From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1763773AbZFOP52 (ORCPT ); Mon, 15 Jun 2009 11:57:28 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1763446AbZFOP5A (ORCPT ); Mon, 15 Jun 2009 11:57:00 -0400 Received: from courier.cs.helsinki.fi ([128.214.9.1]:43907 "EHLO mail.cs.helsinki.fi" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1763494AbZFOP47 (ORCPT ); Mon, 15 Jun 2009 11:56:59 -0400 Subject: Re: [GIT PULL v2] Early SLAB fixes for 2.6.31 From: Pekka Enberg To: Christoph Lameter Cc: torvalds@linux-foundation.org, linux-kernel@vger.kernel.org, akpm@linux-foundation.org, kamezawa.hiroyu@jp.fujitsu.com, lizf@cn.fujitsu.com, mingo@elte.hu, npiggin@suse.de, yinghai@kernel.org, benh@kernel.crashing.org In-Reply-To: References: <1245077891.23207.48.camel@penberg-laptop> <1245078661.23207.50.camel@penberg-laptop> <84144f020906150827k1b8c0ce4q5ed4d140f7ca869@mail.gmail.com> Date: Mon, 15 Jun 2009 18:57:00 +0300 Message-Id: <1245081420.23207.58.camel@penberg-laptop> Mime-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: 7bit X-Mailer: Evolution 2.24.3 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 2009-06-15 at 11:51 -0400, Christoph Lameter wrote: > On Mon, 15 Jun 2009, Pekka Enberg wrote: > > > > OK, I can clean this up, no problem. > > > > Actually, there's a slight complication here. If I push gfp mask to > > __might_sleep(), lockdep_trace_alloc() and so on, the mask is > > effective _everywhere_ even outside of slab. Yes, it makes sense if we > > push the masking right down to the page allocator but I wonder if > > that's something we want to do at this point? > > __might_sleep just should not trigger right? The mask does not need to be > passed. __might_sleep may be called uselessly during bootup if __GFP_WAIT > is set. But it should not trigger any output. Look at the initial > statements of __might_sleep: They are already prepared to simply return in > the early boot case. Oh, yeah, you're right about __might_sleep() so we can just ignore that. But we still have bit of a problem for should_failslab() and __lockdep_trace_alloc(). We might be able to deal with the former by adding system_state check but for the latter, we need to mask the gfp flags. Hmm. Pekka