From mboxrd@z Thu Jan 1 00:00:00 1970 From: Marcelo Tosatti Subject: Re: [PATCH 03/11] KVM: x86: retry non-page-table writing instruction Date: Mon, 22 Aug 2011 17:42:10 -0300 Message-ID: <20110822204209.GA3490@amt.cnet> References: <4E4A10E8.5090705@cn.fujitsu.com> <4E4A113F.5090404@cn.fujitsu.com> <20110822195929.GA2662@amt.cnet> <4E52BA31.4030203@cn.fujitsu.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Avi Kivity , LKML , KVM To: Xiao Guangrong Return-path: Received: from mx1.redhat.com ([209.132.183.28]:29334 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751233Ab1HVUnA (ORCPT ); Mon, 22 Aug 2011 16:43:00 -0400 Content-Disposition: inline In-Reply-To: <4E52BA31.4030203@cn.fujitsu.com> Sender: kvm-owner@vger.kernel.org List-ID: On Tue, Aug 23, 2011 at 04:21:05AM +0800, Xiao Guangrong wrote: > On 08/23/2011 03:59 AM, Marcelo Tosatti wrote: > > >> + if (!vcpu->arch.mmu.direct_map && !mmu_is_nested(vcpu)) > >> + gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL); > > > > Why write? > > > > Since the fault is caused by page table written, and the 'gpa' can > be written after instruction is retied. > > >> + kvm_mmu_unprotect_page(vcpu->kvm, gpa >> PAGE_SHIFT); > >> + > >> + /* > >> + * The shadow pages have been zapped, then we call the page > >> + * fault path to change the mapping to writable. > >> + */ > >> + vcpu->arch.mmu.page_fault(vcpu, cr2, PFERR_WRITE_MASK, true); > > > > I don't see why is this necessary. Just allowing the instruction to > > proceed should be enough? > > > > It used to avoid later VM-exit, since we will retry the instruction > but the mapped is still read-only. So we can it to let the mapping become > writable to avoid page fault again. Its not like this case is performance sensitive. Usually optimizing things without the need for it leads to bad results. So please drop this.