Messages in this thread | | | Date | Mon, 2 Jul 2012 07:59:27 -0700 | From | "Paul E. McKenney" <> | Subject | Re: rcu: BUG: spinlock recursion on CPU#3, trinity-child19/5970 |
| |
On Mon, Jul 02, 2012 at 04:36:53PM +0200, Sasha Levin wrote: > On Mon, 2012-07-02 at 07:22 -0700, Paul E. McKenney wrote: > > > OK, so it looks like the context-switch-time call into preemptible RCU > > > needs to be outside of the runqueue-lock critical section. One way > > > to do this is be reverting 616c310e (Move PREEMPT_RCU preemption to > > > switch_to() invocation). Sasha, could you please try this out? > > > > Hmmmm... Here is a patch for this reversion on top of -rcu with > > conflicts resolved. > > It looks like how I've resolved it.
Good show!!!
> Just for clarity, the only rcu related patch I have on top of -next is > this reversal one. I've dropped anything we tested before. > > Looks like its running fine for a bit now (~10 min). I'll ping if > anything changes, but on the other tests it has usually failed at this > point.
If testing goes well (both yours and mine), I will be submitting this as urgent, since it is a 3.5 regression. If they don't take it, I will queue it for 3.6.
Thanx, Paul
| |