Remove the mutex serializing access to the cache.
Instead, queue up new requests on a bio_list if the driver is busy.

This improves sequential write performance by ca. 2%.
Signed-off-by: Geert Uytterhoeven <Geert.Uytterhoeven@sonycom.com>
Cc: Jim Paris <jim@jtan.com>
Cc: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>