locking, sched: The PREEMPT-RT locking infrastructure

index 4d93b02..906b598 100644
@@ -485,13 +485,13 @@ static int __init test_lockup_init(void)
 		       offsetof(spinlock_t, lock.wait_lock.magic),
 		       SPINLOCK_MAGIC) ||
 	    test_magic(lock_rwlock_ptr,
-		       offsetof(rwlock_t, rtmutex.wait_lock.magic),
+		       offsetof(rwlock_t, rwbase.rtmutex.wait_lock.magic),
 		       SPINLOCK_MAGIC) ||
 	    test_magic(lock_mutex_ptr,
-		       offsetof(struct mutex, lock.wait_lock.magic),
+		       offsetof(struct mutex, rtmutex.wait_lock.magic),
 		       SPINLOCK_MAGIC) ||
 	    test_magic(lock_rwsem_ptr,
-		       offsetof(struct rw_semaphore, rtmutex.wait_lock.magic),
+		       offsetof(struct rw_semaphore, rwbase.rtmutex.wait_lock.magic),
 		       SPINLOCK_MAGIC))
 		return -EINVAL;
 #else

index 21625cb..a1f27fd 100644
@@ -1879,6 +1879,13 @@ futex_proxy_trylock_atomic(u32 __user *pifutex, struct futex_hash_bucket *hb1,
 	if (!top_waiter)
 		return 0;
 
+	/*
+	 * Ensure that this is a waiter sitting in futex_wait_requeue_pi()
+	 * and waiting on the 'waitqueue' futex which is always !PI.
+	 */
+	if (!top_waiter->rt_waiter || top_waiter->pi_state)
+		ret = -EINVAL;
+
 	/* Ensure we requeue to the expected futex. */
 	if (!match_futex(top_waiter->requeue_pi_key, key2))
 		return -EINVAL;

index a1f27fd..bc5395e 100644
@@ -1354,7 +1354,7 @@ static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval)
  *  -  1 - acquired the lock;
  *  - <0 - error
  *
- * The hb->lock and futex_key refs shall be held by the caller.
+ * The hb->lock must be held by the caller.
  *
  * @exiting is only set when the return value is -EBUSY. If so, this holds
  * a refcount on the exiting task on return and the caller needs to drop it

@@ -2621,8 +2621,7 @@ static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q,
  *
  * Setup the futex_q and locate the hash_bucket.  Get the futex value and
  * compare it with the expected value.  Handle atomic faults internally.
- * Return with the hb lock held and a q.key reference on success, and unlocked
- * with no q.key reference on failure.
+ * Return with the hb lock held on success, and unlocked on failure.
  *
  * Return:
  *  -  0 - uaddr contains val and hb has been locked;

@@ -2700,8 +2699,8 @@ static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val,
 			       current->timer_slack_ns);
 retry:
 	/*
-	 * Prepare to wait on uaddr. On success, holds hb lock and increments
-	 * q.key refs.
+	 * Prepare to wait on uaddr. On success, it holds hb->lock and q
+	 * is initialized.
 	 */
 	ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
 	if (ret)

@@ -2712,7 +2711,6 @@ retry:
 
 	/* If we were woken (and unqueued), we succeeded, whatever. */
 	ret = 0;
-	/* unqueue_me() drops q.key ref */
 	if (!unqueue_me(&q))
 		goto out;
 	ret = -ETIMEDOUT;

@@ -3205,8 +3203,8 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 	q.requeue_pi_key = &key2;
 
 	/*
-	 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
-	 * count.
+	 * Prepare to wait on uaddr. On success, it holds hb->lock and q
+	 * is initialized.
 	 */
 	ret = futex_wait_setup(uaddr, val, flags, &q, &hb);
 	if (ret)

@@ -3235,9 +3233,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 	 * In order for us to be here, we know our q.key == key2, and since
 	 * we took the hb->lock above, we also know that futex_requeue() has
 	 * completed and we no longer have to concern ourselves with a wakeup
-	 * race with the atomic proxy lock acquisition by the requeue code. The
-	 * futex_requeue dropped our key1 reference and incremented our key2
-	 * reference count.
+	 * race with the atomic proxy lock acquisition by the requeue code.
 	 */
 
 	/*

index bc5395e..c39d5f1 100644
@@ -1299,27 +1299,6 @@ static int attach_to_pi_owner(u32 __user *uaddr, u32 uval, union futex_key *key,
 	return 0;
 }
 
-static int lookup_pi_state(u32 __user *uaddr, u32 uval,
-			   struct futex_hash_bucket *hb,
-			   union futex_key *key, struct futex_pi_state **ps,
-			   struct task_struct **exiting)
-{
-	struct futex_q *top_waiter = futex_top_waiter(hb, key);
-
-	/*
-	 * If there is a waiter on that futex, validate it and
-	 * attach to the pi_state when the validation succeeds.
-	 */
-	if (top_waiter)
-		return attach_to_pi_state(uaddr, uval, top_waiter->pi_state, ps);
-
-	/*
-	 * We are the first waiter - try to look up the owner based on
-	 * @uval and attach to it.
-	 */
-	return attach_to_pi_owner(uaddr, uval, key, ps, exiting);
-}
-
 static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval)
 {
 	int err;

@@ -2038,8 +2017,8 @@ retry_private:
 		 * At this point the top_waiter has either taken uaddr2 or is
 		 * waiting on it.  If the former, then the pi_state will not
 		 * exist yet, look it up one more time to ensure we have a
-		 * reference to it. If the lock was taken, ret contains the
-		 * vpid of the top waiter task.
+		 * reference to it. If the lock was taken, @ret contains the
+		 * VPID of the top waiter task.
 		 * If the lock was not taken, we have pi_state and an initial
 		 * refcount on it. In case of an error we have nothing.
 		 */

