From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754324Ab3AXO5O (ORCPT ); Thu, 24 Jan 2013 09:57:14 -0500 Received: from mga01.intel.com ([192.55.52.88]:48034 "EHLO mga01.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753890Ab3AXO5M (ORCPT ); Thu, 24 Jan 2013 09:57:12 -0500 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="4.84,530,1355126400"; d="scan'208";a="277925414" Date: Thu, 24 Jan 2013 22:57:07 +0800 From: Fengguang Wu To: paul.szabo@sydney.edu.au Cc: linux-mm@kvack.org, 695182@bugs.debian.org, linux-kernel@vger.kernel.org, Andrew Morton , Jan Kara Subject: Re: [PATCH] Negative (setpoint-dirty) in bdi_position_ratio() Message-ID: <20130124145707.GB12745@localhost> References: <201301200002.r0K02Atl031280@como.maths.usyd.edu.au> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <201301200002.r0K02Atl031280@como.maths.usyd.edu.au> User-Agent: Mutt/1.5.21 (2010-09-15) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Paul, > (This patch does not solve the PAE OOM issue.) You may try the below debug patch. The only way the writeback patches should trigger OOM, I think, is for the number of dirty/writeback pages going out of control. Or more simple, you may show us the OOM dmesg which will contain the number of dirty pages. Or run this in a continuous loop during your tests, and see how the dirty numbers change before OOM: while : do grep -E '(Dirty|Writeback)' /proc/meminfo sleep 1 done Thanks, Fengguang diff --git a/mm/page-writeback.c b/mm/page-writeback.c index 50f0824..cf1165a 100644 --- a/mm/page-writeback.c +++ b/mm/page-writeback.c @@ -1147,6 +1147,16 @@ pause: if (task_ratelimit) break; + if (nr_dirty > dirty_thresh + dirty_thresh / 2) { + if (printk_ratelimit()) + printk(KERN_WARNING "nr_dirty=%lu dirty_thresh=%lu task_ratelimit=%lu dirty_ratelimit=%lu pos_ratio=%lu\n", + nr_dirty, + dirty_thresh, + task_ratelimit, + dirty_ratelimit, + pos_ratio); + } + /* * In the case of an unresponding NFS server and the NFS dirty * pages exceeds dirty_thresh, give the other good bdi's a pipe