We're filling the "remainder" word with little-endian data, then writing
it out to IO registers with endian-correcting writel(). That probably
won't work on big-endian systems.

Let's mark the "remainder" variable as LE32 (since we fill it with
memcpy()) and do the swapping explicitly.

Some of this function could be done more easily without memcpy(), but
the unaligned "remainder" case is a little hard to do without
potentially overrunning 'tx_buf', so I just applied the same solution in
all cases (memcpy() + le32_to_cpu()).

Tested only on a little-endian system.
Signed-off-by: Brian Norris <briannorris@chromium.org>
Reviewed-by: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Archit Taneja <architt@codeaurora.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180109203248.139249-2-briannorris@chromium.org