kernel/locking/rtmutex.c

   1 /*
   2  * RT-Mutexes: simple blocking mutual exclusion locks with PI support
   3  *
   4  * started by Ingo Molnar and Thomas Gleixner.
   5  *
   6  *  Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
   7  *  Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
   8  *  Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
   9  *  Copyright (C) 2006 Esben Nielsen
  10  *
  11  *  See Documentation/rt-mutex-design.txt for details.
  12  */
  13 #include <linux/spinlock.h>
  14 #include <linux/export.h>
  15 #include <linux/sched.h>
  16 #include <linux/sched/rt.h>
  17 #include <linux/sched/deadline.h>
  18 #include <linux/timer.h>
  19
  20 #include "rtmutex_common.h"
  21
  22 /*
  23  * lock->owner state tracking:
  24  *
  25  * lock->owner holds the task_struct pointer of the owner. Bit 0
  26  * is used to keep track of the "lock has waiters" state.
  27  *
  28  * owner        bit0
  29  * NULL         0       lock is free (fast acquire possible)
  30  * NULL         1       lock is free and has waiters and the top waiter
  31  *                              is going to take the lock*
  32  * taskpointer  0       lock is held (fast release possible)
  33  * taskpointer  1       lock is held and has waiters**
  34  *
  35  * The fast atomic compare exchange based acquire and release is only
  36  * possible when bit 0 of lock->owner is 0.
  37  *
  38  * (*) It also can be a transitional state when grabbing the lock
  39  * with ->wait_lock is held. To prevent any fast path cmpxchg to the lock,
  40  * we need to set the bit0 before looking at the lock, and the owner may be
  41  * NULL in this small time, hence this can be a transitional state.
  42  *
  43  * (**) There is a small time when bit 0 is set but there are no
  44  * waiters. This can happen when grabbing the lock in the slow path.
  45  * To prevent a cmpxchg of the owner releasing the lock, we need to
  46  * set this bit before looking at the lock.
  47  */
  48
  49 static void
  50 rt_mutex_set_owner(struct rt_mutex *lock, struct task_struct *owner)
  51 {
  52         unsigned long val = (unsigned long)owner;
  53
  54         if (rt_mutex_has_waiters(lock))
  55                 val |= RT_MUTEX_HAS_WAITERS;
  56
  57         lock->owner = (struct task_struct *)val;
  58 }
  59
  60 static inline void clear_rt_mutex_waiters(struct rt_mutex *lock)
  61 {
  62         lock->owner = (struct task_struct *)
  63                         ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
  64 }
  65
  66 static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
  67 {
  68         if (!rt_mutex_has_waiters(lock))
  69                 clear_rt_mutex_waiters(lock);
  70 }
  71
  72 /*
  73  * We can speed up the acquire/release, if the architecture
  74  * supports cmpxchg and if there's no debugging state to be set up
  75  */
  76 #if defined(__HAVE_ARCH_CMPXCHG) && !defined(CONFIG_DEBUG_RT_MUTEXES)
  77 # define rt_mutex_cmpxchg(l,c,n)        (cmpxchg(&l->owner, c, n) == c)
  78 static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
  79 {
  80         unsigned long owner, *p = (unsigned long *) &lock->owner;
  81
  82         do {
  83                 owner = *p;
  84         } while (cmpxchg(p, owner, owner | RT_MUTEX_HAS_WAITERS) != owner);
  85 }
  86 #else
  87 # define rt_mutex_cmpxchg(l,c,n)        (0)
  88 static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
  89 {
  90         lock->owner = (struct task_struct *)
  91                         ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
  92 }
  93 #endif
  94
  95 static inline int
  96 rt_mutex_waiter_less(struct rt_mutex_waiter *left,
  97                      struct rt_mutex_waiter *right)
  98 {
  99         if (left->task->prio < right->task->prio)
 100                 return 1;
 101
 102         /*
 103          * If both tasks are dl_task(), we check their deadlines.