@@ -2047,19 +2026,25 @@ retry_private:
 			WARN_ON(pi_state);
 			task_count++;
 			/*
-			 * If we acquired the lock, then the user space value
-			 * of uaddr2 should be vpid. It cannot be changed by
-			 * the top waiter as it is blocked on hb2 lock if it
-			 * tries to do so. If something fiddled with it behind
-			 * our back the pi state lookup might unearth it. So
-			 * we rather use the known value than rereading and
-			 * handing potential crap to lookup_pi_state.
+			 * If futex_proxy_trylock_atomic() acquired the
+			 * user space futex, then the user space value
+			 * @uaddr2 has been set to the @hb1's top waiter
+			 * task VPID. This task is guaranteed to be alive
+			 * and cannot be exiting because it is either
+			 * sleeping or blocked on @hb2 lock.
+			 *
+			 * The @uaddr2 futex cannot have waiters either as
+			 * otherwise futex_proxy_trylock_atomic() would not
+			 * have succeeded.
 			 *
-			 * If that call succeeds then we have pi_state and an
-			 * initial refcount on it.
+			 * In order to requeue waiters to @hb2, pi state is
+			 * required. Hand in the VPID value (@ret) and
+			 * allocate PI state with an initial refcount on
+			 * it.
 			 */
-			ret = lookup_pi_state(uaddr2, ret, hb2, &key2,
-					      &pi_state, &exiting);
+			ret = attach_to_pi_owner(uaddr2, ret, &key2, &pi_state,
+						 &exiting);
+			WARN_ON(ret);
 		}
 
 		switch (ret) {

@@ -2183,9 +2168,9 @@ retry_private:
 	}
 
 	/*
-	 * We took an extra initial reference to the pi_state either
-	 * in futex_proxy_trylock_atomic() or in lookup_pi_state(). We
-	 * need to drop it here again.
+	 * We took an extra initial reference to the pi_state either in
+	 * futex_proxy_trylock_atomic() or in attach_to_pi_owner(). We need
+	 * to drop it here again.
 	 */
 	put_pi_state(pi_state);
 

@@ -2364,7 +2349,7 @@ static int __fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
 	 * Modifying pi_state _before_ the user space value would leave the
 	 * pi_state in an inconsistent state when we fault here, because we
 	 * need to drop the locks to handle the fault. This might be observed
-	 * in the PID check in lookup_pi_state.
+	 * in the PID checks when attaching to PI state .
 	 */
 retry:
 	if (!argowner) {

index c39d5f1..8ddc87c 100644
@@ -2000,7 +2000,7 @@ retry_private:
 		}
 	}
 
-	if (requeue_pi && (task_count - nr_wake < nr_requeue)) {
+	if (requeue_pi) {
 		struct task_struct *exiting = NULL;
 
 		/*

index 8ddc87c..5439742 100644
@@ -2116,6 +2116,8 @@ retry_private:
 
 		/* Ensure we requeue to the expected futex for requeue_pi. */
 		if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
+			/* Don't account for it */
+			task_count--;
 			ret = -EINVAL;
 			break;
 		}

@@ -2157,6 +2159,8 @@ retry_private:
 				 */
 				this->pi_state = NULL;
 				put_pi_state(pi_state);
+				/* Don't account for it */
+				task_count--;
 				/*
 				 * We stop queueing more waiters and let user
 				 * space deal with the mess.

index 5439742..6cb6b5d 100644
@@ -2104,20 +2104,17 @@ retry_private:
 			break;
 		}
 
-		/*
-		 * Wake nr_wake waiters.  For requeue_pi, if we acquired the
-		 * lock, we already woke the top_waiter.  If not, it will be
-		 * woken by futex_unlock_pi().
-		 */
-		if (++task_count <= nr_wake && !requeue_pi) {
-			mark_wake_futex(&wake_q, this);
+		/* Plain futexes just wake or requeue and are done */
+		if (!requeue_pi) {
+			if (++task_count <= nr_wake)
+				mark_wake_futex(&wake_q, this);
+			else
+				requeue_futex(this, hb1, hb2, &key2);
 			continue;
 		}
 
 		/* Ensure we requeue to the expected futex for requeue_pi. */
-		if (requeue_pi && !match_futex(this->requeue_pi_key, &key2)) {
-			/* Don't account for it */
-			task_count--;
+		if (!match_futex(this->requeue_pi_key, &key2)) {
 			ret = -EINVAL;
 			break;
 		}

@@ -2125,50 +2122,45 @@ retry_private:
 		/*
 		 * Requeue nr_requeue waiters and possibly one more in the case
 		 * of requeue_pi if we couldn't acquire the lock atomically.
+		 *
+		 * Prepare the waiter to take the rt_mutex. Take a refcount
+		 * on the pi_state and store the pointer in the futex_q
+		 * object of the waiter.
 		 */
-		if (requeue_pi) {
+		get_pi_state(pi_state);
+		this->pi_state = pi_state;
+		ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
+						this->rt_waiter, this->task);
+		if (ret == 1) {
 			/*
-			 * Prepare the waiter to take the rt_mutex. Take a
-			 * refcount on the pi_state and store the pointer in
-			 * the futex_q object of the waiter.
+			 * We got the lock. We do neither drop the refcount
+			 * on pi_state nor clear this->pi_state because the
+			 * waiter needs the pi_state for cleaning up the
+			 * user space value. It will drop the refcount
+			 * after doing so.
 			 */
-			get_pi_state(pi_state);
-			this->pi_state = pi_state;
-			ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
-							this->rt_waiter,
-							this->task);
-			if (ret == 1) {
-				/*
-				 * We got the lock. We do neither drop the
-				 * refcount on pi_state nor clear
-				 * this->pi_state because the waiter needs the
-				 * pi_state for cleaning up the user space
-				 * value. It will drop the refcount after
-				 * doing so.
-				 */
-				requeue_pi_wake_futex(this, &key2, hb2);
-				continue;
-			} else if (ret) {
-				/*
-				 * rt_mutex_start_proxy_lock() detected a
-				 * potential deadlock when we tried to queue
-				 * that waiter. Drop the pi_state reference
-				 * which we took above and remove the pointer
-				 * to the state from the waiters futex_q
-				 * object.
-				 */
-				this->pi_state = NULL;
-				put_pi_state(pi_state);
-				/* Don't account for it */
-				task_count--;
-				/*
-				 * We stop queueing more waiters and let user
-				 * space deal with the mess.
-				 */
-				break;
-			}
+			requeue_pi_wake_futex(this, &key2, hb2);
+			task_count++;
+			continue;
+		} else if (ret) {
+			/*
+			 * rt_mutex_start_proxy_lock() detected a potential
+			 * deadlock when we tried to queue that waiter.
+			 * Drop the pi_state reference which we took above
+			 * and remove the pointer to the state from the
+			 * waiters futex_q object.
+			 */
+			this->pi_state = NULL;
+			put_pi_state(pi_state);
+			/*
+			 * We stop queueing more waiters and let user space
+			 * deal with the mess.
+			 */
+			break;
 		}
+		/* Waiter is queued, move it to hb2 */
 		requeue_futex(this, hb1, hb2, &key2);
+		task_count++;
 	}
 
