[ARM] Fix broken sl82c105 DMA prevention

We must _never_ _ever_ on pain of death enable IDE DMA on SL82C105
chipsets where the southbridge revision is <= 5, otherwise data
corruption will occur.

Strangely this used to work, but something has changed in the upper
echelons of the IDE layer to break the hosts decision to deny DMA.
Let's make it crystal clear to the IDE layer that we know best.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>