This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / kernel / rtmutex.c
diff --git a/kernel/rtmutex.c b/kernel/rtmutex.c
new file mode 100644 (file)
index 0000000..3e13a1e
--- /dev/null
@@ -0,0 +1,991 @@
+/*
+ * RT-Mutexes: simple blocking mutual exclusion locks with PI support
+ *
+ * started by Ingo Molnar and Thomas Gleixner.
+ *
+ *  Copyright (C) 2004-2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
+ *  Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com>
+ *  Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
+ *  Copyright (C) 2006 Esben Nielsen
+ *
+ *  See Documentation/rt-mutex-design.txt for details.
+ */
+#include <linux/spinlock.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/timer.h>
+
+#include "rtmutex_common.h"
+
+#ifdef CONFIG_DEBUG_RT_MUTEXES
+# include "rtmutex-debug.h"
+#else
+# include "rtmutex.h"
+#endif
+
+/*
+ * lock->owner state tracking:
+ *
+ * lock->owner holds the task_struct pointer of the owner. Bit 0 and 1
+ * are used to keep track of the "owner is pending" and "lock has
+ * waiters" state.
+ *
+ * owner       bit1    bit0
+ * NULL                0       0       lock is free (fast acquire possible)
+ * NULL                0       1       invalid state
+ * NULL                1       0       Transitional State*
+ * NULL                1       1       invalid state
+ * taskpointer 0       0       lock is held (fast release possible)
+ * taskpointer 0       1       task is pending owner
+ * taskpointer 1       0       lock is held and has waiters
+ * taskpointer 1       1       task is pending owner and lock has more waiters
+ *
+ * Pending ownership is assigned to the top (highest priority)
+ * waiter of the lock, when the lock is released. The thread is woken
+ * up and can now take the lock. Until the lock is taken (bit 0
+ * cleared) a competing higher priority thread can steal the lock
+ * which puts the woken up thread back on the waiters list.
+ *
+ * The fast atomic compare exchange based acquire and release is only
+ * possible when bit 0 and 1 of lock->owner are 0.
+ *
+ * (*) There's a small time where the owner can be NULL and the
+ * "lock has waiters" bit is set.  This can happen when grabbing the lock.
+ * To prevent a cmpxchg of the owner releasing the lock, we need to set this
+ * bit before looking at the lock, hence the reason this is a transitional
+ * state.
+ */
+
+static void
+rt_mutex_set_owner(struct rt_mutex *lock, struct task_struct *owner,
+                  unsigned long mask)
+{
+       unsigned long val = (unsigned long)owner | mask;
+
+       if (rt_mutex_has_waiters(lock))
+               val |= RT_MUTEX_HAS_WAITERS;
+
+       lock->owner = (struct task_struct *)val;
+}
+
+static inline void clear_rt_mutex_waiters(struct rt_mutex *lock)
+{
+       lock->owner = (struct task_struct *)
+                       ((unsigned long)lock->owner & ~RT_MUTEX_HAS_WAITERS);
+}
+
+static void fixup_rt_mutex_waiters(struct rt_mutex *lock)
+{
+       if (!rt_mutex_has_waiters(lock))
+               clear_rt_mutex_waiters(lock);
+}
+
+/*
+ * We can speed up the acquire/release, if the architecture
+ * supports cmpxchg and if there's no debugging state to be set up
+ */
+#if defined(__HAVE_ARCH_CMPXCHG) && !defined(CONFIG_DEBUG_RT_MUTEXES)
+# define rt_mutex_cmpxchg(l,c,n)       (cmpxchg(&l->owner, c, n) == c)
+static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
+{
+       unsigned long owner, *p = (unsigned long *) &lock->owner;
+
+       do {
+               owner = *p;
+       } while (cmpxchg(p, owner, owner | RT_MUTEX_HAS_WAITERS) != owner);
+}
+#else
+# define rt_mutex_cmpxchg(l,c,n)       (0)
+static inline void mark_rt_mutex_waiters(struct rt_mutex *lock)
+{
+       lock->owner = (struct task_struct *)
+                       ((unsigned long)lock->owner | RT_MUTEX_HAS_WAITERS);
+}
+#endif
+
+/*
+ * Calculate task priority from the waiter list priority
+ *
+ * Return task->normal_prio when the waiter list is empty or when
+ * the waiter is not allowed to do priority boosting
+ */
+int rt_mutex_getprio(struct task_struct *task)
+{
+       if (likely(!task_has_pi_waiters(task)))
+               return task->normal_prio;
+
+       return min(task_top_pi_waiter(task)->pi_list_entry.prio,
+                  task->normal_prio);
+}
+
+/*
+ * Adjust the priority of a task, after its pi_waiters got modified.
