>The only thing that I've seen that is consistent in all the crashes is
>that they always occur during _very_ heavy disk I/O, like when savecore
>or kvm_mkdb is run.  Most of the time, it's a MMU fault like Richard
>...
>The last time I built a kernel that didn't have these problems was back
>on 8-August, so I'd be willing to bet that something has changed since
>that time.  Perhaps there's another egcs gode-gen bug in m68k (as there
>apparently is in the -current, or at least very recent, i386 egcs)?

funny you should mention that.  any recent kernels that i build with
egcs must be built with no optimization (i.e. -g)  if i turn on optimization
the kernel will crash next time i try something like 'make' in /usr/src
however, when built with -g, the kernel is quite stable.  btw, i am
using sbc scsi, not ncr scsi, and i am not sure if that will make
a difference.

there definitly are some optimizer bugs with egcs on the m68k.  it
does miscompile libgnumalloc with optimization turned on.  atleast
the very least its creating bad code when optimizing in pic mode.
just looking at the assembler, it egcs seems to keep around quite
a few 'virtual' registers on the stack.  egcs-current from cygnus.com
appears to have fixed this problem, but its not very stable.