 104          */
 105         if (dl_prio(left->task->prio) && dl_prio(right->task->prio))
 106                 return (left->task->dl.deadline < right->task->dl.deadline);
 107
 108         return 0;
 109 }
 110
 111 static void
 112 rt_mutex_enqueue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
 113 {
 114         struct rb_node **link = &lock->waiters.rb_node;
 115         struct rb_node *parent = NULL;
 116         struct rt_mutex_waiter *entry;
 117         int leftmost = 1;
 118
 119         while (*link) {
 120                 parent = *link;
 121                 entry = rb_entry(parent, struct rt_mutex_waiter, tree_entry);
 122                 if (rt_mutex_waiter_less(waiter, entry)) {
 123                         link = &parent->rb_left;
 124                 } else {
 125                         link = &parent->rb_right;
 126                         leftmost = 0;
 127                 }
 128         }
 129
 130         if (leftmost)
 131                 lock->waiters_leftmost = &waiter->tree_entry;
 132
 133         rb_link_node(&waiter->tree_entry, parent, link);
 134         rb_insert_color(&waiter->tree_entry, &lock->waiters);
 135 }
 136
 137 static void
 138 rt_mutex_dequeue(struct rt_mutex *lock, struct rt_mutex_waiter *waiter)
 139 {
 140         if (RB_EMPTY_NODE(&waiter->tree_entry))
 141                 return;
 142
 143         if (lock->waiters_leftmost == &waiter->tree_entry)
 144                 lock->waiters_leftmost = rb_next(&waiter->tree_entry);
 145
 146         rb_erase(&waiter->tree_entry, &lock->waiters);
 147         RB_CLEAR_NODE(&waiter->tree_entry);
 148 }
 149
 150 static void
 151 rt_mutex_enqueue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
 152 {
 153         struct rb_node **link = &task->pi_waiters.rb_node;
 154         struct rb_node *parent = NULL;
 155         struct rt_mutex_waiter *entry;
 156         int leftmost = 1;
 157
 158         while (*link) {
 159                 parent = *link;
 160                 entry = rb_entry(parent, struct rt_mutex_waiter, pi_tree_entry);
 161                 if (rt_mutex_waiter_less(waiter, entry)) {
 162                         link = &parent->rb_left;
 163                 } else {
 164                         link = &parent->rb_right;
 165                         leftmost = 0;
 166                 }
 167         }
 168
 169         if (leftmost)
 170                 task->pi_waiters_leftmost = &waiter->pi_tree_entry;
 171
 172         rb_link_node(&waiter->pi_tree_entry, parent, link);
 173         rb_insert_color(&waiter->pi_tree_entry, &task->pi_waiters);
 174 }
 175
 176 static void
 177 rt_mutex_dequeue_pi(struct task_struct *task, struct rt_mutex_waiter *waiter)
 178 {
 179         if (RB_EMPTY_NODE(&waiter->pi_tree_entry))
 180                 return;
 181
 182         if (task->pi_waiters_leftmost == &waiter->pi_tree_entry)
 183                 task->pi_waiters_leftmost = rb_next(&waiter->pi_tree_entry);
 184
 185         rb_erase(&waiter->pi_tree_entry, &task->pi_waiters);
 186         RB_CLEAR_NODE(&waiter->pi_tree_entry);
 187 }
 188
 189 /*
 190  * Calculate task priority from the waiter tree priority
 191  *
 192  * Return task->normal_prio when the waiter tree is empty or when
 193  * the waiter is not allowed to do priority boosting
 194  */
 195 int rt_mutex_getprio(struct task_struct *task)
 196 {
 197         if (likely(!task_has_pi_waiters(task)))
 198                 return task->normal_prio;
 199
 200         return min(task_top_pi_waiter(task)->task->prio,
 201                    task->normal_prio);
 202 }
 203
 204 /*
 205  * Adjust the priority of a task, after its pi_waiters got modified.
 206  *
 207  * This can be both boosting and unboosting. task->pi_lock must be held.
 208  */
 209 static void __rt_mutex_adjust_prio(struct task_struct *task)
 210 {
 211         int prio = rt_mutex_getprio(task);
 212
 213         if (task->prio != prio)
 214                 rt_mutex_setprio(task, prio);
 215 }
 216
 217 /*
 218  * Adjust task priority (undo boosting). Called from the exit path of
 219  * rt_mutex_slowunlock() and rt_mutex_slowlock().
 220  *
 221  * (Note: We do this outside of the protection of lock->wait_lock to
 222  * allow the lock to be taken while or before we readjust the priority
 223  * of task. We do not use the spin_xx_mutex() variants here as we are
 224  * outside of the debug path.)
 225  */
 226 static void rt_mutex_adjust_prio(struct task_struct *task)
 227 {
 228         unsigned long flags;
 229
 230         raw_spin_lock_irqsave(&task->pi_lock, flags);
 231         __rt_mutex_adjust_prio(task);
 232         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 233 }
 234
 235 /*
 236  * Max number of times we'll walk the boosting chain:
 237  */
 238 int max_lock_depth = 1024;
 239
 240 /*
 241  * Adjust the priority chain. Also used for deadlock detection.
 242  * Decreases task's usage by one - may thus free the task.
 243  *
 244  * @task: the task owning the mutex (owner) for which a chain walk is probably
 245  *        needed
 246  * @deadlock_detect: do we have to carry out deadlock detection?
 247  * @orig_lock: the mutex (can be NULL if we are walking the chain to recheck
 248  *             things for a task that has just got its priority adjusted, and
 249  *             is waiting on a mutex)
 250  * @orig_waiter: rt_mutex_waiter struct for the task that has just donated
 251  *               its priority to the mutex owner (can be NULL in the case
 252  *               depicted above or if the top waiter is gone away and we are
 253  *               actually deboosting the owner)
 254  * @top_task: the current top waiter
 255  *
 256  * Returns 0 or -EDEADLK.
 257  */
 258 static int rt_mutex_adjust_prio_chain(struct task_struct *task,
 259                                       int deadlock_detect,
 260                                       struct rt_mutex *orig_lock,
 261                                       struct rt_mutex_waiter *orig_waiter,
 262                                       struct task_struct *top_task)
 263 {
 264         struct rt_mutex *lock;
 265         struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
 266         int detect_deadlock, ret = 0, depth = 0;
 267         unsigned long flags;
 268
 269         detect_deadlock = debug_rt_mutex_detect_deadlock(orig_waiter,
 270                                                          deadlock_detect);
 271
 272         /*
 273          * The (de)boosting is a step by step approach with a lot of
 274          * pitfalls. We want this to be preemptible and we want hold a
 275          * maximum of two locks per step. So we have to check
 276          * carefully whether things change under us.