+ *
+ * This can be both boosting and unboosting. task->pi_lock must be held.
+ */
+static void __rt_mutex_adjust_prio(struct task_struct *task)
+{
+       int prio = rt_mutex_getprio(task);
+
+       if (task->prio != prio)
+               rt_mutex_setprio(task, prio);
+}
+
+/*
+ * Adjust task priority (undo boosting). Called from the exit path of
+ * rt_mutex_slowunlock() and rt_mutex_slowlock().
+ *
+ * (Note: We do this outside of the protection of lock->wait_lock to
+ * allow the lock to be taken while or before we readjust the priority
+ * of task. We do not use the spin_xx_mutex() variants here as we are
+ * outside of the debug path.)
+ */
+static void rt_mutex_adjust_prio(struct task_struct *task)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&task->pi_lock, flags);
+       __rt_mutex_adjust_prio(task);
+       spin_unlock_irqrestore(&task->pi_lock, flags);
+}
+
+/*
+ * Max number of times we'll walk the boosting chain:
+ */
+int max_lock_depth = 1024;
+
+/*
+ * Adjust the priority chain. Also used for deadlock detection.
+ * Decreases task's usage by one - may thus free the task.
+ * Returns 0 or -EDEADLK.
+ */
+static int rt_mutex_adjust_prio_chain(struct task_struct *task,
+                                     int deadlock_detect,
+                                     struct rt_mutex *orig_lock,
+                                     struct rt_mutex_waiter *orig_waiter,
+                                     struct task_struct *top_task)
+{
+       struct rt_mutex *lock;
+       struct rt_mutex_waiter *waiter, *top_waiter = orig_waiter;
+       int detect_deadlock, ret = 0, depth = 0;
+       unsigned long flags;
+
+       detect_deadlock = debug_rt_mutex_detect_deadlock(orig_waiter,
+                                                        deadlock_detect);
+
+       /*
+        * The (de)boosting is a step by step approach with a lot of
+        * pitfalls. We want this to be preemptible and we want hold a
+        * maximum of two locks per step. So we have to check
+        * carefully whether things change under us.
+        */
+ again:
+       if (++depth > max_lock_depth) {
+               static int prev_max;
+
+               /*
+                * Print this only once. If the admin changes the limit,
+                * print a new message when reaching the limit again.
+                */
+               if (prev_max != max_lock_depth) {
+                       prev_max = max_lock_depth;
+                       printk(KERN_WARNING "Maximum lock depth %d reached "
+                              "task: %s (%d)\n", max_lock_depth,
+                              top_task->comm, top_task->pid);
+               }
+               put_task_struct(task);
+
+               return deadlock_detect ? -EDEADLK : 0;
+       }
+ retry:
+       /*
+        * Task can not go away as we did a get_task() before !
+        */
+       spin_lock_irqsave(&task->pi_lock, flags);
+
+       waiter = task->pi_blocked_on;
+       /*
+        * Check whether the end of the boosting chain has been
+        * reached or the state of the chain has changed while we
+        * dropped the locks.
+        */
+       if (!waiter || !waiter->task)
+               goto out_unlock_pi;
+
+       if (top_waiter && (!task_has_pi_waiters(task) ||
+                          top_waiter != task_top_pi_waiter(task)))
+               goto out_unlock_pi;
+
+       /*
+        * When deadlock detection is off then we check, if further
+        * priority adjustment is necessary.
