From mboxrd@z Thu Jan 1 00:00:00 1970 From: Paolo Bonzini Subject: Re: [PATCH 07/21] KVM: x86: Emulator considers imm as memory operand Date: Wed, 05 Nov 2014 12:36:52 +0100 Message-ID: <545A0BD4.3070300@redhat.com> References: <1414922101-17626-1-git-send-email-namit@cs.technion.ac.il> <1414922101-17626-8-git-send-email-namit@cs.technion.ac.il> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 8bit Cc: kvm@vger.kernel.org, nadav.amit@gmail.com To: Nadav Amit Return-path: Received: from mx1.redhat.com ([209.132.183.28]:43463 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753888AbaKELhB (ORCPT ); Wed, 5 Nov 2014 06:37:01 -0500 In-Reply-To: <1414922101-17626-8-git-send-email-namit@cs.technion.ac.il> Sender: kvm-owner@vger.kernel.org List-ID: On 02/11/2014 10:54, Nadav Amit wrote: > The emulator mistakenly considers some of the immediate operands as memory > operands, performs memory read and uses the wrong data. By default, every > operand is marked as OP_MEM, so if it is not changed, memory read may be > wrongly emulated and the wrong value would be used. Consider for instance the > ROR instruction - src2 (the number of times) would be read from memory instead > of being used as immediate. > > Mark every immediate operand as such to avoid this problem. > > Signed-off-by: Nadav Amit > --- > arch/x86/kvm/emulate.c | 8 ++++++++ > 1 file changed, 8 insertions(+) > > diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c > index f456783..e624d62 100644 > --- a/arch/x86/kvm/emulate.c > +++ b/arch/x86/kvm/emulate.c > @@ -4269,6 +4269,7 @@ static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op, > fetch_register_operand(op); > break; > case OpCL: > + op->type = OP_IMM; > op->bytes = 1; > op->val = reg_read(ctxt, VCPU_REGS_RCX) & 0xff; > break; > @@ -4276,6 +4277,7 @@ static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op, > rc = decode_imm(ctxt, op, 1, true); > break; > case OpOne: > + op->type = OP_IMM; > op->bytes = 1; > op->val = 1; > break; > @@ -4334,21 +4336,27 @@ static int decode_operand(struct x86_emulate_ctxt *ctxt, struct operand *op, > ctxt->memop.bytes = ctxt->op_bytes + 2; > goto mem_common; > case OpES: > + op->type = OP_IMM; > op->val = VCPU_SREG_ES; > break; > case OpCS: > + op->type = OP_IMM; > op->val = VCPU_SREG_CS; > break; > case OpSS: > + op->type = OP_IMM; > op->val = VCPU_SREG_SS; > break; > case OpDS: > + op->type = OP_IMM; > op->val = VCPU_SREG_DS; > break; > case OpFS: > + op->type = OP_IMM; > op->val = VCPU_SREG_FS; > break; > case OpGS: > + op->type = OP_IMM; > op->val = VCPU_SREG_GS; > break; > case OpImplicit: > I'm including this for stable@ because, until commit c44b4c6ab80e (KVM: emulate: clean up initializations in init_decode_cache, 2014-04-16), this would be harmless. The type would be set to OP_REG instead of OP_IMM, but it would not matter because no writeback is done for Src2 operands. Now, the uninitialized op->type may be an OP_MEM from a previous instruction. Paolo