[ofa-general] Re: [PATCH 2/9 Rev3] [core] Add skb_blist & hard_start_xmit_batch

[Krishna Kumar2]

Hi Evgeniy,

Evgeniy Polyakov <johnpol at 2ka.mipt.ru> wrote on 08/08/2007 05:31:43 PM:

> > +int dev_change_tx_batch_skb(struct net_device *dev, unsigned long
new_batch_skb)
> > +{
> > +   int ret = 0;
> > +   struct sk_buff_head *blist;
> > +
> > +   if (!dev->hard_start_xmit_batch) {
> > +      /* Driver doesn't support batching skb API */
> > +      ret = -ENOTSUPP;
> > +      goto out;
> > +   }
> > +
> > +   /* Handle invalid argument */
> > +   if (new_batch_skb < 0) {
> > +      ret = -EINVAL;
> > +      goto out;
> > +   }
> It is unsigned, how can it be less than zero?

Yuck, originally I had it as int and changed to ulong and forgot to
remove this check.

> And actually you use it just like a binary flag (casted to/from u32 in
> the code, btw), so why not using ethtool_value directly here?

I still need to check if the value is changing, so the one check is needed.
Later I am using it as a value directly.

> > +   /* Check if new value is same as the current */
> > +   if (!!dev->skb_blist == !!new_batch_skb)
> > +      goto out;
> > +
> > +   if (new_batch_skb &&
> > +       (blist = kmalloc(sizeof *blist, GFP_KERNEL)) == NULL) {
> > +      ret = -ENOMEM;
> > +      goto out;
> > +   }
> > +
> > +   spin_lock(&dev->queue_lock);
> > +   if (new_batch_skb) {
> > +      skb_queue_head_init(blist);
> > +      dev->skb_blist = blist;
> > +   } else
> > +      free_batching(dev);
> > +   spin_unlock(&dev->queue_lock);
>
> This needs bh lock too, since blist is accessed by qdisc_restart.

Yes, had it in the code, put it in the list of changes, but missed it for
some reason :(

> > +int dev_add_skb_to_blist(struct sk_buff *skb, struct net_device *dev)
> > +{
> > +   if (!list_empty(&ptype_all))
> > +      dev_queue_xmit_nit(skb, dev);
> > +
> > +   if (netif_needs_gso(dev, skb)) {
> > +      if (unlikely(dev_gso_segment(skb))) {
> > +         kfree(skb);
> > +         return 0;
> > +      }
> > +
> > +      if (skb->next) {
> > +         int count = 0;
> > +
> > +         do {
> > +            struct sk_buff *nskb = skb->next;
> > +
> > +            skb->next = nskb->next;
> > +            __skb_queue_tail(dev->skb_blist, nskb);
> > +            count++;
> > +         } while (skb->next);
>
> Is it possible to move list without iterating over each entry?

Though I cannot see something obvious to do that, let me see if
something is possible as it will make a good difference.

thanks for your suggestions,

- KK