+        */
+       if (!detect_deadlock && waiter->list_entry.prio == task->prio)
+               goto out_unlock_pi;
+
+       lock = waiter->lock;
+       if (!spin_trylock(&lock->wait_lock)) {
+               spin_unlock_irqrestore(&task->pi_lock, flags);
+               cpu_relax();
+               goto retry;
+       }
+
+       /* Deadlock detection */
+       if (lock == orig_lock || rt_mutex_owner(lock) == top_task) {
+               debug_rt_mutex_deadlock(deadlock_detect, orig_waiter, lock);
+               spin_unlock(&lock->wait_lock);
+               ret = deadlock_detect ? -EDEADLK : 0;
+               goto out_unlock_pi;
+       }
+
+       top_waiter = rt_mutex_top_waiter(lock);
+
+       /* Requeue the waiter */
+       plist_del(&waiter->list_entry, &lock->wait_list);
+       waiter->list_entry.prio = task->prio;
+       plist_add(&waiter->list_entry, &lock->wait_list);
+
+       /* Release the task */
+       spin_unlock_irqrestore(&task->pi_lock, flags);
+       put_task_struct(task);
+
+       /* Grab the next task */
+       task = rt_mutex_owner(lock);
+       spin_lock_irqsave(&task->pi_lock, flags);
+
+       if (waiter == rt_mutex_top_waiter(lock)) {
+               /* Boost the owner */
+               plist_del(&top_waiter->pi_list_entry, &task->pi_waiters);
+               waiter->pi_list_entry.prio = waiter->list_entry.prio;
+               plist_add(&waiter->pi_list_entry, &task->pi_waiters);
+               __rt_mutex_adjust_prio(task);
+
+       } else if (top_waiter == waiter) {
+               /* Deboost the owner */
+               plist_del(&waiter->pi_list_entry, &task->pi_waiters);
+               waiter = rt_mutex_top_waiter(lock);
+               waiter->pi_list_entry.prio = waiter->list_entry.prio;
+               plist_add(&waiter->pi_list_entry, &task->pi_waiters);
+               __rt_mutex_adjust_prio(task);
+       }
+
+       get_task_struct(task);
+       spin_unlock_irqrestore(&task->pi_lock, flags);
+
+       top_waiter = rt_mutex_top_waiter(lock);
+       spin_unlock(&lock->wait_lock);
+
+       if (!detect_deadlock && waiter != top_waiter)
+               goto out_put_task;
+
+       goto again;
+
+ out_unlock_pi:
+       spin_unlock_irqrestore(&task->pi_lock, flags);
+ out_put_task:
+       put_task_struct(task);
+
+       return ret;
+}
+
+/*
+ * Optimization: check if we can steal the lock from the
+ * assigned pending owner [which might not have taken the
+ * lock yet]:
+ */
+static inline int try_to_steal_lock(struct rt_mutex *lock)
+{
+       struct task_struct *pendowner = rt_mutex_owner(lock);
+       struct rt_mutex_waiter *next;
+       unsigned long flags;
+
+       if (!rt_mutex_owner_pending(lock))
+               return 0;
+
+       if (pendowner == current)
+               return 1;
+
+       spin_lock_irqsave(&pendowner->pi_lock, flags);
+       if (current->prio >= pendowner->prio) {
+               spin_unlock_irqrestore(&pendowner->pi_lock, flags);
+               return 0;
+       }
+
+       /*
+        * Check if a waiter is enqueued on the pending owners
+        * pi_waiters list. Remove it and readjust pending owners
+        * priority.
+        */
+       if (likely(!rt_mutex_has_waiters(lock))) {
+               spin_unlock_irqrestore(&pendowner->pi_lock, flags);
+               return 1;
+       }
+
+       /* No chain handling, pending owner is not blocked on anything: */
+       next = rt_mutex_top_waiter(lock);
+       plist_del(&next->pi_list_entry, &pendowner->pi_waiters);
+       __rt_mutex_adjust_prio(pendowner);
+       spin_unlock_irqrestore(&pendowner->pi_lock, flags);
+
+       /*
+        * We are going to steal the lock and a waiter was
+        * enqueued on the pending owners pi_waiters queue. So
+        * we have to enqueue this waiter into
+        * current->pi_waiters list. This covers the case,
+        * where current is boosted because it holds another
+        * lock and gets unboosted because the booster is
+        * interrupted, so we would delay a waiter with higher
+        * priority as current->normal_prio.
+        *
+        * Note: in the rare case of a SCHED_OTHER task changing
+        * its priority and thus stealing the lock, next->task
+        * might be current:
+        */
+       if (likely(next->task != current)) {
+               spin_lock_irqsave(&current->pi_lock, flags);
+               plist_add(&next->pi_list_entry, &current->pi_waiters);
+               __rt_mutex_adjust_prio(current);
+               spin_unlock_irqrestore(&current->pi_lock, flags);
+       }
+       return 1;
+}
+
+/*
+ * Try to take an rt-mutex
+ *
+ * This fails
+ * - when the lock has a real owner
+ * - when a different pending owner exists and has higher priority than current
+ *
+ * Must be called with lock->wait_lock held.
