Description of problem: With a Rawhide x86_64 system, last fully updated about a week ago, kernel updated (and rebooted) yesterday (kernel-2.6.17-1.2396.fc6.x86_64). I did a "yum -y update" to upgrade to the latest rawhide, for some reason the system ran out of ram during the upgrade. (512 Mb RAM, 2 Gb swap). Nothing helped but the reset button. After reset I found the attached text in my /var/log/messages. It seems gconf was running wild, but never got killed.
Created attachment 132481 [details] /var/log/messages
gconf is going completely out of control, essentially forkbombing the machine and munching up gobs of ram. I saw this happen on one of my boxes too a few days ago. I somehow ended up with *hundreds* of gconf processes.
ah, what a horrible bug.
So two interesting lines are: Jul 15 07:16:41 shalem gconfd (hans-2265): SIGHUP received, reloading all databases Jul 15 07:16:42 shalem gconfd (hans-2265): Received signal 8, shutting down abnormally. Please file a GConf bug report. It looks like it's getting a hangup signal to reload itself, and somehow in the process of doing that it sets some persistent state that causes a floating point exception everytime it runs.
Hmm, I noticed a few other things randomly dieing with sigfpe around the middle of last week , but then they seemed to disappear, and I didn't see the gconf issue repeat itself either. Given other things were also getting sigfpe, I think gconf may just be a poor victim here. There was an FPU optimisation merged into the kernel a little while earlier (2376 on July 12th), so maybe this is a kernel bug after all. I'll throw it out of the next build, and we can see if this reoccurs. Note, that patch is currently in -mm, and will probably go upstream in 2.6.19, so if this starts happening again around then, we'll know this was the cause.
Okay, i'm going to throw it back your way then. If the kernel theory doesn't pan out, feel free to punt back to me.
For what its worth I haven't seen this since, but yesterday beep-media-player (gtk2 xmms) crashed on me with a SIGFPE.
Closing due to lack of reproducibility