Bug 40886

At the moment, we only output the errors that the GPU reports in the kernel logs. However, these are usually not helpful in any way.

To improve the quality of bug reports, it is also necessary to output meaningful registers values and try to understand roughly were the problem is. If possible, an error code should be generated to help merging bug reports into a meaningful one.

This task is *very* suitable for students who want to learn about nouveau. If you consider applying to this, please know that you will have a lot of documentation to read and you will also be required to ask many questions to the Nouveau developers. The actual implementation should be quite small.

The Ubuntu xorg team proposed us to improve our bug "reportability". Here is what they have available on the intel driver that we could actually try to copy.

# Jesse Barnes on ubuntu-devel@lists.ubuntu.com:
#   You'll get three events, one when the error is detected, one before
#   the reset and one after.  Each has a different environment variable set;
#   the initial error has ERROR=1, the pre-reset event has RESET=1 and the
#   post-reset event has ERROR=0.

# Disable freeze hook.
SUBSYSTEM=="drm", ACTION=="change", ENV{ERROR}=="1", RUN+="/usr/share/apport/apport-gpu-error-intel.py"

The python script copies dmesg, Xorg.0.log, and
/sys/kernel/debug/dri/0/i915_error_state.  The latter is an
intel-specific error dump they use to help diagnose bugs.
We also capture a variety of other data and files, but those three seem
to be what the devs want, mostly.

We extract a couple error codes from the error_state file to use as a
way of automatically detecting dupes.

Here's a few examples of the results of all this:

  - https://bugs.freedesktop.org/show_bug.cgi?id=35854
  - https://bugs.freedesktop.org/show_bug.cgi?id=34014
  - https://bugs.freedesktop.org/show_bug.cgi?id=34307