+ */
+static int try_to_take_rt_mutex(struct rt_mutex *lock)
+{
+       /*
+        * We have to be careful here if the atomic speedups are
+        * enabled, such that, when
+        *  - no other waiter is on the lock
+        *  - the lock has been released since we did the cmpxchg
+        * the lock can be released or taken while we are doing the
+        * checks and marking the lock with RT_MUTEX_HAS_WAITERS.
+        *
+        * The atomic acquire/release aware variant of
+        * mark_rt_mutex_waiters uses a cmpxchg loop. After setting
+        * the WAITERS bit, the atomic release / acquire can not
+        * happen anymore and lock->wait_lock protects us from the
+        * non-atomic case.
+        *
+        * Note, that this might set lock->owner =
+        * RT_MUTEX_HAS_WAITERS in the case the lock is not contended
+        * any more. This is fixed up when we take the ownership.
+        * This is the transitional state explained at the top of this file.
+        */
+       mark_rt_mutex_waiters(lock);
+
+       if (rt_mutex_owner(lock) && !try_to_steal_lock(lock))
+               return 0;
+
+       /* We got the lock. */
+       debug_rt_mutex_lock(lock);
+
+       rt_mutex_set_owner(lock, current, 0);
+
+       rt_mutex_deadlock_account_lock(lock, current);
+
+       return 1;
+}
+
+/*
+ * Task blocks on lock.
+ *
+ * Prepare waiter and propagate pi chain
+ *
+ * This must be called with lock->wait_lock held.
+ */
+static int task_blocks_on_rt_mutex(struct rt_mutex *lock,
+                                  struct rt_mutex_waiter *waiter,
+                                  int detect_deadlock)
+{
+       struct task_struct *owner = rt_mutex_owner(lock);
+       struct rt_mutex_waiter *top_waiter = waiter;
+       unsigned long flags;
+       int boost = 0, res;
+
+       spin_lock_irqsave(&current->pi_lock, flags);
+       __rt_mutex_adjust_prio(current);
+       waiter->task = current;
+       waiter->lock = lock;
+       plist_node_init(&waiter->list_entry, current->prio);
+       plist_node_init(&waiter->pi_list_entry, current->prio);
+
+       /* Get the top priority waiter on the lock */
+       if (rt_mutex_has_waiters(lock))
+               top_waiter = rt_mutex_top_waiter(lock);
+       plist_add(&waiter->list_entry, &lock->wait_list);
+
+       current->pi_blocked_on = waiter;
+
+       spin_unlock_irqrestore(&current->pi_lock, flags);
+
+       if (waiter == rt_mutex_top_waiter(lock)) {
+               spin_lock_irqsave(&owner->pi_lock, flags);
+               plist_del(&top_waiter->pi_list_entry, &owner->pi_waiters);
+               plist_add(&waiter->pi_list_entry, &owner->pi_waiters);
+
+               __rt_mutex_adjust_prio(owner);
+               if (owner->pi_blocked_on) {
+                       boost = 1;
+                       /* gets dropped in rt_mutex_adjust_prio_chain()! */
+                       get_task_struct(owner);
+               }
+               spin_unlock_irqrestore(&owner->pi_lock, flags);
+       }
+       else if (debug_rt_mutex_detect_deadlock(waiter, detect_deadlock)) {
+               spin_lock_irqsave(&owner->pi_lock, flags);
+               if (owner->pi_blocked_on) {
+                       boost = 1;
+                       /* gets dropped in rt_mutex_adjust_prio_chain()! */
+                       get_task_struct(owner);
+               }
+               spin_unlock_irqrestore(&owner->pi_lock, flags);
+       }
+       if (!boost)
+               return 0;
+
+       spin_unlock(&lock->wait_lock);
+
+       res = rt_mutex_adjust_prio_chain(owner, detect_deadlock, lock, waiter,
+                                        current);
+
+       spin_lock(&lock->wait_lock);
+
+       return res;
+}
+
+/*
+ * Wake up the next waiter on the lock.
+ *
+ * Remove the top waiter from the current tasks waiter list and from
+ * the lock waiter list. Set it as pending owner. Then wake it up.
+ *
+ * Called with lock->wait_lock held.
