With the update to the spi_bitbang driver, the transfer setup code is
being called more often, and thus is often re-doing calculations that have
been done before.  The SPI layer allows our driver to add its own data to
each device so add a result cache to each device.

This should also remove the problem where we where directly setting up
registers in the setup call which meant we might overwrite the state of an
extant transfer.,
Signed-off-by: Ben Dooks <ben@simtec.co.uk>
Cc: David Brownell <david-b@pacbell.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>