Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 8 Sep 2021 13:44:11 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     stern@rowland.harvard.edu, alexander.shishkin@linux.intel.com,
        hpa@zytor.com, parri.andrea@gmail.com, mingo@kernel.org,
        paulmck@kernel.org, vincent.weaver@maine.edu, tglx@linutronix.de,
        jolsa@redhat.com, acme@redhat.com, torvalds@linux-foundation.org,
        linux-kernel@vger.kernel.org, eranian@google.com, will@kernel.org
Cc:     linux-tip-commits@vger.kernel.org
Subject: Re: [tip:locking/core] tools/memory-model: Add extra ordering for
 locks and remove it for ordinary release/acquire
Message-ID: <YTiiC1mxzHyUJ47F@hirez.programming.kicks-ass.net>
References: <20180926182920.27644-2-paulmck@linux.ibm.com>
 <tip-6e89e831a90172bc3d34ecbba52af5b9c4a447d1@git.kernel.org>
 <YTiXyiA92dM9726M@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YTiXyiA92dM9726M@hirez.programming.kicks-ass.net>
Precedence: bulk

On Wed, Sep 08, 2021 at 01:00:26PM +0200, Peter Zijlstra wrote:
> On Tue, Oct 02, 2018 at 03:11:10AM -0700, tip-bot for Alan Stern wrote:
> > Commit-ID:  6e89e831a90172bc3d34ecbba52af5b9c4a447d1
> > Gitweb:     https://git.kernel.org/tip/6e89e831a90172bc3d34ecbba52af5b9c4a447d1
> > Author:     Alan Stern <stern@rowland.harvard.edu>
> > AuthorDate: Wed, 26 Sep 2018 11:29:17 -0700
> > Committer:  Ingo Molnar <mingo@kernel.org>
> > CommitDate: Tue, 2 Oct 2018 10:28:01 +0200
> > 
> > tools/memory-model: Add extra ordering for locks and remove it for ordinary release/acquire
> > 
> > More than one kernel developer has expressed the opinion that the LKMM
> > should enforce ordering of writes by locking.  In other words, given
> > the following code:
> > 
> > 	WRITE_ONCE(x, 1);
> > 	spin_unlock(&s):
> > 	spin_lock(&s);
> > 	WRITE_ONCE(y, 1);
> > 
> > the stores to x and y should be propagated in order to all other CPUs,
> > even though those other CPUs might not access the lock s.  In terms of
> > the memory model, this means expanding the cumul-fence relation.
> 
> Let me revive this old thread... recently we ran into the case:
> 
> 	WRITE_ONCE(x, 1);
> 	spin_unlock(&s):
> 	spin_lock(&r);
> 	WRITE_ONCE(y, 1);
> 
> which is distinct from the original in that UNLOCK and LOCK are not on
> the same variable.
> 
> I'm arguing this should still be RCtso by reason of:
> 
>   spin_lock() requires an atomic-acquire which:
> 
>     TSO-arch)		implies smp_mb()
>     ARM64)		is RCsc for any stlr/ldar
>     Power)		LWSYNC
>     Risc-V)		fence r , rw
>     *)			explicitly has smp_mb()
> 
> 
> However, Boqun points out that the memory model disagrees, per:
> 
>   https://lkml.kernel.org/r/YTI2UjKy+C7LeIf+@boqun-archlinux
> 
> Is this an error/oversight of the memory model, or did I miss a subtlety
> somewhere?

Hmm.. that argument isn't strong enough for Risc-V if I read that FENCE
thing right. That's just R->RW ordering, which doesn't constrain the
first WRITE_ONCE().

However, that spin_unlock has "fence rw, w" with a subsequent write. So
the whole thing then becomes something like:


	WRITE_ONCE(x, 1);
	FENCE RW, W
	WRITE_ONCE(s.lock, 0);
	AMOSWAP %0, 1, r.lock
	FENCE R, WR
	WRITE_ONCE(y, 1);


Which I think is still sufficient, irrespective of the whole s!=r thing.