I've attached an old CPU memory latentcy test that ram quite happily on 
linux SMP machines until today.

I have a Dual AMD 2400 with an MSI K7D-Master, IG RAM.

The problem is that the test segfaults in random places.

The code is quite simple, using fork and a shared mmap()ed file.

Code goes like this:

Map a temp file shared.
initialize some ptrs.
fork
    - child busy loops reading data from shared memory
    - parent busy waits (using tsc) then sets mem location
    - child pops out of loop and also reads tsc
    - parent busy loops reading data from shared memory
    - child busy waits (using tsc) then sets mem location
    etc etc

Core files indicate that the memory is unavailable however gdb indicates 
that the memory is readable.

I'm downloading 2.4.23 and I'll give that a go later.

If anyone wants to try it for me on a 2.6 or a 2.4.23 kernel I'd 
appreciate that.  This could concievably be a hardware problem but that 
seems unlikely since the machine has been very stable for quite a while.

Ideas ?