Re: [PATCH tip/core/rcu 06/30] rcu: Add WRITE_ONCE to rcu_node ->exp_seq_rq store

[Paul E. McKenney]

On Mon, Feb 17, 2020 at 04:11:35PM -0500, Joel Fernandes wrote:
> 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@xxxxxxxxxx wrote:
> > > 
> > > > From: "Paul E. McKenney" <paulmck@xxxxxxxxxx>
> > > > 
> > > > 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@xxxxxxxxxx>
> > > > ---
> > > >  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().

That does sounds a bit on the required side!  ;-)

							Thanx, Paul

> 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"));
> > >