It's possibly that we get an reset requestion when interrupts are disabled.
(For example an oops in an interrupt handler).  Therefor, we can't call
ioremap in the reset function.  Moving the ioremap of the registers we
need access to an arch_initcall helps the problem.

However we still have a window between boot and the arch_initcall in
which the register pointer will not be setup and thus we spin if the reset
function is called.  If one needs to ensure even this case is covered, look
at use of the watchdog provided on 83xx to reset the processor.

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>