+ */
+static void wakeup_next_waiter(struct rt_mutex *lock)
+{
+       struct rt_mutex_waiter *waiter;
+       struct task_struct *pendowner;
+       unsigned long flags;
+
+       spin_lock_irqsave(&current->pi_lock, flags);
+
+       waiter = rt_mutex_top_waiter(lock);
+       plist_del(&waiter->list_entry, &lock->wait_list);
+
+       /*
+        * Remove it from current->pi_waiters. We do not adjust a
+        * possible priority boost right now. We execute wakeup in the
+        * boosted mode and go back to normal after releasing
+        * lock->wait_lock.
+        */
+       plist_del(&waiter->pi_list_entry, &current->pi_waiters);
+       pendowner = waiter->task;
+       waiter->task = NULL;
+
+       rt_mutex_set_owner(lock, pendowner, RT_MUTEX_OWNER_PENDING);
+
+       spin_unlock_irqrestore(&current->pi_lock, flags);
+
+       /*
+        * Clear the pi_blocked_on variable and enqueue a possible
+        * waiter into the pi_waiters list of the pending owner. This
+        * prevents that in case the pending owner gets unboosted a
+        * waiter with higher priority than pending-owner->normal_prio
+        * is blocked on the unboosted (pending) owner.
+        */
+       spin_lock_irqsave(&pendowner->pi_lock, flags);
+
+       WARN_ON(!pendowner->pi_blocked_on);
+       WARN_ON(pendowner->pi_blocked_on != waiter);
+       WARN_ON(pendowner->pi_blocked_on->lock != lock);
+
+       pendowner->pi_blocked_on = NULL;
+
+       if (rt_mutex_has_waiters(lock)) {
+               struct rt_mutex_waiter *next;
+
+               next = rt_mutex_top_waiter(lock);
+               plist_add(&next->pi_list_entry, &pendowner->pi_waiters);
+       }
+       spin_unlock_irqrestore(&pendowner->pi_lock, flags);
+
+       wake_up_process(pendowner);
+}
+
+/*
+ * Remove a waiter from a lock
+ *
+ * Must be called with lock->wait_lock held
+ */
+static void remove_waiter(struct rt_mutex *lock,
+                         struct rt_mutex_waiter *waiter)
+{
+       int first = (waiter == rt_mutex_top_waiter(lock));
+       struct task_struct *owner = rt_mutex_owner(lock);
+       unsigned long flags;
+       int boost = 0;
+
+       spin_lock_irqsave(&current->pi_lock, flags);
+       plist_del(&waiter->list_entry, &lock->wait_list);
+       waiter->task = NULL;
+       current->pi_blocked_on = NULL;
+       spin_unlock_irqrestore(&current->pi_lock, flags);
+
+       if (first && owner != current) {
+
+               spin_lock_irqsave(&owner->pi_lock, flags);
+
+               plist_del(&waiter->pi_list_entry, &owner->pi_waiters);
+
+               if (rt_mutex_has_waiters(lock)) {
+                       struct rt_mutex_waiter *next;
+
+                       next = rt_mutex_top_waiter(lock);
+                       plist_add(&next->pi_list_entry, &owner->pi_waiters);
+               }
+               __rt_mutex_adjust_prio(owner);
+
+               if (owner->pi_blocked_on) {
+                       boost = 1;
+                       /* gets dropped in rt_mutex_adjust_prio_chain()! */
+                       get_task_struct(owner);
+               }
+               spin_unlock_irqrestore(&owner->pi_lock, flags);
+       }
+
+       WARN_ON(!plist_node_empty(&waiter->pi_list_entry));
+
+       if (!boost)
+               return;
+
+       spin_unlock(&lock->wait_lock);
+
+       rt_mutex_adjust_prio_chain(owner, 0, lock, NULL, current);
+
+       spin_lock(&lock->wait_lock);
+}
+
+/*
+ * Recheck the pi chain, in case we got a priority setting
+ *
+ * Called from sched_setscheduler
+ */
+void rt_mutex_adjust_pi(struct task_struct *task)
+{
+       struct rt_mutex_waiter *waiter;
+       unsigned long flags;
+
+       spin_lock_irqsave(&task->pi_lock, flags);
+
+       waiter = task->pi_blocked_on;
+       if (!waiter || waiter->list_entry.prio == task->prio) {
+               spin_unlock_irqrestore(&task->pi_lock, flags);
+               return;
+       }
+
+       /* gets dropped in rt_mutex_adjust_prio_chain()! */
+       get_task_struct(task);
