AIX? Feh! I've been working with a client porting some iXmatch code to AIX. Ugh. I've been killing their AIX machine on a regular basis; causing kernel panics. I'm working with java code and some JNI code, not doing anything as root at all. I shouldn't be able to cause the machine to choke at all if I'm not running as root. Ever. I though this was a solved problem, like, 15-20 years ago. I'm a java developer - I don't want to have to learn kdb!


WFT?


If you're interested, here's the stack frame causing the problem:

(0)> f
pvthread+007700 STACK:
[00646324]ld_xatt1+00008C (00000000, 10000000, 00000000, D12C5000, 00000001 [??])
[006B6F18]ld_relocate+000078 (??)
[006B3430]ld_loadmodule+0005D8 (2FF3B300, 2FF3A5CC, 00000031, 00000280, 2FF3A5CD, 00000000, 2FF3B268)
[006B3C5C]ld_proc_load+000530 (??, ??, ??, ??, ??, ??, ??)
[006BE97C]_dl_loadModule+000068 (??, ??, ??, ??, ??, ??, ??)
[006BE538]ld_loadutil+0000D0 (??, ??, ??, ??, ??)
[001070E0]forkstack_load+000060 ()
Not a valid VMM address @ D020F428
(0)>


Yeah - it looks like a dynamic loader problem to me too. It's also tied to the pthreads library somehow. The code works fine when running single-threaded, but in a MT environment, it dies hard. I wrote a Junit test class that spins up a bunch of child threads and then fires up the ixmatch code. If the ixmatch code is loaded from thread 0, it works great. If not, it dies and takes the machine down with it.

Sick, sick, sick.  ¶ 5:45 PM