xref: /linux/kernel/locking/rtmutex.c (revision 42b16d3ac371a2fac9b6f08fd75f23f34ba3955a)
1 // SPDX-License-Identifier: GPL-2.0-only
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
11  * Adaptive Spinlocks:
12  *  Copyright (C) 2008 Novell, Inc., Gregory Haskins, Sven Dietrich,
13  *				     and Peter Morreale,
14  * Adaptive Spinlocks simplification:
15  *  Copyright (C) 2008 Red Hat, Inc., Steven Rostedt <srostedt@redhat.com>
16  *
17  *  See Documentation/locking/rt-mutex-design.rst for details.
18  */
19 #include <linux/sched.h>
20 #include <linux/sched/debug.h>
21 #include <linux/sched/deadline.h>
22 #include <linux/sched/signal.h>
23 #include <linux/sched/rt.h>
24 #include <linux/sched/wake_q.h>
25 #include <linux/ww_mutex.h>
26 
27 #include <trace/events/lock.h>
28 
29 #include "rtmutex_common.h"
30 
31 #ifndef WW_RT
32 # define build_ww_mutex()	(false)
33 # define ww_container_of(rtm)	NULL
34 
__ww_mutex_add_waiter(struct rt_mutex_waiter * waiter,struct rt_mutex * lock,struct ww_acquire_ctx * ww_ctx)35 static inline int __ww_mutex_add_waiter(struct rt_mutex_waiter *waiter,
36 					struct rt_mutex *lock,
37 					struct ww_acquire_ctx *ww_ctx)
38 {
39 	return 0;
40 }
41 
__ww_mutex_check_waiters(struct rt_mutex * lock,struct ww_acquire_ctx * ww_ctx)42 static inline void __ww_mutex_check_waiters(struct rt_mutex *lock,
43 					    struct ww_acquire_ctx *ww_ctx)
44 {
45 }
46 
ww_mutex_lock_acquired(struct ww_mutex * lock,struct ww_acquire_ctx * ww_ctx)47 static inline void ww_mutex_lock_acquired(struct ww_mutex *lock,
48 					  struct ww_acquire_ctx *ww_ctx)
49 {
50 }
51 
__ww_mutex_check_kill(struct rt_mutex * lock,struct rt_mutex_waiter * waiter,struct ww_acquire_ctx * ww_ctx)52 static inline int __ww_mutex_check_kill(struct rt_mutex *lock,
53 					struct rt_mutex_waiter *waiter,
54 					struct ww_acquire_ctx *ww_ctx)
55 {
56 	return 0;
57 }
58 
59 #else
60 # define build_ww_mutex()	(true)
61 # define ww_container_of(rtm)	container_of(rtm, struct ww_mutex, base)
62 # include "ww_mutex.h"
63 #endif
64 
65 /*
66  * lock->owner state tracking:
67  *
68  * lock->owner holds the task_struct pointer of the owner. Bit 0
69  * is used to keep track of the "lock has waiters" state.
70  *
71  * owner	bit0
72  * NULL		0	lock is free (fast acquire possible)
73  * NULL		1	lock is free and has waiters and the top waiter
74  *				is going to take the lock*
75  * taskpointer	0	lock is held (fast release possible)
76  * taskpointer	1	lock is held and has waiters**
77  *
78  * The fast atomic compare exchange based acquire and release is only
79  * possible when bit 0 of lock->owner is 0.
80  *
81  * (*) It also can be a transitional state when grabbing the lock
82  * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
83  * we need to set the bit0 before looking at the lock, and the owner may be
84  * NULL in this small time, hence this can be a transitional state.
85  *
86  * (**) There is a small time when bit 0 is set but there are no
87  * waiters. This can happen when grabbing the lock in the slow path.
88  * To prevent a cmpxchg of the owner releasing the lock, we need to
89  * set this bit before looking at the lock.
90  */
91 
92 static __always_inline struct task_struct *
rt_mutex_owner_encode(struct rt_mutex_base * lock,struct task_struct * owner)93 rt_mutex_owner_encode(struct rt_mutex_base *lock, struct task_struct *owner)
94 {
95 	unsigned long val = (unsigned long)owner;
96 
97 	if (rt_mutex_has_waiters(lock))
98 		val |= RT_MUTEX_HAS_WAITERS;
99 
100 	return (struct task_struct *)val;
101 }
102 
103 static __always_inline void
rt_mutex_set_owner(struct rt_mutex_base * lock,struct task_struct * owner)104 rt_mutex_set_owner(struct rt_mutex_base *lock, struct task_struct *owner)
105 {
106 	/*
107 	 * lock->wait_lock is held but explicit acquire semantics are needed
108 	 * for a new lock owner so WRITE_ONCE is insufficient.
109 	 */
110 	xchg_acquire(&lock->owner, rt_mutex_owner_encode(lock, owner));
111 }
112 
rt_mutex_clear_owner(struct rt_mutex_base * lock)113 static __always_inline void rt_mutex_clear_owner(struct rt_mutex_base *lock)
114 {
115 	/* lock->wait_lock is held so the unlock provides release semantics. */
116 	WRITE_ONCE(lock->owner, rt_mutex_owner_encode(lock, NULL));
117 }
118 
clear_rt_mutex_waiters(struct rt_mutex_base * lock)119 static __always_inline void clear_rt_mutex_waiters(struct rt_mutex_base *lock)
120 {
121 	lock->owner = (struct task_struct *)
122 			((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
123 }
124 
125 static __always_inline void
fixup_rt_mutex_waiters(struct rt_mutex_base * lock,bool acquire_lock)126 fixup_rt_mutex_waiters(struct rt_mutex_base *lock, bool acquire_lock)
127 {
128 	unsigned long owner, *p = (unsigned long *) &lock->owner;
129 
130 	if (rt_mutex_has_waiters(lock))
131 		return;
132 
133 	/*
134 	 * The rbtree has no waiters enqueued, now make sure that the
135 	 * lock->owner still has the waiters bit set, otherwise the
136 	 * following can happen:
137 	 *
138 	 * CPU 0	CPU 1		CPU2
139 	 * l->owner=T1
140 	 *		rt_mutex_lock(l)
141 	 *		lock(l->lock)
142 	 *		l->owner = T1 | HAS_WAITERS;
143 	 *		enqueue(T2)
144 	 *		boost()
145 	 *		  unlock(l->lock)
146 	 *		block()
147 	 *
148 	 *				rt_mutex_lock(l)
149 	 *				lock(l->lock)
150 	 *				l->owner = T1 | HAS_WAITERS;
151 	 *				enqueue(T3)
152 	 *				boost()
153 	 *				  unlock(l->lock)
154 	 *				block()
155 	 *		signal(->T2)	signal(->T3)
156 	 *		lock(l->lock)
157 	 *		dequeue(T2)
158 	 *		deboost()
159 	 *		  unlock(l->lock)
160 	 *				lock(l->lock)
161 	 *				dequeue(T3)
162 	 *				 ==> wait list is empty
163 	 *				deboost()
164 	 *				 unlock(l->lock)
165 	 *		lock(l->lock)
166 	 *		fixup_rt_mutex_waiters()
167 	 *		  if (wait_list_empty(l) {
168 	 *		    l->owner = owner
169 	 *		    owner = l->owner & ~HAS_WAITERS;
170 	 *		      ==> l->owner = T1
171 	 *		  }
172 	 *				lock(l->lock)
173 	 * rt_mutex_unlock(l)		fixup_rt_mutex_waiters()
174 	 *				  if (wait_list_empty(l) {
175 	 *				    owner = l->owner & ~HAS_WAITERS;
176 	 * cmpxchg(l->owner, T1, NULL)
177 	 *  ===> Success (l->owner = NULL)
178 	 *
179 	 *				    l->owner = owner
180 	 *				      ==> l->owner = T1
181 	 *				  }
182 	 *
183 	 * With the check for the waiter bit in place T3 on CPU2 will not
184 	 * overwrite. All tasks fiddling with the waiters bit are
185 	 * serialized by l->lock, so nothing else can modify the waiters
186 	 * bit. If the bit is set then nothing can change l->owner either
187 	 * so the simple RMW is safe. The cmpxchg() will simply fail if it
188 	 * happens in the middle of the RMW because the waiters bit is
189 	 * still set.
190 	 */
191 	owner = READ_ONCE(*p);
192 	if (owner & RT_MUTEX_HAS_WAITERS) {
193 		/*
194 		 * See rt_mutex_set_owner() and rt_mutex_clear_owner() on
195 		 * why xchg_acquire() is used for updating owner for
196 		 * locking and WRITE_ONCE() for unlocking.
197 		 *
198 		 * WRITE_ONCE() would work for the acquire case too, but
199 		 * in case that the lock acquisition failed it might
200 		 * force other lockers into the slow path unnecessarily.
201 		 */
202 		if (acquire_lock)
203 			xchg_acquire(p, owner & ~RT_MUTEX_HAS_WAITERS);
204 		else
205 			WRITE_ONCE(*p, owner & ~RT_MUTEX_HAS_WAITERS);
206 	}
207 }
208 
209 /*
210  * We can speed up the acquire/release, if there's no debugging state to be
211  * set up.
