[vfio-users] GPU passthrough errors with linux 5.1 and newer

[Zoltán Kővágó]

I'm attaching the workaround I use with linux 5.2.1 and a w8.1 guest. 
I'm not saying you should use it, as it probably has a negative effect 
on performance (I never benchmarked it, but I didn't notice anything 
with older games) but at least you could confirm whether you have the 
same problem or not.

Regards,
Zoltan

On 2019-08-03 13:35, Ivan Volosyuk wrote:
> I just hit the problem on Windows 8.1 guest on kernel 5.2.5 (gentoo), tried
> to upgrade from kernel 4.19.57.
> It seems the problem doesn't happen if I use Windows 10 guest or kernel
> 4.19.57.
> There is graphics artifacts on my RTX-2080TI in W8.1 guest. I temporary
> switched to older kernel. Any news about this? Any way to debug it?
> 
> 
> On Thu, Aug 1, 2019 at 5:08 AM Zoltán Kővágó <dirty.ice.hu at gmail.com> wrote:
> 
>> On 2019-07-31 15:41, José Ramón Muñoz Pekkarinen wrote:
>>> On Sun, 21 Jul 2019 at 21:59, Zoltán Kővágó <dirty.ice.hu at gmail.com>
>> wrote:
>>>>
>>>> Hi,
>>>>
>>>> Recently my previously perfectly working GPU passthrough setup (with a
>>>> win8.1 x64 guest with OVMF) started to malfunction in various ways:
>>>> screen randomly turned off for a few seconds, BSOD with
>>>> VIDEO_TDR_FAILURE, 3d apps randomly crashing, not drawing the windows'
>>>> content, and graphical glitches (for example in furmark the OSD text
>>>> flickers).
>>>>
>>>> After fiddling around with various qemu versions, nvidia driver versions
>>>> on the guest, I figured out that with a linux 5.0 kernel it works fine,
>>>> but with 5.1 it randomly fails. I bisected it and it looks like the
>>>> culprit is the commit 4e103134b862 "KVM: x86/mmu: Zap only the relevant
>>>> pages when removing a memslot"[1]. I tried to revert in on top of 5.2.1
>>>> but too many things changed in the meantime. Anyway, if I replace the
>>>> body of kvm_mmu_invalidate_zap_pages_in_memslot with
>>>> kvm_mmu_zap_all(kvm); it works again (probably with horrible performance
>>>> degradation).
>>>>
>>>> Did anyone experience anything like this? I'm using Alex's ACS override
>>>> patch, maybe it violates some assumption that the new code has?
>>>
>>>       Hi,
>>>
>>>       I noticed some changes that made 5.0 not working well when
>>> detecting screen speakers through hdmi, but this I didn't see anytime.
>>> My problem flew away with 5.1.15(the one I currently use), and no
>>> other spread. I never needed the ACS override patch in my setup,
>>> what happen if you try without it, does your groups comes wrong in
>>> any ways?
>>>
>>>       Best regards.
>>>
>>>       José.
>>>
>>
>> Hi,
>>
>> Unfortunately without pcie_acs_override=downstream my iommu groups look
>> like this (i.e. both video cards and their pci bridges are in one
>> group), and I never had a problem with it in the last ~4.5 years.
>>
>> # ls /sys/kernel/iommu_groups/*/devices
>> /sys/kernel/iommu_groups/0/devices:
>> 0000:00:00.0
>>
>> /sys/kernel/iommu_groups/10/devices:
>> 0000:00:1c.3
>>
>> /sys/kernel/iommu_groups/11/devices:
>> 0000:00:1d.0
>>
>> /sys/kernel/iommu_groups/12/devices:
>> 0000:00:1f.0  0000:00:1f.2  0000:00:1f.3
>>
>> /sys/kernel/iommu_groups/1/devices:
>> 0000:00:01.0  0000:00:01.1  0000:01:00.0  0000:01:00.1  0000:02:00.0
>> 0000:02:00.1
>>
>> /sys/kernel/iommu_groups/2/devices:
>> 0000:00:02.0
>>
>> /sys/kernel/iommu_groups/3/devices:
>> 0000:00:03.0
>>
>> /sys/kernel/iommu_groups/4/devices:
>> 0000:00:14.0
>>
>> /sys/kernel/iommu_groups/5/devices:
>> 0000:00:16.0
>>
>> /sys/kernel/iommu_groups/6/devices:
>> 0000:00:19.0
>>
>> /sys/kernel/iommu_groups/7/devices:
>> 0000:00:1a.0
>>
>> /sys/kernel/iommu_groups/8/devices:
>> 0000:00:1b.0
>>
>> /sys/kernel/iommu_groups/9/devices:
>> 0000:00:1c.0
>>
>> Regards,
>> Zoltan
>>
>> _______________________________________________
>> vfio-users mailing list
>> vfio-users at redhat.com
>> https://www.redhat.com/mailman/listinfo/vfio-users
>>
> 

-------------- next part --------------
A non-text attachment was scrubbed...
Name: hack.diff
Type: text/x-patch
Size: 1065 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20190803/e1ca74bb/attachment.bin>