On Wed, Sep 27, 2017 at 06:03:33PM +0000, Eric Dumazet wrote:
> >> diff --git a/net/ipv6/route.c b/net/ipv6/route.c
>
> >
> > Hi Paolo,
> >
> > Eric and I discussed about this issue recently as well :).
> >
> > What about the following change:
> >
> > diff --git a/include/net/dst.h b/include/net/dst.h
> > index 93568bd0a352..33e1d86bcef6 100644
> > --- a/include/net/dst.h
> > +++ b/include/net/dst.h
> > @@ -258,14 +258,18 @@ static inline void dst_hold(struct dst_entry *dst)
> > static inline void dst_use(struct dst_entry *dst, unsigned long time)
> > {
> > dst_hold(dst);
> > - dst->__use++;
> > - dst->lastuse = time;
> > + if (dst->lastuse != time) {
> > + dst->__use++;
> > + dst->lastuse = time;
> > + }
> > }
> >
> > static inline void dst_use_noref(struct dst_entry *dst, unsigned long time)
> > {
> > - dst->__use++;
> > - dst->lastuse = time;
> > + if (dst->lastuse != time) {
> > + dst->__use++;
> > + dst->lastuse = time;
> > + }
> > }
> >
> > static inline struct dst_entry *dst_clone(struct dst_entry *dst)
> > diff --git a/net/ipv6/route.c b/net/ipv6/route.c
> > index 26cc9f483b6d..e195f093add3 100644
> > --- a/net/ipv6/route.c
> > +++ b/net/ipv6/route.c
> > @@ -1170,8 +1170,7 @@ struct rt6_info *ip6_pol_route(struct net *net,
> > struct fib6_table *table,
> >
> > struct rt6_info *pcpu_rt;
> >
> > - rt->dst.lastuse = jiffies;
> > - rt->dst.__use++;
> > + dst_use_noref(rt, jiffies);
> > pcpu_rt = rt6_get_pcpu_route(rt);
> >
> > if (pcpu_rt) {
> >
> >
> > This way we always only update dst->__use and dst->lastuse at most
> > once per jiffy. And we don't really need to update pcpu and then do
> > the copy over from pcpu_rt to rt operation.
> >
> > Another thing is that I don't really see any places making use of
> > dst->__use. So maybe we can also get rid of this dst->__use field?
> >
> > Thanks.
> > Wei
>
> Paolo, given we are very close to send Wei awesome work about IPv6
> routing cache,
> could we ask you to wait few days before doing the same work from your side ?
>
> Main issue is the rwlock, and we are converting it to full RCU.
+1
We can get a better picture of other optimizations once
the rwlock is removed.
>
> You are sending patches that are making our job very difficult IMO.
>
> We chose to mimic the change done in neighbour code years ago
> ( 0ed8ddf4045fcfcac36bad753dc4046118c603ec )
>
> Thanks.