 277          */
 278  again:
 279         if (++depth > max_lock_depth) {
 280                 static int prev_max;
 281
 282                 /*
 283                  * Print this only once. If the admin changes the limit,
 284                  * print a new message when reaching the limit again.
 285                  */
 286                 if (prev_max != max_lock_depth) {
 287                         prev_max = max_lock_depth;
 288                         printk(KERN_WARNING "Maximum lock depth %d reached "
 289                                "task: %s (%d)\n", max_lock_depth,
 290                                top_task->comm, task_pid_nr(top_task));
 291                 }
 292                 put_task_struct(task);
 293
 294                 return deadlock_detect ? -EDEADLK : 0;
 295         }
 296  retry:
 297         /*
 298          * Task can not go away as we did a get_task() before !
 299          */
 300         raw_spin_lock_irqsave(&task->pi_lock, flags);
 301
 302         waiter = task->pi_blocked_on;
 303         /*
 304          * Check whether the end of the boosting chain has been
 305          * reached or the state of the chain has changed while we
 306          * dropped the locks.
 307          */
 308         if (!waiter)
 309                 goto out_unlock_pi;
 310
 311         /*
 312          * Check the orig_waiter state. After we dropped the locks,
 313          * the previous owner of the lock might have released the lock.
 314          */
 315         if (orig_waiter && !rt_mutex_owner(orig_lock))
 316                 goto out_unlock_pi;
 317
 318         /*
 319          * Drop out, when the task has no waiters. Note,
 320          * top_waiter can be NULL, when we are in the deboosting
 321          * mode!
 322          */
 323         if (top_waiter && (!task_has_pi_waiters(task) ||
 324                            top_waiter != task_top_pi_waiter(task)))
 325                 goto out_unlock_pi;
 326
 327         /*
 328          * When deadlock detection is off then we check, if further
 329          * priority adjustment is necessary.
 330          */
 331         if (!detect_deadlock && waiter->task->prio == task->prio)
 332                 goto out_unlock_pi;
 333
 334         lock = waiter->lock;
 335         if (!raw_spin_trylock(&lock->wait_lock)) {
 336                 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 337                 cpu_relax();
 338                 goto retry;
 339         }
 340
 341         /* Deadlock detection */
 342         if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
 343                 debug_rt_mutex_deadlock(deadlock_detect, orig_waiter, lock);
 344                 raw_spin_unlock(&lock->wait_lock);
 345                 ret = deadlock_detect ? -EDEADLK : 0;
 346                 goto out_unlock_pi;
 347         }
 348
 349         top_waiter = rt_mutex_top_waiter(lock);
 350
 351         /* Requeue the waiter */
 352         rt_mutex_dequeue(lock, waiter);
 353         waiter->task->prio = task->prio;
 354         rt_mutex_enqueue(lock, waiter);
 355
 356         /* Release the task */
 357         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 358         if (!rt_mutex_owner(lock)) {
 359                 /*
 360                  * If the requeue above changed the top waiter, then we need
 361                  * to wake the new top waiter up to try to get the lock.
 362                  */
 363
 364                 if (top_waiter != rt_mutex_top_waiter(lock))
 365                         wake_up_process(rt_mutex_top_waiter(lock)->task);
 366                 raw_spin_unlock(&lock->wait_lock);
 367                 goto out_put_task;
 368         }
 369         put_task_struct(task);
 370
 371         /* Grab the next task */
 372         task = rt_mutex_owner(lock);
 373         get_task_struct(task);
 374         raw_spin_lock_irqsave(&task->pi_lock, flags);
 375
 376         if (waiter == rt_mutex_top_waiter(lock)) {
 377                 /* Boost the owner */
 378                 rt_mutex_dequeue_pi(task, top_waiter);
 379                 rt_mutex_enqueue_pi(task, waiter);
 380                 __rt_mutex_adjust_prio(task);
 381
 382         } else if (top_waiter == waiter) {
 383                 /* Deboost the owner */
 384                 rt_mutex_dequeue_pi(task, waiter);
 385                 waiter = rt_mutex_top_waiter(lock);
 386                 rt_mutex_enqueue_pi(task, waiter);
 387                 __rt_mutex_adjust_prio(task);
 388         }
 389
 390         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 391
 392         top_waiter = rt_mutex_top_waiter(lock);
 393         raw_spin_unlock(&lock->wait_lock);
 394
 395         if (!detect_deadlock && waiter != top_waiter)
 396                 goto out_put_task;
 397
 398         goto again;
 399
 400  out_unlock_pi:
 401         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 402  out_put_task:
 403         put_task_struct(task);
 404
 405         return ret;
 406 }
 407
 408 /*
 409  * Try to take an rt-mutex
 410  *
 411  * Must be called with lock->wait_lock held.
 412  *
 413  * @lock:   the lock to be acquired.
