linux-kernel - Re: [PATCH] rcu/nocb: Check rdp_gp->nocb_timer in __call_rcu_nocb

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <2490b90d-ebce-42bf-8a83-1314e8272df8@paulmck-laptop>
Date: Fri, 19 Jan 2024 16:29:52 -0800
From: "Paul E. McKenney" <paulmck@...nel.org>
To: Frederic Weisbecker <frederic@...nel.org>
Cc: Zqiang <qiang.zhang1211@...il.com>, quic_neeraju@...cinc.com,
	joel@...lfernandes.org, rcu@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: Re: [PATCH] rcu/nocb: Check rdp_gp->nocb_timer in
 __call_rcu_nocb_wake()

On Fri, Jan 19, 2024 at 10:47:23PM +0100, Frederic Weisbecker wrote:
> Le Thu, Jan 18, 2024 at 06:51:57AM -0800, Paul E. McKenney a écrit :
> > On Wed, Jan 17, 2024 at 01:07:25PM +0100, Frederic Weisbecker wrote:
> > > Le Wed, Jan 17, 2024 at 06:26:16PM +0800, Zqiang a écrit :
> > > > Currently, only rdp_gp->nocb_timer is used, for nocb_timer of
> > > > no-rdp_gp structure, the timer_pending() is always return false,
> > > > this commit therefore need to check rdp_gp->nocb_timer in
> > > > __call_rcu_nocb_wake().
> > > > 
> > > > Signed-off-by: Zqiang <qiang.zhang1211@...il.com>
> > > > ---
> > > >  kernel/rcu/tree_nocb.h | 3 ++-
> > > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
> > > > index 54971afc3a9b..3f85577bddd4 100644
> > > > --- a/kernel/rcu/tree_nocb.h
> > > > +++ b/kernel/rcu/tree_nocb.h
> > > > @@ -564,6 +564,7 @@ static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_alldone,
> > > >  	long lazy_len;
> > > >  	long len;
> > > >  	struct task_struct *t;
> > > > +	struct rcu_data *rdp_gp = rdp->nocb_gp_rdp;
> > > >  
> > > >  	// If we are being polled or there is no kthread, just leave.
> > > >  	t = READ_ONCE(rdp->nocb_gp_kthread);
> > > > @@ -608,7 +609,7 @@ static void __call_rcu_nocb_wake(struct rcu_data *rdp, bool was_alldone,
> > > >  		smp_mb(); /* Enqueue before timer_pending(). */
> > > >  		if ((rdp->nocb_cb_sleep ||
> > > >  		     !rcu_segcblist_ready_cbs(&rdp->cblist)) &&
> > > > -		    !timer_pending(&rdp->nocb_timer)) {
> > > > +		    !timer_pending(&rdp_gp->nocb_timer)) {
> > > 
> > > Hehe, good eyes ;-)
> > > 
> > > I had that change in mind but while checking that area further I actually
> > > wondered what is the actual purpose of this RCU_NOCB_WAKE_FORCE thing. If
> > > we reach that place, it means that the nocb_gp kthread should be awaken
> > > already (or the timer pending), so what does a force wake up solve in that
> > > case?
> > > 
> > > Paul, any recollection of that?
> > 
> > Huh.  We never actually do RCU_NOCB_WAKE_FORCE in v6.7, if I followed
> > all the code paths correctly.
> > 
> > Historically, I have been worried about lost wakeups.  Also, there
> > used to be code paths in which a wakeup was not needed, for example,
> > because we knew that the ending of the current grace period would take
> > care of things.  Unless there was some huge pile of callbacks, in which
> > case an immediate wakeup could avoid falling behind a callback flood.
> 
> Ok then looks like it's time for me to add RCU_NOCB_WAKE_FORCE removal in
> my TODO list...unless Zqiang would like to give it a try? :-)
> 
> > Given that rcutorture does test callback flooding, we appear to be OK,
> > but maybe it is time to crank up the flooding more.
> > 
> > On the other hand, I have started seeing the (very) occasional OOM
> > on TREE03.
> > (In addition to those that show up from time to time on the
> > single-CPU TREE09 scenario.)
> 
> Interesting, are those recent? Bisectable?

Bisection in progress, got it down to 10 commits.  Yet again about
ten hours per step on 20 systems...

Though maybe I should have put more time into making it happen faster.
Except that I was on travel, so I doubt that I would have made all that
much progress.  ;-)

							Thanx, Paul

							Thanx, Paul