Re: [PATCH v2 net-next] net: generalize skb freeing deferral to per-cpu lists

[Paolo Abeni <pabeni@redhat.com>]

On Tue, 2022-04-26 at 06:11 -0700, Eric Dumazet wrote:
> On Tue, Apr 26, 2022 at 12:38 AM Paolo Abeni <pabeni@redhat.com> wrote:
> > On Fri, 2022-04-22 at 13:12 -0700, Eric Dumazet wrote:
> > [...]
> > > @@ -6571,6 +6577,28 @@ static int napi_threaded_poll(void *data)
> > >       return 0;
> > >  }
> > > 
> > > +static void skb_defer_free_flush(struct softnet_data *sd)
> > > +{
> > > +     struct sk_buff *skb, *next;
> > > +     unsigned long flags;
> > > +
> > > +     /* Paired with WRITE_ONCE() in skb_attempt_defer_free() */
> > > +     if (!READ_ONCE(sd->defer_list))
> > > +             return;
> > > +
> > > +     spin_lock_irqsave(&sd->defer_lock, flags);
> > > +     skb = sd->defer_list;
> > 
> > I *think* that this read can possibly be fused with the previous one,
> > and another READ_ONCE() should avoid that.
> 
> Only the lockless read needs READ_ONCE()
> 
> For the one after spin_lock_irqsave(&sd->defer_lock, flags),
> there is no need for any additional barrier.
> 
> If the compiler really wants to use multiple one-byte-at-a-time loads,
> we are not going to fight, there might be good reasons for that.

I'm unsure I explained my doubt in a clear way: what I fear is that the
compiler could emit a single read instruction, corresponding to the
READ_ONCE() outside the lock, so that the spin-locked section will
operate on "old" defer_list. 

If that happens we could end-up with 'defer_count' underestimating the
list lenght. It looks like that is tolerable, as we will still be
protected vs defer_list growing too much.

Acked-by: Paolo Abeni <pabeni@redhat.com>