Re: [PATCH v3 2/3] locking: Clarify requirements for smp_mb__after_spinlock()

[Andrea Parri <andrea.parri@amarulasolutions.com>]

On Tue, Jul 03, 2018 at 08:39:10AM -0700, Paul E. McKenney wrote:

[...]

> > + * smp_mb__after_spinlock() provides the equivalent of a full memory barrier
> > + * between program-order earlier lock acquisitions and program-order later
> 
> Not just the earlier lock acquisition, but also all program-order earlier
> memory accesses, correct?

I understand: "but also all program-order earlier memory accesses program-order
before that lock acquisition(s) ...".  Yes, but:

  - I considered this as implied by the above (L ->mb M2 and M1 ->po L implies
    M1 ->mb M2, where M1, M2 are memory accesses and L is a lock acquisition);

  - my prose abilities are limited ;-), and I was/am unable to come up with an
    (to me) acceptable or readable enough way to make it explicit; some ideas?


> > + *	  WRITE_ONCE(X, 1);		WRITE_ONCE(Y, 1);
> > + *	  spin_lock(S);			smp_mb();
> > + *	  smp_mb__after_spinlock();	r1 = READ_ONCE(X);
> > + *	  r0 = READ_ONCE(Y);
> > + *	  spin_unlock(S);
> 
> Should we say that this is an instance of the SB pattern?  (Am OK either
> way, just asking the question.)

I don't think we *should* ;-),  but I'm also OK either way.

  Andrea