Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 18 Jan 2023 21:22:35 +0100
From:   Andrea Parri <parri.andrea@gmail.com>
To:     Alan Stern <stern@rowland.harvard.edu>
Cc:     Jonas Oberhauser <jonas.oberhauser@huaweicloud.com>,
        paulmck@kernel.org, will@kernel.org, peterz@infradead.org,
        boqun.feng@gmail.com, npiggin@gmail.com, dhowells@redhat.com,
        j.alglave@ucl.ac.uk, luc.maranget@inria.fr, akiyks@gmail.com,
        dlustig@nvidia.com, joel@joelfernandes.org, urezki@gmail.com,
        quic_neeraju@quicinc.com, frederic@kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] tools/memory-model: Make ppo a subrelation of po
Message-ID: <Y8hS5iq2+vbcYlkT@andrea>
References: <20230117193159.22816-1-jonas.oberhauser@huaweicloud.com>
 <Y8hN7vs/w8LhMasT@rowland.harvard.edu>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y8hN7vs/w8LhMasT@rowland.harvard.edu>
Precedence: bulk

> It would be nice here to have a separate term for a potentially 
> cross-CPU fence.
> 
> In fact, why don't we make a concerted effort to straighten out the 
> terminology more fully?  Now seems like a good time to do it.
> 
> To begin with, let's be more careful about the difference between an 
> order-inducing object (an event or pair of events) and the relation of 
> being ordered by such an object.  For instance, given:
> 
> 	A: WRITE_ONCE(x, 1);
> 	B: smp_mb();
> 	C: r1 = READ_ONCE(y);
> 
> then B is an order-inducing object (a memory barrier), and (A,C) is a 
> pair of events ordered by that object.  In general, an order is related 
> to an order-inducing object by:
> 
> 	order = po ; [order-inducing object] ; po
> 
> with suitable modifications for things like smp_store_release where 
> one of the events being ordered _is_ the order-inducing event.
> 
> So for example, we could consistently refer to all order-inducing events 
> as either barriers or fences, and all order-reflecting relations as 
> orders.  This would require widespread changes to the .cat file, but I 
> think it would be worthwhile.
> 
> (Treating "barrier" and "fence" as synonyms seems to be too deeply 
> entrenched to try and fight against.)
> 
> Once that is straightened out, we can distinguish between fences or 
> orders that are weak vs. strong.  And then we can divide up strong 
> fences/orders into single-CPU vs. cross-CPU, if we want to.
> 
> How does that sound?

Sounds like a lot of work, renaming and review, for no clear win
to me.  :-)  But hey, if other are into it...

  Andrea