futex: Lower the lock contention on the HB lock during wake up

wake_futex_pi() wakes the task before releasing the hash bucket lock
(HB). The first thing the woken up task usually does is to acquire the
lock which requires the HB lock. On SMP Systems this leads to blocking
on the HB lock which is released by the owner shortly after.
This patch rearranges the unlock path by first releasing the HB lock and
then waking up the task.

[ tglx: Fixed up the rtmutex unlock path ]

Originally-from: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Link: http://lkml.kernel.org/r/20150617083350.GA2433@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

kernel/futex.c

@@ -1117,11 +1117,14 @@ static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q)
 	q->lock_ptr = NULL;
 }
 
-static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
+static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this,
+			 struct futex_hash_bucket *hb)
 {
 	struct task_struct *new_owner;
 	struct futex_pi_state *pi_state = this->pi_state;
 	u32 uninitialized_var(curval), newval;
+	WAKE_Q(wake_q);
+	bool deboost;
 	int ret = 0;
 
 	if (!pi_state)
@@ -1173,7 +1176,19 @@ static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_q *this)
 	raw_spin_unlock_irq(&new_owner->pi_lock);
 
 	raw_spin_unlock(&pi_state->pi_mutex.wait_lock);
-	rt_mutex_unlock(&pi_state->pi_mutex);
+
+	deboost = rt_mutex_futex_unlock(&pi_state->pi_mutex, &wake_q);
+
+	/*
+	 * First unlock HB so the waiter does not spin on it once he got woken
+	 * up. Second wake up the waiter before the priority is adjusted. If we
+	 * deboost first (and lose our higher priority), then the task might get
+	 * scheduled away before the wake up can take place.
+	 */
+	spin_unlock(&hb->lock);
+	wake_up_q(&wake_q);
+	if (deboost)
+		rt_mutex_adjust_prio(current);
 
 	return 0;
 }
@@ -2413,13 +2428,23 @@ retry:
 	 */
 	match = futex_top_waiter(hb, &key);
 	if (match) {
-		ret = wake_futex_pi(uaddr, uval, match);
+		ret = wake_futex_pi(uaddr, uval, match, hb);
+		/*
+		 * In case of success wake_futex_pi dropped the hash
+		 * bucket lock.
+		 */
+		if (!ret)
+			goto out_putkey;
 		/*
 		 * The atomic access to the futex value generated a
 		 * pagefault, so retry the user-access and the wakeup:
 		 */
 		if (ret == -EFAULT)
 			goto pi_faulted;
+		/*
+		 * wake_futex_pi has detected invalid state. Tell user
+		 * space.
+		 */
 		goto out_unlock;
 	}
 
@@ -2440,6 +2465,7 @@ retry:
 
 out_unlock:
 	spin_unlock(&hb->lock);
+out_putkey:
 	put_futex_key(&key);
 	return ret;
 

kernel/locking/rtmutex.c

@@ -300,7 +300,7 @@ static void __rt_mutex_adjust_prio(struct task_struct *task)
  * 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)
+void rt_mutex_adjust_prio(struct task_struct *task)
 {
 	unsigned long flags;
 
@@ -1247,13 +1247,12 @@ static inline int rt_mutex_slowtrylock(struct rt_mutex *lock)
 }
 
 /*
- * Slow path to release a rt-mutex:
+ * Slow path to release a rt-mutex.
+ * Return whether the current task needs to undo a potential priority boosting.
  */
-static void __sched
-rt_mutex_slowunlock(struct rt_mutex *lock)
+static bool __sched rt_mutex_slowunlock(struct rt_mutex *lock,
+					struct wake_q_head *wake_q)
 {
-	WAKE_Q(wake_q);
-
 	raw_spin_lock(&lock->wait_lock);
 
 	debug_rt_mutex_unlock(lock);
@@ -1294,7 +1293,7 @@ rt_mutex_slowunlock(struct rt_mutex *lock)
 	while (!rt_mutex_has_waiters(lock)) {
 		/* Drops lock->wait_lock ! */
 		if (unlock_rt_mutex_safe(lock) == true)
-			return;
+			return false;
 		/* Relock the rtmutex and try again */
 		raw_spin_lock(&lock->wait_lock);
 	}
@@ -1305,13 +1304,12 @@ rt_mutex_slowunlock(struct rt_mutex *lock)
 	 *
 	 * Queue the next waiter for wakeup once we release the wait_lock.
 	 */
-	mark_wakeup_next_waiter(&wake_q, lock);
+	mark_wakeup_next_waiter(wake_q, lock);
 
 	raw_spin_unlock(&lock->wait_lock);
-	wake_up_q(&wake_q);
 
-	/* Undo pi boosting if necessary: */
-	rt_mutex_adjust_prio(current);
+	/* check PI boosting */
+	return true;
 }
 
 /*
@@ -1362,12 +1360,23 @@ rt_mutex_fasttrylock(struct rt_mutex *lock,
 
 static inline void
 rt_mutex_fastunlock(struct rt_mutex *lock,
-		    void (*slowfn)(struct rt_mutex *lock))
+		    bool (*slowfn)(struct rt_mutex *lock,
+				   struct wake_q_head *wqh))
 {
-	if (likely(rt_mutex_cmpxchg(lock, current, NULL)))
+	WAKE_Q(wake_q);
+
+	if (likely(rt_mutex_cmpxchg(lock, current, NULL))) {
 		rt_mutex_deadlock_account_unlock(current);
-	else
-		slowfn(lock);
+
+	} else {
+		bool deboost = slowfn(lock, &wake_q);
+
+		wake_up_q(&wake_q);
+
+		/* Undo pi boosting if necessary: */
+		if (deboost)
+			rt_mutex_adjust_prio(current);
+	}
 }
 
 /**
@@ -1461,6 +1470,23 @@ void __sched rt_mutex_unlock(struct rt_mutex *lock)
 }
 EXPORT_SYMBOL_GPL(rt_mutex_unlock);
 
+/**
+ * rt_mutex_futex_unlock - Futex variant of rt_mutex_unlock
+ * @lock: the rt_mutex to be unlocked
+ *
+ * Returns: true/false indicating whether priority adjustment is
+ * required or not.
+ */
+bool __sched rt_mutex_futex_unlock(struct rt_mutex *lock,
+				   struct wake_q_head *wqh)
+{
+	if (likely(rt_mutex_cmpxchg(lock, current, NULL))) {
+		rt_mutex_deadlock_account_unlock(current);
+		return false;
+	}
+	return rt_mutex_slowunlock(lock, wqh);
+}
+
 /**
  * rt_mutex_destroy - mark a mutex unusable
  * @lock: the mutex to be destroyed

kernel/locking/rtmutex_common.h

@@ -131,6 +131,9 @@ extern int rt_mutex_finish_proxy_lock(struct rt_mutex *lock,
 				      struct hrtimer_sleeper *to,
 				      struct rt_mutex_waiter *waiter);
 extern int rt_mutex_timed_futex_lock(struct rt_mutex *l, struct hrtimer_sleeper *to);
+extern bool rt_mutex_futex_unlock(struct rt_mutex *lock,
+				  struct wake_q_head *wqh);
+extern void rt_mutex_adjust_prio(struct task_struct *task);
 
 #ifdef CONFIG_DEBUG_RT_MUTEXES
 # include "rtmutex-debug.h"