 	/*

index 6cb6b5d..8d8bad5 100644
@@ -1939,15 +1939,27 @@ static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
 		 */
 		if (refill_pi_state_cache())
 			return -ENOMEM;
+
 		/*
-		 * requeue_pi must wake as many tasks as it can, up to nr_wake
-		 * + nr_requeue, since it acquires the rt_mutex prior to
-		 * returning to userspace, so as to not leave the rt_mutex with
-		 * waiters and no owner.  However, second and third wake-ups
-		 * cannot be predicted as they involve race conditions with the
-		 * first wake and a fault while looking up the pi_state.  Both
-		 * pthread_cond_signal() and pthread_cond_broadcast() should
-		 * use nr_wake=1.
+		 * futex_requeue() allows the caller to define the number
+		 * of waiters to wake up via the @nr_wake argument. With
+		 * REQUEUE_PI, waking up more than one waiter is creating
+		 * more problems than it solves. Waking up a waiter makes
+		 * only sense if the PI futex @uaddr2 is uncontended as
+		 * this allows the requeue code to acquire the futex
+		 * @uaddr2 before waking the waiter. The waiter can then
+		 * return to user space without further action. A secondary
+		 * wakeup would just make the futex_wait_requeue_pi()
+		 * handling more complex, because that code would have to
+		 * look up pi_state and do more or less all the handling
+		 * which the requeue code has to do for the to be requeued
+		 * waiters. So restrict the number of waiters to wake to
+		 * one, and only wake it up when the PI futex is
+		 * uncontended. Otherwise requeue it and let the unlock of
+		 * the PI futex handle the wakeup.
+		 *
+		 * All REQUEUE_PI users, e.g. pthread_cond_signal() and
+		 * pthread_cond_broadcast() must use nr_wake=1.
 		 */
 		if (nr_wake != 1)
 			return -EINVAL;

index 8d8bad5..a5232f6 100644
@@ -1934,13 +1934,6 @@ static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
 			return -EINVAL;
 
 		/*
-		 * requeue_pi requires a pi_state, try to allocate it now
-		 * without any locks in case it fails.
-		 */
-		if (refill_pi_state_cache())
-			return -ENOMEM;
-
-		/*
 		 * futex_requeue() allows the caller to define the number
 		 * of waiters to wake up via the @nr_wake argument. With
 		 * REQUEUE_PI, waking up more than one waiter is creating

@@ -1963,6 +1956,13 @@ static int futex_requeue(u32 __user *uaddr1, unsigned int flags,
 		 */
 		if (nr_wake != 1)
 			return -EINVAL;
+
+		/*
+		 * requeue_pi requires a pi_state, try to allocate it now
+		 * without any locks in case it fails.
+		 */
+		if (refill_pi_state_cache())
+			return -ENOMEM;
 	}
 
 retry:

index a5232f6..82660b9 100644
@@ -3070,27 +3070,22 @@ pi_faulted:
 }
 
 /**
- * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex
+ * handle_early_requeue_pi_wakeup() - Handle early wakeup on the initial futex
  * @hb:		the hash_bucket futex_q was original enqueued on
  * @q:		the futex_q woken while waiting to be requeued
- * @key2:	the futex_key of the requeue target futex
  * @timeout:	the timeout associated with the wait (NULL if none)
  *
- * Detect if the task was woken on the initial futex as opposed to the requeue
- * target futex.  If so, determine if it was a timeout or a signal that caused
- * the wakeup and return the appropriate error code to the caller.  Must be
- * called with the hb lock held.
+ * Determine the cause for the early wakeup.
  *
  * Return:
- *  -  0 = no early wakeup detected;
- *  - <0 = -ETIMEDOUT or -ERESTARTNOINTR
+ *  -EWOULDBLOCK or -ETIMEDOUT or -ERESTARTNOINTR
  */
 static inline
 int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
-				   struct futex_q *q, union futex_key *key2,
+				   struct futex_q *q,
 				   struct hrtimer_sleeper *timeout)
 {
-	int ret = 0;
+	int ret;
 
 	/*
 	 * With the hb lock held, we avoid races while we process the wakeup.

@@ -3099,22 +3094,21 @@ int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb,
 	 * It can't be requeued from uaddr2 to something else since we don't
 	 * support a PI aware source futex for requeue.
 	 */
-	if (!match_futex(&q->key, key2)) {
-		WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr));
-		/*
-		 * We were woken prior to requeue by a timeout or a signal.
-		 * Unqueue the futex_q and determine which it was.
-		 */
-		plist_del(&q->list, &hb->chain);
-		hb_waiters_dec(hb);
+	WARN_ON_ONCE(&hb->lock != q->lock_ptr);
 
-		/* Handle spurious wakeups gracefully */
-		ret = -EWOULDBLOCK;
-		if (timeout && !timeout->task)
-			ret = -ETIMEDOUT;
-		else if (signal_pending(current))
-			ret = -ERESTARTNOINTR;
-	}
+	/*
+	 * We were woken prior to requeue by a timeout or a signal.
+	 * Unqueue the futex_q and determine which it was.
+	 */
+	plist_del(&q->list, &hb->chain);
+	hb_waiters_dec(hb);
+
+	/* Handle spurious wakeups gracefully */
+	ret = -EWOULDBLOCK;
+	if (timeout && !timeout->task)
+		ret = -ETIMEDOUT;
+	else if (signal_pending(current))
+		ret = -ERESTARTNOINTR;
 	return ret;
 }
 

@@ -3217,7 +3211,9 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 	futex_wait_queue_me(hb, &q, to);
 
 	spin_lock(&hb->lock);
-	ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to);
+	/* Is @q still queued on uaddr1? */
+	if (!match_futex(&q->key, key2))
+		ret = handle_early_requeue_pi_wakeup(hb, &q, to);
 	spin_unlock(&hb->lock);
 	if (ret)
 		goto out;

index 82660b9..e7b4c61 100644
@@ -197,6 +197,8 @@ struct futex_pi_state {
  * @rt_waiter:		rt_waiter storage for use with requeue_pi
  * @requeue_pi_key:	the requeue_pi target futex key
  * @bitset:		bitset for the optional bitmasked wakeup
+ * @requeue_state:	State field for futex_requeue_pi()
+ * @requeue_wait:	RCU wait for futex_requeue_pi() (RT only)
  *
  * We use this hashed waitqueue, instead of a normal wait_queue_entry_t, so
  * we can wake only the relevant ones (hashed queues may be shared).

