Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210617142828.346111-1-sxwjean@me.com> <20210617142828.346111-4-sxwjean@me.com>
 <YMtmbjw5EGdXRdD8@boqun-archlinux>
In-Reply-To: <YMtmbjw5EGdXRdD8@boqun-archlinux>
From:   Xiongwei Song <sxwjean@gmail.com>
Date:   Fri, 18 Jun 2021 09:54:59 +0800
Message-ID: <CAEVVKH8TYoQjZSJr7DjecgEq-EAZEqWqC=EC_pFPzti3VLh6_g@mail.gmail.com>
Subject: Re: [PATCH 3/3] locking/lockdep: print possible warning after
 counting deps
To:     Boqun Feng <boqun.feng@gmail.com>
Cc:     Xiongwei Song <sxwjean@me.com>, peterz@infradead.org,
        mingo@redhat.com, will@kernel.org, longman@redhat.com,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Thu, Jun 17, 2021 at 11:13 PM Boqun Feng <boqun.feng@gmail.com> wrote:
>
> Hi,
>
> On Thu, Jun 17, 2021 at 10:28:28PM +0800, Xiongwei Song wrote:
> > From: Xiongwei Song <sxwjean@gmail.com>
> >
> > The graph walk might hit error when counting dependencies. Once the
> > return value is negative, print a warning to reminder users.
> >
>
> Thanks for the improvement, but please see below:
>
> > Suggested-by: Waiman Long <longman@redhat.com>
> > Signed-off-by: Xiongwei Song <sxwjean@gmail.com>
> > ---
> >  kernel/locking/lockdep.c | 12 ++++++++++--
> >  1 file changed, 10 insertions(+), 2 deletions(-)
> >
> > diff --git a/kernel/locking/lockdep.c b/kernel/locking/lockdep.c
> > index cb94097014d8..cfe0f4374594 100644
> > --- a/kernel/locking/lockdep.c
> > +++ b/kernel/locking/lockdep.c
> > @@ -2028,8 +2028,12 @@ static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
> >  {
> >       unsigned long  count = 0;
> >       struct lock_list *target_entry;
> > +     enum bfs_result ret;
> > +
> > +     ret = __bfs_forwards(this, (void *)&count, noop_count, NULL, &target_entry);
> >
> > -     __bfs_forwards(this, (void *)&count, noop_count, NULL, &target_entry);
> > +     if (bfs_error(ret))
> > +             print_bfs_bug(ret);
>
> Here print_bfs_bug() will eventually call debug_locks_off_graph_unlock()
> to release the graph lock, and the caller (lockdep_count_fowards_deps())
> will also call graph_unlock() afterwards, and that means we unlock
> *twice* if a BFS error happens... although in that case, lockdep should
> stop working so messing up with the graph lock may not hurt anything,
> but still, I don't think we want to do that.
>
> So probably you can open-code __lockdep_count_forward_deps() into
> lockdep_count_forwards_deps(), and call print_bfs_bug() or
> graph_unlock() accordingly. The body of __lockdep_count_forward_deps()
> is really small, so I think it's OK to open-code it into its caller.

Thank you so much for the detailed comments. Let me improve and update
the patch.

Regards,
Xiongwei

>
> Regards,
> Boqun
>
> >
> >       return count;
> >  }
> > @@ -2053,8 +2057,12 @@ static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
> >  {
> >       unsigned long  count = 0;
> >       struct lock_list *target_entry;
> > +     enum bfs_result ret;
> > +
> > +     ret = __bfs_backwards(this, (void *)&count, noop_count, NULL, &target_entry);
> >
> > -     __bfs_backwards(this, (void *)&count, noop_count, NULL, &target_entry);
> > +     if (bfs_error(ret))
> > +             print_bfs_bug(ret);
> >
> >       return count;
> >  }
> > --
> > 2.30.2
> >