Re: futex atomic vs ordering constraints

[Davidlohr Bueso <dave@stgolabs.net>]

On Wed, 2015-08-26 at 20:16 +0200, Peter Zijlstra wrote:
> Of course, if anything else prior to futex_atomic_op_inuser() implies an
> (RCsc) RELEASE or stronger the primitive can do without providing
> anything itself.
> 
> This turns out to be the case, a successful get_futex_key() implies a
> full memory barrier; recent: 1d0dcb3ad9d3 ("futex: Implement lockless
> wakeups").

Hmm while it is certainly true that get_futex_key() implies a full
barrier, I don't see why you're referring to the recent wake_q stuff;
where the futex "wakeup" is done much after futex_atomic_op_inuser. Yes,
that too implies a barrier, but not wrt get_futex_key() -- which
fundamentally relies on get_futex_key_refs().

> 
> And since get_futex_key() is fundamental to doing _anything_ with a
> futex, I think its semi-sane to rely on this.

Right, and it wouldn't be the first thing that relies on get_futex_key()
implying a full barrier.