Re: rcu_preempt self-detected stall on CPU from 4.5-rc3, since 3.17

[Paul E. McKenney]

On Tue, Feb 09, 2016 at 09:11:55PM +1100, Ross Green wrote:
> Continued testing with the latest linux-4.5-rc3 release.
> 
> Please find attached a copy of traces from dmesg:
> 
> There is a lot more debug and trace data so hopefully this will shed
> some light on what might be happening here.
> 
> My testing remains run a series of simple benchmarks, let that run to
> completion and then leave the system idle away with just a few daemons
> running.
> 
> the self detected stalls in this instance turned up after a days run time.
> There were  NO heavy artificial computational loads on the machine.

It does indeed look quiet on that dmesg for a good long time.

The following insanely crude not-for-mainline hack -might- be producing
good results in my testing.  It will take some time before I can claim
statistically different results.  But please feel free to give it a go
in the meantime.  (Thanks to Al Viro for pointing me in this direction.)

							Thanx, Paul

------------------------------------------------------------------------

commit 0c2c8d9fd1641809830a7a75f84dcad69936ef56
Author: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
Date:   Tue Feb 16 15:42:36 2016 -0800

    rcu: Crude exploratory hack
    
    Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>

diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
index 507d0ed48b97..5928e084620d 100644
--- a/kernel/rcu/tree.c
+++ b/kernel/rcu/tree.c
@@ -2194,8 +2194,10 @@ static int __noreturn rcu_gp_kthread(void *arg)
 					       READ_ONCE(rsp->gpnum),
 					       TPS("fqswait"));
 			rsp->gp_state = RCU_GP_WAIT_FQS;
-			ret = wait_event_interruptible_timeout(rsp->gp_wq,
-					rcu_gp_fqs_check_wake(rsp, &gf), j);
+			ret = schedule_timeout_interruptible(j > 0 ? j : 1);
+			rcu_gp_fqs_check_wake(rsp, &gf);
+			// ret = wait_event_interruptible_timeout(rsp->gp_wq,
+			//		rcu_gp_fqs_check_wake(rsp, &gf), j);
 			rsp->gp_state = RCU_GP_DOING_FQS;
 			/* Locking provides needed memory barriers. */
 			/* If grace period done, leave loop. */