From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1030209AbVKCOvN (ORCPT ); Thu, 3 Nov 2005 09:51:13 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1030233AbVKCOvN (ORCPT ); Thu, 3 Nov 2005 09:51:13 -0500 Received: from clock-tower.bc.nu ([81.2.110.250]:29377 "EHLO lxorguk.ukuu.org.uk") by vger.kernel.org with ESMTP id S1030209AbVKCOvM (ORCPT ); Thu, 3 Nov 2005 09:51:12 -0500 Subject: Re: ECC circuitry error / md weirdness? From: Alan Cox To: PinkFreud Cc: linux-kernel@vger.kernel.org In-Reply-To: <20051102155711.GT19490@eriador.mirkwood.net> References: <20051102155711.GT19490@eriador.mirkwood.net> Content-Type: text/plain Content-Transfer-Encoding: 7bit Date: Thu, 03 Nov 2005 15:21:14 +0000 Message-Id: <1131031274.18848.50.camel@localhost.localdomain> Mime-Version: 1.0 X-Mailer: Evolution 2.2.3 (2.2.3-2.fc4) Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Mer, 2005-11-02 at 10:57 -0500, PinkFreud wrote: > We have an md array (RAID5) with 3 disks + 1 spare. Recently, this > appeared in the logs: > > Oct 27 23:44:58 cbs-server kernel: hdk: status timeout: status=0x80 { > Busy } > Oct 27 23:44:58 cbs-server kernel: > Oct 27 23:44:58 cbs-server kernel: hdk: DMA disabled > Oct 27 23:44:58 cbs-server kernel: PDC202XX: Secondary channel reset. > Oct 27 23:44:58 cbs-server kernel: hdk: drive not ready for command > Oct 27 23:45:04 cbs-server kernel: ide5: reset: master: ECC circuitry > error > Oct 27 23:45:04 cbs-server kernel: hdk: status error: status=0x58 { > DriveReady SeekComplete DataRequest } > > After that was just a repetition of the 'drive not ready for command' > and status=0x58 lines. > > What really threw me for a loop, though, was the fact that hdk was one > of the active disks in the array mentioned above. md was happily > writing to a disk that the kernel thought was failing! I had to > manually fail the disk out of the array to convince md to pull the > spare in. > > The end result is one hell of a corrupt filesystem (I'm now seeing > 'ghost' files that won't go away): > > [root@cbs-server cope11.feat]# ls -al | grep example_func.nii.gz > [root@cbs-server cope11.feat]# ls -al example_func.nii.gz > ls: example_func.nii.gz: Input/output error > [root@cbs-server cope11.feat]# > > fsck has had no luck in fixing these errors, though it does find > - and fix - problems every time I run it (ext3 fs). > > I suspect I'm going to have to mkfs the array (unless someone can > recommend something else!). My main concern, though, is figuring out > what went wrong with hdk and md in the first place. I've never seen > the ECC circuitry error that was thrown before. AFAICT, the hard disk > appears to be fine. It's about 3 months old, and both SMART offline > data collection and extended self test were run last night without a > single error being logged by the drive. Likewise, it stopped throwing > errors in the system logs when it was failed out of the array. > > I'm also concerned about why md was writing to a disk that the kernel > saw as having errors. Should it not fail the disk out of the array > automatically? > > Specs on the system in question: > 2.4.31 (vanilla) SMP > 2 Promise 20268 IDE controllers > 4 WDC WD3200SB-01KMA0 disks > >