter(lock, waiter); | |
38d589f2 PZ |
1912 | cleanup = true; |
1913 | } | |
cfafcd11 PZ |
1914 | /* |
1915 | * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might | |
1916 | * have to fix that up. | |
1917 | */ | |
1918 | fixup_rt_mutex_waiters(lock); | |
1919 | ||
38d589f2 PZ |
1920 | raw_spin_unlock_irq(&lock->wait_lock); |
1921 | ||
1922 | return cleanup; | |
1923 | } |