From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from imap.thunk.org ([74.207.234.97]:34026 "EHLO imap.thunk.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932600AbdIFOuI (ORCPT ); Wed, 6 Sep 2017 10:50:08 -0400 Date: Wed, 6 Sep 2017 10:50:04 -0400 From: "Theodore Ts'o" Subject: Re: test the per-inode DAX flag Message-ID: <20170906145004.hlkwta2a4bjlarin@thunk.org> References: <20170903093325.GA16272@lst.de> <20170904072431.GW27835@eguan.usersys.redhat.com> <20170904154453.GA6771@lst.de> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20170904154453.GA6771@lst.de> Sender: fstests-owner@vger.kernel.org To: Christoph Hellwig Cc: Eryu Guan , fstests@vger.kernel.org, linux-xfs@vger.kernel.org List-ID: On Mon, Sep 04, 2017 at 05:44:53PM +0200, Christoph Hellwig wrote: > > Does this test make sense when filesystem was mounted without dax > > option? I saw these failures when testing on normal block device without > > dax mount option. > > And thinking about it - why would we not allow setting the flag, > especially given that right now it doesn't have a meaning either > with or without DAX.. At least for the ext4 patches that have been crossing my screen, the claim from the commit description is that with the dax mount option, all files are treated as if the DAX flag. And the point of the DAX flag is that if the file system is stored on DAX-capable hardware, and you don't want DAX treatment for all files, that you would set the DAX flag only for those files where having the file system actually using the DAX codepaths would make sense --- as opposed to just using the traditional block I/O paths against the /dev/pmem driver. So I *think* the answer is that the tax makes sense when the file system is mounted w/o the dax option, but we do need to check to see whether the block device is DAX capable. It also makes an interesting question how we should be testing DAX capable file systems on /dev/pmem0. Right now I force the use of /dev/pmem0 *and* use the DAX mount options so that all of the other tests will cause file I/O to use the DAX path. But for the per-inode DAX flag, we would want to mask out the DAX mount option so we can test what happens when we toggle the DAX. But I'm not a DAX expert, so I may have some of this wrong. - Ted