ARM: 6998/2: kernel: use proper memory barriers for bitops

Improve scalability by avoiding costly and unnecessary L2 cache sync
in handling bitops.

Signed-off-by: Heechul Yun <hyun@nvidia.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>