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