From mboxrd@z Thu Jan 1 00:00:00 1970 From: James Bottomley Subject: Re: [PATCH 0/19] clean ups on the drivers Date: Tue, 15 May 2007 12:57:25 -0400 Message-ID: <1179248245.3685.23.camel@mulgrave.il.steeleye.com> References: <20070512190534B.fujita.tomonori@lab.ntt.co.jp> <20070512153023.GA8088@infradead.org> <200705141440.l4EEe3dB004194@mbox.iij4u.or.jp> <4648829A.10602@s5r6.in-berlin.de> <20070515090116.GA9297@infradead.org> <1179230056.3685.2.camel@mulgrave.il.steeleye.com> <20070515115748.GA12461@infradead.org> Mime-Version: 1.0 Content-Type: text/plain Content-Transfer-Encoding: 7bit Return-path: Received: from hancock.steeleye.com ([71.30.118.248]:46142 "EHLO hancock.sc.steeleye.com" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1757277AbXEORFy (ORCPT ); Tue, 15 May 2007 13:05:54 -0400 In-Reply-To: <20070515115748.GA12461@infradead.org> Sender: linux-scsi-owner@vger.kernel.org List-Id: linux-scsi@vger.kernel.org To: Christoph Hellwig Cc: Stefan Richter , FUJITA Tomonori , linux-scsi@vger.kernel.org, jens.axboe@oracle.com On Tue, 2007-05-15 at 12:57 +0100, Christoph Hellwig wrote: > On Tue, May 15, 2007 at 06:54:16AM -0500, James Bottomley wrote: > > Er ... I really hope not ... that's exactly how the parisc iommu > > platform code works ... and why I designed the generic dma mapping this > > way. The key thing parisc needed was the ability to walk up different > > busses until it found the iommu (for example the pci bus -> dino -> GSC > > -> IOMMU) which it does by traversing the dev->parent; However, I didn't > > mandate working this way for other architectures. > > Well, the NACK was not for the implementation details but rather the > exported and documented interface. IIRC the rationale was that Dave wants > to keep the sparc implementation super-optimized and avoid indirection. Well, actually, I do understand that. How about the weaker requirement that you be able to call the dma_ functions on any dev (provided it's properly parented) and leave the implementation up to the platform? > If it was up to me I'd have something like the parisc implementation in > generic code. It might make sense to put it in lib ... however, I don't think many architectures have the problems we have ... specifically certain boxes can have >1 IOMMU, then you really have to know *which* iommu you're programming. James