Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Thu, 2 Nov 2017 16:28:45 -0400 (EDT)
From:   Alan Stern <stern@rowland.harvard.edu>
To:     Andrea Parri <parri.andrea@gmail.com>
cc:     Peter Zijlstra <peterz@infradead.org>,
        "Reshetova, Elena" <elena.reshetova@intel.com>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "gregkh@linuxfoundation.org" <gregkh@linuxfoundation.org>,
        "keescook@chromium.org" <keescook@chromium.org>,
        "tglx@linutronix.de" <tglx@linutronix.de>,
        "mingo@redhat.com" <mingo@redhat.com>,
        "ishkamiel@gmail.com" <ishkamiel@gmail.com>,
        Will Deacon <will.deacon@arm.com>,
        Paul McKenney <paulmck@linux.vnet.ibm.com>,
        <boqun.feng@gmail.com>, <dhowells@redhat.com>,
        <david@fromorbit.com>
Subject: Re: [PATCH] refcount: provide same memory ordering guarantees as in
 atomic_t
In-Reply-To: <20171102174504.GA19833@andrea>
Message-ID: <Pine.LNX.4.44L0.1711021622121.1277-100000@iolanthe.rowland.org>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, 2 Nov 2017, Andrea Parri wrote:

> > This is forbidden.  It would remain forbidden even if the smp_mb in P1 
> > were replaced by a similar release/acquire pair for the same memory 
> > location.
> 
> Hopefully, the LKMM does not agree with this assessment... ;-)

No, it doesn't.

> Here's a two-threads example showing that "(w)mb is _not_ rfi-rel-acq":
> 
> C rfi-rel-acq-is-not-mb
> 
> {}
> 
> P0(int *x, int *y, int *a)
> {
> 	WRITE_ONCE(*x, 1);
> 	smp_store_release(a, 1);
> 	r1 = smp_load_acquire(a);
> 	WRITE_ONCE(*y, 1);
> }
> 
> P1(int *x, int *y)
> {
> 	int r0;
> 	int r1;
> 
> 	r0 = READ_ONCE(*y);
> 	smp_rmb();
> 	r1 = READ_ONCE(*x);
> }
> 
> exists (1:r0=1 /\ 1:r1=0)

Right.  There is a happens-before edge between the two WRITE_ONCE calls
in P0 but no cumul-fence edge, and therefore the test is allowed.  
Here is an example where a happens-before edge suffices to provide
ordering:

P0(int *x, int *y)
{
	WRITE_ONCE(*x, 1);
	smp_wmb();
	WRITE_ONCE(*y, 1);
}

P1(int *x, int *y, int *a)
{
	int rx, ry, ra;

	ry = READ_ONCE(*y);
	smp_store_release(a, 1);
	ra = smp_load_acquire(a);
	rx = READ_ONCE(*x);
}

exists (1:rx=0 /\ 1:ry=1)

This test is forbidden, but it would be allowed if the release and 
acquire accessed different locations.

Alan Stern


From 1582987160255460536@xxx Thu Nov 02 20:23:17 +0000 2017
X-GM-THRID: 1582046402032606032
X-Gmail-Labels: Inbox,Category Forums,HistoricalUnread