Re: [RFC] unify semaphore implementations

[Richard Henderson <rth@twiddle.net>]

On Fri, Apr 29, 2005 at 10:44:17AM +1000, Paul Mackerras wrote:
> You have made semaphores bigger and slower on the architectures that
> have load-linked/store-conditional instructions, which is at least
> ppc, ppc64, sparc64 and alpha.

And mips.

While sparc64 doesn't have ll/sc, it does have compare-and-swap and
it's trivial to use that exactly like we use ll/sc.  S390 also has
compare-and-swap as its atomic primitive.

Seems to me that the ppc semaphore implementation is superior to the
i386 implementation that seems to have been propagated here.  Indeed,
I might think it would help i486, ia64, and amd64 to use the ppc style
compare-and-swap instead of the existing implementation.  Care would
have to be taken such that i386 still works, but I suspect the vast
majority of folk don't configure for that.

I would support two or three common implementations, but definitely
not the one implementation presented.


r~