> From: Freddy Xin
> On 2013年12月13日 18:36, David Laight wrote:
> >> From: fre...@asix.com.tw
> > ...
> >> - skb = __netdev_alloc_skb_ip_align(dev->net, size, flags);
> >> + if (dev->driver_info->flags & FLAG_HW_IPALIGN)
> >> + skb = __netdev_alloc_skb(dev->net, size, flags);
> >> + else
> >> + skb = __netdev_alloc_skb_ip_align(dev->net, size, flags);
> > Given the definition:
> > static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device
> > *dev,
> > unsigned int length, gfp_t gfp)
> > {
> > struct sk_buff *skb = __netdev_alloc_skb(dev, length +
> > NET_IP_ALIGN, gfp);
> >
> > if (NET_IP_ALIGN && skb)
> > skb_reserve(skb, NET_IP_ALIGN);
> > return skb;
> > }
> >
> > It really ought to be possible to code that without an extra conditional.
> >
> > David
>
> The AX88179_178A driver do need to know the value of NET_IP_ALIGN
> to determine whether enabling the feature that makes IP header align
> on a dword-aligned address, but according to the comments from David
> Miller, I need to consider all situations, not just for the case that
> NET_IP_ALIGN is zero, so the condition added in rx_submit is just used to
> determine whether reserving NET_IP_ALIGN bytes for each SKB.
I was thinking of something like:
skb = netdev_alloc_skb(dev, length + dev->skb_align, gfp);
if (NET_IP_ALIGN && skb && !(ev->driver_flags & FLAG_HW_IPALIGN))
skb_reserve(skb, NET_IP_ALIGN);
It might even be reasonable to remove the length adjustment - provided
that all the later code uses the skb length.
David
