Rather than using 8bit transfers for everything, use 8/16/32 bit transfers
as usable with the source/destination addresses and the count size.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>

We have to make sure the DMA channel is actually disabled in hardware before
attempting to reprogram it.  Otherwise the new settings are ignored and we
end up with random hangs/failures.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>

Our dcache invalidate function doesn't just invalidate, it also flushes.
So rename the function accordingly and fix the dma_memcpy() function so it
doesn't inadvertently corrupt the data destination.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>

Creating a new dma_memcpy() function that skips all cache checks allows us
to use the function in very early init where the cache is not yet setup.

Signed-off-by: Mike Frysinger <vapier@gentoo.org>

Signed-off-by: Mike Frysinger <vapier@gentoo.org>

All of the duplicated code for Blackfin processors and boot modes have been
unified.  After all, the core is the same for all processors, just the
peripheral set differs (which gets handled in the drivers).

Signed-off-by: Mike Frysinger <vapier@gentoo.org>