From: Andreas Dilger Subject: Re: [Bug 17361] Watchdog detected hard LOCKUP in jbd2_journal_get_write_access Date: Mon, 11 Oct 2010 14:43:58 -0600 Message-ID: References: <201010021642.o92Ggp1A007965@demeter1.kernel.org> Mime-Version: 1.0 (Apple Message framework v1081) Content-Type: text/plain; charset=us-ascii Content-Transfer-Encoding: 8BIT Cc: linux-ext4 development To: Christian Casteyde Return-path: Received: from idcmail-mo2no.shaw.ca ([64.59.134.9]:49689 "EHLO idcmail-mo2no.shaw.ca" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755126Ab0JKUrn convert rfc822-to-8bit (ORCPT ); Mon, 11 Oct 2010 16:47:43 -0400 In-Reply-To: <201010021642.o92Ggp1A007965@demeter1.kernel.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: On 2010-10-02, at 09:42, Theodore Tso wrote: > https://bugzilla.kernel.org/show_bug.cgi?id=17361 > > Except for the initial bug report, none of the other stack traces have anything to do with ext4/jbd2. And in the initial ext4 trace, we see the complaint that we're calling might_sleep() in ext4_mark_inode_dirty(), in a code path where we are manifestly not taking any spinlocks. And in fact we don't see any spinlocks being taken at the point where the complaint is mode in ext4_mark_inode_dirty(). Yet preempt_count > 1. In my experience, this is typically caused by stack overflow smashing the task struct and randomly setting preempt_count to a non-zero value. Cheers, Andreas