212  */
213 #ifndef CONFIG_DEBUG_RT_MUTEXES
rt_mutex_cmpxchg_acquire(struct rt_mutex_base * lock,struct task_struct * old,struct task_struct * new)214 static __always_inline bool rt_mutex_cmpxchg_acquire(struct rt_mutex_base *lock,
215 						     struct task_struct *old,
216 						     struct task_struct *new)
217 {
218 	return try_cmpxchg_acquire(&lock->owner, &old, new);
219 }
220 
rt_mutex_try_acquire(struct rt_mutex_base * lock)221 static __always_inline bool rt_mutex_try_acquire(struct rt_mutex_base *lock)
222 {
223 	return rt_mutex_cmpxchg_acquire(lock, NULL, current);
224 }
225 
rt_mutex_cmpxchg_release(struct rt_mutex_base * lock,struct task_struct * old,struct task_struct * new)226 static __always_inline bool rt_mutex_cmpxchg_release(struct rt_mutex_base *lock,
227 						     struct task_struct *old,
228 						     struct task_struct *new)
229 {
230 	return try_cmpxchg_release(&lock->owner, &old, new);
231 }
232 
233 /*
234  * Callers must hold the ->wait_lock -- which is the whole purpose as we force
235  * all future threads that attempt to [Rmw] the lock to the slowpath. As such
236  * relaxed semantics suffice.
237  */
mark_rt_mutex_waiters(struct rt_mutex_base * lock)238 static __always_inline void mark_rt_mutex_waiters(struct rt_mutex_base *lock)
239 {
240 	unsigned long *p = (unsigned long *) &lock->owner;
241 	unsigned long owner, new;
242 
243 	owner = READ_ONCE(*p);
244 	do {
245 		new = owner | RT_MUTEX_HAS_WAITERS;
246 	} while (!try_cmpxchg_relaxed(p, &owner, new));
247 
248 	/*
249 	 * The cmpxchg loop above is relaxed to avoid back-to-back ACQUIRE
250 	 * operations in the event of contention. Ensure the successful
251 	 * cmpxchg is visible.
252 	 */
253 	smp_mb__after_atomic();
254 }
255 
256 /*
257  * Safe fastpath aware unlock:
258  * 1) Clear the waiters bit
259  * 2) Drop lock->wait_lock
260  * 3) Try to unlock the lock with cmpxchg
261  */
unlock_rt_mutex_safe(struct rt_mutex_base * lock,unsigned long flags)262 static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex_base *lock,
263 						 unsigned long flags)
264 	__releases(lock->wait_lock)
265 {
266 	struct task_struct *owner = rt_mutex_owner(lock);
267 
268 	clear_rt_mutex_waiters(lock);
269 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
270 	/*
271 	 * If a new waiter comes in between the unlock and the cmpxchg
272 	 * we have two situations:
273 	 *
274 	 * unlock(wait_lock);
275 	 *					lock(wait_lock);
276 	 * cmpxchg(p, owner, 0) == owner
277 	 *					mark_rt_mutex_waiters(lock);
278 	 *					acquire(lock);
279 	 * or:
280 	 *
281 	 * unlock(wait_lock);
282 	 *					lock(wait_lock);
283 	 *					mark_rt_mutex_waiters(lock);
284 	 *
285 	 * cmpxchg(p, owner, 0) != owner
286 	 *					enqueue_waiter();
287 	 *					unlock(wait_lock);
288 	 * lock(wait_lock);
289 	 * wake waiter();
290 	 * unlock(wait_lock);
291 	 *					lock(wait_lock);
292 	 *					acquire(lock);
293 	 */
294 	return rt_mutex_cmpxchg_release(lock, owner, NULL);
295 }
296 
297 #else
rt_mutex_cmpxchg_acquire(struct rt_mutex_base * lock,struct task_struct * old,struct task_struct * new)298 static __always_inline bool rt_mutex_cmpxchg_acquire(struct rt_mutex_base *lock,
299 						     struct task_struct *old,
300 						     struct task_struct *new)
301 {
302 	return false;
303 
304 }
305 
306 static int __sched rt_mutex_slowtrylock(struct rt_mutex_base *lock);
307 
rt_mutex_try_acquire(struct rt_mutex_base * lock)308 static __always_inline bool rt_mutex_try_acquire(struct rt_mutex_base *lock)
309 {
310 	/*
311 	 * With debug enabled rt_mutex_cmpxchg trylock() will always fail.
312 	 *
313 	 * Avoid unconditionally taking the slow path by using
314 	 * rt_mutex_slow_trylock() which is covered by the debug code and can
315 	 * acquire a non-contended rtmutex.
316 	 */
317 	return rt_mutex_slowtrylock(lock);
318 }
319 
rt_mutex_cmpxchg_release(struct rt_mutex_base * lock,struct task_struct * old,struct task_struct * new)320 static __always_inline bool rt_mutex_cmpxchg_release(struct rt_mutex_base *lock,
321 						     struct task_struct *old,
322 						     struct task_struct *new)
323 {
324 	return false;
325 }
326 
mark_rt_mutex_waiters(struct rt_mutex_base * lock)327 static __always_inline void mark_rt_mutex_waiters(struct rt_mutex_base *lock)
328 {
329 	lock->owner = (struct task_struct *)
330 			((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
331 }
332 
333 /*
334  * Simple slow path only version: lock->owner is protected by lock->wait_lock.
335  */
unlock_rt_mutex_safe(struct rt_mutex_base * lock,unsigned long flags)336 static __always_inline bool unlock_rt_mutex_safe(struct rt_mutex_base *lock,
337 						 unsigned long flags)
338 	__releases(lock->wait_lock)
339 {
340 	lock->owner = NULL;
341 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
342 	return true;
343 }
344 #endif
345 
__waiter_prio(struct task_struct * task)346 static __always_inline int __waiter_prio(struct task_struct *task)
347 {
348 	int prio = task->prio;
349 
350 	if (!rt_or_dl_prio(prio))
351 		return DEFAULT_PRIO;
352 
353 	return prio;
354 }
355 
356 /*
357  * Update the waiter->tree copy of the sort keys.
358  */
359 static __always_inline void
waiter_update_prio(struct rt_mutex_waiter * waiter,struct task_struct * task)360 waiter_update_prio(struct rt_mutex_waiter *waiter, struct task_struct *task)
361 {
362 	lockdep_assert_held(&waiter->lock->wait_lock);
363 	lockdep_assert(RB_EMPTY_NODE(&waiter->tree.entry));
364 
365 	waiter->tree.prio = __waiter_prio(task);
366 	waiter->tree.deadline = task->dl.deadline;
367 }
368 
369 /*
370  * Update the waiter->pi_tree copy of the sort keys (from the tree copy).
371  */
372 static __always_inline void
waiter_clone_prio(struct rt_mutex_waiter * waiter,struct task_struct * task)373 waiter_clone_prio(struct rt_mutex_waiter *waiter, struct task_struct *task)
374 {
375 	lockdep_assert_held(&waiter->lock->wait_lock);
376 	lockdep_assert_held(&task->pi_lock);
377 	lockdep_assert(RB_EMPTY_NODE(&waiter->pi_tree.entry));
378 
379 	waiter->pi_tree.prio = waiter->tree.prio;
380 	waiter->pi_tree.deadline = waiter->tree.deadline;
381 }
382 
383 /*
384  * Only use with rt_waiter_node_{less,equal}()
385  */
386 #define task_to_waiter_node(p)	\
387 	&(struct rt_waiter_node){ .prio = __waiter_prio(p), .deadline = (p)->dl.deadline }
388 #define task_to_waiter(p)	\
389 	&(struct rt_mutex_waiter){ .tree = *task_to_waiter_node(p) }
390 
rt_waiter_node_less(struct rt_waiter_node * left,struct rt_waiter_node * right)391 static __always_inline int rt_waiter_node_less(struct rt_waiter_node *left,
392 					       struct rt_waiter_node *right)
393 {
394 	if (left->prio < right->prio)
395 		return 1;
396 
397 	/*
398 	 * If both waiters have dl_prio(), we check the deadlines of the
399 	 * associated tasks.
400 	 * If left waiter has a dl_prio(), and we didn't return 1 above,
401 	 * then right waiter has a dl_prio() too.
402 	 */
403 	if (dl_prio(left->prio))
404 		return dl_time_before(left->deadline, right->deadline);
405 
406 	return 0;
407 }
408 
rt_waiter_node_equal(struct rt_waiter_node * left,struct rt_waiter_node * right)409 static __always_inline int rt_waiter_node_equal(struct rt_waiter_node *left,
410 						 struct rt_waiter_node *right)
411 {
412 	if (left->prio != right->prio)
413 		return 0;
414 
415 	/*
416 	 * If both waiters have dl_prio(), we check the deadlines of the
417 	 * associated tasks.
418 	 * If left waiter has a dl_prio(), and we didn't return 0 above,
419 	 * then right waiter has a dl_prio() too.
420 	 */
421 	if (dl_prio(left->prio))
422 		return left->deadline == right->deadline;
423 
424 	return 1;
425 }
426 
rt_mutex_steal(struct rt_mutex_waiter * waiter,struct rt_mutex_waiter * top_waiter)427 static inline bool rt_mutex_steal(struct rt_mutex_waiter *waiter,
428 				  struct rt_mutex_waiter *top_waiter)
429 {
430 	if (rt_waiter_node_less(&waiter->tree, &top_waiter->tree))
431 		return true;
432 
433 #ifdef RT_MUTEX_BUILD_SPINLOCKS
434 	/*
435 	 * Note that RT tasks are excluded from same priority (lateral)
436 	 * steals to prevent the introduction of an unbounded latency.
