Hi there! As proposed on lkml, here's my little contribution to the Linux Test Project. That's for sure, code inspection by 20,000 eyes is better than having the British museum lending us 10,000 monkeys. However, sometimes clever programers leaves stupid bugs after late night hacking. So here are some of my favorite's attached, to check stability: * mem01 Tries to malloc as much memory as possible, and make pages dirty to force RAM allocation. In other words, this is a memory stress test. Makes linux-2.4.2-ac20+am_console_patch+ng_lowlatency panic.. * proc01 Reads all the readable files in /proc. Good to exercise read locks, and good kernel latency killer too. Let me know if it blocks on read (shouldn't). Also "cat /proc/sys/net/ipv4/route/flush" returns EINVAL on read, is it the expected behaviour? * crash01 This one comes from the world famous 'crashme-2.4', written by George J Carrette. Actually, a lot of code has been rewritten, but credits are due. * crash02 As crash01 creates random instructions and executes them, crash02 creates random syscalls and executes them. This may be dangerous since it can actually erase all your data. However, on non-production system, with fs mounted ro and run as nobody, this could be safe (NO WARRANTY). After some testing, I noticed that some random syscalls are killed by SIGSEGV. Aren't they supposed to return nicely with EFAULT instead? I tried to use the ltp library, following the mmap001.c example. If you see something wrong in my code, please fix it, or tell me about it. These programs are more like "please make me crash" tests, and I'm wondering if different tests like "is the OS working the way we're expecting it" would be interresting. Have you ever gave a try to the VSX-PCTS test suite at http://www.opengroup.org/testing/testsuites/ ? I guess there will be some licensing issues. Anyway, keep on the good work, Linux needs ltp. Have fun, -- Stephane PS: OS signaling: "Don't drive and crash"