5.4Pre sched_ule SMP XOrg hangs or reboots
bsd at noogenesis.org
Sun Mar 27 13:07:09 PST 2005
Hi, i have become fascinated with getting sched_ule to run reliably, and
I've been running tests with about 8 different configurations. I've
listed some of them and the results at my website, at
http://noogenesis/freebsd/testing_ule, the index html for which I've
also attached to this email. I have not yet succeeded in getting a core
dump from the kernel, any advice on coaxing a kernel panic instead of a
hang so that a core dump can be obtained, let me know.
The score is this, I get a hard hang (capslock/numlock do not respond,
can't ping) when running XOrg, always after loading the program nicotine
(ports/net/nicotine), which is a python/gtk2 program. Sometimes, it has
done a sudden reboot instead of a hang, and I haven't had much luck
reproducing these results with any other program in X reliably. Also, I
have not had any hang at all when not running X. As far as testing
goes, ive generally recompiled the kernel with -j8 as a stress test,
without XOrg, and have had no problems with any configuration. So the
bug on my system involving sched_ule seems to be brought out by
something nicotine is doing within the first minute or so of loading
up. I do think that nicotine uses gdk threads, which use the posix
threads, so the connection could be there...
I'm going to continue to search/test things as I think of them, but any
advice as to what to check next would be helpful.
One thing that my results there don't list is the following:
Never had a crash or hang with SMP disabled (even with sched_ule enabled).
Also never had a crash or hang with sched_4bsd and SMP enabled.
My system is a Tyan Tiger MP motherboard with dual athlon MPs.
Any other information that you might find helpful to know, let me know.
Also: what other email lists should I post this too, if any? Should I
make this into a problem report? sched_ule is already known to have
More information about the freebsd-stable