437 	 */
438 	if (rt_or_dl_prio(waiter->tree.prio))
439 		return false;
440 
441 	return rt_waiter_node_equal(&waiter->tree, &top_waiter->tree);
442 #else
443 	return false;
444 #endif
445 }
446 
447 #define __node_2_waiter(node) \
448 	rb_entry((node), struct rt_mutex_waiter, tree.entry)
449 
__waiter_less(struct rb_node * a,const struct rb_node * b)450 static __always_inline bool __waiter_less(struct rb_node *a, const struct rb_node *b)
451 {
452 	struct rt_mutex_waiter *aw = __node_2_waiter(a);
453 	struct rt_mutex_waiter *bw = __node_2_waiter(b);
454 
455 	if (rt_waiter_node_less(&aw->tree, &bw->tree))
456 		return 1;
457 
458 	if (!build_ww_mutex())
459 		return 0;
460 
461 	if (rt_waiter_node_less(&bw->tree, &aw->tree))
462 		return 0;
463 
464 	/* NOTE: relies on waiter->ww_ctx being set before insertion */
465 	if (aw->ww_ctx) {
466 		if (!bw->ww_ctx)
467 			return 1;
468 
469 		return (signed long)(aw->ww_ctx->stamp -
470 				     bw->ww_ctx->stamp) < 0;
471 	}
472 
473 	return 0;
474 }
475 
476 static __always_inline void
rt_mutex_enqueue(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter)477 rt_mutex_enqueue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
478 {
479 	lockdep_assert_held(&lock->wait_lock);
480 
481 	rb_add_cached(&waiter->tree.entry, &lock->waiters, __waiter_less);
482 }
483 
484 static __always_inline void
rt_mutex_dequeue(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter)485 rt_mutex_dequeue(struct rt_mutex_base *lock, struct rt_mutex_waiter *waiter)
486 {
487 	lockdep_assert_held(&lock->wait_lock);
488 
489 	if (RB_EMPTY_NODE(&waiter->tree.entry))
490 		return;
491 
492 	rb_erase_cached(&waiter->tree.entry, &lock->waiters);
493 	RB_CLEAR_NODE(&waiter->tree.entry);
494 }
495 
496 #define __node_2_rt_node(node) \
497 	rb_entry((node), struct rt_waiter_node, entry)
498 
__pi_waiter_less(struct rb_node * a,const struct rb_node * b)499 static __always_inline bool __pi_waiter_less(struct rb_node *a, const struct rb_node *b)
500 {
501 	return rt_waiter_node_less(__node_2_rt_node(a), __node_2_rt_node(b));
502 }
503 
504 static __always_inline void
rt_mutex_enqueue_pi(struct task_struct * task,struct rt_mutex_waiter * waiter)505 rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
506 {
507 	lockdep_assert_held(&task->pi_lock);
508 
509 	rb_add_cached(&waiter->pi_tree.entry, &task->pi_waiters, __pi_waiter_less);
510 }
511 
512 static __always_inline void
rt_mutex_dequeue_pi(struct task_struct * task,struct rt_mutex_waiter * waiter)513 rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
514 {
515 	lockdep_assert_held(&task->pi_lock);
516 
517 	if (RB_EMPTY_NODE(&waiter->pi_tree.entry))
518 		return;
519 
520 	rb_erase_cached(&waiter->pi_tree.entry, &task->pi_waiters);
521 	RB_CLEAR_NODE(&waiter->pi_tree.entry);
522 }
523 
rt_mutex_adjust_prio(struct rt_mutex_base * lock,struct task_struct * p)524 static __always_inline void rt_mutex_adjust_prio(struct rt_mutex_base *lock,
525 						 struct task_struct *p)
526 {
527 	struct task_struct *pi_task = NULL;
528 
529 	lockdep_assert_held(&lock->wait_lock);
530 	lockdep_assert(rt_mutex_owner(lock) == p);
531 	lockdep_assert_held(&p->pi_lock);
532 
533 	if (task_has_pi_waiters(p))
534 		pi_task = task_top_pi_waiter(p)->task;
535 
536 	rt_mutex_setprio(p, pi_task);
537 }
538 
539 /* RT mutex specific wake_q wrappers */
rt_mutex_wake_q_add_task(struct rt_wake_q_head * wqh,struct task_struct * task,unsigned int wake_state)540 static __always_inline void rt_mutex_wake_q_add_task(struct rt_wake_q_head *wqh,
541 						     struct task_struct *task,
542 						     unsigned int wake_state)
543 {
544 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && wake_state == TASK_RTLOCK_WAIT) {
545 		if (IS_ENABLED(CONFIG_PROVE_LOCKING))
546 			WARN_ON_ONCE(wqh->rtlock_task);
547 		get_task_struct(task);
548 		wqh->rtlock_task = task;
549 	} else {
550 		wake_q_add(&wqh->head, task);
551 	}
552 }
553 
rt_mutex_wake_q_add(struct rt_wake_q_head * wqh,struct rt_mutex_waiter * w)554 static __always_inline void rt_mutex_wake_q_add(struct rt_wake_q_head *wqh,
555 						struct rt_mutex_waiter *w)
556 {
557 	rt_mutex_wake_q_add_task(wqh, w->task, w->wake_state);
558 }
559 
rt_mutex_wake_up_q(struct rt_wake_q_head * wqh)560 static __always_inline void rt_mutex_wake_up_q(struct rt_wake_q_head *wqh)
561 {
562 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && wqh->rtlock_task) {
563 		wake_up_state(wqh->rtlock_task, TASK_RTLOCK_WAIT);
564 		put_task_struct(wqh->rtlock_task);
565 		wqh->rtlock_task = NULL;
566 	}
567 
568 	if (!wake_q_empty(&wqh->head))
569 		wake_up_q(&wqh->head);
570 
571 	/* Pairs with preempt_disable() in mark_wakeup_next_waiter() */
572 	preempt_enable();
573 }
574 
575 /*
576  * Deadlock detection is conditional:
577  *
578  * If CONFIG_DEBUG_RT_MUTEXES=n, deadlock detection is only conducted
579  * if the detect argument is == RT_MUTEX_FULL_CHAINWALK.
580  *
581  * If CONFIG_DEBUG_RT_MUTEXES=y, deadlock detection is always
582  * conducted independent of the detect argument.
583  *
584  * If the waiter argument is NULL this indicates the deboost path and
585  * deadlock detection is disabled independent of the detect argument
586  * and the config settings.
587  */
588 static __always_inline bool
rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter * waiter,enum rtmutex_chainwalk chwalk)589 rt_mutex_cond_detect_deadlock(struct rt_mutex_waiter *waiter,
590 			      enum rtmutex_chainwalk chwalk)
591 {
592 	if (IS_ENABLED(CONFIG_DEBUG_RT_MUTEXES))
593 		return waiter != NULL;
594 	return chwalk == RT_MUTEX_FULL_CHAINWALK;
595 }
596 
task_blocked_on_lock(struct task_struct * p)597 static __always_inline struct rt_mutex_base *task_blocked_on_lock(struct task_struct *p)
598 {
599 	return p->pi_blocked_on ? p->pi_blocked_on->lock : NULL;
600 }
601 
602 /*
603  * Adjust the priority chain. Also used for deadlock detection.
604  * Decreases task's usage by one - may thus free the task.
605  *
606  * @task:	the task owning the mutex (owner) for which a chain walk is
607  *		probably needed
608  * @chwalk:	do we have to carry out deadlock detection?
609  * @orig_lock:	the mutex (can be NULL if we are walking the chain to recheck
610  *		things for a task that has just got its priority adjusted, and
611  *		is waiting on a mutex)
612  * @next_lock:	the mutex on which the owner of @orig_lock was blocked before
613  *		we dropped its pi_lock. Is never dereferenced, only used for
614  *		comparison to detect lock chain changes.
615  * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
616  *		its priority to the mutex owner (can be NULL in the case
617  *		depicted above or if the top waiter is gone away and we are
618  *		actually deboosting the owner)
619  * @top_task:	the current top waiter
620  *
621  * Returns 0 or -EDEADLK.
622  *
623  * Chain walk basics and protection scope
624  *
625  * [R] refcount on task
626  * [Pn] task->pi_lock held
627  * [L] rtmutex->wait_lock held
628  *
629  * Normal locking order:
630  *
631  *   rtmutex->wait_lock
632  *     task->pi_lock
633  *
634  * Step	Description				Protected by
635  *	function arguments:
636  *	@task					[R]
637  *	@orig_lock if != NULL			@top_task is blocked on it
638  *	@next_lock				Unprotected. Cannot be
639  *						dereferenced. Only used for
640  *						comparison.
641  *	@orig_waiter if != NULL			@top_task is blocked on it
642  *	@top_task				current, or in case of proxy
643  *						locking protected by calling
644  *						code
645  *	again:
646  *	  loop_sanity_check();
647  *	retry:
648  * [1]	  lock(task->pi_lock);			[R] acquire [P1]
649  * [2]	  waiter = task->pi_blocked_on;		[P1]
650  * [3]	  check_exit_conditions_1();		[P1]
651  * [4]	  lock = waiter->lock;			[P1]
652  * [5]	  if (!try_lock(lock->wait_lock)) {	[P1] try to acquire [L]
653  *	    unlock(task->pi_lock);		release [P1]
654  *	    goto retry;
655  *	  }
656  * [6]	  check_exit_conditions_2();		[P1] + [L]
657  * [7]	  requeue_lock_waiter(lock, waiter);	[P1] + [L]
658  * [8]	  unlock(task->pi_lock);		release [P1]
659  *	  put_task_struct(task);		release [R]
660  * [9]	  check_exit_conditions_3();		[L]
661  * [10]	  task = owner(lock);			[L]
662  *	  get_task_struct(task);		[L] acquire [R]
663  *	  lock(task->pi_lock);			[L] acquire [P2]
664  * [11]	  requeue_pi_waiter(tsk, waiters(lock));[P2] + [L]
665  * [12]	  check_exit_conditions_4();		[P2] + [L]
666  * [13]	  unlock(task->pi_lock);		release [P2]
667  *	  unlock(lock->wait_lock);		release [L]
668  *	  goto again;
669  *
670  * Where P1 is the blocking task and P2 is the lock owner; going up one step
671  * the owner becomes the next blocked task etc..
