From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: [PATCH 3/5] aacraid: update dma mapping Date: Wed, 30 Nov 2005 18:49:51 +0000 Message-ID: <20051130184951.GA18583@infradead.org> References: <1133375778.2288.23.camel@markh1.pdx.osdl.net> <1133376327.3312.43.camel@mulgrave> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from pentafluge.infradead.org ([213.146.154.40]:15013 "EHLO pentafluge.infradead.org") by vger.kernel.org with ESMTP id S1751456AbVK3Stz (ORCPT ); Wed, 30 Nov 2005 13:49:55 -0500 Content-Disposition: inline In-Reply-To: <1133376327.3312.43.camel@mulgrave> Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: James Bottomley Cc: Mark Haverkamp , linux-scsi , Mark Salyzyn On Wed, Nov 30, 2005 at 12:45:27PM -0600, James Bottomley wrote: > On Wed, 2005-11-30 at 10:36 -0800, Mark Haverkamp wrote: > > This patch changes map_single to map_page in compliance with > > recommendations made in the IO-Perf howto document to prevent bounce > > buffering of DMA requests. > > Erm, there's a misunderstanding somewhere. You're replacing correct > pci_map_single calls with its more difficult to use pci_map_page. This > has absolutely zero performance impact and adds the risk that because > pci_map_page is more complex to use, someone gets it wrong somewhere. > Really, for block devices pci_map_single is the absolutely correct > interface. pci_map_page should really only be used by the net stack for > packet fragment mapping. And the really nice thing is that this whole non-S/G I/O path will go away completely for 2.6.16 :)