Re: [PATCH v5] net: batch skb dequeueing from softnet input_pkt_queue

[Changli Gao <xiaosuo@gmail.com>]

On Thu, Apr 22, 2010 at 10:33 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
>
> It does make a difference, Damn it.
>
> I really really start to think you dont read what I wrote, or you dont
> care.

I misunderstood it. Sorry.

>
> Damn, cant you update all the things at once, taking this lock only
> once ?
>
> You focus having an ultra precise count of pkt_queue.len, but we dont
> care at all ! We only want a _limit_, or else the box can be killed by
> DOS.
>
> If in practice this limit can be 2*limit, thats OK.
>
> Cant you understand this ?
>
>
> We need one limit. Not two limits.
>
> I already told you how to do it, but you ignored me and started yet
> another convoluted thing.
>
>
> process_backlog() transfert the queue to its own queue and reset pkt_len
> to 0 (Only once)
>
> End of story.
>
> Maximum packet queued to this cpu softnet_data will be 2 * old_limit.
>
> So what ?
>

Now, I think I really understand. We don't need a precious limit. So
only a additional queue is enough.

-- 
Regards，
Changli Gao(xiaosuo@gmail.com)