Greetings. I have already done some high level searchs of the
linux-kernel mailing list, as well as linux24.sourceforge.net (tytso's 2.4
todo list) and haven't seen mention of a problem referencing this problem.
The problem is large numbers of threads in 2.4.0-test8 can result in a
hard crash of the entire kernel. This can be done as a non-root user.
Code to reproduce the problem (using perl-threads), as well as my
kernel-config are available at http://www.psyber.com/~ted/linux24crash/
[1]
The code creates X threads and lets them run Y seconds. When X > 1000
instabilty sets in, with segfaults on exit.
For X > 1025 (or so) a ctrl-c during the run will cause a kernel crash.
This happens even if the process is running as a non-root user. Instant
kernel take-down. (1200 to 1500 threads with a ctrl-c guarantee a crash
on my machine.)
The machine is an Athlon 750 w/ 128mb (crucial memory) on a Asus K7V,
running Debian/potato (partly w/ "unstable" packages") and the 2.4.0-test8
kernel. This test didn't phase a 2.2.13 kernel, but only 253 threads
started sucessfully (limited by "max user processes" I assume).
2.4.0-test8 seems limited by some sort of max-files-per-process problem on
process cleanup/exit, _although_ changing the open-files via ulimit on
2.4.0-test8 or 2.2.13 had no affect (either to a value of 1500 or 130).
Obviously this is a major bug that should hold up the 2.4 release until
it is resolved. I apologize for not having a patch or a C version of the
code... the new job is keeping me more busy than I'd like.
[1] http://www.psyber.com/~ted/linux24crash/
I figured 20k of extra traffic to the list should be avoided as probably
less than 20 of you will look deeply at this.
-- Ted Deppner http://www.psyber.com/~ted/ - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Sat Sep 23 2000 - 21:00:30 EST