Re: 1.3.88 stress testing (was: .. and oh, btw ..)
Bernd Schmidt (crux@pool.informatik.rwth-aachen.de)
Tue, 16 Apr 1996 13:27:29 +0200 (MESZ)
> Aaaaghhhhh!!!! When I boot with mem=8192k with 1.3.88 and the 1 line
> memory patch, I get SIG11's when running the script. Booting with 1.2.13
> and running the script is fine. I didn't retry with 16M again but it did
> run that test 3 or 4 times and it was OK. It's *really* slow in 8M
> compared to 16M. I'll try to do some tests with earlier kernels to see
> where things went wrong but it is probably the async. swap stuff from
> 1.3.5x(?) that is broken. I haven't looked at 1.3.89 yet either ...
I did some more tests. I ran memtest86 on the machine for a few hours (80
passes). No errors. Then I disabled all caches, turned down the timing options
in the BIOS, turned off the Turbo switch in front of the machine and rebooted.
The machine usually has 33.2 BogoMIPS, for this test it was running at 3.2
BogoMIPS. And I was still able to reproduce memory corruption with the patched
1.3.88. The same system is absolutely fine with caches on and all timings set
to maximum when running kernels <=1.3.58, and broken with kernels >=1.3.59.
This is not a flakey SIMM or an overheating CPU. It's a bug in the swapping
code.