672  *
673 *
674  */
rt_mutex_adjust_prio_chain(struct task_struct * task,enum rtmutex_chainwalk chwalk,struct rt_mutex_base * orig_lock,struct rt_mutex_base * next_lock,struct rt_mutex_waiter * orig_waiter,struct task_struct * top_task)675 static int __sched rt_mutex_adjust_prio_chain(struct task_struct *task,
676 					      enum rtmutex_chainwalk chwalk,
677 					      struct rt_mutex_base *orig_lock,
678 					      struct rt_mutex_base *next_lock,
679 					      struct rt_mutex_waiter *orig_waiter,
680 					      struct task_struct *top_task)
681 {
682 	struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
683 	struct rt_mutex_waiter *prerequeue_top_waiter;
684 	int ret = 0, depth = 0;
685 	struct rt_mutex_base *lock;
686 	bool detect_deadlock;
687 	bool requeue = true;
688 
689 	detect_deadlock = rt_mutex_cond_detect_deadlock(orig_waiter, chwalk);
690 
691 	/*
692 	 * The (de)boosting is a step by step approach with a lot of
693 	 * pitfalls. We want this to be preemptible and we want hold a
694 	 * maximum of two locks per step. So we have to check
695 	 * carefully whether things change under us.
696 	 */
697  again:
698 	/*
699 	 * We limit the lock chain length for each invocation.
700 	 */
701 	if (++depth > max_lock_depth) {
702 		static int prev_max;
703 
704 		/*
705 		 * Print this only once. If the admin changes the limit,
706 		 * print a new message when reaching the limit again.
707 		 */
708 		if (prev_max != max_lock_depth) {
709 			prev_max = max_lock_depth;
710 			printk(KERN_WARNING "Maximum lock depth %d reached "
711 			       "task: %s (%d)\n", max_lock_depth,
712 			       top_task->comm, task_pid_nr(top_task));
713 		}
714 		put_task_struct(task);
715 
716 		return -EDEADLK;
717 	}
718 
719 	/*
720 	 * We are fully preemptible here and only hold the refcount on
721 	 * @task. So everything can have changed under us since the
722 	 * caller or our own code below (goto retry/again) dropped all
723 	 * locks.
724 	 */
725  retry:
726 	/*
727 	 * [1] Task cannot go away as we did a get_task() before !
728 	 */
729 	raw_spin_lock_irq(&task->pi_lock);
730 
731 	/*
732 	 * [2] Get the waiter on which @task is blocked on.
733 	 */
734 	waiter = task->pi_blocked_on;
735 
736 	/*
737 	 * [3] check_exit_conditions_1() protected by task->pi_lock.
738 	 */
739 
740 	/*
741 	 * Check whether the end of the boosting chain has been
742 	 * reached or the state of the chain has changed while we
743 	 * dropped the locks.
744 	 */
745 	if (!waiter)
746 		goto out_unlock_pi;
747 
748 	/*
749 	 * Check the orig_waiter state. After we dropped the locks,
750 	 * the previous owner of the lock might have released the lock.
751 	 */
752 	if (orig_waiter && !rt_mutex_owner(orig_lock))
753 		goto out_unlock_pi;
754 
755 	/*
756 	 * We dropped all locks after taking a refcount on @task, so
757 	 * the task might have moved on in the lock chain or even left
758 	 * the chain completely and blocks now on an unrelated lock or
759 	 * on @orig_lock.
760 	 *
761 	 * We stored the lock on which @task was blocked in @next_lock,
762 	 * so we can detect the chain change.
763 	 */
764 	if (next_lock != waiter->lock)
765 		goto out_unlock_pi;
766 
767 	/*
768 	 * There could be 'spurious' loops in the lock graph due to ww_mutex,
769 	 * consider:
770 	 *
771 	 *   P1: A, ww_A, ww_B
772 	 *   P2: ww_B, ww_A
773 	 *   P3: A
774 	 *
775 	 * P3 should not return -EDEADLK because it gets trapped in the cycle
776 	 * created by P1 and P2 (which will resolve -- and runs into
777 	 * max_lock_depth above). Therefore disable detect_deadlock such that
778 	 * the below termination condition can trigger once all relevant tasks
779 	 * are boosted.
780 	 *
781 	 * Even when we start with ww_mutex we can disable deadlock detection,
782 	 * since we would supress a ww_mutex induced deadlock at [6] anyway.
783 	 * Supressing it here however is not sufficient since we might still
784 	 * hit [6] due to adjustment driven iteration.
785 	 *
786 	 * NOTE: if someone were to create a deadlock between 2 ww_classes we'd
787 	 * utterly fail to report it; lockdep should.
788 	 */
789 	if (IS_ENABLED(CONFIG_PREEMPT_RT) && waiter->ww_ctx && detect_deadlock)
790 		detect_deadlock = false;
791 
792 	/*
793 	 * Drop out, when the task has no waiters. Note,
794 	 * top_waiter can be NULL, when we are in the deboosting
795 	 * mode!
796 	 */
797 	if (top_waiter) {
798 		if (!task_has_pi_waiters(task))
799 			goto out_unlock_pi;
800 		/*
801 		 * If deadlock detection is off, we stop here if we
802 		 * are not the top pi waiter of the task. If deadlock
803 		 * detection is enabled we continue, but stop the
804 		 * requeueing in the chain walk.
805 		 */
806 		if (top_waiter != task_top_pi_waiter(task)) {
807 			if (!detect_deadlock)
808 				goto out_unlock_pi;
809 			else
810 				requeue = false;
811 		}
812 	}
813 
814 	/*
815 	 * If the waiter priority is the same as the task priority
816 	 * then there is no further priority adjustment necessary.  If
817 	 * deadlock detection is off, we stop the chain walk. If its
818 	 * enabled we continue, but stop the requeueing in the chain
819 	 * walk.
820 	 */
821 	if (rt_waiter_node_equal(&waiter->tree, task_to_waiter_node(task))) {
822 		if (!detect_deadlock)
823 			goto out_unlock_pi;
824 		else
825 			requeue = false;
826 	}
827 
828 	/*
829 	 * [4] Get the next lock; per holding task->pi_lock we can't unblock
830 	 * and guarantee @lock's existence.
831 	 */
832 	lock = waiter->lock;
833 	/*
834 	 * [5] We need to trylock here as we are holding task->pi_lock,
835 	 * which is the reverse lock order versus the other rtmutex
836 	 * operations.
837 	 *
838 	 * Per the above, holding task->pi_lock guarantees lock exists, so
839 	 * inverting this lock order is infeasible from a life-time
840 	 * perspective.
841 	 */
842 	if (!raw_spin_trylock(&lock->wait_lock)) {
843 		raw_spin_unlock_irq(&task->pi_lock);
844 		cpu_relax();
845 		goto retry;
846 	}
847 
848 	/*
849 	 * [6] check_exit_conditions_2() protected by task->pi_lock and
850 	 * lock->wait_lock.
851 	 *
852 	 * Deadlock detection. If the lock is the same as the original
853 	 * lock which caused us to walk the lock chain or if the
854 	 * current lock is owned by the task which initiated the chain
855 	 * walk, we detected a deadlock.
856 	 */
857 	if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
858 		ret = -EDEADLK;
859 
860 		/*
861 		 * When the deadlock is due to ww_mutex; also see above. Don't
862 		 * report the deadlock and instead let the ww_mutex wound/die
863 		 * logic pick which of the contending threads gets -EDEADLK.
864 		 *
865 		 * NOTE: assumes the cycle only contains a single ww_class; any
866 		 * other configuration and we fail to report; also, see
867 		 * lockdep.
868 		 */
869 		if (IS_ENABLED(CONFIG_PREEMPT_RT) && orig_waiter && orig_waiter->ww_ctx)
870 			ret = 0;
871 
872 		raw_spin_unlock(&lock->wait_lock);
873 		goto out_unlock_pi;
874 	}
875 
876 	/*
877 	 * If we just follow the lock chain for deadlock detection, no
878 	 * need to do all the requeue operations. To avoid a truckload
879 	 * of conditionals around the various places below, just do the
880 	 * minimum chain walk checks.
881 	 */
882 	if (!requeue) {
883 		/*
884 		 * No requeue[7] here. Just release @task [8]
885 		 */
886 		raw_spin_unlock(&task->pi_lock);
887 		put_task_struct(task);
888 
889 		/*
890 		 * [9] check_exit_conditions_3 protected by lock->wait_lock.
891 		 * If there is no owner of the lock, end of chain.
892 		 */
893 		if (!rt_mutex_owner(lock)) {
894 			raw_spin_unlock_irq(&lock->wait_lock);
895 			return 0;
896 		}
897 
898 		/* [10] Grab the next task, i.e. owner of @lock */
899 		task = get_task_struct(rt_mutex_owner(lock));
900 		raw_spin_lock(&task->pi_lock);
901 
902 		/*
903 		 * No requeue [11] here. We just do deadlock detection.
904 		 *
905 		 * [12] Store whether owner is blocked
906 		 * itself. Decision is made after dropping the locks
907 		 */
908 		next_lock = task_blocked_on_lock(task);
909 		/*
910 		 * Get the top waiter for the next iteration
911 		 */
912 		top_waiter = rt_mutex_top_waiter(lock);
913 
914 		/* [13] Drop locks */
915 		raw_spin_unlock(&task->pi_lock);
916 		raw_spin_unlock_irq(&lock->wait_lock);
917 
918 		/* If owner is not blocked, end of chain. */
919 		if (!next_lock)
920 			goto out_put_task;
921 		goto again;
922 	}
923 
924 	/*
925 	 * Store the current top waiter before doing the requeue
926 	 * operation on @lock. We need it for the boost/deboost
927 	 * decision below.
