alan> So you've got a buggy CPU and a random additional patch. When you get AMD to
alan> swap your CPU (they will) see if you can repeat it without the QNX scheduler
I'm still trying to get the guy that sold me the CPU to replace it. I
cannot do without the CPU for the number of weeks it will take for me to
send it to the U.S. and back. Also, as the bug only causes random
segfaults when I'm compiling the kernel (in which case I simply recompile
again) it's not a main priority right now, since the machine has been
rock-solid apart from that
Also, I don't really see why the QNX scheduler would cause OOPSen in
<select_dcache>, and also why most people on this list have an aversion to
the QNX Scheduler patch, which I have used on this machine since version
1.06 on 2.0.31, and which has been rock-solid ever since, as has 2.1.65
apart from the OOPSen I sent in earlier.
What I like most about the QNX scheduler, is that I can leave a password
cracker going in the background, at a low priority, and it will only use
CPU time when nothing else wants the CPU, so it doesn't affect performance
at all.
Also, with the QNX sheduler in place I have been able to push my
load average to 1000 and over (with a forkbomb and ulimit -u 1024) and
the system was still very responsive from my root login. (which I had
niced negatively.)
It's worth looking at, and should IMHO be considered for inclusion in
2.1.X, if only as a config/Makefile option.
- Matt