> I was going to try profiling the same kernel on a 5000/240 and 5900-260
> to try to get a handle on where to start looking for the fork/exec
> slowness, but I'm having problems getting a profiled kernel built and
> running.  With the trailing patches I at least get a kernel built, but
> it barely starts and get's an interesting error on the way through:
>
>	[ ... kernel panics try to access address 0xf ... ]
>
> #define        MCOUNT_ENTER    s = _splhigh()
> #define        MCOUNT_EXIT     _splx(s) 

If the profiler expects to disable and restore interrupts for short
period of time, toggling MIPS_SR_INT_IE bit of status register is a
handy way.  No idea why profiled kernel panics given your code, though.

Tohru Nishimura