Re: [patch net-next-2.6] net: convert bonding to use rx_handler

[Jiri Pirko <jpirko@redhat.com>]

Fri, Feb 18, 2011 at 08:58:26PM CET, eric.dumazet@gmail.com wrote:
>Le vendredi 18 février 2011 à 20:28 +0100, Jiri Pirko a écrit :
>
>> Yes I saw this. We would have to do something like:
>> 
>> struct skb_rx_context {
>> 	struct sk_buff *skb;
>> 	struct net_device *orig_dev;
>> };
>> 
>> static DEFINE_PER_CPU(struct skb_rx_context, skb_rx_context);
>> 
>> and then in __netif_receive_skb():
>> 
>> struct skb_rx_context *cont = __this_cpu_read(skb_rx_context);
>> 
>> if (cont->skb != skb) {
>> 	cont->skb = skb;
>> 	orig_dev = cont->orig_dev = skb->dev;
>> } else {
>> 	orig_dev = cont->orig_dev;
>> }
>> 
>> 
>> Does this make sense?
>
>Well, yes, something like that, but I think you dont need to keep a
>pointer to current skb. (It would not work if one handled/freed, same
>'skb pointer' is reused a bit later)

Well I think I need. How else should I distinguish that new skb (first
time in __netif_receive_skb) is there and I need to remember orig_dev?

>
>Sorry, I wont have time to look at this right now, its now 21h00 in
>France, time to get some time with family ;)

Np, same time here in CZ :)


>
>See you !
>
>