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, 17 Jan 2019 14:43:54 -0500 (EST)
From:   Alan Stern <stern@rowland.harvard.edu>
To:     Andrea Parri <andrea.parri@amarulasolutions.com>
cc:     LKMM Maintainers -- Akira Yokosawa <akiyks@gmail.com>,
        Boqun Feng <boqun.feng@gmail.com>,
        Daniel Lustig <dlustig@nvidia.com>,
        David Howells <dhowells@redhat.com>,
        Jade Alglave <j.alglave@ucl.ac.uk>,
        Luc Maranget <luc.maranget@inria.fr>,
        Nicholas Piggin <npiggin@gmail.com>,
        "Paul E. McKenney" <paulmck@linux.ibm.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Will Deacon <will.deacon@arm.com>,
        Dmitry Vyukov <dvyukov@google.com>,
        Nick Desaulniers <ndesaulniers@google.com>,
        <linux-kernel@vger.kernel.org>
Subject: Re: Plain accesses and data races in the Linux Kernel Memory Model
In-Reply-To: <20190116213658.GA3984@andrea>
Message-ID: <Pine.LNX.4.44L0.1901171121360.1207-100000@iolanthe.rowland.org>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, 16 Jan 2019, Andrea Parri wrote:

> Can the compiler (maybe, it does?) transform, at the C or at the "asm"
> level, LB1's P0 in LB2's P0 (LB1 and LB2 are reported below)?
> 
> C LB1
> 
> {
> 	int *x = &a;
> }
> 
> P0(int **x, int *y)
> {
> 	int *r0;
> 
> 	r0 = rcu_dereference(*x);
> 	*r0 = 0;
> 	smp_wmb();
> 	WRITE_ONCE(*y, 1);
> }
> 
> P1(int **x, int *y, int *b)
> {
> 	int r0;
> 
> 	r0 = READ_ONCE(*y);
> 	rcu_assign_pointer(*x, b);
> }
> 
> exists (0:r0=b /\ 1:r0=1)
> 
> 
> C LB2
> 
> {
> 	int *x = &a;
> }
> 
> P0(int **x, int *y)
> {
> 	int *r0;
> 
> 	r0 = rcu_dereference(*x);
> 	if (*r0)
> 		*r0 = 0;
> 	smp_wmb();
> 	WRITE_ONCE(*y, 1);
> }
> 
> P1(int **x, int *y, int *b)
> {
> 	int r0;
> 
> 	r0 = READ_ONCE(*y);
> 	rcu_assign_pointer(*x, b);
> }
> 
> exists (0:r0=b /\ 1:r0=1)
> 
> LB1 and LB2 are data-race free, according to the patch; LB1's "exists"
> clause is not satisfiable, while LB2's "exists" clause is satisfiable.

Umm.  Transforming

	*r0 = 0;

to

	if (*r0 != 0)
		*r0 = 0;

wouldn't work on Alpha if r0 was assigned from a plain read with no
memory barrier between.  But when r0 is assigned from an
rcu_dereference call, or if there's no indirection (as in "if (a != 0)
a = 0;"), the compiler is indeed allowed to perform this
transformation.

This means my definition of preserved writes was wrong; a write we 
thought had to be preserved could instead be transformed into a read.

This objection throws a serious monkey wrench into my approach.  For
one thing, it implies that (as in the example) we can't expect
smp_wmb() always to order plain writes.  For another, it means we have
to assume a lot more writes need not be preserved.

I don't know.  This may doom the effort to formalize dependencies to
plain accesses.  Or at least, those other than address dependencies
from marked reads.

Alan