928 	 */
929 	prerequeue_top_waiter = rt_mutex_top_waiter(lock);
930 
931 	/* [7] Requeue the waiter in the lock waiter tree. */
932 	rt_mutex_dequeue(lock, waiter);
933 
934 	/*
935 	 * Update the waiter prio fields now that we're dequeued.
936 	 *
937 	 * These values can have changed through either:
938 	 *
939 	 *   sys_sched_set_scheduler() / sys_sched_setattr()
940 	 *
941 	 * or
942 	 *
943 	 *   DL CBS enforcement advancing the effective deadline.
944 	 */
945 	waiter_update_prio(waiter, task);
946 
947 	rt_mutex_enqueue(lock, waiter);
948 
949 	/*
950 	 * [8] Release the (blocking) task in preparation for
951 	 * taking the owner task in [10].
952 	 *
953 	 * Since we hold lock->waiter_lock, task cannot unblock, even if we
954 	 * release task->pi_lock.
955 	 */
956 	raw_spin_unlock(&task->pi_lock);
957 	put_task_struct(task);
958 
959 	/*
960 	 * [9] check_exit_conditions_3 protected by lock->wait_lock.
961 	 *
962 	 * We must abort the chain walk if there is no lock owner even
963 	 * in the dead lock detection case, as we have nothing to
964 	 * follow here. This is the end of the chain we are walking.
965 	 */
966 	if (!rt_mutex_owner(lock)) {
967 		/*
968 		 * If the requeue [7] above changed the top waiter,
969 		 * then we need to wake the new top waiter up to try
970 		 * to get the lock.
971 		 */
972 		top_waiter = rt_mutex_top_waiter(lock);
973 		if (prerequeue_top_waiter != top_waiter)
974 			wake_up_state(top_waiter->task, top_waiter->wake_state);
975 		raw_spin_unlock_irq(&lock->wait_lock);
976 		return 0;
977 	}
978 
979 	/*
980 	 * [10] Grab the next task, i.e. the owner of @lock
981 	 *
982 	 * Per holding lock->wait_lock and checking for !owner above, there
983 	 * must be an owner and it cannot go away.
984 	 */
985 	task = get_task_struct(rt_mutex_owner(lock));
986 	raw_spin_lock(&task->pi_lock);
987 
988 	/* [11] requeue the pi waiters if necessary */
989 	if (waiter == rt_mutex_top_waiter(lock)) {
990 		/*
991 		 * The waiter became the new top (highest priority)
992 		 * waiter on the lock. Replace the previous top waiter
993 		 * in the owner tasks pi waiters tree with this waiter
994 		 * and adjust the priority of the owner.
995 		 */
996 		rt_mutex_dequeue_pi(task, prerequeue_top_waiter);
997 		waiter_clone_prio(waiter, task);
998 		rt_mutex_enqueue_pi(task, waiter);
999 		rt_mutex_adjust_prio(lock, task);
1000 
1001 	} else if (prerequeue_top_waiter == waiter) {
1002 		/*
1003 		 * The waiter was the top waiter on the lock, but is
1004 		 * no longer the top priority waiter. Replace waiter in
1005 		 * the owner tasks pi waiters tree with the new top
1006 		 * (highest priority) waiter and adjust the priority
1007 		 * of the owner.
1008 		 * The new top waiter is stored in @waiter so that
1009 		 * @waiter == @top_waiter evaluates to true below and
1010 		 * we continue to deboost the rest of the chain.
1011 		 */
1012 		rt_mutex_dequeue_pi(task, waiter);
1013 		waiter = rt_mutex_top_waiter(lock);
1014 		waiter_clone_prio(waiter, task);
1015 		rt_mutex_enqueue_pi(task, waiter);
1016 		rt_mutex_adjust_prio(lock, task);
1017 	} else {
1018 		/*
1019 		 * Nothing changed. No need to do any priority
1020 		 * adjustment.
1021 		 */
1022 	}
1023 
1024 	/*
1025 	 * [12] check_exit_conditions_4() protected by task->pi_lock
1026 	 * and lock->wait_lock. The actual decisions are made after we
1027 	 * dropped the locks.
1028 	 *
1029 	 * Check whether the task which owns the current lock is pi
1030 	 * blocked itself. If yes we store a pointer to the lock for
1031 	 * the lock chain change detection above. After we dropped
1032 	 * task->pi_lock next_lock cannot be dereferenced anymore.
1033 	 */
1034 	next_lock = task_blocked_on_lock(task);
1035 	/*
1036 	 * Store the top waiter of @lock for the end of chain walk
1037 	 * decision below.
1038 	 */
1039 	top_waiter = rt_mutex_top_waiter(lock);
1040 
1041 	/* [13] Drop the locks */
1042 	raw_spin_unlock(&task->pi_lock);
1043 	raw_spin_unlock_irq(&lock->wait_lock);
1044 
1045 	/*
1046 	 * Make the actual exit decisions [12], based on the stored
1047 	 * values.
1048 	 *
1049 	 * We reached the end of the lock chain. Stop right here. No
1050 	 * point to go back just to figure that out.
1051 	 */
1052 	if (!next_lock)
1053 		goto out_put_task;
1054 
1055 	/*
1056 	 * If the current waiter is not the top waiter on the lock,
1057 	 * then we can stop the chain walk here if we are not in full
1058 	 * deadlock detection mode.
1059 	 */
1060 	if (!detect_deadlock && waiter != top_waiter)
1061 		goto out_put_task;
1062 
1063 	goto again;
1064 
1065  out_unlock_pi:
1066 	raw_spin_unlock_irq(&task->pi_lock);
1067  out_put_task:
1068 	put_task_struct(task);
1069 
1070 	return ret;
1071 }
1072 
1073 /*
1074  * Try to take an rt-mutex
1075  *
1076  * Must be called with lock->wait_lock held and interrupts disabled
1077  *
1078  * @lock:   The lock to be acquired.
1079  * @task:   The task which wants to acquire the lock
1080  * @waiter: The waiter that is queued to the lock's wait tree if the
1081  *	    callsite called task_blocked_on_lock(), otherwise NULL
1082  */
1083 static int __sched
try_to_take_rt_mutex(struct rt_mutex_base * lock,struct task_struct * task,struct rt_mutex_waiter * waiter)1084 try_to_take_rt_mutex(struct rt_mutex_base *lock, struct task_struct *task,
1085 		     struct rt_mutex_waiter *waiter)
1086 {
1087 	lockdep_assert_held(&lock->wait_lock);
1088 
1089 	/*
1090 	 * Before testing whether we can acquire @lock, we set the
1091 	 * RT_MUTEX_HAS_WAITERS bit in @lock->owner. This forces all
1092 	 * other tasks which try to modify @lock into the slow path
1093 	 * and they serialize on @lock->wait_lock.
1094 	 *
1095 	 * The RT_MUTEX_HAS_WAITERS bit can have a transitional state
1096 	 * as explained at the top of this file if and only if:
1097 	 *
1098 	 * - There is a lock owner. The caller must fixup the
1099 	 *   transient state if it does a trylock or leaves the lock
1100 	 *   function due to a signal or timeout.
1101 	 *
1102 	 * - @task acquires the lock and there are no other
1103 	 *   waiters. This is undone in rt_mutex_set_owner(@task) at
1104 	 *   the end of this function.
1105 	 */
1106 	mark_rt_mutex_waiters(lock);
1107 
1108 	/*
1109 	 * If @lock has an owner, give up.
1110 	 */
1111 	if (rt_mutex_owner(lock))
1112 		return 0;
1113 
1114 	/*
1115 	 * If @waiter != NULL, @task has already enqueued the waiter
1116 	 * into @lock waiter tree. If @waiter == NULL then this is a
1117 	 * trylock attempt.
1118 	 */
1119 	if (waiter) {
1120 		struct rt_mutex_waiter *top_waiter = rt_mutex_top_waiter(lock);
1121 
1122 		/*
1123 		 * If waiter is the highest priority waiter of @lock,
1124 		 * or allowed to steal it, take it over.
1125 		 */
1126 		if (waiter == top_waiter || rt_mutex_steal(waiter, top_waiter)) {
1127 			/*
1128 			 * We can acquire the lock. Remove the waiter from the
1129 			 * lock waiters tree.
1130 			 */
1131 			rt_mutex_dequeue(lock, waiter);
1132 		} else {
1133 			return 0;
1134 		}
1135 	} else {
1136 		/*
1137 		 * If the lock has waiters already we check whether @task is
1138 		 * eligible to take over the lock.
1139 		 *
1140 		 * If there are no other waiters, @task can acquire
1141 		 * the lock.  @task->pi_blocked_on is NULL, so it does
1142 		 * not need to be dequeued.
1143 		 */
1144 		if (rt_mutex_has_waiters(lock)) {
1145 			/* Check whether the trylock can steal it. */
1146 			if (!rt_mutex_steal(task_to_waiter(task),
1147 					    rt_mutex_top_waiter(lock)))
1148 				return 0;
1149 
1150 			/*
1151 			 * The current top waiter stays enqueued. We
1152 			 * don't have to change anything in the lock
1153 			 * waiters order.
1154 			 */
1155 		} else {
1156 			/*
1157 			 * No waiters. Take the lock without the
1158 			 * pi_lock dance.@task->pi_blocked_on is NULL
1159 			 * and we have no waiters to enqueue in @task
1160 			 * pi waiters tree.
1161 			 */
1162 			goto takeit;
1163 		}
1164 	}
1165 
1166 	/*
1167 	 * Clear @task->pi_blocked_on. Requires protection by
1168 	 * @task->pi_lock. Redundant operation for the @waiter == NULL
1169 	 * case, but conditionals are more expensive than a redundant
1170 	 * store.
