From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754082Ab3LPM4T (ORCPT ); Mon, 16 Dec 2013 07:56:19 -0500 Received: from mail-qe0-f41.google.com ([209.85.128.41]:43446 "EHLO mail-qe0-f41.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753806Ab3LPM4Q (ORCPT ); Mon, 16 Dec 2013 07:56:16 -0500 Date: Mon, 16 Dec 2013 07:56:04 -0500 From: Tejun Heo To: Dave Chinner Cc: "Rafael J. Wysocki" , Jens Axboe , tomaz.solc@tablix.org, aaron.lu@intel.com, linux-kernel@vger.kernel.org, Oleg Nesterov , Greg Kroah-Hartman , Fengguang Wu Subject: Re: Writeback threads and freezable Message-ID: <20131216125604.GD32509@htj.dyndns.org> References: <20131213174932.GA27070@htj.dyndns.org> <20131214015343.GP31386@dastard> <20131214202324.GA4020@htj.dyndns.org> <20131216035652.GY31386@dastard> <20131216125124.GC32509@htj.dyndns.org> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20131216125124.GC32509@htj.dyndns.org> 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 On Mon, Dec 16, 2013 at 07:51:24AM -0500, Tejun Heo wrote: > On Mon, Dec 16, 2013 at 02:56:52PM +1100, Dave Chinner wrote: > > > What are you suggesting? Implementing separate warm and hot unplug > > > paths? That makes no sense whatsoever. Device hot unplug is just a > > > sub operation of general device unplug which should be able to succeed > > > whether the underlying device is failing IOs or not. > > > > I don't care. Trying to issue IO from an an IO error handling path > > where the device has just been removed is fundamentally broken. > > What? Have you even read the original message? IO error handling > path isn't issuing the IO here. The hot unplug operation is > completely asynchronous to the IO path. What's dead locking is not > the filesystem and IO path but device driver layer and hot unplug > path. IOs are not stalled. In fact, this deadlock can be reproduced without hotunplug at all. If you initiate warm unplug and warm unplugging races with suspend/resume cycle, it'll behave exactly the same - the IOs from flushing in that scenario would succeed but IOs failing or not has *NOTHING* to do with this deadlock. It'd still happen. It's freezer behaving as a giant lock and other locks of course getting dragged into giant dependency loop. Can you please at least *try* to understand what's going on before throwing strong assertions? -- tejun