Disclaimer: This comment is written by me and only contains information
I attest to myself. The attachments  (patch and investigation report)
are LLM-written and I am not qualified to assess their accuracy. They
are shared in the spirit of documenting an experimental fix that worked
for me to aid the investigation of this issue. If this is not acceptable
to you, feel free to ignore the attachments or my comment entirely.


This bug also affects my system:

Ubuntu 24.04.4 LTS, 7.0.0-31-generic, RTX 3060, NVIDIA open 595.84,
PreserveVideoMemoryAllocations=1

After a suspend freeze, I let a current LLM investigate the issue and
fix it for me locally.

The first step was a script which reproduces the hang on my system by
performing lots of GPU memory operations while running a freezer-only
suspend test (systemctl --no-block suspend after writing "freezer" to
/sys/power/pm_test). This worked immediately to induce a hang which
behaved the same way as the "organic" one (system fans and lights still
running, capslock reacting, video output dead) and reportedly the system
logs pointed to the same underlying cause.

The second step was configuring the options nvidia
NVreg_UseKernelSuspendNotifiers=1 workaround already mentioned earlier
in this issue. After that, the reproducer script still caused an error
log involving virt_mem_allocator_gm107.c:2026, but the system recovered.

After that, I asked for a root cause fix and got a test patch for
nvidia-drm which fixes the deadlock issues my suspend was running into
(multiple rounds of both "normal" suspend and the repro script without
kernel errors). Reportedly, there are more errors of the same class
though. I also had it add fixes for the simpler ones of those for my
personal build, but don't include that here because those paths are
entirely untested.

Investigation report from the LLM with all the technical details but
none of the human attestation is attached. I read through it but would
only have spotted glaring errors. I can supply additional artifacts of
the investigation like the repro script and perform additional tests if
that's useful.

** Attachment added: "Investigation report (LLM-written)"
   
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2158993/+attachment/5999569/+files/INVESTIGATION-REPORT.md

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2158993

Title:
  System hard-freezes after failed suspend (Xwayland freeze timeout)
  with nvidia-driver-595-open — deadlock in nvidia_modeset/console

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux/+bug/2158993/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to