@@ -219,12 +221,68 @@ struct futex_q {
 	struct rt_mutex_waiter *rt_waiter;
 	union futex_key *requeue_pi_key;
 	u32 bitset;
+	atomic_t requeue_state;
+#ifdef CONFIG_PREEMPT_RT
+	struct rcuwait requeue_wait;
+#endif
 } __randomize_layout;
 
+/*
+ * On PREEMPT_RT, the hash bucket lock is a 'sleeping' spinlock with an
+ * underlying rtmutex. The task which is about to be requeued could have
+ * just woken up (timeout, signal). After the wake up the task has to
+ * acquire hash bucket lock, which is held by the requeue code.  As a task
+ * can only be blocked on _ONE_ rtmutex at a time, the proxy lock blocking
+ * and the hash bucket lock blocking would collide and corrupt state.
+ *
+ * On !PREEMPT_RT this is not a problem and everything could be serialized
+ * on hash bucket lock, but aside of having the benefit of common code,
+ * this allows to avoid doing the requeue when the task is already on the
+ * way out and taking the hash bucket lock of the original uaddr1 when the
+ * requeue has been completed.
+ *
+ * The following state transitions are valid:
+ *
+ * On the waiter side:
+ *   Q_REQUEUE_PI_NONE		-> Q_REQUEUE_PI_IGNORE
+ *   Q_REQUEUE_PI_IN_PROGRESS	-> Q_REQUEUE_PI_WAIT
+ *
+ * On the requeue side:
+ *   Q_REQUEUE_PI_NONE		-> Q_REQUEUE_PI_INPROGRESS
+ *   Q_REQUEUE_PI_IN_PROGRESS	-> Q_REQUEUE_PI_DONE/LOCKED
+ *   Q_REQUEUE_PI_IN_PROGRESS	-> Q_REQUEUE_PI_NONE (requeue failed)
+ *   Q_REQUEUE_PI_WAIT		-> Q_REQUEUE_PI_DONE/LOCKED
+ *   Q_REQUEUE_PI_WAIT		-> Q_REQUEUE_PI_IGNORE (requeue failed)
+ *
+ * The requeue side ignores a waiter with state Q_REQUEUE_PI_IGNORE as this
+ * signals that the waiter is already on the way out. It also means that
+ * the waiter is still on the 'wait' futex, i.e. uaddr1.
+ *
+ * The waiter side signals early wakeup to the requeue side either through
+ * setting state to Q_REQUEUE_PI_IGNORE or to Q_REQUEUE_PI_WAIT depending
+ * on the current state. In case of Q_REQUEUE_PI_IGNORE it can immediately
+ * proceed to take the hash bucket lock of uaddr1. If it set state to WAIT,
+ * which means the wakeup is interleaving with a requeue in progress it has
+ * to wait for the requeue side to change the state. Either to DONE/LOCKED
+ * or to IGNORE. DONE/LOCKED means the waiter q is now on the uaddr2 futex
+ * and either blocked (DONE) or has acquired it (LOCKED). IGNORE is set by
+ * the requeue side when the requeue attempt failed via deadlock detection
+ * and therefore the waiter q is still on the uaddr1 futex.
+ */
+enum {
+	Q_REQUEUE_PI_NONE		=  0,
+	Q_REQUEUE_PI_IGNORE,
+	Q_REQUEUE_PI_IN_PROGRESS,
+	Q_REQUEUE_PI_WAIT,
+	Q_REQUEUE_PI_DONE,
+	Q_REQUEUE_PI_LOCKED,
+};
+
 static const struct futex_q futex_q_init = {
 	/* list gets initialized in queue_me()*/
-	.key = FUTEX_KEY_INIT,
-	.bitset = FUTEX_BITSET_MATCH_ANY
+	.key		= FUTEX_KEY_INIT,
+	.bitset		= FUTEX_BITSET_MATCH_ANY,
+	.requeue_state	= ATOMIC_INIT(Q_REQUEUE_PI_NONE),
 };
 
 /*

@@ -1772,6 +1830,108 @@ void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1,
 	q->key = *key2;
 }
 
+static inline bool futex_requeue_pi_prepare(struct futex_q *q,
+					    struct futex_pi_state *pi_state)
+{
+	int old, new;
+
+	/*
+	 * Set state to Q_REQUEUE_PI_IN_PROGRESS unless an early wakeup has
+	 * already set Q_REQUEUE_PI_IGNORE to signal that requeue should
+	 * ignore the waiter.
+	 */
+	old = atomic_read_acquire(&q->requeue_state);
+	do {
+		if (old == Q_REQUEUE_PI_IGNORE)
+			return false;
+
+		/*
+		 * futex_proxy_trylock_atomic() might have set it to
+		 * IN_PROGRESS and a interleaved early wake to WAIT.
+		 *
+		 * It was considered to have an extra state for that
+		 * trylock, but that would just add more conditionals
+		 * all over the place for a dubious value.
+		 */
+		if (old != Q_REQUEUE_PI_NONE)
+			break;
+
+		new = Q_REQUEUE_PI_IN_PROGRESS;
+	} while (!atomic_try_cmpxchg(&q->requeue_state, &old, new));
+
+	q->pi_state = pi_state;
+	return true;
+}
+
+static inline void futex_requeue_pi_complete(struct futex_q *q, int locked)
+{
+	int old, new;
+
+	old = atomic_read_acquire(&q->requeue_state);
+	do {
+		if (old == Q_REQUEUE_PI_IGNORE)
+			return;
+
+		if (locked >= 0) {
+			/* Requeue succeeded. Set DONE or LOCKED */
+			WARN_ON_ONCE(old != Q_REQUEUE_PI_IN_PROGRESS &&
+				     old != Q_REQUEUE_PI_WAIT);
+			new = Q_REQUEUE_PI_DONE + locked;
+		} else if (old == Q_REQUEUE_PI_IN_PROGRESS) {
+			/* Deadlock, no early wakeup interleave */
+			new = Q_REQUEUE_PI_NONE;
+		} else {
+			/* Deadlock, early wakeup interleave. */
+			WARN_ON_ONCE(old != Q_REQUEUE_PI_WAIT);
+			new = Q_REQUEUE_PI_IGNORE;
+		}
+	} while (!atomic_try_cmpxchg(&q->requeue_state, &old, new));
+
+#ifdef CONFIG_PREEMPT_RT
+	/* If the waiter interleaved with the requeue let it know */
+	if (unlikely(old == Q_REQUEUE_PI_WAIT))
+		rcuwait_wake_up(&q->requeue_wait);
+#endif
+}
+
+static inline int futex_requeue_pi_wakeup_sync(struct futex_q *q)
+{
+	int old, new;
+
+	old = atomic_read_acquire(&q->requeue_state);
+	do {
+		/* Is requeue done already? */
+		if (old >= Q_REQUEUE_PI_DONE)
+			return old;
+
+		/*
+		 * If not done, then tell the requeue code to either ignore
+		 * the waiter or to wake it up once the requeue is done.
+		 */
+		new = Q_REQUEUE_PI_WAIT;
+		if (old == Q_REQUEUE_PI_NONE)
+			new = Q_REQUEUE_PI_IGNORE;
+	} while (!atomic_try_cmpxchg(&q->requeue_state, &old, new));
+
+	/* If the requeue was in progress, wait for it to complete */
+	if (old == Q_REQUEUE_PI_IN_PROGRESS) {
+#ifdef CONFIG_PREEMPT_RT
+		rcuwait_wait_event(&q->requeue_wait,
+				   atomic_read(&q->requeue_state) != Q_REQUEUE_PI_WAIT,
+				   TASK_UNINTERRUPTIBLE);
+#else
+		(void)atomic_cond_read_relaxed(&q->requeue_state, VAL != Q_REQUEUE_PI_WAIT);
+#endif
+	}
+
+	/*
+	 * Requeue is now either prohibited or complete. Reread state
+	 * because during the wait above it might have changed. Nothing
+	 * will modify q->requeue_state after this point.
+	 */
+	return atomic_read(&q->requeue_state);
+}
+
 /**
  * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue
  * @q:		the futex_q

