Currently, Xen doesn't support any direct hardware access
from fully virtualized domains [aka HVM domain => AMD-V (SVM,
Pacificia) or VMX (Intel VT, Vanderpool, etct)].
If there was support for IOMMU in Xen (and in hardware),
and the relevant software architecture to support it is present, then in theory
you could pass the entire hardware component to the HVM domain. This means the
netire nVidia/ATI/3DLabs/S3/etc graphics card, not a window thereof. To get a
window, you need some software to manage it, and it would be easiest to
implement that as a paravirtual driver that sends drawing commands to the owner
of the actual hardware (Dom0 usually).
Migration of fully virtualized domains is also not yet
supported. Only way to do that is a full shutdown and startup in the new
location. There is work in progress for this, initially to save/restore the
domain, which is "slow migration [but faster than shutdown+start]".
Sorry if this has come up before, but the xen-users search function
is totally broken right now (a search for "xen" yields 0 results,
As I understand it, I can have a xen kernel that
exposes the raw video card device to an underlying virtual machine. Will
this allow windows to use the normal nVidia graphics drivers to have hardware
acceleration, provided that I'm running on a Pacifica machine (assuming the
iommu is necessary)?
If that works, then I have another
question. Does xen machine migration between identical hardware and xen
kernel configurations allow for Windows to be migrated if it has raw hardware
access to the video card? That sounds like some really dark magic if it
works, but I can't find any reference that says it's impossible. Of
course, I don't even see how machine migration can keep network connections
up, so there's probably something I'm missing...
Thanks for any
Xen-users mailing list