Re: futex: clarification needed with drop_futex_key_refs and memory barriers

From: Peter Zijlstra
Date: Thu Mar 31 2016 - 03:36:56 EST

On Wed, Mar 30, 2016 at 08:45:10PM -0500, Petros Koutoupis wrote:
> > But this is not a correctness (nor ordering) issue; but purely an
> > architectural side-effect. Furthermore; some proposed changes:
> >
> >
> >
> > might change this side-effect.
> >
> Has there been an update to this patch? The last I see, the conversation
> ended at the end of January, and there hasn't been a change in the
> mainline.

Peter Anvin was going to look at this with some of the Intel hardware
people to fully explore the ramifications of this change, we're waiting
on feedback from that.

> >> Your adjustments here make complete sense. Are you preparing it for
> submission in the near future?

I'll think about it, adding the extra barrier for decrement is of course
not really nice if not strictly required. And while it will not impact
x86 the weakly ordered archs will be affected.