[ofa-general] Re: [PATCH 3/9 Rev3] [sched] Modify qdisc_run to support batching

[Krishna Kumar2]

Evgeniy Polyakov <johnpol at 2ka.mipt.ru> wrote on 08/08/2007 05:44:02 PM:

> On Wed, Aug 08, 2007 at 03:01:45PM +0530, Krishna Kumar
(krkumar2 at in.ibm.com) wrote:
> > +static inline int get_skb(struct net_device *dev, struct Qdisc *q,
> > +           struct sk_buff_head *blist, struct sk_buff **skbp)
> > +{
> > +   if (likely(!blist || (!skb_queue_len(blist) && qdisc_qlen(q) <=
1))) {
> > +      return likely((*skbp = dev_dequeue_skb(dev, q)) != NULL);
> > +   } else {
> > +      int max = dev->tx_queue_len - skb_queue_len(blist);
> > +      struct sk_buff *skb;
> > +
> > +      while (max > 0 && (skb = dev_dequeue_skb(dev, q)) != NULL)
> > +         max -= dev_add_skb_to_blist(skb, dev);
> > +
> > +      *skbp = NULL;
> > +      return 1;   /* we have atleast one skb in blist */
> > +   }
> > +}
>
> Same here - is it possible to get a list in one go instead of pulling
> one-by-one, since it forces quite a few additional unneded lock
> get/releases. What about dev_dequeue_number_skb(dev, q, num), which will
> grab the lock and move a list of skbs from one queue to provided head.

OK, I will try this out.

> > @@ -158,7 +198,10 @@ static inline int qdisc_restart(struct n
> >     /* And release queue */
> >     spin_unlock(&dev->queue_lock);
> >
> > -   ret = dev_hard_start_xmit(skb, dev);
> > +   if (likely(skb))
> > +      ret = dev_hard_start_xmit(skb, dev);
> > +   else
> > +      ret = dev->hard_start_xmit_batch(dev);
>
> Perfectionism says that having array of two functions and calling one of
> them via array_func_pointer[!!skb] will be much faster. Just a though.
> It is actually much faster than if/else on x86 at least.

Thinking about this - I will have to store the 2 pointer array in dev
itself wasting some space, and also fn pointer will have wrong signature
as one takes an extra argument. Will ponder some more :)

thanks,

- KK