The DMA descriptors used by the DW MMC block must be aligned to cacheline
size, otherwise we are unable to properly flush/inval cache over them and
we get data corruption.

The reason I chose this approach of expanding the structure is because
the driver allocates the descriptors in bulk. This approach does waste
space by inserting slop inbetween the descriptors, but it makes access
to the descriptors easy as the compiler does know the real size of the
structure. It also makes cache operations easy, since the size of the
structure is cache aligned and the structure start address is as well.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Chin Liang See <clsee@altera.com>
Cc: Dinh Nguyen <dinguyen@altera.com>
Cc: Albert Aribaud <albert.u.boot@aribaud.net>
Cc: Tom Rini <trini@ti.com>
Cc: Wolfgang Denk <wd@denx.de>
Cc: Pavel Machek <pavel@denx.de>
Cc: Pantelis Antoniou <panto@antoniou-consulting.com>
Acked-by: Pavel Machek <pavel@denx.de>