From mboxrd@z Thu Jan 1 00:00:00 1970 From: Mike Snitzer Subject: linux-next dm-thinp regression due to "bio: skip atomic inc/dec of ->bi_remaining for non-chains" Date: Wed, 13 May 2015 15:28:27 -0400 Message-ID: <20150513192827.GA15831@redhat.com> Reply-To: device-mapper development Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: Content-Disposition: inline List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: dm-devel-bounces@redhat.com Errors-To: dm-devel-bounces@redhat.com To: axboe@kernel.dk Cc: dm-devel@redhat.com, hch@lst.de List-Id: dm-devel.ids The device-mapper-test-suite (dmts) thinp tests no longer run due to linux-block.git commit c4cf5261 ("bio: skip atomic inc/dec of ->bi_remaining for non-chains") That commit has been the focus of some other work I have pending because I was forced to adapt it due to bi_remaining no longer being accessible directly. So it was surprising/interesting for this commit to take center stage in the context of dm-thinp regression. bios don't get submitted (simplest reproducer is the 'dd_benchmark' test in the dmts thin-provisioning suite which just uses dd with oflag=direct immediately after the DM thin device is created). The top-level DM device stays at 100% utilization but no progress is made. Reverting the commit in question resolves the problem. But I'm able to use lvm2 to create thin devices that allow IO to complete; so I haven't figured out what is so special about dmts (which uses dmsetup directly instead of lvm2). I'll see if I can figure out what might be happening... but would welcome more eyes on this to see if anything stands out relative to the dm-thin.c changes.