From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jeff Garzik Subject: Re: qla2xxx and feral ISP updates in their respective BK trees Date: Fri, 20 Jun 2003 17:59:45 -0400 Sender: linux-scsi-owner@vger.kernel.org Message-ID: <3EF383D1.7070900@pobox.com> References: <1055971780.2075.484.camel@mulgrave> <20030620101224.B305@wonky.in0.lcl> <1056129548.2102.19.camel@mulgrave> <20030620102448.O305@wonky.in0.lcl> <1056133068.1804.22.camel@mulgrave> <20030620183417.GB9164@gtf.org> <20030620113759.B305@wonky.in0.lcl> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii; format=flowed Content-Transfer-Encoding: 7bit Return-path: Received: from parcelfarce.linux.theplanet.co.uk ([195.92.249.252]:14771 "EHLO www.linux.org.uk") by vger.kernel.org with ESMTP id S264899AbTFTVp6 (ORCPT ); Fri, 20 Jun 2003 17:45:58 -0400 In-Reply-To: <20030620113759.B305@wonky.in0.lcl> List-Id: linux-scsi@vger.kernel.org To: mjacob@feral.com Cc: James Bottomley , SCSI Mailing List , Andrew Vasquez , Andrew Morton Matthew Jacob wrote: > a) Historically, MMIO has been buggy on ia32. This is no longer the > case, but there definitely were a lot of platforms with this broken. Could you elaborate? What I've mostly found is (in order of frequency): coders unfamiliar with x86 CPU write coalescing/posting rules, and PCI boards whose hardware designs contain inherent races when driven in MMIO mode (particularly on CPUs with deeply posted writes). > b) MMIO is consumption of mapping resources, while PIO isn't. agreed Jeff