Dear Maintainer,

I just switched from Bookworm to Trixie and now experiencing a serious stability issue.

After a while, the driver (or kernel) reports a “GPU lockup” then the screen goes black every few seconds, and at some point everything goes haywire and a forced reboot becomes
necessary (switching to a console does not work). It's hard to pinpoint exactly what triggers this problem, but I've noticed that switching between windows using the Alt-Tab
combination is a very recurring cause of the graphics server crashing (I use Gnome). And of course, I use this function all the time!

I even reinstalled the system from scratch to be sure, but the symptoms are exactly the same. This is very annoying, because with Bookworm, my system was very stable.

Note : A dmesg log just after the crash is attached.

Thanks.

Just a little feedback after a few days' "use". Unfortunately, at this
stage, my system is unusable because there are far too many crashes. It's
switching between windows that at some point will inevitably crash the
system.

Question: could using a previous kernel (while keeping the latest official
kernel, with the option of switching from grub), provide a temporary
solution (1), to regain a stable configuration available? I'm talking about
an older kernel officially provided by Trixie.

(1) I'm aware that this kind of bug is often tricky or time-consuming to
fix.

Thank you.

Hello,

The most time-consuming task is on determining the culprit and you're in
the best position to debug that.

There are many kernel packages available at
https://snapshot.debian.org/package/linux/, can you do a package
bisection and pinpoint the oldest kernel that has this issue? (That is,
pick a kernel somewhat in the middle between 6.1 (from bookworm, good)
and 6.12 (from trixie, bad), say 6.7-1 and test that. Depending on
the result, continue to search between 6.1 and 6.7-1 (if 6.7-1 is
bad) or 6.7-1 and 6.12 (if 6.7-1 still works). Tell us if you need more
guidance here.

Make sure to uninstall the old kernels at the end to not be exposed to
security issues that are already fixed in later kernels.

Best regards
Uwe

Another thing you can test is installing a 6.16 kernel from
experimental.

Best regards
Uwe

1) Kernel 6.16 caused two crashes in half a day (but this seems to be
related to the processor and not the GPU).

2) I didn't notice it at first, but version 6.12.38 is provided in addition
to version 6.12.41, and the former is considered a little more stable, so
I'm testing that one for now.

3) OK, so I'll try this dichotomous process to identify the culprit. It may
take a long time!

4) In addition to the lockups, I see a lot of messages like this (I don't
know if it's anecdotal):

August 22 18:45:20 station1 firefox-esr.desktop[6173]: [GFX1-]: Unable to
clean up GL_RENDERER “KABINI (radeonsi, , ACO, DRM 2.50, 6.12.38+deb13-amd64
)”.

How many days can be considered long enough to conclude that a kernel does
not have this bug? Provided that I test each kernel by doing my usual work
and that I also frantically switch between windows from time to time with
Gnome, knowing that this is critical?

Please refer to the sources.list file attached to the ‘snapshot’ section,
where I have compiled what I have done. The bug was therefore introduced
between versions 6.1.147 and 6.3.0.

From the list of snapshots you provided, I tested the oldest kernel after
Bookworm that I could find, namely 6.3.0. And so 6.3.0 also crashes,
although it is more stable than 6.12.41.

I used reportbug, but I don't know why it didn't automatically provide the
usual information about my environment. If you need more information, please
let me know.

Basically, it's the same for every kernel, but here is the dmesg log
generated just after the crashes.

So it's the driver... what else could it be?

Things are becoming clearer...

I removed *all* non-free packages, namely:
ii  amd64-microcode                         3.20250311.1
      amd64        Platform firmware and microcode for AMD CPUs and SoCs
ii  intel-microcode                         3.20250512.1
      amd64        Processor microcode firmware for Intel CPUs
ii  firmware-intel-graphics                 20250410-2
      all          Binary firmware for Intel iGPUs and IPUs
ii  firmware-intel-misc                     20250410-2
      all          Binary firmware for miscellaneous Intel devices and chips
ii  firmware-mediatek                       20250410-2
      all          Binary firmware for MediaTek and Ralink chips for
networking, SoCs and media
ii  firmware-misc-nonfree                   20250410-2
      all          Binary firmware for various drivers in the Linux kernel
ii  firmware-realtek                        20250410-2
      all          Binary firmware for Realtek network and audio chips

I now use the “official” linux-image-6.12.43+deb13-amd64 kernel and it no
longer crashes at all, even though my system is slightly slower with the
open source module. But in any case, it's perfectly usable for my work, as
I don't need exceptional GPU performance.

Hi,
firmware-amd-graphics is *still* installed but all the other firmware
packages removed or ar you saying firmware-amd-graphics is as well
deinstalled? What is the case here?

But if the situation is as follows: With installed
firmwware-amd-graphics: GPU lock, black screen and crashes.

With firmware-amd-graphics deinstalled, system running more stable.

If this is the case, then please update to the most recent 6.12.48-1,
confirm the above statement, and then please provide in both cases
full kernel log.

Thanks in sadvance,

Regards,
Salvatore

Sorry, my last message was indeed ambiguous. The list showed the non-free
packages on my PC *before* I removed them all. In fact, I even added #non-
free-firmware as a comment in /etc/apt/sources.list and performed an update,
etc., in order to have a completely free system.

And with the 6.12.48+deb13-amd64 kernel reinstalled, my system is rock
solid again (no crashes or problems in a week of use).

So, in reality, the problem comes from amd64-microcode, what else? If case
you want to reassign this bug. For my part, I will no longer use non-free
packages.

Hi

Should this bug be reassigned back to package 'firmware-amd-graphics'?
I have this package installed and the bug has only been happening since
this package was updated on 2025-11-30.
My Kernel: Linux 6.17.13+deb14-amd64 (x86_64)
Card: Kabini [Radeon HD 8280E]

Regards
John