Having similar problem as reported in Bug# 10509,
but am running from updates/6.2/i686/kernel-smp-2.2.16-3.i686.rpm.
Lilo boots with no initrd (IDE boot disk).
Seti@home processes will eventually (usually within 4 hours)
terminate with a floating point exception error. The same behavior
can be seen in Quake3.
Machine is Tyan Tiger 100 motherboard with 2x PIII 700e processors
installed. 128MB RAM.
I have not tried other kernels to this point, and will be glad to do some
troubleshooting per your guidance.
What happens if you install the i586 or i386 kernel
instead of the i686 kernel?
28 Jul - kernel 2.2.16-3smp compiled for 386
(updates/6.2/i386.kernel-smp-2.2.16-3.i386.rpm) has not
demonstrated the problem in about 10 hours of operation
where it used to happen generally within 4. It appears
this compiled kernel does not exhibit the problem.
All my other machines have custom-built kernels, and
so it the plan for this machine. Do you have an idea
of the element of the 686 kernel that is causing this issue,
so that I can attempt to avoid it? I do definitely prefer
to have code compiled to take advantage of the 686
instruction set if possible.
We did not enable the PIII patches (KNI instructions) in the
2.2.16-3 kernel, so that isn't the problem, at any rate. It
could possibly be slightly flaky hardware that isn't stressed
as hard with the other kernels, so there's no guarantee of
a kernel config option that would change it. You can try
building kernels with specific changes, such as turning off
bigmem support -- that's one of the differences between
the i686 and lower kernels.