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

Re: [Xen-users] vga passthrough with amd radeon r9 285



Brian,

I have an AMD R7 250 successfully being passed through to a Windows
8.1 DomU (as well as a USB 3.0 root hub); here are some important
details on how I did it:

* I am not passing through the R7 as the primary graphics adapter.
  Instead, it's passed through as an ordinary PCI device, with the
main graphics adapter being "stdvga" (accessible via VNC).
  Windows is configured to consider the R7 as its main display, and
*not* to extend the display onto the virtual graphics adapter.

  This allows me to use qemu-xen (or uptream) instead of qemu-xen-traditional.

* Make sure you have Xen 4.5 and a recent version of qemu.
  (You'll need to pull them from 'experimental' instead of just sid.)
  Otherwise, there may not be enough address space reserved for the
very large MMIO regions the GPU requires.
  (This will show in Windows as a "code 12", something about not
enough resources being available)

* Make sure your card is not behind an NF200 chip (a popular chip used
to allow multiple PCIe graphics cards several years ago).
  The NF200 chip does not play well with passthrough.
  From what I've gathered, the PLX PCIe switch chip is a proper PCIe switch.
  I have an ASRock motherboard; from what I've gathered, they've got
the most IOMMU-friendly hardware and software (BIOS).

* Your graphics card almost certainly has separate functions for the
GPU and for the HDMI audio circuitry.  Be sure to pass through both
functions (e.g. 00:07.0 and 00:07.1)

* Be sure to configure your kernel command line to hide the graphics
card (using xen-pciback.hide)

The result works very well; all I really lose is the ability to see
early BIOS messages on the monitor (because they show up on the
virtual card).


> On 04/02/2015 00:54, Brian Paterni wrote:
>
> Hi all,
>
<snip>
> I'm trying to passthrough an AMD Radeon R9 285 to a windows domu, but so
> far have had no luck.
>
> dom0 is debian unstable (linux 3.16, xen 4.4.1). Though with the
> qemu-traditional device model removed in debian [1], gfx_passthru is
> currently not possible with debian-provided packages (?)
>
> 1: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=776742
>
> As a result, I've begun to compile from source, but xen git master
> results in the fallowing error message:
>
>    The kernel doesn't support reset from sysfs for PCI device
>
> Since the message lists the 'kernel', I thought it might have something
> to do with the kernel in debian. So I compiled a linux kernel
> (3.19.0-rc6+) from source. However I still experience the same error
> message.
>
> Now I'm at a loss for what to try next. The r9 285 is a relatively new
> gpu. In fact, there currenty is no FOSS drivers for for it (amdgpu has
> yet to be released). Therefore, could the non-working gfx_passthru be
> the result of missing bits in either the linux kernel or xen that have
> yet to be dropped in? Or am I just 'doing it wrong'? The fact that the
> r9 2xx family is listed as supported at [2] leads me to believe it's the
> latter.
>
> 2:
> http://wiki.xen.org/wiki/Xen_VGA_Passthrough_Tested_Adapters#ATI.2FAMD_display_adapters
>
> Attached should be some more information I thought to include. Let me
> know if you need anything else.
>
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxx
> http://lists.xen.org/xen-users
>
>
>
>
> _______________________________________________
> Xen-users mailing list
> Xen-users@xxxxxxxxxxxxx
> http://lists.xen.org/xen-users



-- 
-- Stevie-O
Real programmers use COPY CON PROGRAM.EXE

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users


 


Rackspace

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