 414  * @task:   the task which wants to acquire the lock
 415  * @waiter: the waiter that is queued to the lock's wait list. (could be NULL)
 416  */
 417 static int try_to_take_rt_mutex(struct rt_mutex *lock, struct task_struct *task,
 418                 struct rt_mutex_waiter *waiter)
 419 {
 420         /*
 421          * We have to be careful here if the atomic speedups are
 422          * enabled, such that, when
 423          *  - no other waiter is on the lock
 424          *  - the lock has been released since we did the cmpxchg
 425          * the lock can be released or taken while we are doing the
 426          * checks and marking the lock with RT_MUTEX_HAS_WAITERS.
 427          *
 428          * The atomic acquire/release aware variant of
 429          * mark_rt_mutex_waiters uses a cmpxchg loop. After setting
 430          * the WAITERS bit, the atomic release / acquire can not
 431          * happen anymore and lock->wait_lock protects us from the
 432          * non-atomic case.
 433          *
 434          * Note, that this might set lock->owner =
 435          * RT_MUTEX_HAS_WAITERS in the case the lock is not contended
 436          * any more. This is fixed up when we take the ownership.
 437          * This is the transitional state explained at the top of this file.
 438          */
 439         mark_rt_mutex_waiters(lock);
 440
 441         if (rt_mutex_owner(lock))
 442                 return 0;
 443
 444         /*
 445          * It will get the lock because of one of these conditions:
 446          * 1) there is no waiter
 447          * 2) higher priority than waiters
 448          * 3) it is top waiter
 449          */
 450         if (rt_mutex_has_waiters(lock)) {
 451                 if (task->prio >= rt_mutex_top_waiter(lock)->task->prio) {
 452                         if (!waiter || waiter != rt_mutex_top_waiter(lock))
 453                                 return 0;
 454                 }
 455         }
 456
 457         if (waiter || rt_mutex_has_waiters(lock)) {
 458                 unsigned long flags;
 459                 struct rt_mutex_waiter *top;
 460
 461                 raw_spin_lock_irqsave(&task->pi_lock, flags);
 462
 463                 /* remove the queued waiter. */
 464                 if (waiter) {
 465                         rt_mutex_dequeue(lock, waiter);
 466                         task->pi_blocked_on = NULL;
 467                 }
 468
 469                 /*
 470                  * We have to enqueue the top waiter(if it exists) into
 471                  * task->pi_waiters list.
 472                  */
 473                 if (rt_mutex_has_waiters(lock)) {
 474                         top = rt_mutex_top_waiter(lock);
 475                         rt_mutex_enqueue_pi(task, top);
 476                 }
 477                 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 478         }
 479
 480         /* We got the lock. */
 481         debug_rt_mutex_lock(lock);
 482
 483         rt_mutex_set_owner(lock, task);
 484
 485         rt_mutex_deadlock_account_lock(lock, task);
 486
 487         return 1;
 488 }
 489
 490 /*
 491  * Task blocks on lock.
 492  *
 493  * Prepare waiter and propagate pi chain
 494  *
 495  * This must be called with lock->wait_lock held.
 496  */
 497 static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
 498                                    struct rt_mutex_waiter *waiter,
 499                                    struct task_struct *task,
 500                                    int detect_deadlock)
 501 {
 502         struct task_struct *owner = rt_mutex_owner(lock);
 503         struct rt_mutex_waiter *top_waiter = waiter;
 504         unsigned long flags;
 505         int chain_walk = 0, res;
 506
 507         raw_spin_lock_irqsave(&task->pi_lock, flags);
 508         __rt_mutex_adjust_prio(task);
 509         waiter->task = task;
 510         waiter->lock = lock;
 511
 512         /* Get the top priority waiter on the lock */
 513         if (rt_mutex_has_waiters(lock))
 514                 top_waiter = rt_mutex_top_waiter(lock);
 515         rt_mutex_enqueue(lock, waiter);
 516
 517         task->pi_blocked_on = waiter;
 518
 519         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 520
 521         if (!owner)
 522                 return 0;
 523
 524         if (waiter == rt_mutex_top_waiter(lock)) {
 525                 raw_spin_lock_irqsave(&owner->pi_lock, flags);
 526                 rt_mutex_dequeue_pi(owner, top_waiter);
 527                 rt_mutex_enqueue_pi(owner, waiter);
 528
 529                 __rt_mutex_adjust_prio(owner);
 530                 if (owner->pi_blocked_on)
 531                         chain_walk = 1;
 532                 raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
 533         }
 534         else if (debug_rt_mutex_detect_deadlock(waiter, detect_deadlock))
 535                 chain_walk = 1;
 536
 537         if (!chain_walk)
 538                 return 0;
 539
 540         /*
 541          * The owner can't disappear while holding a lock,
 542          * so the owner struct is protected by wait_lock.
 543          * Gets dropped in rt_mutex_adjust_prio_chain()!
 544          */
 545         get_task_struct(owner);
 546
 547         raw_spin_unlock(&lock->wait_lock);
 548
 549         res = rt_mutex_adjust_prio_chain(owner, detect_deadlock, lock, waiter,
 550                                          task);
 551
 552         raw_spin_lock(&lock->wait_lock);
 553
 554         return res;
 555 }
 556
 557 /*
 558  * Wake up the next waiter on the lock.
