mmc: sdhci: convert ADMA descriptors to a coherent allocation

Rather than using the streaming API, use the coherent allocator to
provide this memory, thereby eliminating cache flushing of it each
time we map and unmap it.  This results in a 7.5% increase in
transfer speed with a UHS-1 card operating in 3.3v mode at a clock
of 49.5MHz.

CRs-fixed: 684185
Change-Id: I82aae6264db1ebfb87badbb223893b066ad53652
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Tested-by: Markus Pargmann <mpa@pengutronix.de>
Tested-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Chris Ball <chris@printf.net>
Git-commit: d1e49f77d7c7b75fdc022e1d46c1549bbc91c5b7
Git-repo: git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git


[venkatg@codeaurora.org: use adma_desc_sz for ADMA_SIZE & align_bytes - 1
for alignment check]
Signed-off-by: Venkat Gopalakrishnan <venkatg@codeaurora.org>