From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([208.118.235.92]:58137) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ShSCn-0001YT-SQ for qemu-devel@nongnu.org; Wed, 20 Jun 2012 17:12:51 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1ShSCm-0001Z0-3Q for qemu-devel@nongnu.org; Wed, 20 Jun 2012 17:12:49 -0400 Received: from mail-pb0-f45.google.com ([209.85.160.45]:59395) by eggs.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ShSCl-0001YY-UD for qemu-devel@nongnu.org; Wed, 20 Jun 2012 17:12:48 -0400 Received: by pbbro12 with SMTP id ro12so1180578pbb.4 for ; Wed, 20 Jun 2012 14:12:46 -0700 (PDT) Message-ID: <4FE23CCA.20700@codemonkey.ws> Date: Wed, 20 Jun 2012 16:12:42 -0500 From: Anthony Liguori MIME-Version: 1.0 References: <1340087992-2399-1-git-send-email-benh@kernel.crashing.org> In-Reply-To: <1340087992-2399-1-git-send-email-benh@kernel.crashing.org> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Subject: Re: [Qemu-devel] [PATCH 00/13] iommu series List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Benjamin Herrenschmidt Cc: qemu-devel@nongnu.org On 06/19/2012 01:39 AM, Benjamin Herrenschmidt wrote: > This is a rebase of the iommu series and the barrier patch together > on top of current qemu. > > As for our discussions about doing things with Memory Regions etc > I eventually came to the conclusion that we should just apply this > first :-) > > My reasons (other than it makes my life much easier which it does) > are that: This series sucks pretty bad. I don't think it can be a lot better though without major rearchitecting so I'm in favor of applying this now and dealing with the fall-out later. I'll respond in detail where all of the problems are. I don't have easy solutions to offer though. Regards, Anthony Liguori > > - We already have PCI DMA accessors, so devices using those > will be unaffected by further changes > > - The few devices that are modified in this series to use the > DMA accessors directly are ... few, and need to do it essentially > because they either deal with multiple bus types (AHCI, EHCI,...) > or because they are in a separate layer (bdev). Fixing them to > use some other interfaces would be easy (they are few)) and might > be unnecessary as well as we might want (or can) easily keep an > object of type "DMAContext" to represent the DMA capabilities of a > device as the head of the chain of MemoryRegions in a future > more flexible design. > > - It provides a good spot to stick our memory barrier > > - It gives us something working now for 1.2, I know that at least > freescale powerpc and a number of ARM folks are waiting for it. > > Cheers, > Ben. >