Allocate the qTD items all at once instead of allocating them
separately. Moreover, make sure each qTD is properly aligned
to 32-bytes boundary and that cache can be safely flushed over
each qTD touple.

Signed-off-by: Marek Vasut <marex@denx.de>
Cc: Fabio Estevam <fabio.estevam@freescale.com>
Cc: Lei Wen <leiwen@marvell.com>
Cc: Otavio Salvador <otavio@ossystems.com.br>
Cc: Stefano Babic <sbabic@denx.de>