Arch Linux

prurigro

Member

Registered: 2008-03-14

Posts: 18

eee users, intel 915 users and any intel users /w opengl kernel panics

Theres a bug (I've already posted it) in the drm source of the linux kernel that causes a hard system lock, leaving the most recent image frozen on the screen until manually rebooting. The system is completely unresponsive.

This issue appears when running a vsynced opengl application and can occur between 5 minutes and a few hours of running the program. The issue is a race condition that eventually ends badly.

People effected are at least all eee and other desktop/laptop intel 915 users using at least the i810 and intel xorg drivers. It likely also effect the IEGD driver, other x environments and *may* effect other intel devices, but those haven't been tested against by me or the person who discovered the problem. The issue has not been fixed in head and effects at least 2.4.23 up to the current kernels including the patched ARCH kernel.

Happily, this bug does have a patch; created by a very intelligent programmer named Mike Isely who was commissioned by his company to find the problem and solve it. The patch has been tested extensively by both him and I and has been shown to be a complete fix; it is available here : http://bugs.freedesktop.org/show_bug.cgi?id=14937 This fix has not been looked at yet by dri or kernel developers (its rather new), so it may still be some time before its included in the kernel sources (hopefully not too long as this is a fairly critical problem)-- I'm hoping that prior to kernel developers adding the patch, arch linux will create a point release with it included, but for now there is no fix in the repos.

for the benefit of eee users and other people with slow processors (it took me about 2 hours to recompile the kernel with the patch) I have released a torrent with the patch + my edited PKGBUILD (for future compiling) and the latest arch kernel package rebuilt with the patch and no other changes (which I've been using and pushing hard for the last 36 hours) here : http://thepiratebay.org/tor/4078731/drm … kernel-eee -- obviously this will be out of date soon enough, but hopefully the next kernel release has the patch...

Encase anyone is curious, Mr Isely noted in an email to me that the problem was created because old features, not expecting the design of new features weren't coded in a way that allowed them to behave in a stable manner with each other. His fix was to stop the old features from leaving the holes that the new features were assuming didn't exist, so there shouldn't be any conflicts with the patch (the old code still works as intended). If anyone would like more in depth detail he outlines a lot in the patch but he sent me even more- ask and I'll paste the nitty gritty details

My test of the patch was 10 hours without error with a vsynced (via driconf) zsnes running a demo. Previously a vsynced zsnes instance would not last more than an hour with vsync on. Since then I have been using apps which have caused the lock up frequently in the past without error. Mr Isely ran many tests to ensure his patch was working but his longest was 20 hours; it is most definitely a solid fix.

The purpose of this post (ontop of providing the package) was to get the issue and fix out there; apparently my eee is the only computer to have been blessed with his patch thus far aside from the ones at his work. This needs attention because lot of people on the web are having problems; many seeming to think its a problem with the intel graphics driver and suggesting disabling dri as a fix (not a very good one!)

Anyway, feel free to leave comments if the torrent has problems or if anyone has any questions/comments about the issue.

ciao

Last edited by prurigro (2008-03-14 03:36:08)