From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: PATCH: Further aacraid work Date: Thu, 17 Jun 2004 14:19:03 +0100 Sender: linux-scsi-owner@vger.kernel.org Message-ID: <20040617131903.GA31775@infradead.org> References: <547AF3BD0F3F0B4CBDC379BAC7E4189FD23FF9@otce2k03.adaptec.com> <20040617130713.GX20511@parcelfarce.linux.theplanet.co.uk> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from [213.146.154.40] ([213.146.154.40]:31915 "EHLO pentafluge.infradead.org") by vger.kernel.org with ESMTP id S266494AbUFQNTG (ORCPT ); Thu, 17 Jun 2004 09:19:06 -0400 Content-Disposition: inline In-Reply-To: <20040617130713.GX20511@parcelfarce.linux.theplanet.co.uk> List-Id: linux-scsi@vger.kernel.org To: Matthew Wilcox Cc: "Salyzyn, Mark" , Alan Cox , linux-kernel@vger.kernel.org, linux-scsi@vger.kernel.org On Thu, Jun 17, 2004 at 02:07:13PM +0100, Matthew Wilcox wrote: > It's a hard problem. Where's the right tradeoff? I have a system that > has a memory map that puts the first 3.75GB of memory at 0, then the > next 256MB at 64GB, then continues from 4GB. If there's only 4GB of > RAM in that system, I'm sure you'd rather use 32-bit descriptors and > anything in that 256MB gets bounce-buffered. With the SGI iommu it doesn't matter how much memory you have anyway, if you're using DAC you'll always get high bits set in your dma address. I suspect many other iommus work the same.