Re: [RFC][PATCH 1/3] locking: Introduce smp_acquire__after_ctrl_dep

["Paul E. McKenney" <paulmck@linux.vnet.ibm.com>]

On Tue, Jun 07, 2016 at 07:48:53PM +0200, Peter Zijlstra wrote:
> On Tue, Jun 07, 2016 at 08:23:15AM -0700, Paul E. McKenney wrote:
> > and if the hardware is not excessively clever (bad bet, by the
> > way, long term),
> 
> This ^
> 
> > > Is there something else than conditional move instructions that could
> > > come to play here? Obviously a much smarter CPU could evaluate all the
> > > jumps and come to the conclusion that the write to c is never depending
> > > on the load from a, but is this implemented somewhere in hardware?
> > 
> > I don't know of any hardware that does that, but given that conditional
> > moves are supported by some weakly ordered hardware, it looks to me
> > that we are stuck with the possibility of "a"-"c" reordering.
> 
> Is why I'm scared of relying on the non-condition.
> 
> The if and else branches are obviously dependent on the completion of
> the load; anything after that, not so much.
> 
> You could construct an argument against this program order speculation
> based on interrupts, which should not observe the stores out of order
> etc.. but if the hardware is that clever, it can also abort the entire
> speculation on interrupt (much like hardware transactions already can).
> 
> So even if today no hardware is this clever (and that isn't proven)
> there's nothing saying it will not ever be.
> 
> This is why I really do not want to advertise and or rely on this
> behaviour.

What Peter said!  ;-)

							Thanx, Paul