AMD RX 5700 XT suddenly stopped working with Qubes

[Jarrah]

Good Morning,

About a month ago, kernel-latest 5.6.4 was released which resolved the
issues booting Qubes on the RX 5700 XT. This has been working and stable
since this time.

Yesterday I installed a number of updates (including Xen and
kernel-latest) and (after the updates had completed) had an issue which
reset my bios. Since then, I have been unable to boot Qubes using the
5700 XT. System works fine in another OS.


The only error output I have seen is as follows (observed on different
boots):

"[Firmware Bug]: cpu 2, try to use APIC520 (LVT offset 2) for vector
0xF4, but the register is already in use for vector 0x0 on this cpu."

A failure to start IOMMUv2 (does not occur with the other graphics card)


After fixing the bios settings, my next thought was to roll back the Xen
and kernel update. However, this did not solve the issue.

Other attempts:

* Boot the system with another GPU: Works fine

* Boot the system without Xen (just the dom0 kernel): works fine.

* Boot the system outputing to the hypervisor console: system boots to
login screen. All autostarted VMs boot. Keyboard entry does not get sent
to dom0, but ctl-alt-del reboots the system and outputs to screen.

* Boot the system and let it sit after the screen locks up: All
autostarted VMs boot but I cannot interact with the system. I can tell
by the hvc above and the fact that my mouse turns on.

* The error above has been connected to an ACPI issue. Adding the
"acpi=off" boot parameter resulted in a kernel panic probably unrelated
to the GPU issue.

* Tested the just-released 5.6.13 kernel: Same issue now found on 5.6.4
(previously good) through 5.6.13.

* Set every IOMMU/ACS/SVM setting I can find in my bios.


I'm stuck for ideas at this point. Has anyone experienced something similar?

Thanks.

[Frédéric Pierret]

At this point, I would rather suggest you to check in changelog of kernel
if there would be related commits but still, post a BZ issue on kernel.

> Thanks.
>

Frédéric

[Jarrah]

> At this point, I would rather suggest you to check in changelog of kernel
> if there would be related commits but still, post a BZ issue on kernel.
>

There are a couple in 5.6.11 that I will have a better look at tomorrow.
Not sure this will be it though. The system fails to boot on both 5.6.4
(previously working) and 5.6.11 (new).

[Mike Keehan]

Doesn't seem likely that it's a kernel problem if 5.6.4 used to work
and now it doesn't. What was the bios issue?

[Jarrah]

> Doesn't seem likely that it's a kernel problem if 5.6.4 used to work
> and now it doesn't.  What was the bios issue?
>

A power fault caused it to drop all settings. I believe I have reset it
to the previous config and, at a minimum, it is compatible with Qubes
without the problem card.

[Mike Keehan]

Ah, OK. Not sure what I'd do now in your situation. About all I can
suggest is to research bios issues with that card via google. See if
anyone else has had problems. Not necessarily exactly what you had,
but just with that card. It might give you some clues.

Mike.