@@ -1799,6 +1959,8 @@ void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key,
 
 	q->lock_ptr = &hb->lock;
 
+	/* Signal locked state to the waiter */
+	futex_requeue_pi_complete(q, 1);
 	wake_up_state(q->task, TASK_NORMAL);
 }
 

@@ -1869,6 +2031,10 @@ futex_proxy_trylock_atomic(u32 __user *pifutex, struct futex_hash_bucket *hb1,
 	if (!match_futex(top_waiter->requeue_pi_key, key2))
 		return -EINVAL;
 
+	/* Ensure that this does not race against an early wakeup */
+	if (!futex_requeue_pi_prepare(top_waiter, NULL))
+		return -EAGAIN;
+
 	/*
 	 * Try to take the lock for top_waiter.  Set the FUTEX_WAITERS bit in
 	 * the contended case or if set_waiters is 1.  The pi_state is returned

@@ -1878,8 +2044,22 @@ futex_proxy_trylock_atomic(u32 __user *pifutex, struct futex_hash_bucket *hb1,
 	ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task,
 				   exiting, set_waiters);
 	if (ret == 1) {
+		/* Dequeue, wake up and update top_waiter::requeue_state */
 		requeue_pi_wake_futex(top_waiter, key2, hb2);
 		return vpid;
+	} else if (ret < 0) {
+		/* Rewind top_waiter::requeue_state */
+		futex_requeue_pi_complete(top_waiter, ret);
+	} else {
+		/*
+		 * futex_lock_pi_atomic() did not acquire the user space
+		 * futex, but managed to establish the proxy lock and pi
+		 * state. top_waiter::requeue_state cannot be fixed up here
+		 * because the waiter is not enqueued on the rtmutex
+		 * yet. This is handled at the callsite depending on the
+		 * result of rt_mutex_start_proxy_lock() which is
+		 * guaranteed to be reached with this function returning 0.
+		 */
 	}
 	return ret;
 }

@@ -2020,6 +2200,8 @@ retry_private:
 		 * intend to requeue waiters, force setting the FUTEX_WAITERS
 		 * bit.  We force this here where we are able to easily handle
 		 * faults rather in the requeue loop below.
+		 *
+		 * Updates topwaiter::requeue_state if a top waiter exists.
 		 */
 		ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1,
 						 &key2, &pi_state,

@@ -2033,6 +2215,24 @@ retry_private:
 		 * VPID of the top waiter task.
 		 * If the lock was not taken, we have pi_state and an initial
 		 * refcount on it. In case of an error we have nothing.
+		 *
+		 * The top waiter's requeue_state is up to date:
+		 *
+		 *  - If the lock was acquired atomically (ret > 0), then
+		 *    the state is Q_REQUEUE_PI_LOCKED.
+		 *
+		 *  - If the trylock failed with an error (ret < 0) then
+		 *    the state is either Q_REQUEUE_PI_NONE, i.e. "nothing
+		 *    happened", or Q_REQUEUE_PI_IGNORE when there was an
+		 *    interleaved early wakeup.
+		 *
+		 *  - If the trylock did not succeed (ret == 0) then the
+		 *    state is either Q_REQUEUE_PI_IN_PROGRESS or
+		 *    Q_REQUEUE_PI_WAIT if an early wakeup interleaved.
+		 *    This will be cleaned up in the loop below, which
+		 *    cannot fail because futex_proxy_trylock_atomic() did
+		 *    the same sanity checks for requeue_pi as the loop
+		 *    below does.
 		 */
 		if (ret > 0) {
 			WARN_ON(pi_state);

@@ -2064,7 +2264,10 @@ retry_private:
 			/* We hold a reference on the pi state. */
 			break;
 
-			/* If the above failed, then pi_state is NULL */
+		/*
+		 * If the above failed, then pi_state is NULL and
+		 * waiter::requeue_state is correct.
+		 */
 		case -EFAULT:
 			double_unlock_hb(hb1, hb2);
 			hb_waiters_dec(hb2);

@@ -2140,21 +2343,39 @@ retry_private:
 		 * object of the waiter.
 		 */
 		get_pi_state(pi_state);
-		this->pi_state = pi_state;
+
+		/* Don't requeue when the waiter is already on the way out. */
+		if (!futex_requeue_pi_prepare(this, pi_state)) {
+			/*
+			 * Early woken waiter signaled that it is on the
+			 * way out. Drop the pi_state reference and try the
+			 * next waiter. @this->pi_state is still NULL.
+			 */
+			put_pi_state(pi_state);
+			continue;
+		}
+
 		ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex,
-						this->rt_waiter, this->task);
+						this->rt_waiter,
+						this->task);
+
 		if (ret == 1) {
 			/*
 			 * We got the lock. We do neither drop the refcount
 			 * on pi_state nor clear this->pi_state because the
 			 * waiter needs the pi_state for cleaning up the
 			 * user space value. It will drop the refcount
-			 * after doing so.
+			 * after doing so. this::requeue_state is updated
+			 * in the wakeup as well.
 			 */
 			requeue_pi_wake_futex(this, &key2, hb2);
 			task_count++;
-			continue;
-		} else if (ret) {
+		} else if (!ret) {
+			/* Waiter is queued, move it to hb2 */
+			requeue_futex(this, hb1, hb2, &key2);
+			futex_requeue_pi_complete(this, 0);
+			task_count++;
+		} else {
 			/*
 			 * rt_mutex_start_proxy_lock() detected a potential
 			 * deadlock when we tried to queue that waiter.

