From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from bombadil.infradead.org ([198.137.202.9]:42620 "EHLO bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755940AbcLTGkS (ORCPT ); Tue, 20 Dec 2016 01:40:18 -0500 Date: Mon, 19 Dec 2016 22:40:17 -0800 From: Christoph Hellwig To: Dan Williams Cc: Dave Jiang , linux-fsdevel , "linux-nvdimm@lists.01.org" Subject: Re: [PATCH v2 3/3] dev-dax: add fallocate support to clear poison Message-ID: <20161220064017.GA8390@infradead.org> References: <148219023506.15885.3919571427691230959.stgit@djiang5-desk3.ch.intel.com> <148219024781.15885.1883996216561314302.stgit@djiang5-desk3.ch.intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Mon, Dec 19, 2016 at 04:46:00PM -0800, Dan Williams wrote: > In the case of filesystem-dax (xfs or ext4) a punch hole operation is > effective for clearing a media error. Which at best is a third order side effect of what it actually does: release allocated space. Which is something we can do on various block devices (SCSI with thin provisioning, device mapper) as well. > It will deallocate blocks and > the zeroing that is performed when the blocks are reallocated will > clear media errors. We want the same semantic for device-dax. > FALLOC_FL_PUNCH_HOLE triggers the driver stack to clear any media > errors, although unlike the filesystem case there are no physical > ranges de-allocated from the device. With support for this subset of > fallocate we can avoid needing to add a custom ioctl for similar > functionality. It's in no way similar functionality. It's absuing an interface for something entirely different. So NAK to this series.