ARM: 7983/1: atomics: implement a better __atomic_add_unless for v6+

Looking at perf profiles of multi-threaded hackbench runs, a significant
performance hit appears to manifest from the cmpxchg loop used to
implement the 32-bit atomic_add_unless function. This can be mitigated
by writing a direct implementation of __atomic_add_unless which doesn't
require iteration outside of the atomic operation.

Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>