From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Kiszka Subject: Re: [ANNOUNCE] qemu-kvm-0.13.0-rc1 Date: Mon, 13 Sep 2010 13:32:18 +0200 Message-ID: <4C8E0BC2.4030500@siemens.com> References: <20100908162918.GA27948@amt.cnet> <4C87EC78.9030800@xutrox.com> <4C87F302.7050702@codemonkey.ws> <4C887C12.2060701@redhat.com> <4C88DA62.4050304@codemonkey.ws> <4C8A1ACD.6090903@redhat.com> <20100910193101.GB23988@sequoia.sous-sol.org> <4C8A8B8E.6060806@codemonkey.ws> <4C8C6EF4.9020906@redhat.com> <4C8CF258.906@codemonkey.ws> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Avi Kivity , Chris Wright , Arjan Koers <0h61vkll2ly8@xutrox.com>, kvm@vger.kernel.org To: Anthony Liguori Return-path: Received: from david.siemens.de ([192.35.17.14]:15086 "EHLO david.siemens.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754972Ab0IMLcw (ORCPT ); Mon, 13 Sep 2010 07:32:52 -0400 In-Reply-To: <4C8CF258.906@codemonkey.ws> Sender: kvm-owner@vger.kernel.org List-ID: Am 12.09.2010 17:31, Anthony Liguori wrote: > On 09/12/2010 01:11 AM, Avi Kivity wrote: >> On 09/10/2010 10:48 PM, Anthony Liguori wrote: >>>> I agree, is there any reason not to enable compiling less into the >>>> binary? >>>> There are folks interested in eliminating as much as possible to reduce >>>> the attack surface and auditing requirements, for example. >>> >>> It's not a bad idea, it's just that what --disable-cpu-emulation does >>> is evil. Being that I wrote the implementation, I'm quite confident >>> in declare it as such :-) >>> >> >> Oh, I thought you were against the idea in itself for some reason. >> >> I'll patch it for 0.13, but any ideas on how it should be rework for >> master? > > Glauber's old Accel interface was close to the right approach. We need > to abstract the exec.c interfaces to use a function pointer table and > have a TCG and KVM implementation. The function pointer tables can then > be registered by a module_init() and we can simply not include the kvm > or TCG files are build time to disable the functionality. Even more flexible would be to have them linked in on demand (as specified on the command line). Additional, in certain contexts unsupported features could then be shipped separately without having to provide n versions of the common core. A second step to build-time configurability, of course. Jan -- Siemens AG, Corporate Technology, CT T DE IT 1 Corporate Competence Center Embedded Linux