 559  *
 560  * Remove the top waiter from the current tasks waiter list and wake it up.
 561  *
 562  * Called with lock->wait_lock held.
 563  */
 564 static void wakeup_next_waiter(struct rt_mutex *lock)
 565 {
 566         struct rt_mutex_waiter *waiter;
 567         unsigned long flags;
 568
 569         raw_spin_lock_irqsave(&current->pi_lock, flags);
 570
 571         waiter = rt_mutex_top_waiter(lock);
 572
 573         /*
 574          * Remove it from current->pi_waiters. We do not adjust a
 575          * possible priority boost right now. We execute wakeup in the
 576          * boosted mode and go back to normal after releasing
 577          * lock->wait_lock.
 578          */
 579         rt_mutex_dequeue_pi(current, waiter);
 580
 581         rt_mutex_set_owner(lock, NULL);
 582
 583         raw_spin_unlock_irqrestore(&current->pi_lock, flags);
 584
 585         wake_up_process(waiter->task);
 586 }
 587
 588 /*
 589  * Remove a waiter from a lock and give up
 590  *
 591  * Must be called with lock->wait_lock held and
 592  * have just failed to try_to_take_rt_mutex().
 593  */
 594 static void remove_waiter(struct rt_mutex *lock,
 595                           struct rt_mutex_waiter *waiter)
 596 {
 597         int first = (waiter == rt_mutex_top_waiter(lock));
 598         struct task_struct *owner = rt_mutex_owner(lock);
 599         unsigned long flags;
 600         int chain_walk = 0;
 601
 602         raw_spin_lock_irqsave(&current->pi_lock, flags);
 603         rt_mutex_dequeue(lock, waiter);
 604         current->pi_blocked_on = NULL;
 605         raw_spin_unlock_irqrestore(&current->pi_lock, flags);
 606
 607         if (!owner)
 608                 return;
 609
 610         if (first) {
 611
 612                 raw_spin_lock_irqsave(&owner->pi_lock, flags);
 613
 614                 rt_mutex_dequeue_pi(owner, waiter);
 615
 616                 if (rt_mutex_has_waiters(lock)) {
 617                         struct rt_mutex_waiter *next;
 618
 619                         next = rt_mutex_top_waiter(lock);
 620                         rt_mutex_enqueue_pi(owner, next);
 621                 }
 622                 __rt_mutex_adjust_prio(owner);
 623
 624                 if (owner->pi_blocked_on)
 625                         chain_walk = 1;
 626
 627                 raw_spin_unlock_irqrestore(&owner->pi_lock, flags);
 628         }
 629
 630         if (!chain_walk)
 631                 return;
 632
 633         /* gets dropped in rt_mutex_adjust_prio_chain()! */
 634         get_task_struct(owner);
 635
 636         raw_spin_unlock(&lock->wait_lock);
 637
 638         rt_mutex_adjust_prio_chain(owner, 0, lock, NULL, current);
 639
 640         raw_spin_lock(&lock->wait_lock);
 641 }
 642
 643 /*
 644  * Recheck the pi chain, in case we got a priority setting
 645  *
 646  * Called from sched_setscheduler
 647  */
 648 void rt_mutex_adjust_pi(struct task_struct *task)
 649 {
 650         struct rt_mutex_waiter *waiter;
 651         unsigned long flags;
 652
 653         raw_spin_lock_irqsave(&task->pi_lock, flags);
 654
 655         waiter = task->pi_blocked_on;
 656         if (!waiter || waiter->task->prio == task->prio) {
 657                 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 658                 return;
 659         }
 660
 661         raw_spin_unlock_irqrestore(&task->pi_lock, flags);
 662
 663         /* gets dropped in rt_mutex_adjust_prio_chain()! */
 664         get_task_struct(task);
 665         rt_mutex_adjust_prio_chain(task, 0, NULL, NULL, task);
 666 }
 667
 668 /**
 669  * __rt_mutex_slowlock() - Perform the wait-wake-try-to-take loop
 670  * @lock:                the rt_mutex to take
 671  * @state:               the state the task should block in (TASK_INTERRUPTIBLE
 672  *                       or TASK_UNINTERRUPTIBLE)
 673  * @timeout:             the pre-initialized and started timer, or NULL for none
 674  * @waiter:              the pre-initialized rt_mutex_waiter
 675  *
 676  * lock->wait_lock must be held by the caller.
 677  */
 678 static int __sched
 679 __rt_mutex_slowlock(struct rt_mutex *lock, int state,
 680                     struct hrtimer_sleeper *timeout,
 681                     struct rt_mutex_waiter *waiter)
 682 {
 683         int ret = 0;
 684
 685         for (;;) {
 686                 /* Try to acquire the lock: */
 687                 if (try_to_take_rt_mutex(lock, current, waiter))
 688                         break;
 689
 690                 /*
 691                  * TASK_INTERRUPTIBLE checks for signals and
 692                  * timeout. Ignored otherwise.
