Dear Maintainer,
* What led up to the situation?
I was watching a Youtube video in firefox-esr 140.9.0esr-1~deb13u1 on
2026-04-06, in Xfce. Somewhere around 22:15 (I don't know the exact
time), all graphics froze. I could hear firefox continuing to play the
video. It responded to keyboard input and I could pause the video with
the spacebar, and close Firefox with Alt-F4.
* What exactly did you do (or not do) that was effective (or
ineffective)?
I was able to switch to VT2 and save some logs. I switched back and
forth between VT2 and VT7 (where X was running) a few times, and
sometimes during the switch the graphics would update a little bit, but
then go back to being frozen.
I restarted lightdm and was able to log in again, but the graphics
immediately froze again before the desktop appeared. It took a full
reboot to get things back to normal. This is why I think it's a kernel
bug and not an xorg bug.
I did not try /sys/kernel/debug/dri/1/amdgpu_gpu_recover, because I
didn't know about it at the time.
I have not tried any newer kernels or firmware from trixie-backports.
The freeze also happened under the same circumstances a week earlier,
with the same Firefox version. Earlier Firefox versions didn't trigger
the freeze.
I turned off "Use hardware acceleration when available" in Firefox, and
turned off the compositor in Xfce, and haven't had any freezes since.
The "(WW) AMDGPU(0): flip queue failed: Invalid argument" and
"(WW) AMDGPU(0): Page flip failed: Invalid argument" in the Xorg log are
associated with VT switching, not the freeze. The warnings seem to have
stopped happening now that I've turned the compositor off.
(Incidentally, turning off the Xfce compositor also made the graphics
more responsive and less laggy. With it on, there was a slight lag to
almost everything, most noticeable when dragging windows. It was like
that in both Bookworm and Trixie.)
The kernel is tainted because of the system76-io driver that I got from
Ubuntu and rebuilt for Debian. It's necessary to control fan speed on
the System76 Thelio Mira. Trying to reproduce the bug without it would
mean running for an extended period with the fans out of control and
stuck at full speed, and I'm reluctant to do that.
The Ubuntu packges I rebuilt and installed are:
system76-io-dkms 1.0.4~1724333961~24.04~968f68c~dev
system76-power 1.2.8~1758643655~24.04~2ebd0d2~dev
I'd be surprised if this is fixable with the information I have right
now. I'm wondering if there's anything more I can do to debug this if I
can reproduce it. Unfortunately it's not reliably reproducible, so it
could take a while.