+       spin_unlock_irqrestore(&task->pi_lock, flags);
+
+       rt_mutex_adjust_prio_chain(task, 0, NULL, NULL, task);
+}
+
+/*
+ * Slow path lock function:
+ */
+static int __sched
+rt_mutex_slowlock(struct rt_mutex *lock, int state,
+                 struct hrtimer_sleeper *timeout,
+                 int detect_deadlock)
+{
+       struct rt_mutex_waiter waiter;
+       int ret = 0;
+
+       debug_rt_mutex_init_waiter(&waiter);
+       waiter.task = NULL;
+
+       spin_lock(&lock->wait_lock);
+
+       /* Try to acquire the lock again: */
+       if (try_to_take_rt_mutex(lock)) {
+               spin_unlock(&lock->wait_lock);
+               return 0;
+       }
+
+       set_current_state(state);
+
+       /* Setup the timer, when timeout != NULL */
+       if (unlikely(timeout))
+               hrtimer_start(&timeout->timer, timeout->timer.expires,
+                             HRTIMER_ABS);
+
+       for (;;) {
+               /* Try to acquire the lock: */
+               if (try_to_take_rt_mutex(lock))
+                       break;
+
+               /*
+                * TASK_INTERRUPTIBLE checks for signals and
+                * timeout. Ignored otherwise.
+                */
+               if (unlikely(state == TASK_INTERRUPTIBLE)) {
+                       /* Signal pending? */
+                       if (signal_pending(current))
+                               ret = -EINTR;
+                       if (timeout && !timeout->task)
+                               ret = -ETIMEDOUT;
+                       if (ret)
+                               break;
+               }
+
+               /*
+                * waiter.task is NULL the first time we come here and
+                * when we have been woken up by the previous owner
+                * but the lock got stolen by a higher prio task.
+                */
+               if (!waiter.task) {
+                       ret = task_blocks_on_rt_mutex(lock, &waiter,
+                                                     detect_deadlock);
+                       /*
+                        * If we got woken up by the owner then start loop
+                        * all over without going into schedule to try
+                        * to get the lock now:
+                        */
+                       if (unlikely(!waiter.task))
+                               continue;
+
+                       if (unlikely(ret))
+                               break;
+               }
+
+               spin_unlock(&lock->wait_lock);
+
+               debug_rt_mutex_print_deadlock(&waiter);
+
+               if (waiter.task)
+                       schedule_rt_mutex(lock);
+
+               spin_lock(&lock->wait_lock);
+               set_current_state(state);
+       }
+
+       set_current_state(TASK_RUNNING);
+
+       if (unlikely(waiter.task))
+               remove_waiter(lock, &waiter);
+
+       /*
+        * try_to_take_rt_mutex() sets the waiter bit
+        * unconditionally. We might have to fix that up.
+        */
+       fixup_rt_mutex_waiters(lock);
+
+       spin_unlock(&lock->wait_lock);
+
+       /* Remove pending timer: */
+       if (unlikely(timeout))
+               hrtimer_cancel(&timeout->timer);
+
+       /*
+        * Readjust priority, when we did not get the lock. We might
+        * have been the pending owner and boosted. Since we did not
+        * take the lock, the PI boost has to go.
+        */
+       if (unlikely(ret))
+               rt_mutex_adjust_prio(current);
+
+       debug_rt_mutex_free_waiter(&waiter);
+
+       return ret;
+}
+
+/*
+ * Slow path try-lock function:
+ */
+static inline int
+rt_mutex_slowtrylock(struct rt_mutex *lock)
+{
+       int ret = 0;
+
+       spin_lock(&lock->wait_lock);
+
+       if (likely(rt_mutex_owner(lock) != current)) {
+
+               ret = try_to_take_rt_mutex(lock);
+               /*
+                * try_to_take_rt_mutex() sets the lock waiters
+                * bit unconditionally. Clean this up.
