[PATCH v2] locking/rtmutex: Limit # of lock stealing for non-RT waiters

[Waiman Long]

Commit 48eb3f4fcfd3 ("locking/rtmutex: Implement equal priority lock
stealing") allows unlimited number of lock stealing's for non-RT
tasks. That can lead to lock starvation of non-RT top waiter tasks if
there is a constant incoming stream of non-RT lockers. This can cause
task lockup in PREEMPT_RT kernel. For example,

[ 1249.921363] INFO: task systemd:2178 blocked for more than 622 seconds.
[ 1872.984225] INFO: task kworker/6:4:63401 blocked for more than 622 seconds.

Avoiding this problem and ensuring forward progress by limiting the
number of times that a lock can be stolen from each waiter. This patch
sets a threshold of 10. That number is arbitrary and can be changed
if needed.

With that change, the task lockups previously observed when running
stressful workloads on PREEMPT_RT kernel disappeared.

Fixes: 48eb3f4fcfd3 ("locking/rtmutex: Implement equal priority lock stealing")
Reported-by: Mike Stowell <mstowell@xxxxxxxxxx>
Signed-off-by: Waiman Long <longman@xxxxxxxxxx>
---
 kernel/locking/rtmutex.c        | 9 ++++++---
 kernel/locking/rtmutex_common.h | 8 ++++++++
 2 files changed, 14 insertions(+), 3 deletions(-)

diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
index 7779ee8abc2a..bdddb3dc36c2 100644
--- a/kernel/locking/rtmutex.c
+++ b/kernel/locking/rtmutex.c
@@ -359,10 +359,13 @@ static inline bool rt_mutex_steal(struct rt_mutex_waiter *waiter,
 	if (rt_prio(waiter->prio) || dl_prio(waiter->prio))
 		return false;
 
-	return rt_mutex_waiter_equal(waiter, top_waiter);
-#else
-	return false;
+	if (rt_mutex_waiter_equal(waiter, top_waiter) &&
+	   (top_waiter->nr_steals < RT_MUTEX_LOCK_STEAL_MAX)) {
+		top_waiter->nr_steals++;
+		return true;
+	}
 #endif
+	return false;
 }
 
 #define __node_2_waiter(node) \
diff --git a/kernel/locking/rtmutex_common.h b/kernel/locking/rtmutex_common.h
index c47e8361bfb5..5858efe5cb0e 100644
--- a/kernel/locking/rtmutex_common.h
+++ b/kernel/locking/rtmutex_common.h
@@ -26,6 +26,7 @@
  * @task:		task reference to the blocked task
  * @lock:		Pointer to the rt_mutex on which the waiter blocks
  * @wake_state:		Wakeup state to use (TASK_NORMAL or TASK_RTLOCK_WAIT)
+ * @nr_steals:		Number of times the lock is stolen
  * @prio:		Priority of the waiter
  * @deadline:		Deadline of the waiter if applicable
  * @ww_ctx:		WW context pointer
@@ -36,11 +37,17 @@ struct rt_mutex_waiter {
 	struct task_struct	*task;
 	struct rt_mutex_base	*lock;
 	unsigned int		wake_state;
+	unsigned int		nr_steals;
 	int			prio;
 	u64			deadline;
 	struct ww_acquire_ctx	*ww_ctx;
 };
 
+/*
+ * The maximum number of times where lock can be stolen per waiter.
+ */
+#define	RT_MUTEX_LOCK_STEAL_MAX	10
+
 /**
  * rt_wake_q_head - Wrapper around regular wake_q_head to support
  *		    "sleeping" spinlocks on RT
@@ -194,6 +201,7 @@ static inline void rt_mutex_init_waiter(struct rt_mutex_waiter *waiter)
 	RB_CLEAR_NODE(&waiter->tree_entry);
 	waiter->wake_state = TASK_NORMAL;
 	waiter->task = NULL;
+	waiter->nr_steals = 0;
 }
 
 static inline void rt_mutex_init_rtlock_waiter(struct rt_mutex_waiter *waiter)
-- 
2.31.1