From mboxrd@z Thu Jan 1 00:00:00 1970 From: viro@parcelfarce.linux.theplanet.co.uk Subject: Re: Fw: [Bugme-new] [Bug 3125] New: Can't read data from zip250 reiserfs partition Date: Tue, 3 Aug 2004 02:33:15 +0100 Sender: linux-scsi-owner@vger.kernel.org Message-ID: <20040803013315.GP12308@parcelfarce.linux.theplanet.co.uk> References: <20040727125712.71011b65.akpm@osdl.org> <20040802213323.GO12308@parcelfarce.linux.theplanet.co.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from parcelfarce.linux.theplanet.co.uk ([195.92.249.252]:37796 "EHLO www.linux.org.uk") by vger.kernel.org with ESMTP id S264929AbUHCBd0 (ORCPT ); Mon, 2 Aug 2004 21:33:26 -0400 Content-Disposition: inline In-Reply-To: <20040802213323.GO12308@parcelfarce.linux.theplanet.co.uk> List-Id: linux-scsi@vger.kernel.org To: Andrew Morton Cc: linux-scsi@vger.kernel.org On Mon, Aug 02, 2004 at 10:33:23PM +0100, viro@parcelfarce.linux.theplanet.co.uk wrote: > On Tue, Jul 27, 2004 at 12:57:12PM -0700, Andrew Morton wrote: > > > > Looks like a repeatable crash in the imm.c driver. > > Can't reproduce it here. OTOH, randy had been able to; looks like we > might get somewhere tonight (it's on his home box). OK... Here's what's going on: imm is PIO-only (hardly a surprise, since it's a parport ZIP drive). If we have highmem enabled, eventually we get a request with sglist refering to unmapped page. Normally we wouldn't give a fsck - we know that it's accessible to DMA and that's it. However, in this driver we really need CPU access to the damn thing. So imm.c and ppa.c are screwed on highmem kernels and I would expect even more fun on non-x86. AFAICS, other non-DMA drivers are no better - e.g. eata_pio() is broken in the same way, ditto for fdomain.c, etc. Any suggestions re clean fix? I'm seriously tempted to do atomic kmap of pages we are accessing, ugly as it is...