From: James Bottomley Subject: Re: I/O topology fixes for big physical block size Date: Tue, 28 Sep 2010 06:58:04 +0900 Message-ID: <1285624684.2888.120.camel@mulgrave.site> References: <1285605664-27027-1-git-send-email-martin.petersen@oracle.com> <4CA0CC38.5010804@fusionio.com> <20100927172309.GA13874@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Return-path: Received: from cantor2.suse.de ([195.135.220.15]:44382 "EHLO mx2.suse.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750780Ab0I0V6R (ORCPT ); Mon, 27 Sep 2010 17:58:17 -0400 In-Reply-To: <20100927172309.GA13874@redhat.com> Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: Mike Snitzer Cc: Jens Axboe , "Martin K. Petersen" , "linux-scsi@vger.kernel.org" On Mon, 2010-09-27 at 13:23 -0400, Mike Snitzer wrote: > On Mon, Sep 27 2010 at 12:54pm -0400, > Jens Axboe wrote: > > > On 2010-09-28 01:41, Martin K. Petersen wrote: > > > Mike Snitzer reported that he has access to a device that supports thin > > > provisioning but does not use the Block Limits VPD page to indicate > > > discard granularity. Instead it reports a huge (1MB) physical block > > > size. That caused a bit of fallout in the topology stack which assumed a > > > physical block size of 4KiB or less. > > > > Fixing the overflow aside, I question the validity of setting the physical > > block size to something larger than PAGE_SIZE as there's no way that that > > could really work in the current kernel. > > > > I would suggest doing something similar as we do with other 'invalid' > > settings that we cannot honor, print a warning and drop the queue > > limits to PAGE_SIZE. > > I'm inclined to agree. Doesn't make a whole lot of sense. > > But could this cap of PAGE_SIZE be enforced with a follow-on patch? Or > would you rather see it be dealt with in a single revised 2/2 patch? It's up to Jens, but I'd prefer, unless there's a very good reason, that the patches we put into the kernel be right first time around, since that generates a cleaner history and a better example should anyone go looking for one in the tree. James