From mboxrd@z Thu Jan 1 00:00:00 1970 From: Avi Kivity Subject: Re: [PATCH 07/13] KVM: SVM: Add intercept checks for descriptor table accesses Date: Mon, 28 Mar 2011 16:34:37 +0200 Message-ID: <4D909C7D.3060201@redhat.com> References: <1301309210-11120-1-git-send-email-joerg.roedel@amd.com> <1301309210-11120-8-git-send-email-joerg.roedel@amd.com> <4D9080AA.6080304@redhat.com> <20110328135608.GE2085@amd.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit Cc: Marcelo Tosatti , "kvm@vger.kernel.org" To: "Roedel, Joerg" Return-path: Received: from mx1.redhat.com ([209.132.183.28]:28847 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753245Ab1C1Oem (ORCPT ); Mon, 28 Mar 2011 10:34:42 -0400 In-Reply-To: <20110328135608.GE2085@amd.com> Sender: kvm-owner@vger.kernel.org List-ID: On 03/28/2011 03:56 PM, Roedel, Joerg wrote: > On Mon, Mar 28, 2011 at 08:35:54AM -0400, Avi Kivity wrote: > > On 03/28/2011 12:46 PM, Joerg Roedel wrote: > > > This patch add intercept checks into the KVM instruction > > > emulator to check for the 8 instructions that access the > > > descriptor table addresses. > > > > > > +static struct opcode group6[] = { > > > + DI(ModRM, sldt), > > > + DI(ModRM, str), > > > + DI(ModRM | Priv, lldt), > > > + DI(ModRM | Priv, ltr), > > > + N, N, N, N, > > > +}; > > > + > > > static struct group_dual group7 = { { > > > - N, N, DI(ModRM | SrcMem | Priv, lgdt), DI(ModRM | SrcMem | Priv, lidt), > > > + DI(ModRM | DstMem | Priv, sgdt), DI(ModRM | DstMem | Priv, sidt), > > > + DI(ModRM | SrcMem | Priv, lgdt), DI(ModRM | SrcMem | Priv, lidt), > > > > | Mov, to avoid RMW for SIDT, for example. Also need to indicate the > > operand size correctly. > > > > > diff --git a/arch/x86/kvm/svm.c b/arch/x86/kvm/svm.c > > > index 381b038..485a09f 100644 > > > --- a/arch/x86/kvm/svm.c > > > +++ b/arch/x86/kvm/svm.c > > > @@ -3871,6 +3871,10 @@ static void svm_fpu_deactivate(struct kvm_vcpu *vcpu) > > > #define POST_EX(exit) { .exit_code = (exit), \ > > > .stage = X86_ICPT_POST_EXCEPT, \ > > > .valid = true } > > > +#define POST_MEM(exit) { .exit_code = (exit), \ > > > + .stage = X86_ICPT_POST_MEMACCESS, \ > > > + .valid = true } > > > + > > > > > > static struct __x86_intercept { > > > u32 exit_code; > > > @@ -3884,9 +3888,18 @@ static struct __x86_intercept { > > > [x86_intercept_smsw] = POST_EX(SVM_EXIT_READ_CR0), > > > [x86_intercept_dr_read] = POST_EX(SVM_EXIT_READ_DR0), > > > [x86_intercept_dr_write] = POST_EX(SVM_EXIT_WRITE_DR0), > > > + [x86_intercept_sldt] = POST_MEM(SVM_EXIT_LDTR_READ), > > > + [x86_intercept_str] = POST_MEM(SVM_EXIT_TR_READ), > > > + [x86_intercept_lldt] = POST_MEM(SVM_EXIT_LDTR_WRITE), > > > + [x86_intercept_ltr] = POST_MEM(SVM_EXIT_TR_WRITE), > > > + [x86_intercept_sgdt] = POST_MEM(SVM_EXIT_GDTR_READ), > > > + [x86_intercept_sidt] = POST_MEM(SVM_EXIT_IDTR_READ), > > > + [x86_intercept_lgdt] = POST_MEM(SVM_EXIT_GDTR_WRITE), > > > + [x86_intercept_lidt] = POST_MEM(SVM_EXIT_IDTR_WRITE), > > > }; > > > > Spec says POST_EX()? > > Well, not entirely clear. Spec says that #GP takes precedence before the > intercept and the intruction reference says the #GP fires if the > supplied address is not within segment limits or the segment itself is > not valid, which, in my interpretation, made them POST_MEM. My interpretation would be that a #GP(0) due to CPL check takes precedence over the intercept, which takes precedence over a #GP(selector) due to the actual memory reference. -- error compiling committee.c: too many arguments to function