On Tue, Apr 2, 2013 at 11:24 PM, Rajahalme, Jarno (NSN - FI/Espoo) <
jarno.rajaha...@nsn.com> wrote:
>
> On Mar 30, 2013, at 18:20 , ext Pravin B Shelar wrote:
>
> > diff --git a/datapath/linux/compat/include/linux/netdevice.h
> b/datapath/linux/compat/include/linux/netdevice.h
> > index 71aad87..4ffdac4 100644
> > --- a/datapath/linux/compat/include/linux/netdevice.h
> > +++ b/datapath/linux/compat/include/linux/netdevice.h
> > @@ -167,4 +167,30 @@ static inline struct sk_buff
> *__skb_gso_segment(struct sk_buff *skb,
> > }
> > #endif
> >
> > +#if LINUX_VERSION_CODE >= KERNEL_VERSION(3,9,0)
> > +static inline struct sk_buff *rpl__skb_gso_segment(struct sk_buff *skb,
> > + netdev_features_t features,
> > + bool tx_path)
> > +{
> > + struct sk_buff *nskb;
> > + char cb[48];
> > +
> > + /* From 3.9 kernel skb->cb is used by skb gso. Therefore
> > + * make copy of it to restore it. */
> > +
> > + memcpy(cb, skb->cb, sizeof(skb->cb));
> > + nskb = __skb_gso_segment(skb, 0, false);
> > + if (IS_ERR(nskb))
> > + return nskb;
> > +
> > + skb = nskb;
> > + while (nskb) {
> > + memcpy(nskb->cb, cb, sizeof(nskb->cb));
> > + nskb = nskb->next;
> > + }
> > + return skb;
> > +}
> > +#define __skb_gso_segment rpl__skb_gso_segment
> > +#endif
> > +
>
> Is it only the struct ovs_skb_cb we are concerned with here? If so, it
> would be more efficient just to copy "sizeof (struct ovs_skb_cb)" bytes
> back and forth, which is currently less than the full 48 bytes.
>
>
It is part of linux compatibility code, therefore I do not have access to
'struct ovs_skb_cb'. I can move it to datapath directory, but I am not sure
if it is worth doing it if lisp starts using UDP TUNNEL GSO. Since then
upcall handler will be the only user of skb_gso_segment() where GSO packets
should be rare.
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
