linux-kernel - Re: [PATCH tip/core/rcu 06/30] rcu: Add WRITE_ONCE to rcu_node ->exp_seq

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: <20200217211135.GA207704@google.com>
Date:   Mon, 17 Feb 2020 16:11:35 -0500
From:   Joel Fernandes <joel@...lfernandes.org>
To:     "Paul E. McKenney" <paulmck@...nel.org>
Cc:     Steven Rostedt <rostedt@...dmis.org>, rcu@...r.kernel.org,
        linux-kernel@...r.kernel.org, kernel-team@...com, mingo@...nel.org,
        jiangshanlai@...il.com, dipankar@...ibm.com,
        akpm@...ux-foundation.org, mathieu.desnoyers@...icios.com,
        josh@...htriplett.org, tglx@...utronix.de, peterz@...radead.org,
        dhowells@...hat.com, edumazet@...gle.com, fweisbec@...il.com,
        oleg@...hat.com, Linus Torvalds <torvalds@...ux-foundation.org>
Subject: Re: [PATCH tip/core/rcu 06/30] rcu: Add WRITE_ONCE to rcu_node
 ->exp_seq_rq store

On Sat, Feb 15, 2020 at 02:58:03AM -0800, Paul E. McKenney wrote:
> On Fri, Feb 14, 2020 at 10:47:43PM -0500, Steven Rostedt wrote:
> > On Fri, 14 Feb 2020 15:55:43 -0800
> > paulmck@...nel.org wrote:
> > 
> > > From: "Paul E. McKenney" <paulmck@...nel.org>
> > > 
> > > The rcu_node structure's ->exp_seq_rq field is read locklessly, so
> > > this commit adds the WRITE_ONCE() to a load in order to provide proper
> > > documentation and READ_ONCE()/WRITE_ONCE() pairing.
> > > 
> > > This data race was reported by KCSAN.  Not appropriate for backporting
> > > due to failure being unlikely.
> > > 
> > > Signed-off-by: Paul E. McKenney <paulmck@...nel.org>
> > > ---
> > >  kernel/rcu/tree_exp.h | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/kernel/rcu/tree_exp.h b/kernel/rcu/tree_exp.h
> > > index d7e0484..85b009e 100644
> > > --- a/kernel/rcu/tree_exp.h
> > > +++ b/kernel/rcu/tree_exp.h
> > > @@ -314,7 +314,7 @@ static bool exp_funnel_lock(unsigned long s)
> > >  				   sync_exp_work_done(s));
> > >  			return true;
> > >  		}
> > > -		rnp->exp_seq_rq = s; /* Followers can wait on us. */
> > > +		WRITE_ONCE(rnp->exp_seq_rq, s); /* Followers can wait on us. */
> > 
> > Didn't Linus say this is basically bogus?
> > 
> > Perhaps just using it as documenting that it's read locklessly, but is
> > it really needed?
> 
> Yes, Linus explicitly stated that WRITE_ONCE() is not required in
> this case, but he also said that he was OK with it being there for
> documentation purposes.

Just to add, PeterZ does approve of WRITE_ONCE() to prevent store-tearing
where applicable.

And I have reproduced Will's example [1] with the arm64 Clang compiler
shipping with the latest Android NDK just now. It does break up stores when
writing zeroes to 64-bit valyes, this is a real problem which WRITE_ONCE()
resolves. And I've verified GCC on arm64 does break up 64-bit immediate value
writes without WRITE_ONCE().

thanks,

 - Joel

[1] https://lore.kernel.org/lkml/20190821103200.kpufwtviqhpbuv2n@willie-the-truck/


> And within RCU, I -do- need it because I absolutely need to see if a
> given patch introduced new KCSAN reports.  So I need it for the same
> reason that I need the build to proceed without warnings.
> 
> Others who are working with less concurrency-intensive code might quite
> reasonably make other choices, of course.  And my setting certain KCSAN
> config options in my own builds doesn't inconvenience them, so we should
> all be happy, right?  :-)
> 
> 							Thanx, Paul
> 
> > -- Steve
> > 
> > 
> > 
> > >  		spin_unlock(&rnp->exp_lock);
> > >  		trace_rcu_exp_funnel_lock(rcu_state.name, rnp->level,
> > >  					  rnp->grplo, rnp->grphi, TPS("nxtlvl"));
> >