Robert Swindells, with addition of serial console, other assorted cleanup
and much headbanging by me. Tested only on a iPaq 3970 so far.
NB: So far this uses the SA-11xx-targetted assembly code (including cache
handling, etc). Fortunately, none of that's required since hpcboot allocs
cache-inhibited pages from WinCE to copy the 2nd-stage loader and kernel
into. I've split the code into SA-11xx and PXA-2x0 bits in my tree, but
that needs help/review from someone with ARM/XScale clue.
3af75740f5