[Fedora-xen] HVM DomU on F9?

[Dustin Henning]

	Great information, and straight from the source!  Can you tell me on
which version of KVM support for TPR was implemented?  I assume I am not up
to said version, but I want to confirm, because when I start my Windows
guests (currently all uniprocessor) in KVM, csrss.exe uses all available
processor constantly until I follow the KVM Windows ACPI workaround
directions.  I have updated since December, but I am running Fedora 7 rpms,
so I may have to move up a version or two to get that support.  Thanks,
	Dustin

-----Original Message-----
From: Avi Kivity [mailto:avi at qumranet.com] 
Sent: Tuesday, May 20, 2008 11:55
To: Dustin.Henning at prd-inc.com
Cc: fedora-xen at redhat.com
Subject: Re: [Fedora-xen] HVM DomU on F9?

Dustin Henning wrote:
>        I actually prefer to run HVM DomUs in a regular kernel if possible
> because I have always had mouse problems in Dom0 of F7, but not in the
base
> kernel.  I understand that KVM is an option, but since Xen now supports
> Windows with ACPI (without the degradation that was once common in Xen and
> is still common in KVM  because of a certain ACPI register apparently
> regularly polled by Windows [I would love to know what register, as I
could
>   

It's called that Task Priority Register, or TPR.

> then try to find a way to prevent Windows from doing that and KVM might
> perform as well as Xen]), I feel that KVM would be a big step backwards.
>   

For uniprocessor guests, kvm supports TPR optimization (since December 
2007).  It's disabled on SMP guests due to stability issues, unfortunately.

> Additionally, there are apparently more paravirtual drivers available for
> Windows (though they are not readily available in binary form and stable,
> more on that in my next message), though using std-vga in KVM is nice
(with
> standard VESA 2.0, a generic Windows driver already exists outside of the
> project).

A paravirtualized network driver is freely available for kvm.  It is 
still not fully stable under heavy loads, though.  We expect to fix this 
soon.

-- 
error compiling committee.c: too many arguments to function