Re-work the udelay() function. On fast ColdFire CPU's we are
overflowing in the calculation and getting totally bogus
results.

On old 68k and ColdFire CPU's we have no 32*32->64 multiply
instruction. So we resort to a mul/shift calculation similar
to what ARM uses.
Signed-off-by: Greg Ungerer <gerg@snapgear.com>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>