The ppc64 implementation of memcpy_to/from_io was bogus (used memcpy
which uses cache hints and thus is broken on non cacheable IO space).

This re-implements them with some simple/gross C code doing 32 bits
accesses when aligned and bytes accesses when not. 
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>