From: Keir Fraser Subject: Re: [PATCH 06/16] vmx: nest: handling VMX instruction exits Date: Wed, 15 Sep 2010 08:38:33 +0100 Message-ID: References: <20100915071706.GA23507@qhe2-db> Mime-Version: 1.0 Content-Type: text/plain; charset="US-ASCII" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20100915071706.GA23507@qhe2-db> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Qing He Cc: Tim Deegan , "xen-devel@lists.xensource.com" , "Dong, Eddie" List-Id: xen-devel@lists.xenproject.org On 15/09/2010 08:17, "Qing He" wrote: >> What is wrong with simply extending x86_emulate to handle these VMX-related >> instructions? We've dealt with emulators provided by Intel guys in the past >> and frankly they were full of holes. > > That needs additional callback when handling vmcs and state change, > doesn't it? I'm a little worried that it's too vmx-specific to get > into x86_emulate, and that's why we used a separate decoder in the > first place (I know it's ugly, though). A few VMX-specific callbacks would be fine. Extra callbacks are cheap. Just focus on making the callback interface clean and tidy. I'd *much* rather have VMX-specific callbacks than an extra emulator. > And if we are to use x86_emulate, is it possible to avoid redecoding the > opcode, because exit reason is already there? If vmexit reason fully decodes the instruction for you then I would agree that skipping x86_emulate could make sense. And then your instruction emulator would be really simple and fast -- vmx_io_instruction() is a good example of this. If you still need to parse the instruction to decode ModRM and the like, then I don't see that the partial decode from vmexit reason saves you much at all, and you might as well go the whole hog and do full decode. I don't see much saving from a hacky middle-ground. -- Keir