 693                  */
 694                 if (unlikely(state == TASK_INTERRUPTIBLE)) {
 695                         /* Signal pending? */
 696                         if (signal_pending(current))
 697                                 ret = -EINTR;
 698                         if (timeout && !timeout->task)
 699                                 ret = -ETIMEDOUT;
 700                         if (ret)
 701                                 break;
 702                 }
 703
 704                 raw_spin_unlock(&lock->wait_lock);
 705
 706                 debug_rt_mutex_print_deadlock(waiter);
 707
 708                 schedule_rt_mutex(lock);
 709
 710                 raw_spin_lock(&lock->wait_lock);
 711                 set_current_state(state);
 712         }
 713
 714         return ret;
 715 }
 716
 717 /*
 718  * Slow path lock function:
 719  */
 720 static int __sched
 721 rt_mutex_slowlock(struct rt_mutex *lock, int state,
 722                   struct hrtimer_sleeper *timeout,
 723                   int detect_deadlock)
 724 {
 725         struct rt_mutex_waiter waiter;
 726         int ret = 0;
 727
 728         debug_rt_mutex_init_waiter(&waiter);
 729         RB_CLEAR_NODE(&waiter.pi_tree_entry);
 730         RB_CLEAR_NODE(&waiter.tree_entry);
 731
 732         raw_spin_lock(&lock->wait_lock);
 733
 734         /* Try to acquire the lock again: */
 735         if (try_to_take_rt_mutex(lock, current, NULL)) {
 736                 raw_spin_unlock(&lock->wait_lock);
 737                 return 0;
 738         }
 739
 740         set_current_state(state);
 741
 742         /* Setup the timer, when timeout != NULL */
 743         if (unlikely(timeout)) {
 744                 hrtimer_start_expires(&timeout->timer, HRTIMER_MODE_ABS);
 745                 if (!hrtimer_active(&timeout->timer))
 746                         timeout->task = NULL;
 747         }
 748
 749         ret = task_blocks_on_rt_mutex(lock, &waiter, current, detect_deadlock);
 750
 751         if (likely(!ret))
 752                 ret = __rt_mutex_slowlock(lock, state, timeout, &waiter);
 753
 754         set_current_state(TASK_RUNNING);
 755
 756         if (unlikely(ret))
 757                 remove_waiter(lock, &waiter);
 758
 759         /*
 760          * try_to_take_rt_mutex() sets the waiter bit
 761          * unconditionally. We might have to fix that up.
 762          */
 763         fixup_rt_mutex_waiters(lock);
 764
 765         raw_spin_unlock(&lock->wait_lock);
 766
 767         /* Remove pending timer: */
 768         if (unlikely(timeout))
 769                 hrtimer_cancel(&timeout->timer);
 770
 771         debug_rt_mutex_free_waiter(&waiter);
 772
 773         return ret;
 774 }
 775
 776 /*
 777  * Slow path try-lock function:
 778  */
 779 static inline int
 780 rt_mutex_slowtrylock(struct rt_mutex *lock)
 781 {
 782         int ret = 0;
 783
 784         raw_spin_lock(&lock->wait_lock);
 785
 786         if (likely(rt_mutex_owner(lock) != current)) {
 787
 788                 ret = try_to_take_rt_mutex(lock, current, NULL);
 789                 /*
 790                  * try_to_take_rt_mutex() sets the lock waiters
 791                  * bit unconditionally. Clean this up.
 792                  */
 793                 fixup_rt_mutex_waiters(lock);
 794         }
 795
 796         raw_spin_unlock(&lock->wait_lock);
 797
 798         return ret;
 799 }
 800
 801 /*
 802  * Slow path to release a rt-mutex:
 803  */
 804 static void __sched
 805 rt_mutex_slowunlock(struct rt_mutex *lock)
 806 {
 807         raw_spin_lock(&lock->wait_lock);
 808
 809         debug_rt_mutex_unlock(lock);
 810
 811         rt_mutex_deadlock_account_unlock(current);
 812
 813         if (!rt_mutex_has_waiters(lock)) {
 814                 lock->owner = NULL;
 815                 raw_spin_unlock(&lock->wait_lock);
 816                 return;
 817         }
 818
 819         wakeup_next_waiter(lock);
 820
 821         raw_spin_unlock(&lock->wait_lock);
 822
 823         /* Undo pi boosting if necessary: */
 824         rt_mutex_adjust_prio(current);
 825 }
 826
 827 /*
 828  * debug aware fast / slowpath lock,trylock,unlock
 829  *
 830  * The atomic acquire/release ops are compiled away, when either the
 831  * architecture does not support cmpxchg or when debugging is enabled.
