From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753861Ab2F2XEa (ORCPT ); Fri, 29 Jun 2012 19:04:30 -0400 Received: from e35.co.us.ibm.com ([32.97.110.153]:39465 "EHLO e35.co.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751850Ab2F2XE2 (ORCPT ); Fri, 29 Jun 2012 19:04:28 -0400 Date: Fri, 29 Jun 2012 16:01:52 -0700 From: "Paul E. McKenney" To: Sasha Levin Cc: Dave Jones , "linux-kernel@vger.kernel.org" , Peter Zijlstra Subject: Re: rcu: BUG: spinlock recursion on CPU#3, trinity-child19/5970 Message-ID: <20120629230152.GF2416@linux.vnet.ibm.com> Reply-To: paulmck@linux.vnet.ibm.com References: <1340964584.2936.40.camel@lappy> <20120629172320.GA2416@linux.vnet.ibm.com> <1341006040.26928.4.camel@lappy> <20120629222744.GD2416@linux.vnet.ibm.com> <1341009607.27537.0.camel@lappy> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1341009607.27537.0.camel@lappy> User-Agent: Mutt/1.5.21 (2010-09-15) x-cbid: 12062923-6148-0000-0000-000007336ADC Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sat, Jun 30, 2012 at 12:40:07AM +0200, Sasha Levin wrote: > On Fri, 2012-06-29 at 15:27 -0700, Paul E. McKenney wrote: > > On Fri, Jun 29, 2012 at 11:40:40PM +0200, Sasha Levin wrote: > > > On Fri, 2012-06-29 at 10:23 -0700, Paul E. McKenney wrote: > > > > On Fri, Jun 29, 2012 at 12:09:44PM +0200, Sasha Levin wrote: > > > > > Hi Paul, > > > > > > > > > > I think I've stumbled on another bug that will increase your paranoia levels even further. > > > > > > > > > > I got the following lockup when fuzzing with trinity inside a KVM tools guest, using latest linux-next. > > > > > > > > > > It appears that it was caused by a03d6178 ("rcu: Move RCU grace-period cleanup into kthread"). This issue doesn't reproduce easily though, it took some fuzzing before hitting it. > > > > > > > > Hmmm... If the preemption at that point in __rcu_read_unlock() is > > > > required to make this happen, then it would be pretty hard to hit. > > > > I suspect that you can make it reproduce more quickly by putting > > > > a udelay(10) or similar right after the assignment of INT_MIN to > > > > t->rcu_read_lock_nesting in __rcu_read_unlock() in kernel/rcupdate.c. > > > > Can this be reproduced while running with lockdep enabled? > > > > > > The good news are that it is much easier to reproduce it by adding a udelay(10) at the point you've mentioned. > > > > How quickly does it reproduce? > > 10 seconds more or less. Very good -- if an alleged fix survives for 10 minutes, we have excellent statistical confidence that it is in fact a fix. Thanx, Paul