@@ -2164,15 +2385,13 @@ retry_private:
 			 */
 			this->pi_state = NULL;
 			put_pi_state(pi_state);
+			futex_requeue_pi_complete(this, ret);
 			/*
 			 * We stop queueing more waiters and let user space
 			 * deal with the mess.
 			 */
 			break;
 		}
-		/* Waiter is queued, move it to hb2 */
-		requeue_futex(this, hb1, hb2, &key2);
-		task_count++;
 	}
 
 	/*

@@ -3161,6 +3380,7 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 	struct futex_hash_bucket *hb;
 	union futex_key key2 = FUTEX_KEY_INIT;
 	struct futex_q q = futex_q_init;
+	struct rt_mutex_base *pi_mutex;
 	int res, ret;
 
 	if (!IS_ENABLED(CONFIG_FUTEX_PI))

@@ -3210,32 +3430,22 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 	/* Queue the futex_q, drop the hb lock, wait for wakeup. */
 	futex_wait_queue_me(hb, &q, to);
 
-	spin_lock(&hb->lock);
-	/* Is @q still queued on uaddr1? */
-	if (!match_futex(&q->key, key2))
+	switch (futex_requeue_pi_wakeup_sync(&q)) {
+	case Q_REQUEUE_PI_IGNORE:
+		/* The waiter is still on uaddr1 */
+		spin_lock(&hb->lock);
 		ret = handle_early_requeue_pi_wakeup(hb, &q, to);
-	spin_unlock(&hb->lock);
-	if (ret)
-		goto out;
-
-	/*
-	 * In order for us to be here, we know our q.key == key2, and since
-	 * we took the hb->lock above, we also know that futex_requeue() has
-	 * completed and we no longer have to concern ourselves with a wakeup
-	 * race with the atomic proxy lock acquisition by the requeue code.
-	 */
+		spin_unlock(&hb->lock);
+		break;
 
-	/*
-	 * Check if the requeue code acquired the second futex for us and do
-	 * any pertinent fixup.
-	 */
-	if (!q.rt_waiter) {
+	case Q_REQUEUE_PI_LOCKED:
+		/* The requeue acquired the lock */
 		if (q.pi_state && (q.pi_state->owner != current)) {
 			spin_lock(q.lock_ptr);
 			ret = fixup_owner(uaddr2, &q, true);
 			/*
-			 * Drop the reference to the pi state which
-			 * the requeue_pi() code acquired for us.
+			 * Drop the reference to the pi state which the
+			 * requeue_pi() code acquired for us.
 			 */
 			put_pi_state(q.pi_state);
 			spin_unlock(q.lock_ptr);

@@ -3245,18 +3455,14 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 			 */
 			ret = ret < 0 ? ret : 0;
 		}
-	} else {
-		struct rt_mutex_base *pi_mutex;
+		break;
 
-		/*
-		 * We have been woken up by futex_unlock_pi(), a timeout, or a
-		 * signal.  futex_unlock_pi() will not destroy the lock_ptr nor
-		 * the pi_state.
-		 */
-		WARN_ON(!q.pi_state);
+	case Q_REQUEUE_PI_DONE:
+		/* Requeue completed. Current is 'pi_blocked_on' the rtmutex */
 		pi_mutex = &q.pi_state->pi_mutex;
 		ret = rt_mutex_wait_proxy_lock(pi_mutex, to, &rt_waiter);
 
+		/* Current is not longer pi_blocked_on */
 		spin_lock(q.lock_ptr);
 		if (ret && !rt_mutex_cleanup_proxy_lock(pi_mutex, &rt_waiter))
 			ret = 0;

@@ -3276,17 +3482,21 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags,
 
 		unqueue_me_pi(&q);
 		spin_unlock(q.lock_ptr);
-	}
 
-	if (ret == -EINTR) {
-		/*
-		 * We've already been requeued, but cannot restart by calling
-		 * futex_lock_pi() directly. We could restart this syscall, but
-		 * it would detect that the user space "val" changed and return
-		 * -EWOULDBLOCK.  Save the overhead of the restart and return
-		 * -EWOULDBLOCK directly.
-		 */
-		ret = -EWOULDBLOCK;
+		if (ret == -EINTR) {
+			/*
+			 * We've already been requeued, but cannot restart
+			 * by calling futex_lock_pi() directly. We could
+			 * restart this syscall, but it would detect that
+			 * the user space "val" changed and return
+			 * -EWOULDBLOCK.  Save the overhead of the restart
+			 * and return -EWOULDBLOCK directly.
+			 */
+			ret = -EWOULDBLOCK;
+		}
+		break;
+	default:
+		BUG();
 	}
 
 out:

index 92b7d28..5c9299a 100644
@@ -214,7 +214,19 @@ EXPORT_SYMBOL_GPL(__rt_mutex_init);
 void __sched rt_mutex_init_proxy_locked(struct rt_mutex_base *lock,
 					struct task_struct *proxy_owner)
 {
+	static struct lock_class_key pi_futex_key;
+
 	__rt_mutex_base_init(lock);
+	/*
+	 * On PREEMPT_RT the futex hashbucket spinlock becomes 'sleeping'
+	 * and rtmutex based. That causes a lockdep false positive, because
+	 * some of the futex functions invoke spin_unlock(&hb->lock) with
+	 * the wait_lock of the rtmutex associated to the pi_futex held.
+	 * spin_unlock() in turn takes wait_lock of the rtmutex on which
+	 * the spinlock is based, which makes lockdep notice a lock
+	 * recursion. Give the futex/rtmutex wait_lock a separate key.
+	 */
+	lockdep_set_class(&lock->wait_lock, &pi_futex_key);
 	rt_mutex_set_owner(lock, proxy_owner);
 }
 

index 9881eac..4d244e2 100644
@@ -121,7 +121,11 @@
 /*
  * The preempt_count offset after spin_lock()
  */
+#if !defined(CONFIG_PREEMPT_RT)
 #define PREEMPT_LOCK_OFFSET	PREEMPT_DISABLE_OFFSET
+#else
+#define PREEMPT_LOCK_OFFSET	0
+#endif
 