+                */
+               fixup_rt_mutex_waiters(lock);
+       }
+
+       spin_unlock(&lock->wait_lock);
+
+       return ret;
+}
+
+/*
+ * Slow path to release a rt-mutex:
+ */
+static void __sched
+rt_mutex_slowunlock(struct rt_mutex *lock)
+{
+       spin_lock(&lock->wait_lock);
+
+       debug_rt_mutex_unlock(lock);
+
+       rt_mutex_deadlock_account_unlock(current);
+
+       if (!rt_mutex_has_waiters(lock)) {
+               lock->owner = NULL;
+               spin_unlock(&lock->wait_lock);
+               return;
+       }
+
+       wakeup_next_waiter(lock);
+
+       spin_unlock(&lock->wait_lock);
+
+       /* Undo pi boosting if necessary: */
+       rt_mutex_adjust_prio(current);
+}
+
+/*
+ * debug aware fast / slowpath lock,trylock,unlock
+ *
+ * The atomic acquire/release ops are compiled away, when either the
+ * architecture does not support cmpxchg or when debugging is enabled.
+ */
+static inline int
+rt_mutex_fastlock(struct rt_mutex *lock, int state,
+                 int detect_deadlock,
+                 int (*slowfn)(struct rt_mutex *lock, int state,
+                               struct hrtimer_sleeper *timeout,
+                               int detect_deadlock))
+{
+       if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
+               rt_mutex_deadlock_account_lock(lock, current);
+               return 0;
+       } else
+               return slowfn(lock, state, NULL, detect_deadlock);
+}
+
+static inline int
+rt_mutex_timed_fastlock(struct rt_mutex *lock, int state,
+                       struct hrtimer_sleeper *timeout, int detect_deadlock,
+                       int (*slowfn)(struct rt_mutex *lock, int state,
+                                     struct hrtimer_sleeper *timeout,
+                                     int detect_deadlock))
+{
+       if (!detect_deadlock && likely(rt_mutex_cmpxchg(lock, NULL, current))) {
+               rt_mutex_deadlock_account_lock(lock, current);
+               return 0;
+       } else
+               return slowfn(lock, state, timeout, detect_deadlock);
+}
+
+static inline int
+rt_mutex_fasttrylock(struct rt_mutex *lock,
+                    int (*slowfn)(struct rt_mutex *lock))
+{
+       if (likely(rt_mutex_cmpxchg(lock, NULL, current))) {
+               rt_mutex_deadlock_account_lock(lock, current);
+               return 1;
+       }
+       return slowfn(lock);
+}
+
+static inline void
+rt_mutex_fastunlock(struct rt_mutex *lock,
+                   void (*slowfn)(struct rt_mutex *lock))
+{
+       if (likely(rt_mutex_cmpxchg(lock, current, NULL)))
+               rt_mutex_deadlock_account_unlock(current);
+       else
+               slowfn(lock);
+}
+
+/**
+ * rt_mutex_lock - lock a rt_mutex
+ *
+ * @lock: the rt_mutex to be locked
+ */
+void __sched rt_mutex_lock(struct rt_mutex *lock)
+{
+       might_sleep();
+
+       rt_mutex_fastlock(lock, TASK_UNINTERRUPTIBLE, 0, rt_mutex_slowlock);
+}
+EXPORT_SYMBOL_GPL(rt_mutex_lock);
+
+/**
+ * rt_mutex_lock_interruptible - lock a rt_mutex interruptible
+ *
+ * @lock:              the rt_mutex to be locked
+ * @detect_deadlock:   deadlock detection on/off
+ *
+ * Returns:
+ *  0          on success
+ * -EINTR      when interrupted by a signal
+ * -EDEADLK    when the lock would deadlock (when deadlock detection is on)
+ */
+int __sched rt_mutex_lock_interruptible(struct rt_mutex *lock,
+                                                int detect_deadlock)
+{
+       might_sleep();
+
+       return rt_mutex_fastlock(lock, TASK_INTERRUPTIBLE,
+                                detect_deadlock, rt_mutex_slowlock);
+}
+EXPORT_SYMBOL_GPL(rt_mutex_lock_interruptible);
+
+/**
+ * rt_mutex_lock_interruptible_ktime - lock a rt_mutex interruptible
+ *                                    the timeout structure is provided
+ *                                    by the caller
