From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [Qemu-devel] [PATCH 1/2] Type-safe ioport callbacks Date: Tue, 26 Oct 2010 19:18:21 +0200 Message-ID: <4CC70D5D.4020109@redhat.com> References: <1287934469-16624-1-git-send-email-avi@redhat.com> <1287934469-16624-2-git-send-email-avi@redhat.com> <4CC55554.4060103@redhat.com> <4CC68BE1.5010207@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit Cc: Anthony Liguori , qemu-devel@nongnu.org, kvm@vger.kernel.org To: Blue Swirl Return-path: Received: from mx1.redhat.com ([209.132.183.28]:60193 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1759807Ab0JZRS3 (ORCPT ); Tue, 26 Oct 2010 13:18:29 -0400 In-Reply-To: Sender: kvm-owner@vger.kernel.org List-ID: On 10/26/2010 05:09 PM, Blue Swirl wrote: > On Tue, Oct 26, 2010 at 8:05 AM, Avi Kivity wrote: > > On 10/25/2010 08:38 PM, Blue Swirl wrote: > >> > >> > > >> > I don't really see why we need registration; cpu_register_io() takes > >> > function pointers, a size, and an opaque, and gives an integer handle > >> > in > >> > return. With the IOPort object approach, you set up the IOPort with > >> > function pointers, size is implied, and the opaque is derived using > >> > container_of(); the handle is simply the address of the object. > >> > >> With the handle, we can separate setting up the structures at device > >> level, and mapping the object using only the handle at bus or other > >> higher level. Can this be done with the object approach? > > > > I believe so. The handle is simply an indirect pointer, no? > > Yes, but then the object should also contain size information. That > should not be needed for mapping at higher level. Sorry, I don't follow your meaning. When I said "size is implied" I meant that the IOPort object has a separate function pointer for sizes 1, 2, and 4, so it ioport_register() doesn't need a size parameter. But I don't see how that relates to your comment. > >> The purpose of that patch series was to perform the separation for PCI > >> BARs. I wasn't so happy with the series, so I never pushed. > > > > In fact I think an IOPort is even more suitable; if we need additional > > attributes we can use a derived object: > > > > struct PCIIOPort { > > IOPort ioport; > > /* additional fields */ > > }; > > One issue with my series was that it would be great if the devices > just had some BAR structures (used by PCI layer to map the devices) > inside PCI/qdev structures, but I invented that too late. Maybe this > can be addressed in your design? It looks to be orthogonal. It would be great to have a BAR object; that object can then use your API, my API, or the existing API. -- error compiling committee.c: too many arguments to function