Re: Nested SVM

[Christoph Egger <Christoph.Egger@amd.com>]

On 03/13/12 12:56, Ian Campbell wrote:
> On Tue, 2012-03-13 at 11:40 +0000, Christoph Egger wrote:
>> On 03/12/12 13:11, Ian Campbell wrote:
>>>         * Nested-virtualisation (currently should be marked experimental,
>>>           likely to release that way? Consider nested-svm separate to
>>>           nested-vmx. Nested-svm is in better shape)
>>
>> I tested the following scenarios:
>>
>> - Linux on Xen on Xen
>> - NetBSD on Xen on Xen
>> - KVM (+ KVM-guest) on Xen
>> - KVM (+ KVM-guest) on Xen on Xen (yeah, l3 guest works!)
>> - Windows 7 (both 32bit and 64bit) on Xen on Xen
>> - Win7 XP mode on Windows 7 (both 32bit and 64bit) on Xen
>> - Hyper-V (64bit) on Xen
>> - VMware ESX (64bit) on Xen
>
> Wow, that's far more combinations than I expected.
>
> I wonder where we can document this, is there a nested-HVM page on the
> wiki perhaps?
>
>> Issues:
>> - Hyper-V SMP not tested, only UP
>> - Win7 XP mode often quits with BSOD,
>
> Hrm, this one sounds quite important? I expect Win7 XP mode is the
> primary real world use case for Nested-HVM?

yes, but I have no idea what might be wrong.

>> but when it boots playing Solitair makes fun :-)
>
> ;-)

Playing Solitair gives a good impression on the end-user feeling
in how fast it response. It is also a good test case if interrupts
come in fast enough. If that is not the case a double click is
interpreted as two single clicks.

Playing Solitair in Win7 XP mode runs even faster than in Win7
thus makes more fun to play in Win7 XP mode.

>> - L2 guest screen doesn't refresh in graphical mode
>>     (VGA text mode works),
>>     workaround: dis- and re-establish the VNC connection,
>>     between two VNC reconnects you can work like with your
>>     monitor powered off
>
> Uh, that also sounds pretty critical. Does it also effect the Win7 XP
> mode usecase?

No, because in the Win7 XP mode usecase the graphic output is done by
the L1 guest (Windows 7) and that works. This problem only occurs when
the L2 guest runs in graphic mode.

I think, this issue is neither AMD nor Intel specific.
I think, this has something to do with the non-buffered io mode.
IIRC, VGA text mode uses a buffered io mode.

> Between those two issues I think removing the experimental tag sounds
> premature?

Yes.

Christoph


-- 
---to satisfy European Law for business letters:
Advanced Micro Devices GmbH
Einsteinring 24, 85689 Dornach b. Muenchen
Geschaeftsfuehrer: Alberto Bozzo, Andrew Bowd
Sitz: Dornach, Gemeinde Aschheim, Landkreis Muenchen
Registergericht Muenchen, HRB Nr. 43632