Make sure dma_alloc_coherent allocates memory from the local node.  This
is important on Cell where we avoid going through the slow cpu
interconnect.

Note:  I could only test this patch on Cell, it should be verified on
some pseries machine by those that have the hardware.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>