During the livepatch loading process, several RCU warnings were triggered on
our production servers, as shown below:
[20329161.705294] livepatch: enabling patch 'livepatch_61_release6'
[20329161.713184] livepatch: 'livepatch_61_release6': starting patching transition
[20329172.998661] rcu_tasks_wait_gp: rcu_tasks grace period 1109713 is 10099 jiffies old.
[20329193.049536] rcu_tasks_wait_gp: rcu_tasks grace period 1109717 is 10059 jiffies old.
[20329213.131403] rcu_tasks_wait_gp: rcu_tasks grace period 1109725 is 10037 jiffies old.
[20329213.934005] livepatch: 'livepatch_61_release6': patching complete
The cause of these warnings was that the KLP transition was holding the
tasklist_lock (which is part of the RCU read-side critical section) for
too long, triggering the warning. To resolve this, we should avoid holding
the lock for an extended period. By checking need_resched(), we can ensure
the RCU warning no longer appears.
Signed-off-by: Yafang Shao <laoar.shao@xxxxxxxxx>
---
kernel/livepatch/transition.c | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/kernel/livepatch/transition.c b/kernel/livepatch/transition.c
index 04704a19dcfe..3d1f8d3d0f5a 100644
--- a/kernel/livepatch/transition.c
+++ b/kernel/livepatch/transition.c
@@ -491,9 +491,18 @@ void klp_try_complete_transition(void)
complete = false;
break;
}
+
+ /* Avoid potential RCU stalls */
+ if (need_resched()) {
+ complete = false;
+ break;
+ }
}
read_unlock(&tasklist_lock);
+ /* The above operation might be expensive. */
+ cond_resched();
+
/*
* Ditto for the idle "swapper" tasks.
*/
--
2.43.5
