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, 10 Jan 2019 21:20:28 +0100
From:   Peter Zijlstra <peterz@infradead.org>
To:     Florian Westphal <fw@strlen.de>
Cc:     Anatol Pomozov <anatol.pomozov@gmail.com>,
        Dmitry Vyukov <dvyukov@google.com>, paulmck@linux.ibm.com,
        LKML <linux-kernel@vger.kernel.org>
Subject: Re: seqcount usage in xt_replace_table()
Message-ID: <20190110202028.GJ2861@worktop.programming.kicks-ass.net>
References: <CAOMFOmU=t9YbE68YpoQn0Ug0XN1F+0iuJp9J=htCRRrbdaPSVA@mail.gmail.com>
 <20190108223746.shuwx3ro7cgwz7hh@breakpoint.cc>
 <20190110124123.GA21224@hirez.programming.kicks-ass.net>
 <20190110144812.mkbokbj2iyryj6lv@breakpoint.cc>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20190110144812.mkbokbj2iyryj6lv@breakpoint.cc>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Jan 10, 2019 at 03:48:12PM +0100, Florian Westphal wrote:
> Peter Zijlstra <peterz@infradead.org> wrote:
> > 	/*
> > 	 * Ensure contents of newinfo are visible before assigning to
> > 	 * private.
> > 	 */
> > 	smp_wmb();
> > 	table->private = newinfo;
> > 
> > we have:
> > 
> > 	smp_store_release(&table->private, newinfo);
> > 
> > But what store does that second smp_wmb() order against? The comment:
> > 
> > 	/* make sure all cpus see new ->private value */
> > 	smp_wmb();
> > 
> > makes no sense what so ever, no smp_*() barrier can provide such
> > guarantees.
> 
> IIRC I added this at the request of a reviewer of an earlier iteration
> of that patch.
> 
> IIRC the concern was that compiler/hw could re-order
> 
> smb_wmb();
> table->private = newinfo;
> /* wait until all cpus are done with old table */
> 
>  into:
> 
> smb_wmb();
> /* wait until all cpus are done with old table */
> ...
> table->private = newinfo;
> 
> and that (obviously) makes the wait-loop useless.

The thing is, the 'wait for all cpus' thing is pure loads, not stores,
so smp_wmb() is a complete NOP there.

If you want to ensure those loads happen after that store (which does
indeed seem like a sensible thing), you're going to have to use
smp_mb().