1171 	 */
1172 	raw_spin_lock(&task->pi_lock);
1173 	task->pi_blocked_on = NULL;
1174 	/*
1175 	 * Finish the lock acquisition. @task is the new owner. If
1176 	 * other waiters exist we have to insert the highest priority
1177 	 * waiter into @task->pi_waiters tree.
1178 	 */
1179 	if (rt_mutex_has_waiters(lock))
1180 		rt_mutex_enqueue_pi(task, rt_mutex_top_waiter(lock));
1181 	raw_spin_unlock(&task->pi_lock);
1182 
1183 takeit:
1184 	/*
1185 	 * This either preserves the RT_MUTEX_HAS_WAITERS bit if there
1186 	 * are still waiters or clears it.
1187 	 */
1188 	rt_mutex_set_owner(lock, task);
1189 
1190 	return 1;
1191 }
1192 
1193 /*
1194  * Task blocks on lock.
1195  *
1196  * Prepare waiter and propagate pi chain
1197  *
1198  * This must be called with lock->wait_lock held and interrupts disabled
1199  */
task_blocks_on_rt_mutex(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter,struct task_struct * task,struct ww_acquire_ctx * ww_ctx,enum rtmutex_chainwalk chwalk)1200 static int __sched task_blocks_on_rt_mutex(struct rt_mutex_base *lock,
1201 					   struct rt_mutex_waiter *waiter,
1202 					   struct task_struct *task,
1203 					   struct ww_acquire_ctx *ww_ctx,
1204 					   enum rtmutex_chainwalk chwalk)
1205 {
1206 	struct task_struct *owner = rt_mutex_owner(lock);
1207 	struct rt_mutex_waiter *top_waiter = waiter;
1208 	struct rt_mutex_base *next_lock;
1209 	int chain_walk = 0, res;
1210 
1211 	lockdep_assert_held(&lock->wait_lock);
1212 
1213 	/*
1214 	 * Early deadlock detection. We really don't want the task to
1215 	 * enqueue on itself just to untangle the mess later. It's not
1216 	 * only an optimization. We drop the locks, so another waiter
1217 	 * can come in before the chain walk detects the deadlock. So
1218 	 * the other will detect the deadlock and return -EDEADLOCK,
1219 	 * which is wrong, as the other waiter is not in a deadlock
1220 	 * situation.
1221 	 *
1222 	 * Except for ww_mutex, in that case the chain walk must already deal
1223 	 * with spurious cycles, see the comments at [3] and [6].
1224 	 */
1225 	if (owner == task && !(build_ww_mutex() && ww_ctx))
1226 		return -EDEADLK;
1227 
1228 	raw_spin_lock(&task->pi_lock);
1229 	waiter->task = task;
1230 	waiter->lock = lock;
1231 	waiter_update_prio(waiter, task);
1232 	waiter_clone_prio(waiter, task);
1233 
1234 	/* Get the top priority waiter on the lock */
1235 	if (rt_mutex_has_waiters(lock))
1236 		top_waiter = rt_mutex_top_waiter(lock);
1237 	rt_mutex_enqueue(lock, waiter);
1238 
1239 	task->pi_blocked_on = waiter;
1240 
1241 	raw_spin_unlock(&task->pi_lock);
1242 
1243 	if (build_ww_mutex() && ww_ctx) {
1244 		struct rt_mutex *rtm;
1245 
1246 		/* Check whether the waiter should back out immediately */
1247 		rtm = container_of(lock, struct rt_mutex, rtmutex);
1248 		res = __ww_mutex_add_waiter(waiter, rtm, ww_ctx);
1249 		if (res) {
1250 			raw_spin_lock(&task->pi_lock);
1251 			rt_mutex_dequeue(lock, waiter);
1252 			task->pi_blocked_on = NULL;
1253 			raw_spin_unlock(&task->pi_lock);
1254 			return res;
1255 		}
1256 	}
1257 
1258 	if (!owner)
1259 		return 0;
1260 
1261 	raw_spin_lock(&owner->pi_lock);
1262 	if (waiter == rt_mutex_top_waiter(lock)) {
1263 		rt_mutex_dequeue_pi(owner, top_waiter);
1264 		rt_mutex_enqueue_pi(owner, waiter);
1265 
1266 		rt_mutex_adjust_prio(lock, owner);
1267 		if (owner->pi_blocked_on)
1268 			chain_walk = 1;
1269 	} else if (rt_mutex_cond_detect_deadlock(waiter, chwalk)) {
1270 		chain_walk = 1;
1271 	}
1272 
1273 	/* Store the lock on which owner is blocked or NULL */
1274 	next_lock = task_blocked_on_lock(owner);
1275 
1276 	raw_spin_unlock(&owner->pi_lock);
1277 	/*
1278 	 * Even if full deadlock detection is on, if the owner is not
1279 	 * blocked itself, we can avoid finding this out in the chain
1280 	 * walk.
1281 	 */
1282 	if (!chain_walk || !next_lock)
1283 		return 0;
1284 
1285 	/*
1286 	 * The owner can't disappear while holding a lock,
1287 	 * so the owner struct is protected by wait_lock.
1288 	 * Gets dropped in rt_mutex_adjust_prio_chain()!
1289 	 */
1290 	get_task_struct(owner);
1291 
1292 	raw_spin_unlock_irq(&lock->wait_lock);
1293 
1294 	res = rt_mutex_adjust_prio_chain(owner, chwalk, lock,
1295 					 next_lock, waiter, task);
1296 
1297 	raw_spin_lock_irq(&lock->wait_lock);
1298 
1299 	return res;
1300 }
1301 
1302 /*
1303  * Remove the top waiter from the current tasks pi waiter tree and
1304  * queue it up.
1305  *
1306  * Called with lock->wait_lock held and interrupts disabled.
1307  */
mark_wakeup_next_waiter(struct rt_wake_q_head * wqh,struct rt_mutex_base * lock)1308 static void __sched mark_wakeup_next_waiter(struct rt_wake_q_head *wqh,
1309 					    struct rt_mutex_base *lock)
1310 {
1311 	struct rt_mutex_waiter *waiter;
1312 
1313 	lockdep_assert_held(&lock->wait_lock);
1314 
1315 	raw_spin_lock(&current->pi_lock);
1316 
1317 	waiter = rt_mutex_top_waiter(lock);
1318 
1319 	/*
1320 	 * Remove it from current->pi_waiters and deboost.
1321 	 *
1322 	 * We must in fact deboost here in order to ensure we call
1323 	 * rt_mutex_setprio() to update p->pi_top_task before the
1324 	 * task unblocks.
1325 	 */
1326 	rt_mutex_dequeue_pi(current, waiter);
1327 	rt_mutex_adjust_prio(lock, current);
1328 
1329 	/*
1330 	 * As we are waking up the top waiter, and the waiter stays
1331 	 * queued on the lock until it gets the lock, this lock
1332 	 * obviously has waiters. Just set the bit here and this has
1333 	 * the added benefit of forcing all new tasks into the
1334 	 * slow path making sure no task of lower priority than
1335 	 * the top waiter can steal this lock.
1336 	 */
1337 	lock->owner = (void *) RT_MUTEX_HAS_WAITERS;
1338 
1339 	/*
1340 	 * We deboosted before waking the top waiter task such that we don't
1341 	 * run two tasks with the 'same' priority (and ensure the
1342 	 * p->pi_top_task pointer points to a blocked task). This however can
1343 	 * lead to priority inversion if we would get preempted after the
1344 	 * deboost but before waking our donor task, hence the preempt_disable()
1345 	 * before unlock.
1346 	 *
1347 	 * Pairs with preempt_enable() in rt_mutex_wake_up_q();
1348 	 */
1349 	preempt_disable();
1350 	rt_mutex_wake_q_add(wqh, waiter);
1351 	raw_spin_unlock(&current->pi_lock);
1352 }
1353 
__rt_mutex_slowtrylock(struct rt_mutex_base * lock)1354 static int __sched __rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1355 {
1356 	int ret = try_to_take_rt_mutex(lock, current, NULL);
1357 
1358 	/*
1359 	 * try_to_take_rt_mutex() sets the lock waiters bit
1360 	 * unconditionally. Clean this up.
1361 	 */
1362 	fixup_rt_mutex_waiters(lock, true);
1363 
1364 	return ret;
1365 }
1366 
1367 /*
1368  * Slow path try-lock function:
1369  */
rt_mutex_slowtrylock(struct rt_mutex_base * lock)1370 static int __sched rt_mutex_slowtrylock(struct rt_mutex_base *lock)
1371 {
1372 	unsigned long flags;
1373 	int ret;
1374 
1375 	/*
1376 	 * If the lock already has an owner we fail to get the lock.
1377 	 * This can be done without taking the @lock->wait_lock as
1378 	 * it is only being read, and this is a trylock anyway.
1379 	 */
1380 	if (rt_mutex_owner(lock))
1381 		return 0;
1382 
1383 	/*
1384 	 * The mutex has currently no owner. Lock the wait lock and try to
1385 	 * acquire the lock. We use irqsave here to support early boot calls.
1386 	 */
1387 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
1388 
1389 	ret = __rt_mutex_slowtrylock(lock);
1390 
1391 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1392 
1393 	return ret;
1394 }
1395 
__rt_mutex_trylock(struct rt_mutex_base * lock)1396 static __always_inline int __rt_mutex_trylock(struct rt_mutex_base *lock)
1397 {
1398 	if (likely(rt_mutex_cmpxchg_acquire(lock, NULL, current)))
1399 		return 1;
1400 
1401 	return rt_mutex_slowtrylock(lock);
1402 }
1403 
1404 /*
1405  * Slow path to release a rt-mutex.
