On Mon, Jul 29, 2019 at 01:50:44PM +0800, Boqun Feng wrote:
> On Sun, Jul 28, 2019 at 11:35:44AM -0400, Joel Fernandes wrote:
> [...]
> > > > > > +load of y (rfe link), P2's smp_store_release() ensures that P2's 
> > > > > > load
> > > > > > +of y executes before P2's store to z (second fence), which implies 
> > > > > > that
> > > > > > +that stores to x and y propagate to P2 before the 
> > > > > > smp_store_release(), which
> > > > > > +means that P2's smp_store_release() will propagate stores to x and 
> > > > > > y to all
> > > > > > +CPUs before the store to z propagates (A-cumulative property of 
> > > > > > this fence).
> > > > > > +Finally P0's load of z executes after P2's store to z has 
> > > > > > propagated to
> > > > > > +P0 (rfe link).
> > > > > 
> > > > > Again, a better change would be simply to replace the two instances of
> > > > > "fence" in the original text with "cumul-fence".
> > > > 
> > > > Ok that's fine. But I still feel the rfe is not a part of the 
> > > > cumul-fence.
> > > > The fences have nothing to do with the rfe. Or, I am missing something 
> > > > quite
> > > > badly.
> > > > 
> > > > Boqun, did you understand what Alan is saying?
> > > > 
> > > 
> > > I think 'cumul-fence' that Alan mentioned is not a fence, but a
> > > relation, which could be the result of combining a rfe relation and a
> > > A-cumulative fence relation. Please see the section "PROPAGATION ORDER
> > > RELATION: cumul-fence" or the definition of cumul-fence in
> > > linux-kernel.cat.
> > > 
> > > Did I get you right, Alan? If so, your suggestion is indeed a better
> > > change.
> > 
> > To be frank, I don't think it is better if that's what Alan meant. It is
> > better to be explicit about the ->rfe so that the reader walking through the
> > example can clearly see the ordering and make sense of it.
> > 
> > Just saying 'cumul-fence' and then hoping the reader sees the light is quite
> > a big assumption and makes the document less readable.
> > 
> 
> After a bit more rereading of the document, I still think Alan's way is
> better ;-)

I think I finally understood. What I was missing was this definition of
cumul-fence involves an rf relation (with writes being possibly on different
CPUs).

E ->cumul-fence F
        F is a release fence and some X comes before F in program order,
        where either X = E or else E ->rf X; or

So I think what Alan meant is there is a cumul-fence between y=1 and z=1
because fo the ->rfe of y. Thus making it not necessary to mention the rfe.

Labeling E and F in the example...

        P1()
        {
                WRITE_ONCE(x, 2);
                smp_wmb();
                WRITE_ONCE(y, 1);               // This is E
        }

        P2()
        {
                int r2;

                r2 = READ_ONCE(y);              // This is X
                smp_store_release(&z, 1);       // This is F
        }

Here, E ->rf X ->release-fence -> F
implies,
      E ->cumul-fence F

Considering that, I agree with Alan's suggestion.

> 
>       The formal definition of the prop relation involves a coe or
>       fre link, followed by an arbitrary number of cumul-fence links,
>       ending with an rfe link.
> 
> , so using "cumul-fence" actually matches the definition of ->prop.
> 
> For the ease of readers, I can think of two ways:
> 
> 1.    Add a backwards reference to cumul-fence section here, right
>       before using its name.

Instead of that, a reference to the fact that the rfe causes a cumul-fence
between y=1 and z=1 may be helpful. No need backward reference IMO.

> 2.    Use "->cumul-fence" notation rather than "cumul-fence" here,
>       i.e. add an arrow "->" before the name to call it out that name
>       "cumul-fence" here stands for links/relations rather than a
>       fence/barrier. Maybe it's better to convert all references to 
>       links/relations to the "->" notations in the whole doc.

No, it is a fence that causes the relation in this example.

I don't think there is a distinction between ->cumul-fence and cumul-fence at
least for this example. The smp_store_release() is a FENCE which in this
example is really a cumul-fence between y=1 and z=1 because the release fence
affects propogation order of y and z on all CPUs. For any given CPU, y
propagates to that CPU before z does, even though y and z executed on
different CPUs.

I think what you're talking about is some other definition of cumul-fence
that is not mentioned in the formal definitions. May be you can update the
document with such definition then? AFAIU, cumul-fence is always a fence.

thanks,

 - Joel

Reply via email to