On Wed, Aug 15, 2007 at 10:13:49PM +0200, Segher Boessenkool wrote:
> >>>>>Well if there is only one memory location involved, then smp_rmb()
> >>>>>isn't
> >>>>>going to really do anything anyway, so it would be incorrect to use
> >>>>>it.
> >>>>
> >>>>rmb() orders *any* two reads; that includes two reads from the same
> >>>>location.
> >>>
> >>>If the two reads are to the same location, all CPUs I am aware of
> >>>will maintain the ordering without need for a memory barrier.
> >>
> >>That's true of course, although there is no real guarantee for that.
> >
> >A CPU that did not provide this property ("cache coherence") would be
> >most emphatically reviled.
>
> That doesn't have anything to do with coherency as far as I can see.
>
> It's just about the order in which a CPU (speculatively) performs the
> loads
> (which isn't necessarily the same as the order in which it executes the
> corresponding instructions, even).
Please check the definition of "cache coherence".
Summary: the CPU is indeed within its rights to execute loads and stores
to a single variable out of order, -but- only if it gets the same result
that it would have obtained by executing them in order. Which means that
any reordering of accesses by a single CPU to a single variable will be
invisible to the software.
> >So we are pretty safe assuming that CPUs
> >will provide it.
>
> Yeah, pretty safe. I just don't like undocumented assumptions :-)
Can't help you there! ;-)
Thanx, Paul
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [EMAIL PROTECTED]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
