I noticed David posted a patch to do something similar the
other day. With a generic atomic_cmpxchg available, such a
patch is no longer objectionable to those architectures
that don't #define __HAVE_ARCH_CMPXCHG

Do we want this David? Any other architecture have a super
optimised atomic_dec_and_lock that surpasses this
implementation?

-- 
SUSE Labs, Novell Inc.