 832  */
 833 static inline int
 834 rt_mutex_fastlock(struct rt_mutex *lock, int state,
 835                   int detect_deadlock,
 836                   int (*slowfn)(struct rt_mutex *lock, int state,
 837                                 struct hrtimer_sleeper *timeout,
 838                                 int detect_deadlock))
 839 {
 840         if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
 841                 rt_mutex_deadlock_account_lock(lock, current);
 842                 return 0;
 843         } else
 844                 return slowfn(lock, state, NULL, detect_deadlock);
 845 }
 846
 847 static inline int
 848 rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
 849                         struct hrtimer_sleeper *timeout, int detect_deadlock,
 850                         int (*slowfn)(struct rt_mutex *lock, int state,
 851                                       struct hrtimer_sleeper *timeout,
 852                                       int detect_deadlock))
 853 {
 854         if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
 855                 rt_mutex_deadlock_account_lock(lock, current);
 856                 return 0;
 857         } else
 858                 return slowfn(lock, state, timeout, detect_deadlock);
 859 }
 860
 861 static inline int
 862 rt_mutex_fasttrylock(struct rt_mutex *lock,
 863                      int (*slowfn)(struct rt_mutex *lock))
 864 {
 865         if (likely(rt_mutex_cmpxchg(lock, NULL, current))) {
 866                 rt_mutex_deadlock_account_lock(lock, current);
 867                 return 1;
 868         }
 869         return slowfn(lock);
 870 }
 871
 872 static inline void
 873 rt_mutex_fastunlock(struct rt_mutex *lock,
 874                     void (*slowfn)(struct rt_mutex *lock))
 875 {
 876         if (likely(rt_mutex_cmpxchg(lock, current, NULL)))
 877                 rt_mutex_deadlock_account_unlock(current);
 878         else
 879                 slowfn(lock);
 880 }
 881
 882 /**
 883  * rt_mutex_lock - lock a rt_mutex
 884  *
 885  * @lock: the rt_mutex to be locked
 886  */
 887 void __sched rt_mutex_lock(struct rt_mutex *lock)
 888 {
 889         might_sleep();
 890
 891         rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, 0, rt_mutex_slowlock);
 892 }
 893 EXPORT_SYMBOL_GPL(rt_mutex_lock);
 894
 895 /**
 896  * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
 897  *
 898  * @lock:               the rt_mutex to be locked
 899  * @detect_deadlock:    deadlock detection on/off
 900  *
 901  * Returns:
 902  *  0           on success
 903  * -EINTR       when interrupted by a signal
 904  * -EDEADLK     when the lock would deadlock (when deadlock detection is on)
 905  */
 906 int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock,
 907                                                  int detect_deadlock)
 908 {
 909         might_sleep();
 910
 911         return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE,
 912                                  detect_deadlock, rt_mutex_slowlock);
 913 }
 914 EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
 915
 916 /**
 917  * rt_mutex_timed_lock - lock a rt_mutex interruptible
 918  *                      the timeout structure is provided
 919  *                      by the caller
 920  *
 921  * @lock:               the rt_mutex to be locked
 922  * @timeout:            timeout structure or NULL (no timeout)
 923  * @detect_deadlock:    deadlock detection on/off
 924  *
 925  * Returns:
 926  *  0           on success
 927  * -EINTR       when interrupted by a signal
 928  * -ETIMEDOUT   when the timeout expired
 929  * -EDEADLK     when the lock would deadlock (when deadlock detection is on)
 930  */
 931 int
 932 rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout,
 933                     int detect_deadlock)
 934 {
 935         might_sleep();
 936
 937         return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
 938                                        detect_deadlock, rt_mutex_slowlock);
 939 }
 940 EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
 941
 942 /**
 943  * rt_mutex_trylock - try to lock a rt_mutex
 944  *
 945  * @lock:       the rt_mutex to be locked
 946  *
 947  * Returns 1 on success and 0 on contention
 948  */
 949 int __sched rt_mutex_trylock(struct rt_mutex *lock)
 950 {
 951         return rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
 952 }
 953 EXPORT_SYMBOL_GPL(rt_mutex_trylock);
 954
 955 /**
 956  * rt_mutex_unlock - unlock a rt_mutex
 957  *
 958  * @lock: the rt_mutex to be unlocked
 959  */
 960 void __sched rt_mutex_unlock(struct rt_mutex *lock)
 961 {
 962         rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
 963 }
 964 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
 965
 966 /**
 967  * rt_mutex_destroy - mark a mutex unusable
 968  * @lock: the mutex to be destroyed
 969  *
 970  * This function marks the mutex uninitialized, and any subsequent
 971  * use of the mutex is forbidden. The mutex must not be locked when
 972  * this function is called.
 973  */
 974 void rt_mutex_destroy(struct rt_mutex *lock)
 975 {
 976         WARN_ON(rt_mutex_is_locked(lock));
 977 #ifdef CONFIG_DEBUG_RT_MUTEXES
 978         lock->magic = NULL;
 979 #endif
 980 }
 981
 982 EXPORT_SYMBOL_GPL(rt_mutex_destroy);
 983
 984 /**
 985  * __rt_mutex_init - initialize the rt lock
 986  *
 987  * @lock: the rt lock to be initialized
 988  *
 989  * Initialize the rt lock to unlocked state.
