From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754181Ab0EFCFh (ORCPT ); Wed, 5 May 2010 22:05:37 -0400 Received: from mail.openrapids.net ([64.15.138.104]:41331 "EHLO blackscsi.openrapids.net" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1752727Ab0EFCFf (ORCPT ); Wed, 5 May 2010 22:05:35 -0400 Date: Wed, 5 May 2010 22:05:32 -0400 From: Mathieu Desnoyers To: "Paul E. McKenney" Cc: linux-kernel@vger.kernel.org, mingo@elte.hu, laijs@cn.fujitsu.com, dipankar@in.ibm.com, akpm@linux-foundation.org, josh@joshtriplett.org, dvhltc@us.ibm.com, niv@us.ibm.com, tglx@linutronix.de, peterz@infradead.org, rostedt@goodmis.org, Valdis.Kletnieks@vt.edu, dhowells@redhat.com, eric.dumazet@gmail.com Subject: Re: [PATCH tip/core/rcu 01/48] rcu: optionally leave lockdep enabled after RCU lockdep splat Message-ID: <20100506020532.GA16547@Krystal> References: <20100504201934.GA19234@linux.vnet.ibm.com> <1273004398-19760-1-git-send-email-paulmck@linux.vnet.ibm.com> <20100505224641.GA15359@Krystal> <20100505230557.GD2439@linux.vnet.ibm.com> <20100505232457.GA1313@Krystal> <20100505233601.GG2439@linux.vnet.ibm.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20100505233601.GG2439@linux.vnet.ibm.com> X-Editor: vi X-Info: http://www.efficios.com X-Operating-System: Linux/2.6.26-2-686 (i686) X-Uptime: 22:03:42 up 103 days, 4:40, 3 users, load average: 0.12, 0.06, 0.01 User-Agent: Mutt/1.5.18 (2008-05-17) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org * Paul E. McKenney (paulmck@linux.vnet.ibm.com) wrote: > On Wed, May 05, 2010 at 07:24:57PM -0400, Mathieu Desnoyers wrote: > > * Paul E. McKenney (paulmck@linux.vnet.ibm.com) wrote: > > > On Wed, May 05, 2010 at 06:46:41PM -0400, Mathieu Desnoyers wrote: > > > > * Paul E. McKenney (paulmck@linux.vnet.ibm.com) wrote: > > > > > From: Lai Jiangshan > > > > > > > > > > There is no need to disable lockdep after an RCU lockdep splat, > > > > > so remove the debug_lockdeps_off() from lockdep_rcu_dereference(). > > > > > To avoid repeated lockdep splats, use a static variable in the inlined > > > > > rcu_dereference_check() and rcu_dereference_protected() macros so that > > > > > a given instance splats only once, but so that multiple instances can > > > > > be detected per boot. > > > > > > > > > > This is controlled by a new config variable CONFIG_PROVE_RCU_REPEATEDLY, > > > > > which is disabled by default. This provides the normal lockdep behavior > > > > > by default, but permits people who want to find multiple RCU-lockdep > > > > > splats per boot to easily do so. > > > > > > > > I'll play the devil's advocate here. (just because that's so much fun) > > > > ;-) > > > > > > > > If we look at: > > > > > > > > include/linux/debug_locks.h: > > > > > > > > static inline int __debug_locks_off(void) > > > > { > > > > return xchg(&debug_locks, 0); > > > > } > > > > > > > > We see that all code following a call to "debug_locks_off()" can assume > > > > that it cannot possibly run concurrently with other code following > > > > "debug_locks_off()". Now, I'm not saying that the code we currently have > > > > will necessarily break, but I think it is worth asking if there is any > > > > assumption in lockdep (or RCU lockdep more specifically) about mutual > > > > exclusion after debug_locks_off() ? > > > > > > > > Because if there is, then this patch is definitely breaking something by > > > > not protecting lockdep against multiple concurrent executions. > > > > > > So what in lockdep_rcu_dereference() needs to be protected from concurrent > > > execution? All that happens beyond that point is a bunch of printk()s, > > > printing the locks held by this task, and dumping this task's stack. > > > > > > Thanx, Paul > > > > I agree with you that printing the current task information should be safe. > > However, I am not sure that concurrent updates to the lock_class while printk() > > is showing its information will end up doing what we expect it to do. > > > > It could be acceptable to have unreliable information in these rare cases, but > > the important thing would be to ensure that the kernel does not OOPS. > > But any races other than the printk()s can already happen as follows: > > o CPU 0 needs to update some information about the lock. It > checks debug_locks and finds that it is non-zero. > > o CPU 1 detects a deadlock, and invokes __debug_locks_off(), > which atomically sets debug_locks to zero. > > o CPU 1 then proceeds to printk() information that CPU 0 > is concurrently modifying. Which looks to be OK in any case. > > Or is there some other race that cannot already happen that I am > introducing? Nope, I don't think so. Although it's probably worth putting a comment in lockdep_rcu_dereference() to state that lockdep can be used by multiple concurrent instances here, just in case someone ever consider adding code to this splat handler thinking lockdep is always only used by a single "splat" at a time. Thanks, Mathieu > > Thanx, Paul -- Mathieu Desnoyers Operating System Efficiency R&D Consultant EfficiOS Inc. http://www.efficios.com