[PATCH 5/6] locking/rtmutex: Use rt_mutex specific scheduler helpers

[Peter Zijlstra]

From: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>

Have rt_mutex use the rt_mutex specific scheduler helpers to avoid
recursion vs rtlock on the PI state.

[[ peterz: adapted to new names ]]

Reported-by: Crystal Wood <swood@xxxxxxxxxx>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
---
 kernel/locking/rtmutex.c     |   14 ++++++++++++--
 kernel/locking/rwbase_rt.c   |    2 ++
 kernel/locking/rwsem.c       |    8 +++++++-
 kernel/locking/spinlock_rt.c |    4 ++++
 4 files changed, 25 insertions(+), 3 deletions(-)

--- a/kernel/locking/rtmutex.c
+++ b/kernel/locking/rtmutex.c
@@ -1636,7 +1636,7 @@ static int __sched rt_mutex_slowlock_blo
 		raw_spin_unlock_irq(&lock->wait_lock);
 
 		if (!owner || !rtmutex_spin_on_owner(lock, waiter, owner))
-			schedule();
+			rt_mutex_schedule();
 
 		raw_spin_lock_irq(&lock->wait_lock);
 		set_current_state(state);
@@ -1665,7 +1665,7 @@ static void __sched rt_mutex_handle_dead
 	WARN(1, "rtmutex deadlock detected\n");
 	while (1) {
 		set_current_state(TASK_INTERRUPTIBLE);
-		schedule();
+		rt_mutex_schedule();
 	}
 }
 
@@ -1761,6 +1761,15 @@ static int __sched rt_mutex_slowlock(str
 	int ret;
 
 	/*
+	 * Do all pre-schedule work here, before we queue a waiter and invoke
+	 * PI -- any such work that trips on rtlock (PREEMPT_RT spinlock) would
+	 * otherwise recurse back into task_blocks_on_rt_mutex() through
+	 * rtlock_slowlock() and will then enqueue a second waiter for this
+	 * same task and things get really confusing real fast.
+	 */
+	rt_mutex_pre_schedule();
+
+	/*
 	 * Technically we could use raw_spin_[un]lock_irq() here, but this can
 	 * be called in early boot if the cmpxchg() fast path is disabled
 	 * (debug, no architecture support). In this case we will acquire the
@@ -1771,6 +1780,7 @@ static int __sched rt_mutex_slowlock(str
 	raw_spin_lock_irqsave(&lock->wait_lock, flags);
 	ret = __rt_mutex_slowlock_locked(lock, ww_ctx, state);
 	raw_spin_unlock_irqrestore(&lock->wait_lock, flags);
+	rt_mutex_post_schedule();
 
 	return ret;
 }
--- a/kernel/locking/rwbase_rt.c
+++ b/kernel/locking/rwbase_rt.c
@@ -71,6 +71,7 @@ static int __sched __rwbase_read_lock(st
 	struct rt_mutex_base *rtm = &rwb->rtmutex;
 	int ret;
 
+	rwbase_pre_schedule();
 	raw_spin_lock_irq(&rtm->wait_lock);
 
 	/*
@@ -125,6 +126,7 @@ static int __sched __rwbase_read_lock(st
 		rwbase_rtmutex_unlock(rtm);
 
 	trace_contention_end(rwb, ret);
+	rwbase_post_schedule();
 	return ret;
 }
 
--- a/kernel/locking/rwsem.c
+++ b/kernel/locking/rwsem.c
@@ -1427,8 +1427,14 @@ static inline void __downgrade_write(str
 #define rwbase_signal_pending_state(state, current)	\
 	signal_pending_state(state, current)
 
+#define rwbase_pre_schedule()				\
+	rt_mutex_pre_schedule()
+
 #define rwbase_schedule()				\
-	schedule()
+	rt_mutex_schedule()
+
+#define rwbase_post_schedule()				\
+	rt_mutex_post_schedule()
 
 #include "rwbase_rt.c"
 
--- a/kernel/locking/spinlock_rt.c
+++ b/kernel/locking/spinlock_rt.c
@@ -184,9 +184,13 @@ static __always_inline int  rwbase_rtmut
 
 #define rwbase_signal_pending_state(state, current)	(0)
 
+#define rwbase_pre_schedule()
+
 #define rwbase_schedule()				\
 	schedule_rtlock()
 
+#define rwbase_post_schedule()
+
 #include "rwbase_rt.c"
 /*
  * The common functions which get wrapped into the rwlock API.