1406  */
rt_mutex_slowunlock(struct rt_mutex_base * lock)1407 static void __sched rt_mutex_slowunlock(struct rt_mutex_base *lock)
1408 {
1409 	DEFINE_RT_WAKE_Q(wqh);
1410 	unsigned long flags;
1411 
1412 	/* irqsave required to support early boot calls */
1413 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
1414 
1415 	debug_rt_mutex_unlock(lock);
1416 
1417 	/*
1418 	 * We must be careful here if the fast path is enabled. If we
1419 	 * have no waiters queued we cannot set owner to NULL here
1420 	 * because of:
1421 	 *
1422 	 * foo->lock->owner = NULL;
1423 	 *			rtmutex_lock(foo->lock);   <- fast path
1424 	 *			free = atomic_dec_and_test(foo->refcnt);
1425 	 *			rtmutex_unlock(foo->lock); <- fast path
1426 	 *			if (free)
1427 	 *				kfree(foo);
1428 	 * raw_spin_unlock(foo->lock->wait_lock);
1429 	 *
1430 	 * So for the fastpath enabled kernel:
1431 	 *
1432 	 * Nothing can set the waiters bit as long as we hold
1433 	 * lock->wait_lock. So we do the following sequence:
1434 	 *
1435 	 *	owner = rt_mutex_owner(lock);
1436 	 *	clear_rt_mutex_waiters(lock);
1437 	 *	raw_spin_unlock(&lock->wait_lock);
1438 	 *	if (cmpxchg(&lock->owner, owner, 0) == owner)
1439 	 *		return;
1440 	 *	goto retry;
1441 	 *
1442 	 * The fastpath disabled variant is simple as all access to
1443 	 * lock->owner is serialized by lock->wait_lock:
1444 	 *
1445 	 *	lock->owner = NULL;
1446 	 *	raw_spin_unlock(&lock->wait_lock);
1447 	 */
1448 	while (!rt_mutex_has_waiters(lock)) {
1449 		/* Drops lock->wait_lock ! */
1450 		if (unlock_rt_mutex_safe(lock, flags) == true)
1451 			return;
1452 		/* Relock the rtmutex and try again */
1453 		raw_spin_lock_irqsave(&lock->wait_lock, flags);
1454 	}
1455 
1456 	/*
1457 	 * The wakeup next waiter path does not suffer from the above
1458 	 * race. See the comments there.
1459 	 *
1460 	 * Queue the next waiter for wakeup once we release the wait_lock.
1461 	 */
1462 	mark_wakeup_next_waiter(&wqh, lock);
1463 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1464 
1465 	rt_mutex_wake_up_q(&wqh);
1466 }
1467 
__rt_mutex_unlock(struct rt_mutex_base * lock)1468 static __always_inline void __rt_mutex_unlock(struct rt_mutex_base *lock)
1469 {
1470 	if (likely(rt_mutex_cmpxchg_release(lock, current, NULL)))
1471 		return;
1472 
1473 	rt_mutex_slowunlock(lock);
1474 }
1475 
1476 #ifdef CONFIG_SMP
rtmutex_spin_on_owner(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter,struct task_struct * owner)1477 static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1478 				  struct rt_mutex_waiter *waiter,
1479 				  struct task_struct *owner)
1480 {
1481 	bool res = true;
1482 
1483 	rcu_read_lock();
1484 	for (;;) {
1485 		/* If owner changed, trylock again. */
1486 		if (owner != rt_mutex_owner(lock))
1487 			break;
1488 		/*
1489 		 * Ensure that @owner is dereferenced after checking that
1490 		 * the lock owner still matches @owner. If that fails,
1491 		 * @owner might point to freed memory. If it still matches,
1492 		 * the rcu_read_lock() ensures the memory stays valid.
1493 		 */
1494 		barrier();
1495 		/*
1496 		 * Stop spinning when:
1497 		 *  - the lock owner has been scheduled out
1498 		 *  - current is not longer the top waiter
1499 		 *  - current is requested to reschedule (redundant
1500 		 *    for CONFIG_PREEMPT_RCU=y)
1501 		 *  - the VCPU on which owner runs is preempted
1502 		 */
1503 		if (!owner_on_cpu(owner) || need_resched() ||
1504 		    !rt_mutex_waiter_is_top_waiter(lock, waiter)) {
1505 			res = false;
1506 			break;
1507 		}
1508 		cpu_relax();
1509 	}
1510 	rcu_read_unlock();
1511 	return res;
1512 }
1513 #else
rtmutex_spin_on_owner(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter,struct task_struct * owner)1514 static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
1515 				  struct rt_mutex_waiter *waiter,
1516 				  struct task_struct *owner)
1517 {
1518 	return false;
1519 }
1520 #endif
1521 
1522 #ifdef RT_MUTEX_BUILD_MUTEX
1523 /*
1524  * Functions required for:
1525  *	- rtmutex, futex on all kernels
1526  *	- mutex and rwsem substitutions on RT kernels
1527  */
1528 
1529 /*
1530  * Remove a waiter from a lock and give up
1531  *
1532  * Must be called with lock->wait_lock held and interrupts disabled. It must
1533  * have just failed to try_to_take_rt_mutex().
1534  */
remove_waiter(struct rt_mutex_base * lock,struct rt_mutex_waiter * waiter)1535 static void __sched remove_waiter(struct rt_mutex_base *lock,
1536 				  struct rt_mutex_waiter *waiter)
1537 {
1538 	bool is_top_waiter = (waiter == rt_mutex_top_waiter(lock));
1539 	struct task_struct *owner = rt_mutex_owner(lock);
1540 	struct rt_mutex_base *next_lock;
1541 
1542 	lockdep_assert_held(&lock->wait_lock);
1543 
1544 	raw_spin_lock(&current->pi_lock);
1545 	rt_mutex_dequeue(lock, waiter);
1546 	current->pi_blocked_on = NULL;
1547 	raw_spin_unlock(&current->pi_lock);
1548 
1549 	/*
1550 	 * Only update priority if the waiter was the highest priority
1551 	 * waiter of the lock and there is an owner to update.
1552 	 */
1553 	if (!owner || !is_top_waiter)
1554 		return;
1555 
1556 	raw_spin_lock(&owner->pi_lock);
1557 
1558 	rt_mutex_dequeue_pi(owner, waiter);
1559 
1560 	if (rt_mutex_has_waiters(lock))
1561 		rt_mutex_enqueue_pi(owner, rt_mutex_top_waiter(lock));
1562 
1563 	rt_mutex_adjust_prio(lock, owner);
1564 
1565 	/* Store the lock on which owner is blocked or NULL */
1566 	next_lock = task_blocked_on_lock(owner);
1567 
1568 	raw_spin_unlock(&owner->pi_lock);
1569 
1570 	/*
1571 	 * Don't walk the chain, if the owner task is not blocked
1572 	 * itself.
1573 	 */
1574 	if (!next_lock)
1575 		return;
1576 
1577 	/* gets dropped in rt_mutex_adjust_prio_chain()! */
1578 	get_task_struct(owner);
1579 
1580 	raw_spin_unlock_irq(&lock->wait_lock);
1581 
1582 	rt_mutex_adjust_prio_chain(owner, RT_MUTEX_MIN_CHAINWALK, lock,
1583 				   next_lock, NULL, current);
1584 
1585 	raw_spin_lock_irq(&lock->wait_lock);
1586 }
1587 
1588 /**
1589  * rt_mutex_slowlock_block() - Perform the wait-wake-try-to-take loop
1590  * @lock:		 the rt_mutex to take
1591  * @ww_ctx:		 WW mutex context pointer
1592  * @state:		 the state the task should block in (TASK_INTERRUPTIBLE
1593  *			 or TASK_UNINTERRUPTIBLE)
1594  * @timeout:		 the pre-initialized and started timer, or NULL for none
1595  * @waiter:		 the pre-initialized rt_mutex_waiter
1596  *
1597  * Must be called with lock->wait_lock held and interrupts disabled
1598  */
rt_mutex_slowlock_block(struct rt_mutex_base * lock,struct ww_acquire_ctx * ww_ctx,unsigned int state,struct hrtimer_sleeper * timeout,struct rt_mutex_waiter * waiter)1599 static int __sched rt_mutex_slowlock_block(struct rt_mutex_base *lock,
1600 					   struct ww_acquire_ctx *ww_ctx,
1601 					   unsigned int state,
1602 					   struct hrtimer_sleeper *timeout,
1603 					   struct rt_mutex_waiter *waiter)
1604 {
1605 	struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1606 	struct task_struct *owner;
1607 	int ret = 0;
1608 
1609 	for (;;) {
1610 		/* Try to acquire the lock: */
1611 		if (try_to_take_rt_mutex(lock, current, waiter))
1612 			break;
1613 
1614 		if (timeout && !timeout->task) {
1615 			ret = -ETIMEDOUT;
1616 			break;
1617 		}
1618 		if (signal_pending_state(state, current)) {
1619 			ret = -EINTR;
1620 			break;
1621 		}
1622 
1623 		if (build_ww_mutex() && ww_ctx) {
1624 			ret = __ww_mutex_check_kill(rtm, waiter, ww_ctx);
1625 			if (ret)
1626 				break;
1627 		}
1628 
1629 		if (waiter == rt_mutex_top_waiter(lock))
1630 			owner = rt_mutex_owner(lock);
1631 		else
1632 			owner = NULL;
1633 		raw_spin_unlock_irq(&lock->wait_lock);
1634 
1635 		if (!owner || !rtmutex_spin_on_owner(lock, waiter, owner))
1636 			rt_mutex_schedule();
1637 
1638 		raw_spin_lock_irq(&lock->wait_lock);
1639 		set_current_state(state);
1640 	}
1641 
1642 	__set_current_state(TASK_RUNNING);
1643 	return ret;
1644 }
1645 
rt_mutex_handle_deadlock(int res,int detect_deadlock,struct rt_mutex_base * lock,struct rt_mutex_waiter * w)1646 static void __sched rt_mutex_handle_deadlock(int res, int detect_deadlock,
1647 					     struct rt_mutex_base *lock,
1648 					     struct rt_mutex_waiter *w)
1649 {
1650 	/*
1651 	 * If the result is not -EDEADLOCK or the caller requested
1652 	 * deadlock detection, nothing to do here.