 /*
  * The preempt_count offset needed for things like:

index af7e3af..3eaf636 100644
@@ -338,6 +338,26 @@ static __always_inline int rt_mutex_waiter_equal(struct rt_mutex_waiter *left,
 	return 1;
 }
 
+static inline bool rt_mutex_steal(struct rt_mutex_waiter *waiter,
+				  struct rt_mutex_waiter *top_waiter)
+{
+	if (rt_mutex_waiter_less(waiter, top_waiter))
+		return true;
+
+#ifdef RT_MUTEX_BUILD_SPINLOCKS
+	/*
+	 * Note that RT tasks are excluded from same priority (lateral)
+	 * steals to prevent the introduction of an unbounded latency.
+	 */
+	if (rt_prio(waiter->prio) || dl_prio(waiter->prio))
+		return false;
+
+	return rt_mutex_waiter_equal(waiter, top_waiter);
+#else
+	return false;
+#endif
+}
+
 #define __node_2_waiter(node) \
 	rb_entry((node), struct rt_mutex_waiter, tree_entry)
 

@@ -932,19 +952,21 @@ try_to_take_rt_mutex(struct rt_mutex_base *lock, struct task_struct *task,
 	 * trylock attempt.
 	 */
 	if (waiter) {
-		/*
-		 * If waiter is not the highest priority waiter of
-		 * @lock, give up.
-		 */
-		if (waiter != rt_mutex_top_waiter(lock))
-			return 0;
+		struct rt_mutex_waiter *top_waiter = rt_mutex_top_waiter(lock);
 
 		/*
-		 * We can acquire the lock. Remove the waiter from the
-		 * lock waiters tree.
+		 * If waiter is the highest priority waiter of @lock,
+		 * or allowed to steal it, take it over.
 		 */
-		rt_mutex_dequeue(lock, waiter);
-
+		if (waiter == top_waiter || rt_mutex_steal(waiter, top_waiter)) {
+			/*
+			 * We can acquire the lock. Remove the waiter from the
+			 * lock waiters tree.
+			 */
+			rt_mutex_dequeue(lock, waiter);
+		} else {
+			return 0;
+		}
 	} else {
 		/*
 		 * If the lock has waiters already we check whether @task is

@@ -955,13 +977,9 @@ try_to_take_rt_mutex(struct rt_mutex_base *lock, struct task_struct *task,
 		 * not need to be dequeued.
 		 */
 		if (rt_mutex_has_waiters(lock)) {
-			/*
-			 * If @task->prio is greater than or equal to
-			 * the top waiter priority (kernel view),
-			 * @task lost.
-			 */
-			if (!rt_mutex_waiter_less(task_to_waiter(task),
-						  rt_mutex_top_waiter(lock)))
+			/* Check whether the trylock can steal it. */
+			if (!rt_mutex_steal(task_to_waiter(task),
+					    rt_mutex_top_waiter(lock)))
 				return 0;
 
 			/*

index 3eaf636..8aaa352 100644
@@ -8,6 +8,11 @@
  *  Copyright (C) 2005-2006 Timesys Corp., Thomas Gleixner <[email protected]>
  *  Copyright (C) 2005 Kihon Technologies Inc., Steven Rostedt
  *  Copyright (C) 2006 Esben Nielsen
+ * Adaptive Spinlocks:
+ *  Copyright (C) 2008 Novell, Inc., Gregory Haskins, Sven Dietrich,
+ *				     and Peter Morreale,
+ * Adaptive Spinlocks simplification:
+ *  Copyright (C) 2008 Red Hat, Inc., Steven Rostedt <[email protected]>
  *
  *  See Documentation/locking/rt-mutex-design.rst for details.
  */

@@ -1297,6 +1302,52 @@ static __always_inline void __rt_mutex_unlock(struct rt_mutex_base *lock)
 	rt_mutex_slowunlock(lock);
 }
 
+#ifdef CONFIG_SMP
+static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
+				  struct rt_mutex_waiter *waiter,
+				  struct task_struct *owner)
+{
+	bool res = true;
+
+	rcu_read_lock();
+	for (;;) {
+		/* If owner changed, trylock again. */
+		if (owner != rt_mutex_owner(lock))
+			break;
+		/*
+		 * Ensure that @owner is dereferenced after checking that
+		 * the lock owner still matches @owner. If that fails,
+		 * @owner might point to freed memory. If it still matches,
+		 * the rcu_read_lock() ensures the memory stays valid.
+		 */
+		barrier();
+		/*
+		 * Stop spinning when:
+		 *  - the lock owner has been scheduled out
+		 *  - current is not longer the top waiter
+		 *  - current is requested to reschedule (redundant
+		 *    for CONFIG_PREEMPT_RCU=y)
+		 *  - the VCPU on which owner runs is preempted
+		 */
+		if (!owner->on_cpu || waiter != rt_mutex_top_waiter(lock) ||
+		    need_resched() || vcpu_is_preempted(task_cpu(owner))) {
+			res = false;
+			break;
+		}
+		cpu_relax();
+	}
+	rcu_read_unlock();
+	return res;
+}
+#else
+static bool rtmutex_spin_on_owner(struct rt_mutex_base *lock,
+				  struct rt_mutex_waiter *waiter,
+				  struct task_struct *owner)
+{
+	return false;
+}
+#endif
+
 #ifdef RT_MUTEX_BUILD_MUTEX
 /*
  * Functions required for:

@@ -1381,6 +1432,7 @@ static int __sched rt_mutex_slowlock_block(struct rt_mutex_base *lock,
 					   struct rt_mutex_waiter *waiter)
 {
 	struct rt_mutex *rtm = container_of(lock, struct rt_mutex, rtmutex);
+	struct task_struct *owner;
 	int ret = 0;
 
 	for (;;) {

@@ -1403,9 +1455,14 @@ static int __sched rt_mutex_slowlock_block(struct rt_mutex_base *lock,
 				break;
 		}
 
+		if (waiter == rt_mutex_top_waiter(lock))
+			owner = rt_mutex_owner(lock);
+		else
+			owner = NULL;
 		raw_spin_unlock_irq(&lock->wait_lock);
 
-		schedule();
+		if (!owner || !rtmutex_spin_on_owner(lock, waiter, owner))
+			schedule();
 
 		raw_spin_lock_irq(&lock->wait_lock);
 		set_current_state(state);

