Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S267926AbUJHNOA (ORCPT ); Fri, 8 Oct 2004 09:14:00 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S269952AbUJHNOA (ORCPT ); Fri, 8 Oct 2004 09:14:00 -0400 Received: from parcelfarce.linux.theplanet.co.uk ([195.92.249.252]:28346 "EHLO www.linux.org.uk") by vger.kernel.org with ESMTP id S267926AbUJHNN4 (ORCPT ); Fri, 8 Oct 2004 09:13:56 -0400 Date: Fri, 8 Oct 2004 08:06:37 -0300 From: Marcelo Tosatti To: Leonid Petrov Cc: linux-kernel@vger.kernel.org, user@lpetrov.net Subject: Re: Out of Memory: Killed process while 1 Gb is free Message-ID: <20041008110636.GF16028@logos.cnet> References: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.5.5.1i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2081 Lines: 49 On Thu, Sep 30, 2004 at 10:50:36AM -0400, Leonid Petrov wrote: > Summary: > "Out of Memory: Killed process" when more than 1 Gb of free memory > is available. > > Description: > My Linux-server after 2-10 days starts killing processes claiming > that it is "Out of memory". Top does not show any processess which > occupy more than several percent of memory. It always occur when > the amount of available memory (1Gb) is about half of the total > memory (2Gb). Killing a process does not help and the kernel > continue mass murder till the state then it becomes non-operational. > During agony vmstat showed that the occupied 1Gb of memory was used > mainly by cache and buffers. Analazing kernel log I found 245 events > for last 3 weeks. All events occured when anount of Normal memory > was in the range [0,2052]Kb (and about 1Gb of free high memory). > It did not occur before upgrade to 2.6.8.1. Out of memory events > occured in both Hyperthreading enabled and Hyperthreading disabled > mode. > > Keywords: > Virtual memory managment Hi, There have been several changes in this area since 2.6.8.1, it would be nice if you can repeat the tests with a recent -mm kernel (where most of the changes are). One information is that there are not many freeable pages in the normal zone (active+inactive) lists: DMA free:1904kB min:16kB low:32kB high:48kB active:112kB inactive:0kB present:16384kB Sep 30 08:53:22 lacerta kernel: protections[]: 8 476 732 Sep 30 08:53:22 lacerta kernel: Normal free:1512kB min:936kB low:1872kB high:2808kB active:216kB inactive:436kB present:901120kB For me it looks like you might be seeing a kernel memory leak (what else could cause such condition?) IIRC quite some kernel memory leaks have been fixed since 2.6.8.1 (cdrom driver, NFS, ...) - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/