Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <6438d903-ab97-48c7-c338-9f0bc2686f94@efficios.com>
 <7A9876BA-C375-42A7-A5C9-FD940D2898D7@joelfernandes.org> <5bd5ee4a-710a-96bc-abe8-772b2e60f478@efficios.com>
 <CAEXW_YRFbsCzT9iPdVfmeZ5qK+2fnVAwSzxbj1EXmU+vepOKdg@mail.gmail.com> <20221220230521.GC26563@lothringen>
In-Reply-To: <20221220230521.GC26563@lothringen>
From:   Joel Fernandes <joel@joelfernandes.org>
Date:   Tue, 20 Dec 2022 18:46:10 -0500
Message-ID: <CAEXW_YRw76oCBevJwFvwRRTyNGJZW2H0Zbg1WNX7D4R3bTvsHw@mail.gmail.com>
Subject: Re: [RFC 0/2] srcu: Remove pre-flip memory barrier
To:     Frederic Weisbecker <frederic@kernel.org>
Cc:     Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
        linux-kernel@vger.kernel.org,
        Josh Triplett <josh@joshtriplett.org>,
        Lai Jiangshan <jiangshanlai@gmail.com>,
        "Paul E. McKenney" <paulmck@kernel.org>, rcu@vger.kernel.org,
        Steven Rostedt <rostedt@goodmis.org>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, Dec 20, 2022 at 6:05 PM Frederic Weisbecker <frederic@kernel.org> w=
rote:
>
> On Tue, Dec 20, 2022 at 07:06:57PM +0000, Joel Fernandes wrote:
> > On Tue, Dec 20, 2022 at 7:01 PM Mathieu Desnoyers
> > <mathieu.desnoyers@efficios.com> wrote:
> > >
> > > On 2022-12-20 13:29, Joel Fernandes wrote:
> > > >
> > >
> > > > I do want to finish my memory barrier studies of SRCU over the holi=
days since I have been deep in the hole with that already. Back to the post=
 flip memory barrier here since I think now even that might not be needed=
=E2=80=A6
> > >
> > > I strongly suspect the memory barrier after flip is useless for the s=
ame
> > > reasons I mentioned explaining why the barrier before the flip is use=
less.
> > >
> > > However, we need to double-check that we have memory barriers at the
> > > beginning and end of synchronize_srcu, and between load of "unlock"
> > > counters and load of "lock" counters.
> > >
> > > Where is the barrier at the beginning of synchronize_srcu ?
> >
> > I believe we don't need another memory barrier at the beginning of
> > synchronize_srcu() (but this part of my SRCU study is still pending
> > ;)) . The grace period guarantee (read-side critical sections don't
> > span the GP) is already enforced by the memory barrier between
> > scanning for all unlocks, and scanning for all locks (Paired with
> > corresponding memory barriers on the read-side).
> >
> > Accordingly, before we scan all locks and match lock =3D=3D unlock, the=
re
> > is an smp_mb(). Why is that not sufficient?
>
> That's not enough, you still need a barrier between the updater's pre-GP
> accesses and the scans, so that post-GP read side sees the updater's pre-=
GP
> accesses:
>
>
>             UPDATER                        READER
>             -------                        ------
>             WRITE A                        WRITE srcu_read_lock
>             smp_mb() //rcu_seq_snap()      smp_mb()
>             READ srcu_read_lock //scans    READ A

But see the comments also in srcu_readers_active_idx_check()

* Needs to be a smp_mb() as the read side may
* contain a read from a variable that is written to before the
* synchronize_srcu() in the write side

So that appears to be already covered. Or is your point that the scans
are not happening on the same CPU as the pre-GP writer, as scans are
happening from workqueue ?

Perhaps that comment misled me.

Confused,

 - Joel