On Aug 29, 2011 3:56 PM, "Justin Pettit" <jpet...@nicira.com> wrote:
> diff --git a/datapath/linux/compat/include/linux/netdevice.h
b/datapath/linux/compat/include/linux/netdevice.h
> index 04ebd89..664ff2e 100644
> --- a/datapath/linux/compat/include/linux/netdevice.h
> +++ b/datapath/linux/compat/include/linux/netdevice.h
> @@ -74,7 +74,14 @@ extern void unregister_netdevice_many(struct list_head
*head);
> #endif
>
> #ifndef HAVE_DEV_DISABLE_LRO
> -extern void dev_disable_lro(struct net_device *dev);
> +/* Some distributions (e.g., RHEL5) backported LRO support, but not the
> + * userspace interface to adjust them, so it is necessary to call
> + * dev_disable_lro() from the kernel. If this is an older kernel with
> + * LRO support, then we assume that they either backported
> + * dev_disable_lro() or provided the userspace interface. In either
> + * case, it's a no-op.
> + */
> +static inline void dev_disable_lro(struct net_device *dev) { }
I don't think it's a very good idea to have a backport that doesn't do the
same thing as the upstream version (where possible). So while in this case
we know that if the function is called it will be a no-op, it's confusing to
enforce that and unnecessary because of the version guard around the actual
call.
I would keep the backport and move the comment to call site. The compiled
result should be the same, of course.
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
