Atomic operations on the ARM

[Richard Earnshaw]

> I don't think the two approaches overlap much even today, if at all.
> I've occasionally thought about making the generic atomicity.h routines
> call into the gthr.h layer, thus bringing them together and providing a
> fallback atomicity header that is actually atomic, if terribly slow.

Surely the cost would only be noticeable if there was another thread 
trying to update an atomic variable at the same time.  And that's the case 
that we are getting wrong at present with the 'don't do atomic operations' 
code that we have at present.

R.