From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754341Ab0CDAUt (ORCPT ); Wed, 3 Mar 2010 19:20:49 -0500 Received: from ogre.sisk.pl ([217.79.144.158]:44887 "EHLO ogre.sisk.pl" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753153Ab0CDAUm (ORCPT ); Wed, 3 Mar 2010 19:20:42 -0500 From: "Rafael J. Wysocki" To: Alan Stern Subject: Re: [linux-pm] Is it supposed to be ok to call del_gendisk while userspace is frozen? Date: Thu, 4 Mar 2010 01:23:13 +0100 User-Agent: KMail/1.12.4 (Linux/2.6.33-git-rjw; KDE/4.3.5; x86_64; ; ) Cc: Pavel Machek , Jens Axboe , Maxim Levitsky , "linux-pm" , "linux-kernel" , Andrew Morton References: In-Reply-To: MIME-Version: 1.0 Content-Type: Text/Plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Message-Id: <201003040123.13617.rjw@sisk.pl> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wednesday 03 March 2010, Alan Stern wrote: > On Wed, 3 Mar 2010, Pavel Machek wrote: > > > Hi! > > > > > > > The reason for freezing those tasks is to avoid writebacks at random > > > > > times during a system sleep transition, when the underlying device may > > > > > already be suspended, right? > > > > > > > > It is also there to avoid inconsistency between in-filesystem data and > > > > snapshot in hibernation image. > > > > > > A good point, although in this case I think it won't matter. Writing > > > out a dirty page twice (once right after taking the snapshot and then > > > again after resuming from hibernation) will leave the disk in a correct > > > state. > > > > No, I don't think so. Have you considered all the various journalling > > systems? > > > > Definitely not in presence of I/O errors. Commit block can only be > > written after previous blocks are successfully writen to the journal. > > > > So lets see: > > > > > > > > Write previous block, write commit block, write more blocks > > > > > > > > Error writing previous block (block now contains garbage), leading to > > kernel panic > > > > > > > > journalling assumptions broken: commit block is there, but previous > > blocks are not intact. Data loss. > > > > ...and that was the first I could think about. Lets not do > > this. Barriers were invented for a reason. > > Very well. Then we still need a solution to the original problem: > Devices sometimes need to be unregistered during resume, but > del_gendisk() blocks on the writeback thread, which is frozen until > after the resume finishes. How do you suggest this be fixed? I thought about thawing the writeback thread earlier in such cases. Would that makes sense / is it doable at all? Rafael