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:   Tue, 19 Nov 2019 16:58:26 +0100
From:   Oleg Nesterov <oleg@redhat.com>
To:     Waiman Long <longman@redhat.com>
Cc:     Peter Zijlstra <peterz@infradead.org>, mingo@kernel.org,
        will@kernel.org, tglx@linutronix.de, linux-kernel@vger.kernel.org,
        bigeasy@linutronix.de, juri.lelli@redhat.com, williams@redhat.com,
        bristot@redhat.com, dave@stgolabs.net, jack@suse.com
Subject: Re: [PATCH 5/5] locking/percpu-rwsem: Remove the embedded rwsem
Message-ID: <20191119155826.GA4739@redhat.com>
References: <20191113102115.116470462@infradead.org>
 <20191113102855.925208237@infradead.org>
 <ee75fc38-c3c8-3f9e-13ba-5c8312d61325@redhat.com>
MIME-Version: 1.0
In-Reply-To: <ee75fc38-c3c8-3f9e-13ba-5c8312d61325@redhat.com>
User-Agent: Mutt/1.5.24 (2015-08-30)
Content-Type: text/plain; charset=WINDOWS-1252
Content-Transfer-Encoding: quoted-printable
Content-Disposition: inline
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 11/19, Waiman Long wrote:
>
> On 11/13/19 5:21 AM, Peter Zijlstra wrote:
> > +static int percpu_rwsem_wake_function(struct wait_queue_entry *wq_entr=
y,
> > +=09=09=09=09      unsigned int mode, int wake_flags,
> > +=09=09=09=09      void *key)
> > +{
> > +=09struct task_struct *p =3D get_task_struct(wq_entry->private);
> > +=09bool reader =3D wq_entry->flags & WQ_FLAG_CUSTOM;
> > +=09struct percpu_rw_semaphore *sem =3D key;
> > +
> > +=09/* concurrent against percpu_down_write(), can get stolen */
> > +=09if (!__percpu_rwsem_trylock(sem, reader))
> > +=09=09return 1;
> > +
> > +=09list_del_init(&wq_entry->entry);
> > +=09smp_store_release(&wq_entry->private, NULL);
> > +
> > +=09wake_up_process(p);
> > +=09put_task_struct(p);
> > +
> > +=09return !reader; /* wake 'all' readers and 1 writer */
> > +}
> > +
>
> If I read the function correctly, you are setting the WQ_FLAG_EXCLUSIVE
> for both readers and writers and __wake_up() is called with an exclusive
> count of one. So only one reader or writer is woken up each time.

This depends on what percpu_rwsem_wake_function() returns. If it returns 1,
__wake_up_common() stops, exactly because all waiters have WQ_FLAG_EXCLUSIV=
E.

> However, the comment above said we wake 'all' readers and 1 writer. That
> doesn't match the actual code, IMO.

Well, "'all' readers" probably means "all readers before writer",

> To match the comments, you should
> have set WQ_FLAG_EXCLUSIVE flag only on writer. In this case, you
> probably don't need WQ_FLAG_CUSTOM to differentiate between readers and
> writers.

See above...

note also the

=09if (!__percpu_rwsem_trylock(sem, reader))
=09=09return 1;

at the start of percpu_rwsem_wake_function(). We want to stop wake_up_commo=
n()
as soon as percpu_rwsem_trylock() fails. Because we know that if it fails o=
nce
it can't succeed later. Although iiuc this can only happen if another (new)
writer races with __wake_up(&sem->waiters).


I guess WQ_FLAG_CUSTOM can be avoided, percpu_rwsem_wait() could do

=09if (read)
=09=09__add_wait_queue_entry_tail(...);
=09else {
=09=09wq_entry.flags |=3D WQ_FLAG_EXCLUSIVE;
=09=09__add_wait_queue(...);
=09}

but this is "unfair".

Oleg.