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:   Wed, 27 Mar 2019 08:04:02 -0700
From:   "Paul E. McKenney" <paulmck@linux.ibm.com>
To:     NeilBrown <neilb@suse.com>
Cc:     Herbert Xu <herbert@gondor.apana.org.au>,
        Thomas Graf <tgraf@suug.ch>, netdev@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH 3/4] rhashtable: use bit_spin_locks to protect hash
 bucket.
Reply-To: paulmck@linux.ibm.com
References: <155349021177.1111.15681654355431465791.stgit@noble.brown>
 <155349033961.1111.18247269615646768227.stgit@noble.brown>
 <20190326050320.gwk3tgtqwl5csivt@gondor.apana.org.au>
 <874l7p463d.fsf@notabene.neil.brown.name>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <874l7p463d.fsf@notabene.neil.brown.name>
User-Agent: Mutt/1.5.21 (2010-09-15)
Message-Id: <20190327150402.GX4102@linux.ibm.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, Mar 27, 2019 at 09:35:18AM +1100, NeilBrown wrote:
> On Tue, Mar 26 2019, Herbert Xu wrote:
> 
> > On Mon, Mar 25, 2019 at 04:05:39PM +1100, NeilBrown wrote:
> >>
> >> + * Sometimes we unlock a bucket by writing a new pointer there.  In that
> >> + * case we don't need to unlock, but we do need to reset state such as
> >> + * local_bh. For that we have rht_unlocked().  This doesn't include
> >> + * the memory barrier that bit_spin_unlock() provides, but rcu_assign_pointer()
> >> + * will have provided that.
> >
> > Hmm, are you sure that's enough? IIRC rcu_assign_pointer only
> > provides a write barrier compared to the more complete (but one-way)
> > barrier that a spin-lock provides.
> >
> 
> The bit_spin_unlock(), which I am avoiding as unnecessary, would have
> provided release semantics.
> i.e. any write by this CPU that happened before the releasing write
> will be visible to other CPUs before (or when) they see the result of
> the releasing write.
> This is (as I understand it) exactly that rcu_assign_pointer() promises
> - even before acquire semantics were added as Paul just reported.
> 
> So yes, I am sure (surer now that I've walked through it carefully).

But why not construct a litmus test and apply tools/memory-model?  ;-)

							Thanx, Paul