The default ARMv4 method consisting of reading through some memory
area isn't compatible with the cache replacement policy of some
ARMv5TEJ compatible cache implementations.  It is also a bit wasteful
when a dedicated instruction can do the needed work optimally.

It is hard to tell if all ARMv5TEJ cores will support the used CP15
instruction, but at least all those implementations Linux currently
knows about (ARM926 and ARM1026) do support it.

Tested on an OMAP1610 H2 target.
Signed-off-by: Nicolas Pitre <nico@marvell.com>
Tested-by: George G. Davis <gdavis@mvista.com>
Acked-by: Russell King <rmk+kernel@arm.linux.org.uk>