Re: [RFC PATCH v2] memory-barriers: remove smp_mb__after_unlock_lock()

From: Will Deacon
Date: Mon Aug 17 2015 - 04:58:11 EST


On Mon, Aug 17, 2015 at 07:15:01AM +0100, Paul E. McKenney wrote:
> On Mon, Aug 17, 2015 at 02:06:07PM +1000, Michael Ellerman wrote:
> > On Wed, 2015-08-12 at 08:43 -0700, Paul E. McKenney wrote:
> > > On Wed, Aug 12, 2015 at 02:44:15PM +0100, Will Deacon wrote:
> > > > On Fri, Jul 24, 2015 at 04:30:46PM +0100, Paul E. McKenney wrote:
> > > > > On Fri, Jul 24, 2015 at 12:31:01PM +0100, Will Deacon wrote:
> > > > > > On Wed, Jul 15, 2015 at 02:12:21PM +0100, Paul E. McKenney wrote:
> > > > > > > > > commit 695c05d4b9666c50b40a1c022678b5f6e2e3e771
> > > > > > > > > Author: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> > > > > > > > > Date: Tue Jul 14 18:35:23 2015 -0700
> > > > > > > > >
> > > > > > > > > rcu,locking: Privatize smp_mb__after_unlock_lock()
> > > > > > > > >
> > > > > > > > > RCU is the only thing that uses smp_mb__after_unlock_lock(), and is
> > > > > > > > > likely the only thing that ever will use it, so this commit makes this
> > > > > > > > > macro private to RCU.
> > > > > > > > >
> > > > > > > > > Signed-off-by: Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
> > > > > > > > > Cc: Will Deacon <will.deacon@xxxxxxx>
> > > > > > > > > Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> > > > > > > > > Cc: Benjamin Herrenschmidt <benh@xxxxxxxxxxxxxxxxxxx>
> > > > > > > > > Cc: "linux-arch@xxxxxxxxxxxxxxx" <linux-arch@xxxxxxxxxxxxxxx>
> > > > > >
> > > > > > Are you planning to queue this somewhere? I think it makes sense regardless
> > > > > > of whether we change PowerPc or not and ideally it would be merged around
> > > > > > the same time as my relaxed atomics series.
> > > > >
> > > > > I have is in -rcu. By default, I will push it to the 4.4 merge window.
> > > > > Please let me know if you need it sooner.
> > > >
> > > > The generic relaxed atomics are now queued in -tip, so it would be really
> > > > good to see this Documentation update land in 4.3 if at all possible. I
> > > > appreciate it's late in the cycle, but it's always worth asking.
> > >
> > > Can't hurt to give it a try. I have set -rcu's rcu/next branch to this
> > > commit, and if it passes a few day's worth of testing, I will see what
> > > Ingo has to say about a pull request.
> > >
> > > This commit also privatizes smp_mb__after_unlock_lock() as well as
> > > updating documentation. Looks like we need to strengthen powerpc's
> > > locking primitives, then get rid of smp_mb__after_unlock_lock() entirely.
> > > Or did that already happen and I just missed it?
> >
> > No it didn't.
> >
> > I thought the end result of this thread was that we didn't *need* to change the
> > powerpc lock semantics? Or did I read it wrong?
> >
> > ie. the docs now say that RELEASE+ACQUIRE is not a full barrier, which is
> > consistent with our current implementation.
>
> That change happened about 1.5 years ago, and I thought that the
> current discussion was about reversing it, based in part on the
> recent powerpc benchmarks of locking primitives with and without the
> sync instruction. But regardless, I clearly cannot remove either the
> smp_mb__after_unlock_lock() or the powerpc definition of it to be smp_mb()
> if powerpc unlock/lock is not strengthened.

Yup. Peter and I would really like to get rid of smp_mb__after_unlock_lock
entirely, which would mean strengthening the ppc spinlocks. Moving the
barrier primitive into RCU is a good step to prevent more widespread usage
of the barrier, but we'd really like to go further if the performance impact
is deemed acceptable (which is what this thread is about).

Will
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/