> I don't want to be part of this flamewar, but ...
> A divide by a constant (which essentially is what we have) can be
> turned into a multiply. Is a multiply really too expensive to have in
> hardclock()?  Then you must be running a truly pityful machine.

It's really not a 'divide by a constant', it's 'divide by a variable
that doesn't change,' if HZ is not speicified via a #define.

Yeah, some machines implement software division by certain numbers
to be muliplies then shifts, but not all, and that doesn't help if you
actually have a 'divide' instruction that's slow.

you really do want to avoid any extra overhead at all, in hardclock.