Now that mb() is an instruction barrier, it will slow performance if we issue
unnecessary barriers.

The spinlock defines have a number of unnecessary barriers.  The __ldcw()
define is both a hardware and compiler barrier.  The mb() barriers in the
routines using __ldcw() serve no purpose.

The only barrier needed is the one in arch_spin_unlock().  We need to ensure
all accesses are complete prior to releasing the lock.
Signed-off-by: John David Anglin <dave.anglin@bell.net>
Cc: stable@vger.kernel.org # 4.0+
Signed-off-by: Helge Deller <deller@gmx.de>