 990  *
 991  * Initializing of a locked rt lock is not allowed
 992  */
 993 void __rt_mutex_init(struct rt_mutex *lock, const char *name)
 994 {
 995         lock->owner = NULL;
 996         raw_spin_lock_init(&lock->wait_lock);
 997         lock->waiters = RB_ROOT;
 998         lock->waiters_leftmost = NULL;
 999
1000         debug_rt_mutex_init(lock, name);
1001 }
1002 EXPORT_SYMBOL_GPL(__rt_mutex_init);
1003
1004 /**
1005  * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
1006  *                              proxy owner
1007  *
1008  * @lock:       the rt_mutex to be locked
1009  * @proxy_owner:the task to set as owner
1010  *
1011  * No locking. Caller has to do serializing itself
1012  * Special API call for PI-futex support
1013  */
1014 void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
1015                                 struct task_struct *proxy_owner)
1016 {
1017         __rt_mutex_init(lock, NULL);
1018         debug_rt_mutex_proxy_lock(lock, proxy_owner);
1019         rt_mutex_set_owner(lock, proxy_owner);
1020         rt_mutex_deadlock_account_lock(lock, proxy_owner);
1021 }
1022
1023 /**
1024  * rt_mutex_proxy_unlock - release a lock on behalf of owner
1025  *
1026  * @lock:       the rt_mutex to be locked
1027  *
1028  * No locking. Caller has to do serializing itself
1029  * Special API call for PI-futex support
1030  */
1031 void rt_mutex_proxy_unlock(struct rt_mutex *lock,
1032                            struct task_struct *proxy_owner)
1033 {
1034         debug_rt_mutex_proxy_unlock(lock);
1035         rt_mutex_set_owner(lock, NULL);
1036         rt_mutex_deadlock_account_unlock(proxy_owner);
1037 }
1038
1039 /**
1040  * rt_mutex_start_proxy_lock() - Start lock acquisition for another task
1041  * @lock:               the rt_mutex to take
1042  * @waiter:             the pre-initialized rt_mutex_waiter
1043  * @task:               the task to prepare
1044  * @detect_deadlock:    perform deadlock detection (1) or not (0)
1045  *
1046  * Returns:
1047  *  0 - task blocked on lock
1048  *  1 - acquired the lock for task, caller should wake it up
1049  * <0 - error
1050  *
1051  * Special API call for FUTEX_REQUEUE_PI support.
1052  */
1053 int rt_mutex_start_proxy_lock(struct rt_mutex *lock,
1054                               struct rt_mutex_waiter *waiter,
1055                               struct task_struct *task, int detect_deadlock)
1056 {
1057         int ret;
1058
1059         raw_spin_lock(&lock->wait_lock);
1060
1061         if (try_to_take_rt_mutex(lock, task, NULL)) {
1062                 raw_spin_unlock(&lock->wait_lock);
1063                 return 1;
1064         }
1065
1066         ret = task_blocks_on_rt_mutex(lock, waiter, task, detect_deadlock);
1067
1068         if (ret && !rt_mutex_owner(lock)) {
1069                 /*
1070                  * Reset the return value. We might have
1071                  * returned with -EDEADLK and the owner
1072                  * released the lock while we were walking the
1073                  * pi chain.  Let the waiter sort it out.
1074                  */
1075                 ret = 0;
1076         }
1077
1078         if (unlikely(ret))
1079                 remove_waiter(lock, waiter);
1080
1081         raw_spin_unlock(&lock->wait_lock);
1082
1083         debug_rt_mutex_print_deadlock(waiter);
1084
1085         return ret;
1086 }
1087
1088 /**
1089  * rt_mutex_next_owner - return the next owner of the lock
1090  *
1091  * @lock: the rt lock query
1092  *
1093  * Returns the next owner of the lock or NULL
1094  *
1095  * Caller has to serialize against other accessors to the lock
1096  * itself.
1097  *
1098  * Special API call for PI-futex support
1099  */
1100 struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock)
1101 {
1102         if (!rt_mutex_has_waiters(lock))
1103                 return NULL;
1104
1105         return rt_mutex_top_waiter(lock)->task;
1106 }
1107
1108 /**
1109  * rt_mutex_finish_proxy_lock() - Complete lock acquisition
1110  * @lock:               the rt_mutex we were woken on
1111  * @to:                 the timeout, null if none. hrtimer should already have
1112  *                      been started.
1113  * @waiter:             the pre-initialized rt_mutex_waiter
1114  * @detect_deadlock:    perform deadlock detection (1) or not (0)
1115  *
1116  * Complete the lock acquisition started our behalf by another thread.
1117  *
1118  * Returns:
1119  *  0 - success
1120  * <0 - error, one of -EINTR, -ETIMEDOUT, or -EDEADLK
1121  *
1122  * Special API call for PI-futex requeue support
1123  */
1124 int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
1125                                struct hrtimer_sleeper *to,
1126                                struct rt_mutex_waiter *waiter,
1127                                int detect_deadlock)
1128 {
1129         int ret;
1130
1131         raw_spin_lock(&lock->wait_lock);
1132
1133         set_current_state(TASK_INTERRUPTIBLE);
1134
1135         ret = __rt_mutex_slowlock(lock, TASK_INTERRUPTIBLE, to, waiter);
1136
1137         set_current_state(TASK_RUNNING);
1138
1139         if (unlikely(ret))
1140                 remove_waiter(lock, waiter);
1141
1142         /*
1143          * try_to_take_rt_mutex() sets the waiter bit unconditionally. We might
1144          * have to fix that up.
1145          */
1146         fixup_rt_mutex_waiters(lock);
1147
1148         raw_spin_unlock(&lock->wait_lock);
1149
1150         return ret;
1151 }