Re: RCU stalls -> lockup.

[Dave Jones <davej@redhat.com>]

On Sat, Oct 04, 2014 at 10:15:56PM -0400, Tejun Heo wrote:
 > On Thu, Oct 02, 2014 at 12:36:55PM -0700, Paul E. McKenney wrote:
 > > On Thu, Oct 02, 2014 at 01:55:15PM -0400, Dave Jones wrote:
 > > > I just hit this on my box running 3.17rc7
 > > > It was followed by a userspace lockup. (Could still ping, and sysrq
 > > > from the console, but even getty wasn't responding on the console).
 > > > 
 > > > I was trying to reproduce another bug faster, and had ramped up the
 > > > number of processes trinity to uses to 512. This didn't take long
 > > > to fall out..
 > > 
 > > This might be related to an exchange I had with Tejun (CCed), where
 > > the work queues were running all out, preventing any quiescent states
 > > from happening.  One fix under consideration is to add a quiescent state,
 > > similar to the one in softirq handling.
 > 
 > Dave, can you please test whether the following patch makes a
 > difference if the problem is reproducible?
 > 
 >   http://lkml.kernel.org/r/20141003153701.7c7da030@jlaw-desktop.mno.stratus.com

initial tests look good, haven't seen any reoccurance of the problem so far.

	Dave