From: Steven Rostedt <rostedt@goodmis.org>
Subject: futex: Fix bug on when a requeued RT task times out

Requeue with timeout causes a bug with PREEMPT_RT_FULL.

The bug comes from a timed out condition.


	TASK 1				TASK 2
	------				------
    futex_wait_requeue_pi()
	futex_wait_queue_me()
	<timed out>

					double_lock_hb();

	raw_spin_lock(pi_lock);
	if (current->pi_blocked_on) { 
	} else {
	    current->pi_blocked_on = PI_WAKE_INPROGRESS;
	    run_spin_unlock(pi_lock);
	    spin_lock(hb->lock); <-- blocked!


					plist_for_each_entry_safe(this) {
					    rt_mutex_start_proxy_lock();
						task_blocks_on_rt_mutex();
						BUG_ON(task->pi_blocked_on)!!!!

The BUG_ON() actually has a check for PI_WAKE_INPROGRESS, but the
problem is that, after TASK 1 sets PI_WAKE_INPROGRESS, it then tries to
grab the hb->lock, which it fails to do so. As the hb->lock is a mutex,
it will block and set the "pi_blocked_on" to the hb->lock.

When TASK 2 goes to requeue it, the check for PI_WAKE_INPROGESS fails
because the task1's pi_blocked_on is no longer set to that, but instead,
set to the hb->lock.

The fix:

When calling rt_mutex_start_proxy_lock() a check is made to see
if the proxy tasks pi_blocked_on is set. If so, exit out early.
Otherwise set it to a new flag PI_REQUEUE_INPROGRESS, which notifies
the proxy task that it is being requeued, and will handle things
appropriately.

Cc: stable-rt@vger.kernel.org
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>


---
 kernel/rtmutex.c        |   32 +++++++++++++++++++++++++++++++-
 kernel/rtmutex_common.h |    1 +
 2 files changed, 32 insertions(+), 1 deletion(-)

Index: linux-stable/kernel/rtmutex.c
===================================================================
--- linux-stable.orig/kernel/rtmutex.c
+++ linux-stable/kernel/rtmutex.c
@@ -69,7 +69,8 @@ static void fixup_rt_mutex_waiters(struc
 
 static int rt_mutex_real_waiter(struct rt_mutex_waiter *waiter)
 {
-	return waiter && waiter != PI_WAKEUP_INPROGRESS;
+	return waiter && waiter != PI_WAKEUP_INPROGRESS &&
+		waiter != PI_REQUEUE_INPROGRESS;
 }
 
 /*
@@ -1002,6 +1003,35 @@ int rt_mutex_start_proxy_lock(struct rt_
 		return 1;
 	}
 
+#ifdef CONFIG_PREEMPT_RT_FULL
+	/*
+	 * In PREEMPT_RT there's an added race.
+	 * If the task, that we are about to requeue, times out,
+	 * it can set the PI_WAKEUP_INPROGRESS. This tells the requeue
+	 * to skip this task. But right after the task sets
+	 * its pi_blocked_on to PI_WAKEUP_INPROGRESS it can then
+	 * block on the spin_lock(&hb->lock), which in RT is an rtmutex.
+	 * This will replace the PI_WAKEUP_INPROGRESS with the actual
+	 * lock that it blocks on. We *must not* place this task
+	 * on this proxy lock in that case.
+	 *
+	 * To prevent this race, we first take the task's pi_lock
+	 * and check if it has updated its pi_blocked_on. If it has,
+	 * we assume that it woke up and we return -EAGAIN.
+	 * Otherwise, we set the task's pi_blocked_on to
+	 * PI_REQUEUE_INPROGRESS, so that if the task is waking up
+	 * it will know that we are in the process of requeuing it.
+	 */
+	raw_spin_lock_irq(&task->pi_lock);
+	if (task->pi_blocked_on) {
+		raw_spin_unlock_irq(&task->pi_lock);
+		raw_spin_unlock(&lock->wait_lock);
+		return -EAGAIN;
+	}
+	task->pi_blocked_on = PI_REQUEUE_INPROGRESS;
+	raw_spin_unlock_irq(&task->pi_lock);
+#endif
+
 	ret = task_blocks_on_rt_mutex(lock, waiter, task, detect_deadlock);
 
 	if (ret && !rt_mutex_owner(lock)) {
Index: linux-stable/kernel/rtmutex_common.h
===================================================================
--- linux-stable.orig/kernel/rtmutex_common.h
+++ linux-stable/kernel/rtmutex_common.h
@@ -104,6 +104,7 @@ static inline struct task_struct *rt_mut
  * PI-futex support (proxy locking functions, etc.):
  */
 #define PI_WAKEUP_INPROGRESS	((struct rt_mutex_waiter *) 1)
+#define PI_REQUEUE_INPROGRESS	((struct rt_mutex_waiter *) 2)
 
 extern struct task_struct *rt_mutex_next_owner(struct rt_mutex *lock);
 extern void rt_mutex_init_proxy_locked(struct rt_mutex *lock,