From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753623AbZBXDhe (ORCPT ); Mon, 23 Feb 2009 22:37:34 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751906AbZBXDhS (ORCPT ); Mon, 23 Feb 2009 22:37:18 -0500 Received: from e4.ny.us.ibm.com ([32.97.182.144]:47826 "EHLO e4.ny.us.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751592AbZBXDhQ (ORCPT ); Mon, 23 Feb 2009 22:37:16 -0500 Date: Mon, 23 Feb 2009 19:37:12 -0800 From: "Paul E. McKenney" To: Nick Piggin Cc: Ingo Molnar , Andrew Morton , Vegard Nossum , stable@kernel.org, Nick Piggin , Pekka Enberg , linux-kernel@vger.kernel.org Subject: Re: [PATCH] mm: fix lazy vmap purging (use-after-free error) Message-ID: <20090224033712.GA7173@linux.vnet.ibm.com> Reply-To: paulmck@linux.vnet.ibm.com References: <20090220140157.GA12799@elte.hu> <20090223193351.GA11921@elte.hu> <20090223204359.GT6751@linux.vnet.ibm.com> <200902241423.21091.nickpiggin@yahoo.com.au> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <200902241423.21091.nickpiggin@yahoo.com.au> User-Agent: Mutt/1.5.15+20070412 (2007-04-11) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Tue, Feb 24, 2009 at 02:23:19PM +1100, Nick Piggin wrote: > On Tuesday 24 February 2009 07:43:59 Paul E. McKenney wrote: > > On Mon, Feb 23, 2009 at 08:33:51PM +0100, Ingo Molnar wrote: > > > * Andrew Morton wrote: > > > > On Mon, 23 Feb 2009 08:17:26 -0800 "Paul E. McKenney" > wrote: > > > > > On Tue, Feb 24, 2009 at 12:29:36AM +1100, Nick Piggin wrote: > > > > > > On Monday 23 February 2009 16:17:09 Paul E. McKenney wrote: > > > > > > > The boot CPU runs in the context of its idle thread during > > > > > > > boot-up. During this time, idle_cpu(0) will always return > > > > > > > nonzero, which will fool Classic and Hierarchical RCU into > > > > > > > deciding that a large chunk of the boot-up sequence is a big long > > > > > > > quiescent state. This in turn causes RCU to prematurely end > > > > > > > grace periods during this time. > > > > > > > > > > > > > > This patch creates a new global variable that is set to 1 just > > > > > > > before the boot CPU first enters the scheduler, after which the > > > > > > > idle task really is idle. > > > > > > > > > > > > Nice work all (btw. if this patch goes in rather than using > > > > > > system_state, then please make the variable __read_mostly). > > > > > > > > > > Hmmm... I misread this and made system_state be __read_mostly. Let > > > > > me know if this is bad, easy to fix if needed. > > > > > > > > Please don't use system_state. The whole thing is just bad > > > > design. It's a global variable, breaks encapsulation, creates > > > > interactions etc. CS-101 stuff. > > > > > > ok, i've removed the patch - Paul, would you mind to re-send > > > your original flag solution, with it marked __read_mostly and > > > with the extern declarations put into a suitable header file? > > > > > > Paul, incidentally, this very minute i tracked down that the > > > patch is also causing boot lockups in -tip testing. I havent yet > > > fully debugged it, but a question comes immediately: if there's > > > no grace periods during bootup, wont rcu_sync() & friends just > > > hang indefinitely? > > > > Ouch!!! Indeed they would. > > > > > More thought is needed. > > > > One fix would be to sprinkle calls to rcu_qsctr_inc() through the > > boot process. But a much better approach would be for me to make > > synchronize_rcu() check this same flag, and simply return if called > > during early boot. The rationale for this is that there is but a single > > CPU during early boot, so tinyrcu.c's optimization can be used. ;-) > > Well can you simply return if called if num_online_cpus() == 1, regardless > of the state of boot? Yep! And that is indeed what I do in http://lkml.org/lkml/2009/2/23/305. Thanx, Paul > > Out of both paranoia and self defense, I would check num_online_cpus() > > in my proposed call into RCU. ;-) > > > > Seem reasonable? And does synchronize_sched() also need the UP-only > > optimization? >