From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [PATCH 1/7] KVM: VMX: Allocate MSR Bitmap for each vcpu Date: Wed, 08 Oct 2008 13:10:46 +0200 Message-ID: <48EC9536.4060507@redhat.com> References: <1223439478-26515-1-git-send-email-sheng@linux.intel.com> <1223439478-26515-2-git-send-email-sheng@linux.intel.com> <48EC87FE.3060006@redhat.com> <200810081841.30880.sheng@linux.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: kvm@vger.kernel.org To: Sheng Yang Return-path: Received: from mx2.redhat.com ([66.187.237.31]:34578 "EHLO mx2.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754186AbYJHLKn (ORCPT ); Wed, 8 Oct 2008 07:10:43 -0400 In-Reply-To: <200810081841.30880.sheng@linux.intel.com> Sender: kvm-owner@vger.kernel.org List-ID: Sheng Yang wrote: > On Wednesday 08 October 2008 18:14:22 Avi Kivity wrote: > >> Sheng Yang wrote: >> >>> Separate msr_bitmap for each vcpu, prepared for guest PAT support. >>> >> Why is this necessary? True, it reduces the overhead of the guest >> reading and writing the PAT MSRs, but is such access frequent? >> >> I would think guests set the PAT once, and never change it later. >> > > Yeah. In fact, I just think msr_bitmap for each vcpu would be done sooner or > later, so get it done here. And it's natural to go with GUEST_PAT. And Xen > use it for another purpose(DEBUGCTLMSR related, I haven't checked it through) > (Also svm.c in KVM use per-vcpu msr bitmap) > svm.c uses it for last branch record; I think these are also rarely accessed from the guest. > And without that, a callback should be implement to hook MSR write and update > guest pat write for both vmx and svm, or we should update GUEST PAT every > vmentry according to the vcpu->pat. Either way seems not that natural with > GUEST_PAT support. > We need the callbacks (vmx_set_msr and vmx_get_msr, or did you mean something else?) anyway for save/restore support. > However, if you think msr_bitmap for each vcpu is a waste, I'd like to add > callbacks. > I agree that we will likely need msr bitmap support one day; but let's start without it as this way we test the pat msr callbacks. -- I have a truly marvellous patch that fixes the bug which this signature is too narrow to contain.