arm: Kirkwood: Fix compiler optimization bug for kwgbe_send

arm: Kirkwood: Fix compiler optimization bug for kwgbe_send

kwgbe_send/recv both have loops waiting for the hardware to set  a bit.
GCC 4.3.3 cleverly optimizes the send case to ... a while(1); loop. This
patch uses readl to force a read from device memory. Other volatile
accesses have also been replaced with readl/writel where appropriate
(as per suggestions on the U-boot mailing list).

Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
Signed-off-by: Ben Warren <biggerbadderben@gmail.com>