@@ -1561,6 +1618,7 @@ static __always_inline int __rt_mutex_lock(struct rt_mutex_base *lock,
 static void __sched rtlock_slowlock_locked(struct rt_mutex_base *lock)
 {
 	struct rt_mutex_waiter waiter;
+	struct task_struct *owner;
 
 	lockdep_assert_held(&lock->wait_lock);
 

@@ -1579,9 +1637,14 @@ static void __sched rtlock_slowlock_locked(struct rt_mutex_base *lock)
 		if (try_to_take_rt_mutex(lock, current, &waiter))
 			break;
 
+		if (&waiter == rt_mutex_top_waiter(lock))
+			owner = rt_mutex_owner(lock);
+		else
+			owner = NULL;
 		raw_spin_unlock_irq(&lock->wait_lock);
 
-		schedule_rtlock();
+		if (!owner || !rtmutex_spin_on_owner(lock, &waiter, owner))
+			schedule_rtlock();
 
 		raw_spin_lock_irq(&lock->wait_lock);
 		set_current_state(TASK_RTLOCK_WAIT);

index 4fc7219..835aeda 100644
@@ -8,20 +8,28 @@
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 extern void __rt_spin_lock_init(spinlock_t *lock, const char *name,
-				struct lock_class_key *key);
+				struct lock_class_key *key, bool percpu);
 #else
 static inline void __rt_spin_lock_init(spinlock_t *lock, const char *name,
-				       struct lock_class_key *key)
+				struct lock_class_key *key, bool percpu)
 {
 }
 #endif
 
-#define spin_lock_init(slock)				\
-do {							\
-	static struct lock_class_key __key;		\
-							\
-	rt_mutex_base_init(&(slock)->lock);		\
-	__rt_spin_lock_init(slock, #slock, &__key);	\
+#define spin_lock_init(slock)					\
+do {								\
+	static struct lock_class_key __key;			\
+								\
+	rt_mutex_base_init(&(slock)->lock);			\
+	__rt_spin_lock_init(slock, #slock, &__key, false);	\
+} while (0)
+
+#define local_spin_lock_init(slock)				\
+do {								\
+	static struct lock_class_key __key;			\
+								\
+	rt_mutex_base_init(&(slock)->lock);			\
+	__rt_spin_lock_init(slock, #slock, &__key, true);	\
 } while (0)
 
 extern void rt_spin_lock(spinlock_t *lock);

index 8a9aadb..2dfa35f 100644
@@ -60,6 +60,12 @@ typedef struct spinlock {
 		SPIN_DEP_MAP_INIT(name)				\
 	}
 
+#define __LOCAL_SPIN_LOCK_UNLOCKED(name)			\
+	{							\
+		.lock = __RT_MUTEX_BASE_INITIALIZER(name.lock),	\
+		LOCAL_SPIN_DEP_MAP_INIT(name)			\
+	}
+
 #define DEFINE_SPINLOCK(name)					\
 	spinlock_t name = __SPIN_LOCK_UNLOCKED(name)
 

index a8a4330..91cb36b 100644
@@ -37,9 +37,17 @@ typedef struct raw_spinlock {
 		.name = #lockname,			\
 		.wait_type_inner = LD_WAIT_CONFIG,	\
 	}
+
+# define LOCAL_SPIN_DEP_MAP_INIT(lockname)		\
+	.dep_map = {					\
+		.name = #lockname,			\
+		.wait_type_inner = LD_WAIT_CONFIG,	\
+		.lock_type = LD_LOCK_PERCPU,		\
+	}
 #else
 # define RAW_SPIN_DEP_MAP_INIT(lockname)
 # define SPIN_DEP_MAP_INIT(lockname)
+# define LOCAL_SPIN_DEP_MAP_INIT(lockname)
 #endif
 
 #ifdef CONFIG_DEBUG_SPINLOCK

index c36648b..d2912e4 100644
@@ -120,10 +120,13 @@ EXPORT_SYMBOL(rt_spin_trylock_bh);
 
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 void __rt_spin_lock_init(spinlock_t *lock, const char *name,
-			 struct lock_class_key *key)
+			 struct lock_class_key *key, bool percpu)
 {
+	u8 type = percpu ? LD_LOCK_PERCPU : LD_LOCK_NORMAL;
+
 	debug_check_no_locks_freed((void *)lock, sizeof(*lock));
-	lockdep_init_map_wait(&lock->dep_map, name, key, 0, LD_WAIT_CONFIG);
+	lockdep_init_map_type(&lock->dep_map, name, key, 0, LD_WAIT_CONFIG,
+			      LD_WAIT_INV, type);
 }
 EXPORT_SYMBOL(__rt_spin_lock_init);
 #endif

index 3f02b81..975e33b 100644
@@ -6,6 +6,8 @@
 #include <linux/percpu-defs.h>
 #include <linux/lockdep.h>
 
+#ifndef CONFIG_PREEMPT_RT
+
 typedef struct {
 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 	struct lockdep_map	dep_map;

@@ -95,3 +97,45 @@ do {								\
 		local_lock_release(this_cpu_ptr(lock));		\
 		local_irq_restore(flags);			\
 	} while (0)
+
+#else /* !CONFIG_PREEMPT_RT */
+
+/*
+ * On PREEMPT_RT local_lock maps to a per CPU spinlock, which protects the
+ * critical section while staying preemptible.
+ */
+typedef spinlock_t local_lock_t;
+
+#define INIT_LOCAL_LOCK(lockname) __LOCAL_SPIN_LOCK_UNLOCKED((lockname))
+
+#define __local_lock_init(l)					\
+	do {							\
+		local_spin_lock_init((l));			\
+	} while (0)
+
+#define __local_lock(__lock)					\
+	do {							\
+		migrate_disable();				\
+		spin_lock(this_cpu_ptr((__lock)));		\
+	} while (0)
+
+#define __local_lock_irq(lock)			__local_lock(lock)
+
+#define __local_lock_irqsave(lock, flags)			\
+	do {							\
+		typecheck(unsigned long, flags);		\
+		flags = 0;					\
+		__local_lock(lock);				\
+	} while (0)
+
+#define __local_unlock(__lock)					\
+	do {							\
+		spin_unlock(this_cpu_ptr((__lock)));		\
+		migrate_enable();				\
+	} while (0)
+
+#define __local_unlock_irq(lock)		__local_unlock(lock)
+
+#define __local_unlock_irqrestore(lock, flags)	__local_unlock(lock)
+
+#endif /* CONFIG_PREEMPT_RT */