From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Martin K. Petersen" Subject: Re: I/O topology fixes for big physical block size Date: Mon, 27 Sep 2010 18:36:24 -0400 Message-ID: References: <1285605664-27027-1-git-send-email-martin.petersen@oracle.com> <4CA0CC38.5010804@fusionio.com> <4CA118FF.1080100@fusionio.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from rcsinet10.oracle.com ([148.87.113.121]:22232 "EHLO rcsinet10.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932558Ab0I0Wgn (ORCPT ); Mon, 27 Sep 2010 18:36:43 -0400 In-Reply-To: <4CA118FF.1080100@fusionio.com> (Jens Axboe's message of "Tue, 28 Sep 2010 07:21:51 +0900") Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: Jens Axboe Cc: "Martin K. Petersen" , "James.Bottomley@hansenpartnership.com" , "snitzer@redhat.com" , "linux-scsi@vger.kernel.org" >>>>> "Jens" == Jens Axboe writes: Jens> So it's just the hint, not the actual hardware sector size. The Jens> naming is pretty bad on that, physical and logic... This is the official T10/T13 terminology. Logical block size is the unit used when talking to the device via ACS/SBC. Physical block size is the unit used by the device internally. There's a pretty vague description of what that means in SBC. But it was obviously conceived mainly to handle disk drives with 512-byte logical and 4096-byte physical (hardware) sectors. The notion of using the physical block size as an indicator of erase block size is something this unknown vendor came up with. And 1MB is quite the ways out of range for what we expect in this field. But my point is that devices with physical block sizes bigger than the system page size are already out there. So we need to do the right thing here. And since this is not something affecting runtime behavior I'm in favor of deferring the decision what to do about it to userland. Jens> So that does look better, but in that case I don't think that sd Jens> should dump a warning. That's fine by me. Jens> Does mkfs do the right thing? Depends on which mkfs it is. Mike has tested things and can chip in here... -- Martin K. Petersen Oracle Linux Engineering