[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] RE: pciback+front needed for hvm ?



Passing-through a graphic card to HVM guest is much more challenging than to PV guest.
 
The graphic card (especially an integrated graphic card) may highly depend on the host plaform: e.g., system BIOS, ACPI, PCI hierarchy... And, PCI config space of graphic card may have virtualizion-unfriendly assumption(e.g., may require vBAR == pBAR).
There are lots of graphic cards, of which some may have all such special requirements, and some may have not (so they might be easily passed through to HVM guest correctly).
 
I didn't try the pass-through of graphics card at all. The following is only my understanding to some discussions in the mail list.
IIRC, some people have successfully passed through *some* kind of graphic card on *some* platform to HVM guest according to my memory of the old emails in this mail list, but I'm not sure about the details...
 
PS, in May 2008, Jean Guyader posted some patches about "pass-through a graphic card". You might want to find the archive of the mails and have a look.
 
Good luck! :-)

Thanks,
-- Dexuan


From: Tim Moore [mailto:timothy.moore@xxxxxxxxxxx]
Sent: Tuesday, April 14, 2009 7:55 AM
To: Cui, Dexuan; Kay, Allen M; xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: RE: pciback+front needed for hvm ?

This works with non-HVM guest easy, I am using it for an X server and the Nvidia drivers work great ..

 

All the I/O and Mem resources map through correctly with HVM guest but driver fails to load (Code 12). Yes, I agree ioemu is doing the work to present the PCI Bus to HVM so what is missing from ioemu to allow this to work with HVM ?

 

Thanks,

Tim

 


From: Cui, Dexuan [mailto:dexuan.cui@xxxxxxxxx]
Sent: 14 April 2009 00:47
To: Tim Moore; Kay, Allen M; xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: RE: pciback+front needed for hvm ?

 

For HVM PCI passthrough, ioemu (a modified qemu as Device Model of HVM guest) emulates the PCI bus and does the proper operations. PCI passthrough for HVM guest by VT-d has been supported for years. :-)  Per my experience, almost all the PCI/PCIe devices can work -- however, the passthrough of Graphics Card could be complicated and the community is doing more work.

 

Thanks,
-- Dexuan


From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx [mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Tim Moore
Sent: Tuesday, April 14, 2009 7:29 AM
To: Kay, Allen M; xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: [Xen-devel] RE: pciback+front needed for hvm ?

Please could you point me to the patch ?

 

Thanks !

 


From: Kay, Allen M [mailto:allen.m.kay@xxxxxxxxx]
Sent: 13 April 2009 22:00
To: Tim Moore; xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: RE: pciback+front needed for hvm ?

 

HVM guest uses PCI bus provided by QEMU.  The functionality in QEMU is complete for PCI passthrough support most devices.   Additional patch is needed for graphics device passthrough.

 

From: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx [mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx] On Behalf Of Tim Moore
Sent: Monday, April 13, 2009 6:21 AM
To: xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: [Xen-devel] pciback+front needed for hvm ?

 

Dom0 with pciback + DomU with pcifront. pcifront grabs device during DomU boot and vendor driver then loads against passthrough device and works great!

 

In the same scenario with HVM DomU, is pcifront needed or does qemu provide the PCI Bus ?

 

Both Linux and Win32 hvm fail to load the vendor driver against the PCI device. Is this a problem with qemu or is further support needed in hvm domU for PCI passthrough?

 

Thanks,

Tim

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

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.