Re: [lttng-dev] [PATCH v2 05/12] urcu/uatomic: Add CMM memory model - Paul E. McKenney via lttng-dev

From: "Paul E. McKenney via lttng-dev" <lttng-dev@lists.lttng.org>
To: Olivier Dion <odion@efficios.com>
Cc: Tony Finch <dot@dotat.at>, lttng-dev@lists.lttng.org
Subject: Re: [lttng-dev] [PATCH v2 05/12] urcu/uatomic: Add CMM memory model
Date: Thu, 29 Jun 2023 11:40:08 -0700	[thread overview]
Message-ID: <5a440c51-43d3-4be6-ba5a-215c9d3e960f@paulmck-laptop> (raw)
In-Reply-To: <87bkgyjjsz.fsf@laura>

On Thu, Jun 29, 2023 at 12:49:00PM -0400, Olivier Dion wrote:
> On Wed, 21 Jun 2023, "Paul E. McKenney" <paulmck@kernel.org> wrote:
> > On Wed, Jun 07, 2023 at 02:53:52PM -0400, Olivier Dion wrote:
> >> -#ifdef __URCU_DEREFERENCE_USE_ATOMIC_CONSUME
> >> -# define _rcu_dereference(p) __extension__ ({						\
> >> -				__typeof__(__extension__ ({				\
> >> -					__typeof__(p) __attribute__((unused)) _________p0 = { 0 }; \
> >> -					_________p0;					\
> >> -				})) _________p1;					\
> >> -				__atomic_load(&(p), &_________p1, __ATOMIC_CONSUME);	\
> >
> > There is talk of getting rid of memory_order_consume.  But for the moment,
> > it is what there is.  Another alternative is to use a volatile load,
> > similar to old-style CMM_LOAD_SHARED() or in-kernel READ_ONCE().
> 
> I think we can stick to __ATOMIC_CONSUME for now.  Hopefully getting rid
> of it means it will be an alias for __ATOMIC_ACQUIRE for ever.

That seems emininently reasonable to me!

							Thanx, Paul
_______________________________________________
lttng-dev mailing list
lttng-dev@lists.lttng.org
https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev