From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755010Ab3FFVXJ (ORCPT ); Thu, 6 Jun 2013 17:23:09 -0400 Received: from mail-pd0-f175.google.com ([209.85.192.175]:59456 "EHLO mail-pd0-f175.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753117Ab3FFVXG (ORCPT ); Thu, 6 Jun 2013 17:23:06 -0400 Date: Thu, 6 Jun 2013 14:23:03 -0700 From: Tejun Heo To: =?utf-8?B?6Z+m5aWH?= Cc: torvalds , linux-kernel Subject: Re: race condition in schedule_on_each_cpu() Message-ID: <20130606212303.GH5045@htj.dyndns.org> References: MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: 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 Hello, On Thu, Jun 06, 2013 at 06:14:46PM +0800, 韦奇 wrote: > Hello, Tejun Heo > thanks for your help, > 1) I've test the two kernel version on this problem: > latest 3.10-rc3：(https://www.kernel.org/pub/linux/kernel/v3.x/testing/linux-3.10-rc3.tar.xz) > latest 3.0branch - 3.0.80：(https://www.kernel.org/pub/linux/kernel/v3.x/linux-3.0.80.tar.xz > > they all work fine when hot remove raid disk.. Thanks for verifying. > | preemption | machine 1 | machine 2 | kversion | > ------------------------------------------------------------------------------- > | Fully Preemptible | stuck | no stuck | 3.0.30-rt50 | > | Low-Latency Desktop | no stuck | no stuck | 3.0.30-rt50 | > | Low-Latency Desktop | no stuck | -- | 3.0.30 | > | default | no stuck | -- | 3.0.80 | > | default | no stuck | -- | 3.10-rc3 | > > could you tell me some way to debug this problem. for example, how > to debug workqueue deadlock? I want to find the deadlock point. I looked through the logs but the only worker depletion related patches which pop up are around CPU hotplugs, so I don't think they apply here. If the problem is relatively easy to reproduce && you can't move onto a newer kernel, I'm afraid bisection probably is the best option. Thanks! -- tejun