Re: [Xen VGA Passthrough] Cannot Complete All 3dmark Tests in Windows 8 HVM domU

[Gordan Bobic <gordan@bobich.net>]

On 09/17/2014 06:49 PM, Zytaruk, Kelly wrote:
> Gordan,
>
> Can you tell me exactly what the problem is that you are seeing that prompted you to ask for a bus reset?  I am trying to reproduce your issue but I am not having any luck.  I tried a few different scenarios ranging from clean shutdown to destroying the domain with 'xl des 5' and restarting.  In all cases I was able to cleanly reboot with the graphics driver enabled.

The issue is quite widely documented in the archives of this list. What 
happens is that upon a domU reboot or shutdown + re-create, one of the 
following happens, usually relatively non-deterministically in terms of 
what will happen:

1) BSOD in the ATI driver.

2) domU comes up but the performance is painfully slow and there is 
corruption in the output image, usually manifesting as short white lines 
on the screen.

3) dom0 crash

and every once in a while with extreme luck

4) Works fine - I only saw this happen maybe 5-6 times in all the ATI 
reboots I tried over the months I'd spent trying to work around the issue.

> I am testing with Win7 as the guest OS with Xen 4.2 and Linux 3.4.9.  Graphics card is 'AMD Radeon R9 200' on an AMD Kaveri APU.

My testing was done on a HD4850, HD6450 and HD7970. HD6450 seemed to 
fare better than the others, possibly because it doesn't have an 
auxiliary power input and is powered purely from the slot. I 
hypothesised that it might be plausible that even though it doesn't 
report supporting the power states required for a PCI level power-off it 
still works if you manually write the registers with setpci. I played 
with it a bit on the HD6450 and it seemed to work (no errors), and the 
HD6450 did seem to come up reliably after a domU reboot much more often 
than the others.

But I didn't want to spend too much time on investigating that since the 
HD6450 wasn't sufficient for my requirements, so my findings didn't 
reach a conclusion.

> What are your exact repro steps and are you using any graphics/pci related patches.

I am not using any PCI or graphics related patches. I use Xen 4.3.0 and 
have been using various kernels between 3.8.x and currently 3.14.12. I 
haven't tried ATI cards with kernels more recent than 3.9. Unless you 
have extra patches in your kernel or Xen, I would expect the issue to be 
very reproducible on Xen 4.2 and kernel 3.4.9.

I'm also pretty sure the ATI reboot issue has been reported a few days 
ago. The manifestation of the problem was display corruption while 
playing flash videos, and it didn't happen on a freshly rebooted dom0.

Unfortunately, although I have all the parts to build a second machine 
that is exactly the same as my current server for testing and 
troubleshooting purposes, I am not sure when I will be able to get 
around to doing it. Also, the only spare ATI GPU I still have is the 
HD4850, so any ATI GPU testing I can do will be limited to using that.

Gordan