Hello,

I fear I start to see a pattern here:

I build a kernel using setting COPTS in mk.conf to -mcpu=7400 and -O3
and I reported it crashing within a couple of minutes of load.

I make clean, set the optimizaiton to -O1... and rebuild. the kernel is
much much more stable!
I run it for two hours at extreme high load (two or more parallel
compilations, desktop usage, irc, and just apps open for fun) and after
that time it crashed, but being in X I cannot confirm it was the same
kind of crash.

Today again with the same kernel, I'm up 44minutes and doing all kind of
works. Iexport display for now so I can check what happens.. If it stays
stable for more than 90 mins I will start X too.

the kernel was built using the *same* conf file, I didn't even
reconfigure. Thus I really think something in the kernel is flakey and
sensible to optimizations, apparently my experiences are different from
othes. I haven't dare to try a -O0 kernel yet to see if it is 100%
reliable.
It is erally strange that the stability varies such way. My guess is
that there could be a wrong pointer somewhere or the code has problems
in more than one point andthus is sensible to optimization. Otherwise I
can't explain this "analog beavior" with optiization. It is not just
with/zithout it works/don't works.
It may also be a cache problem and the different optimizations produce
code that uses cache differently...

I hope others can get more defiitive results. I'm pretty convinced now
that the problem is inside the netbsd kernel and not in my hardware. I
again tested MacOS a bit after the new official utility set he cache to
"WB" and I had absolutely no problems.

-Riccardo