RE: [Xen-devel] Fwd: [PATCH 0/18] Nested Virtualization: Overview

["Dong, Eddie" <eddie.dong@xxxxxxxxx>]

Keir Fraser wrote:
> On 16/04/2010 11:27, "Tim Deegan" <Tim.Deegan@xxxxxxxxxx> wrote:
> 
>>> Please read the XenNestedHVM.pdf paper, particularly the section
>>> "Software Architecture". This describes how this is made to be
>>> generic and what needs to be done to adapt to Intel.
>> 
>> Your PDFs suggest that even on Intel CPUs, the nested hypervisor
>> should always see SVM, not VMX.  You shouldn't be surprised or
>> offended if that isn't popular with Intel. :)
> 
> I don't see any good argument for it either. I.e., I don't think we
> care about migrating between AMD and Intel hosts with nestedhvm
> enabled, which I think would be the only argument for it. I know we
> added support for cross-emulating SYSENTER and SYSCALL, but that's
> needed for cross-migration of any 64-bit guest running compat-mode
> apps (i.e., really need to make cross-migration possible at all). I'm
> sceptical enough of the utility of cross-vendor migration *at all*,
> let alone supporting in tandem with advanced features also of dubious
> utility (at least in enterprise space), like nestedhvm.
> 

Although SVM on VMX is possible in theory, I doutb on the feasibility given 
that there are many semantics difference between VMCB & VMCS, which will 
eventually have to be emulated with extra complexity.
Qing will post his natural VMX on VMX patch this week, base on his Xen summit 
talk on http://www.xen.org/files/xensummit_intel09/xensummit-nested-virt.pdf :) 
We can have next level of discussion/comparation then.
BTW, If somebody has implemented SVM on VMX solution already, we can have 
performance comparation between 2 approaches to assist the discussion.

Thx, Eddie

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel