Looking for help on strange Kernel Trap 12

Hello everyone,

I am looking for help on producing a core-dump or some other solution to
survive a strange kernel panic, which begins to appear from the beginning of
August, and persists in recent kernels, including the kernel made from
yesterday's source. It seemed that the panic was triggered by a heavy swap
load, and I were failed to make a coredump or drop into ddb(1) because it
finally stops to respond when sync'ing buffers with another panic, instead
of rebooting or dumping.

When I have cvsup'ed to the recent -CURRENT code and installed a new kernel,
it seemed to be frangible when having a heavy swap load, which results in a
kernel trap 12, which seemed to be occuring on ATA or NIC software
interrupt. What's worse, the panic always triggers a more panic, the
dual-panic will finally cause the system stop to respond, instead of dump
the core and restart, this generates a bit trouble for me because I can not
get access to the direct console often when I am not at the hosting room.

I think the stability, especially stability under heavy load is important
for FreeBSD, so I would like to provide some help to solve, or at least,
locate the problem. I have tried to make a core dump, however I failed
because the two panic simply eliminated the possiblity of giving me a chance
to fall into ddb(1).

Currently I can provide the following information, and of course if that's
not enough, just fell free to write me.

Software installed on the computer are: FreeBSD 5.1-CURRENT, pure-ftpd,
apache2 installed from port with default options.

Hardware: Pentium II-266, 64MB RAM, Intel Pro-100 NIC, ACPI enabled.

The kernel configuration is tweaked, among other things, I have enabled
AUTO_EOI_1, DEVICE_POLLING, and CC option -fno-builtin. Also, I have enabled
-O6 with WANT_FORCEDOWNGRADE=1 in /etc/make.conf, however, when I have
rebuilt the system with only -O1 the problem persists.

Any ideas?

Thanks in advance!

Cheers,
Xin LI
Frontfree Technology Network