Xen users - May 2006 - Multiple VMs Using Multiple Video Cards w/Pacifica

Hi All,

I''ve read a few references to the imminent AMD AM2 CPUs including
Pacifica
technology and that these CPUs have an io-mmu (I may have spelt that wrong or 
used incorrectly terminology, if it''s not an io-mmu it''s
something that
allows devices on the PCI/PCI-express bus to have their memory remapped.)  
WIll this enable XenU domains to have access to multiple video cards?  I 
would like to be able to have two XenU domains have a PCI-express video card 
each to run two independent displays in either Windows XP or Linux.

Thanks,

Steve.

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

> -----Original Message-----
> From: xen-users-bounces@lists.xensource.com 
> [mailto:xen-users-bounces@lists.xensource.com] On Behalf Of 
> Steve Higton
> Sent: 02 May 2006 22:41
> To: Xen-users@lists.xensource.com
> Subject: [Xen-users] Multiple VMs Using Multiple Video Cards 
> w/Pacifica
> 
> Hi All,
> 
> I''ve read a few references to the imminent AMD AM2 CPUs 
> including Pacifica technology and that these CPUs have an 
> io-mmu (I may have spelt that wrong or used incorrectly 
> terminology, if it''s not an io-mmu it''s something that
allows
> devices on the PCI/PCI-express bus to have their memory 
> remapped.) WIll this enable XenU domains to have access to 
> multiple video cards?  I would like to be able to have two 
> XenU domains have a PCI-express video card each to run two 
> independent displays in either Windows XP or Linux.
The short answer is: Probably not immediately.

Let''s first sort out the naming of things: The code-name pacifica is
now
called AMD Virtulization Technology or Secure Virtual Machine extension,
SVM for short. 

IOMMU is a technology that isn''t implemented in the first generation
SVM
processors, but will come in future processors [I don''t honestly know
at
which point in the future]. But yes, this technology will, amongst other
things, allow the mapping of virtualized "pseudo-physical" memory to
"real physical" memory. 

We do have an alternative, which is to use the GART for this. This has
been implemented in the x86-64 kernel for quite some time [to remap
IO-buffers from above 4GB to below 4GB so that 32-bit hardware can
access memory above 4GB], but so far there is no solution for Xen to use
this method, although I believe there is work in progress to make this. 

The second problem is that the PCI device hiding for Dom0 isn''t
implemented for HVM systems. I think this is also being worked on (or at
least it''s on someones list to be done when they get to that item on
the
list or something like that). 

--
Mats

> 
> Thanks,
> 
> Steve.
> 
> _______________________________________________
> Xen-users mailing list
> Xen-users@lists.xensource.com
> http://lists.xensource.com/xen-users
> 
> 

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