1653 	 */
1654 	if (res != -EDEADLOCK || detect_deadlock)
1655 		return;
1656 
1657 	if (build_ww_mutex() && w->ww_ctx)
1658 		return;
1659 
1660 	raw_spin_unlock_irq(&lock->wait_lock);
1661 
1662 	WARN(1, "rtmutex deadlock detected\n");
1663 
1664 	while (1) {
1665 		set_current_state(TASK_INTERRUPTIBLE);
1666 		rt_mutex_schedule();
1667 	}
1668 }
1669 
1670 /**
1671  * __rt_mutex_slowlock - Locking slowpath invoked with lock::wait_lock held
1672  * @lock:	The rtmutex to block lock
1673  * @ww_ctx:	WW mutex context pointer
1674  * @state:	The task state for sleeping
1675  * @chwalk:	Indicator whether full or partial chainwalk is requested
1676  * @waiter:	Initializer waiter for blocking
1677  */
__rt_mutex_slowlock(struct rt_mutex_base * lock,struct ww_acquire_ctx * ww_ctx,unsigned int state,enum rtmutex_chainwalk chwalk,struct rt_mutex_waiter * waiter)1678 static int __sched __rt_mutex_slowlock(struct rt_mutex_base *lock,
1679 				       struct ww_acquire_ctx *ww_ctx,
1680 				       unsigned int state,
1681 				       enum rtmutex_chainwalk chwalk,
1682 				       struct rt_mutex_waiter *waiter)
1683 {
1684 	struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
1685 	struct ww_mutex *ww = ww_container_of(rtm);
1686 	int ret;
1687 
1688 	lockdep_assert_held(&lock->wait_lock);
1689 
1690 	/* Try to acquire the lock again: */
1691 	if (try_to_take_rt_mutex(lock, current, NULL)) {
1692 		if (build_ww_mutex() && ww_ctx) {
1693 			__ww_mutex_check_waiters(rtm, ww_ctx);
1694 			ww_mutex_lock_acquired(ww, ww_ctx);
1695 		}
1696 		return 0;
1697 	}
1698 
1699 	set_current_state(state);
1700 
1701 	trace_contention_begin(lock, LCB_F_RT);
1702 
1703 	ret = task_blocks_on_rt_mutex(lock, waiter, current, ww_ctx, chwalk);
1704 	if (likely(!ret))
1705 		ret = rt_mutex_slowlock_block(lock, ww_ctx, state, NULL, waiter);
1706 
1707 	if (likely(!ret)) {
1708 		/* acquired the lock */
1709 		if (build_ww_mutex() && ww_ctx) {
1710 			if (!ww_ctx->is_wait_die)
1711 				__ww_mutex_check_waiters(rtm, ww_ctx);
1712 			ww_mutex_lock_acquired(ww, ww_ctx);
1713 		}
1714 	} else {
1715 		__set_current_state(TASK_RUNNING);
1716 		remove_waiter(lock, waiter);
1717 		rt_mutex_handle_deadlock(ret, chwalk, lock, waiter);
1718 	}
1719 
1720 	/*
1721 	 * try_to_take_rt_mutex() sets the waiter bit
1722 	 * unconditionally. We might have to fix that up.
1723 	 */
1724 	fixup_rt_mutex_waiters(lock, true);
1725 
1726 	trace_contention_end(lock, ret);
1727 
1728 	return ret;
1729 }
1730 
__rt_mutex_slowlock_locked(struct rt_mutex_base * lock,struct ww_acquire_ctx * ww_ctx,unsigned int state)1731 static inline int __rt_mutex_slowlock_locked(struct rt_mutex_base *lock,
1732 					     struct ww_acquire_ctx *ww_ctx,
1733 					     unsigned int state)
1734 {
1735 	struct rt_mutex_waiter waiter;
1736 	int ret;
1737 
1738 	rt_mutex_init_waiter(&waiter);
1739 	waiter.ww_ctx = ww_ctx;
1740 
1741 	ret = __rt_mutex_slowlock(lock, ww_ctx, state, RT_MUTEX_MIN_CHAINWALK,
1742 				  &waiter);
1743 
1744 	debug_rt_mutex_free_waiter(&waiter);
1745 	return ret;
1746 }
1747 
1748 /*
1749  * rt_mutex_slowlock - Locking slowpath invoked when fast path fails
1750  * @lock:	The rtmutex to block lock
1751  * @ww_ctx:	WW mutex context pointer
1752  * @state:	The task state for sleeping
1753  */
rt_mutex_slowlock(struct rt_mutex_base * lock,struct ww_acquire_ctx * ww_ctx,unsigned int state)1754 static int __sched rt_mutex_slowlock(struct rt_mutex_base *lock,
1755 				     struct ww_acquire_ctx *ww_ctx,
1756 				     unsigned int state)
1757 {
1758 	unsigned long flags;
1759 	int ret;
1760 
1761 	/*
1762 	 * Do all pre-schedule work here, before we queue a waiter and invoke
1763 	 * PI -- any such work that trips on rtlock (PREEMPT_RT spinlock) would
1764 	 * otherwise recurse back into task_blocks_on_rt_mutex() through
1765 	 * rtlock_slowlock() and will then enqueue a second waiter for this
1766 	 * same task and things get really confusing real fast.
1767 	 */
1768 	rt_mutex_pre_schedule();
1769 
1770 	/*
1771 	 * Technically we could use raw_spin_[un]lock_irq() here, but this can
1772 	 * be called in early boot if the cmpxchg() fast path is disabled
1773 	 * (debug, no architecture support). In this case we will acquire the
1774 	 * rtmutex with lock->wait_lock held. But we cannot unconditionally
1775 	 * enable interrupts in that early boot case. So we need to use the
1776 	 * irqsave/restore variants.
1777 	 */
1778 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
1779 	ret = __rt_mutex_slowlock_locked(lock, ww_ctx, state);
1780 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1781 	rt_mutex_post_schedule();
1782 
1783 	return ret;
1784 }
1785 
__rt_mutex_lock(struct rt_mutex_base * lock,unsigned int state)1786 static __always_inline int __rt_mutex_lock(struct rt_mutex_base *lock,
1787 					   unsigned int state)
1788 {
1789 	lockdep_assert(!current->pi_blocked_on);
1790 
1791 	if (likely(rt_mutex_try_acquire(lock)))
1792 		return 0;
1793 
1794 	return rt_mutex_slowlock(lock, NULL, state);
1795 }
1796 #endif /* RT_MUTEX_BUILD_MUTEX */
1797 
1798 #ifdef RT_MUTEX_BUILD_SPINLOCKS
1799 /*
1800  * Functions required for spin/rw_lock substitution on RT kernels
1801  */
1802 
1803 /**
1804  * rtlock_slowlock_locked - Slow path lock acquisition for RT locks
1805  * @lock:	The underlying RT mutex
1806  */
rtlock_slowlock_locked(struct rt_mutex_base * lock)1807 static void __sched rtlock_slowlock_locked(struct rt_mutex_base *lock)
1808 {
1809 	struct rt_mutex_waiter waiter;
1810 	struct task_struct *owner;
1811 
1812 	lockdep_assert_held(&lock->wait_lock);
1813 
1814 	if (try_to_take_rt_mutex(lock, current, NULL))
1815 		return;
1816 
1817 	rt_mutex_init_rtlock_waiter(&waiter);
1818 
1819 	/* Save current state and set state to TASK_RTLOCK_WAIT */
1820 	current_save_and_set_rtlock_wait_state();
1821 
1822 	trace_contention_begin(lock, LCB_F_RT);
1823 
1824 	task_blocks_on_rt_mutex(lock, &waiter, current, NULL, RT_MUTEX_MIN_CHAINWALK);
1825 
1826 	for (;;) {
1827 		/* Try to acquire the lock again */
1828 		if (try_to_take_rt_mutex(lock, current, &waiter))
1829 			break;
1830 
1831 		if (&waiter == rt_mutex_top_waiter(lock))
1832 			owner = rt_mutex_owner(lock);
1833 		else
1834 			owner = NULL;
1835 		raw_spin_unlock_irq(&lock->wait_lock);
1836 
1837 		if (!owner || !rtmutex_spin_on_owner(lock, &waiter, owner))
1838 			schedule_rtlock();
1839 
1840 		raw_spin_lock_irq(&lock->wait_lock);
1841 		set_current_state(TASK_RTLOCK_WAIT);
1842 	}
1843 
1844 	/* Restore the task state */
1845 	current_restore_rtlock_saved_state();
1846 
1847 	/*
1848 	 * try_to_take_rt_mutex() sets the waiter bit unconditionally.
1849 	 * We might have to fix that up:
1850 	 */
1851 	fixup_rt_mutex_waiters(lock, true);
1852 	debug_rt_mutex_free_waiter(&waiter);
1853 
1854 	trace_contention_end(lock, 0);
1855 }
1856 
rtlock_slowlock(struct rt_mutex_base * lock)1857 static __always_inline void __sched rtlock_slowlock(struct rt_mutex_base *lock)
1858 {
1859 	unsigned long flags;
1860 
1861 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
1862 	rtlock_slowlock_locked(lock);
1863 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
1864 }
1865 
1866 #endif /* RT_MUTEX_BUILD_SPINLOCKS */
1867