Comment 34 for bug 660405

Bryce Harrington (bryce) wrote :

Hi,

Thanks for reporting this X gpu lockup bug in Ubuntu maverick.

In my opinion GPU lockups are one of the most frustrating kinds of bugs, both as a user and a developer. I want take a few moments of your time to explain the situation with these types of bugs.

For some reason, -intel has been plagued with these freeze bugs for a long time. In the past -intel has had options to switch to legacy memory or rendering technologies as workarounds, but Intel tends to drop obsolete code pretty aggressively and unfortunately in maverick these workarounds are not available.

As a policy, Intel engineers look only at bugs that are seen in the most recent release of their drivers. With older versions of their drivers, such as the version in maverick, we're on our own.

On the plus side, they provide good tools for gathering debug data about the bugs, such as the intel_gpu_dump too. Unfortunately, the version of Apport in maverick doesn't capture the dmesg or X logs correctly (it collects the files at the time of freeze, but then overwrites them with clean, unhelpful versions when it goes to file the report for you.)

The version of these diagnostic tools in natty has been fixed up, and is allowing us to analyze GPU lockups a lot easier. This has helped us solve several of the bugs, and given us a handle on some of the more challenging ones.

Backporting these fixes may not be feasible though; by their nature the fixes for these type of bugs tend to be very low level, esoteric, and risky of causing something else to regress. Because the freezes are often hard to reproduce, it's not always possible to verify fixes anyway. It may be that our best bet is to focus on natty.

But for now, since these freezes typically require kernel patches to solve, I'm moving your bug report to the kernel package. They may be able to point you to a newer kernel likely to help with this issue.