+ *
+ * @lock:              the rt_mutex to be locked
+ * @timeout:           timeout structure or NULL (no timeout)
+ * @detect_deadlock:   deadlock detection on/off
+ *
+ * Returns:
+ *  0          on success
+ * -EINTR      when interrupted by a signal
+ * -ETIMEOUT   when the timeout expired
+ * -EDEADLK    when the lock would deadlock (when deadlock detection is on)
+ */
+int
+rt_mutex_timed_lock(struct rt_mutex *lock, struct hrtimer_sleeper *timeout,
+                   int detect_deadlock)
+{
+       might_sleep();
+
+       return rt_mutex_timed_fastlock(lock, TASK_INTERRUPTIBLE, timeout,
+                                      detect_deadlock, rt_mutex_slowlock);
+}
+EXPORT_SYMBOL_GPL(rt_mutex_timed_lock);
+
+/**
+ * rt_mutex_trylock - try to lock a rt_mutex
+ *
+ * @lock:      the rt_mutex to be locked
+ *
+ * Returns 1 on success and 0 on contention
+ */
+int __sched rt_mutex_trylock(struct rt_mutex *lock)
+{
+       return rt_mutex_fasttrylock(lock, rt_mutex_slowtrylock);
+}
+EXPORT_SYMBOL_GPL(rt_mutex_trylock);
+
+/**
+ * rt_mutex_unlock - unlock a rt_mutex
+ *
+ * @lock: the rt_mutex to be unlocked
+ */
+void __sched rt_mutex_unlock(struct rt_mutex *lock)
+{
+       rt_mutex_fastunlock(lock, rt_mutex_slowunlock);
+}
+EXPORT_SYMBOL_GPL(rt_mutex_unlock);
+
+/***
+ * rt_mutex_destroy - mark a mutex unusable
+ * @lock: the mutex to be destroyed
+ *
+ * This function marks the mutex uninitialized, and any subsequent
+ * use of the mutex is forbidden. The mutex must not be locked when
+ * this function is called.
+ */
+void rt_mutex_destroy(struct rt_mutex *lock)
+{
+       WARN_ON(rt_mutex_is_locked(lock));
+#ifdef CONFIG_DEBUG_RT_MUTEXES
+       lock->magic = NULL;
+#endif
+}
+
+EXPORT_SYMBOL_GPL(rt_mutex_destroy);
+
+/**
+ * __rt_mutex_init - initialize the rt lock
+ *
+ * @lock: the rt lock to be initialized
+ *
+ * Initialize the rt lock to unlocked state.
+ *
+ * Initializing of a locked rt lock is not allowed
+ */
+void __rt_mutex_init(struct rt_mutex *lock, const char *name)
+{
+       lock->owner = NULL;
+       spin_lock_init(&lock->wait_lock);
+       plist_head_init(&lock->wait_list, &lock->wait_lock);
+
+       debug_rt_mutex_init(lock, name);
+}
+EXPORT_SYMBOL_GPL(__rt_mutex_init);
+
+/**
+ * rt_mutex_init_proxy_locked - initialize and lock a rt_mutex on behalf of a
+ *                             proxy owner
+ *
+ * @lock:      the rt_mutex to be locked
+ * @proxy_owner:the task to set as owner
+ *
+ * No locking. Caller has to do serializing itself
+ * Special API call for PI-futex support
+ */
+void rt_mutex_init_proxy_locked(struct rt_mutex *lock,
+                               struct task_struct *proxy_owner)
+{
+       __rt_mutex_init(lock, NULL);
+       debug_rt_mutex_proxy_lock(lock, proxy_owner);
+       rt_mutex_set_owner(lock, proxy_owner, 0);
+       rt_mutex_deadlock_account_lock(lock, proxy_owner);
+}
+
+/**
+ * rt_mutex_proxy_unlock - release a lock on behalf of owner
+ *
+ * @lock:      the rt_mutex to be locked
+ *
+ * No locking. Caller has to do serializing itself
+ * Special API call for PI-futex support
+ */
+void rt_mutex_proxy_unlock(struct rt_mutex *lock,
+                          struct task_struct *proxy_owner)
+{
+       debug_rt_mutex_proxy_unlock(lock);
+       rt_mutex_set_owner(lock, NULL, 0);
+       rt_mutex_deadlock_account_unlock(proxy_owner);
+}
+
+/**
+ * rt_mutex_next_owner - return the next owner of the lock
+ *
+ * @lock: the rt lock query
+ *
+ * Returns the next owner of the lock or NULL
+ *
+ * Caller has to serialize against other accessors to the lock
+ * itself.
+ *
+ * Special API call for PI-futex support
+ */
+struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock)
+{
+       if (!rt_mutex_has_waiters(lock))
+               return NULL;
+
+       return rt_mutex_top_waiter(lock)->task;
+}