On Sat, 2012-06-30 at 06:14 -0700, Paul E. McKenney wrote: > On Sat, Jun 30, 2012 at 01:36:48PM +0200, Sasha Levin wrote: > > Hi Paul, > > > > On Fri, 2012-06-29 at 10:23 -0700, Paul E. McKenney wrote: > > > Please see below for an untested patch that gets RCU out of this loop, > > > but it is quite possible that something else is involved here, so it would > > > be very good to get a lockdep run, if possible. My concern stems from the > > > fact that interrupts were enabled during the call to __rcu_read_unlock() > > > -- otherwise it would not have been preempted -- so the runqueue locks > > > were presumably not held on entry to __rcu_read_unlock(). > > > > I'll continue with the good news/bad news theme. > > > > Good news - the original problem I've reported seems to be gone. > > > > Bad news - I'm getting the following instead: > > Interesting... > > You are still running with lockdep? It should be on, yes. I don't see anything in dmesg which would suggest it got disabled before this spew. I've attached my .config. I'm starting the kernel with "sched_debug slub_debug=FZPU numa=fake=10" as well.