From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S932415AbWEMMxx (ORCPT ); Sat, 13 May 2006 08:53:53 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S932419AbWEMMxx (ORCPT ); Sat, 13 May 2006 08:53:53 -0400 Received: from mail.gmx.net ([213.165.64.20]:62645 "HELO mail.gmx.net") by vger.kernel.org with SMTP id S932415AbWEMMxw (ORCPT ); Sat, 13 May 2006 08:53:52 -0400 X-Authenticated: #14349625 Subject: Re: swapping and oom-killer: gfp_mask=0x201d2, order=0 From: Mike Galbraith To: Al Boldi Cc: linux-kernel@vger.kernel.org In-Reply-To: <200605131511.05723.a1426z@gawab.com> References: <200605111514.45503.a1426z@gawab.com> <200605121517.59988.a1426z@gawab.com> <1147447913.7520.6.camel@homer> <200605131511.05723.a1426z@gawab.com> Content-Type: text/plain Date: Sat, 13 May 2006 14:54:07 +0200 Message-Id: <1147524848.9829.20.camel@homer> Mime-Version: 1.0 X-Mailer: Evolution 2.4.0 Content-Transfer-Encoding: 7bit X-Y-GMX-Trusted: 0 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Sat, 2006-05-13 at 15:11 +0300, Al Boldi wrote: > Mike Galbraith wrote: > > On Fri, 2006-05-12 at 15:17 +0300, Al Boldi wrote: > > > Note that this is not specific to mem=8M, but rather a general oom > > > observation even for mem=4G, where it is only much later to occur. > > > > An oom situation with 4G ram would be more interesting than this one. > > Agreed, but can you tell me what readahead has to do with this oom? > > oom-killer: gfp_mask=0x201d2, order=0 > [] out_of_memory+0xa5/0xc0 > [] __alloc_pages+0x279/0x310 > [] __do_page_cache_readahead+0xe9/0x120 > [] max_sane_readahead+0x2f/0x50 > [] filemap_nopage+0x2eb/0x370 > [] do_no_page+0x65/0x220 > [] __handle_mm_fault+0xec/0x200 > [] do_page_fault+0x188/0x5c5 > [] do_page_fault+0x0/0x5c5 > [] error_code+0x4f/0x54 Nothing except that it asked for a page at a bad time, triggering the bad-hair-day reaction. That being said, the readahead allocation mask should have probably included GFP_NORETRY. (though with 8MB, if the readahead didn't get you, the subsequent read probably would anyway) -Mike