Re: [PATCH] xen-netback: use netdev_alloc_skb_ip_align

[Stephen Hemminger <stephen@networkplumber.org>]

On Tue, 9 Apr 2013 12:27:31 +0100
Ian Campbell <Ian.Campbell@citrix.com> wrote:

> (apologies for the late reply, I've been away)
> 
> On Wed, 2013-03-20 at 19:21 +0000, Stephen Hemminger wrote:
> > Use standard helper function to allocate and align received packet.
> > Compile tested only
> > 
> > Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
> > 
> > 
> > --- a/drivers/net/xen-netback/netback.c	2013-03-07 18:12:52.825300956 -0800
> > +++ b/drivers/net/xen-netback/netback.c	2013-03-20 12:09:09.052034865 -0700
> > @@ -1357,8 +1357,8 @@ static unsigned xen_netbk_tx_build_gops(
> >  			    ret < MAX_SKB_FRAGS) ?
> >  			PKT_PROT_LEN : txreq.size;
> >  
> > -		skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
> > -				GFP_ATOMIC | __GFP_NOWARN);
> > +		skb = __netdev_alloc_skb_ip_align(vif->dev, data_len,
> > +						  GFP_ATOMIC | __GFP_NOWARN);
> 
> __netdev_alloc_skb_ip_align accounts for NET_IP_ALIGN but not
> NET_SKB_PAD, is this aspect of the change intentional? (I'm not really
> sure how much NET_SKB_PAD is worth in practice).

__nedev_alloc_skb_ip_align(dev, length, gfp)
 calls __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp)
   calls __netdev_alloc_skb

which adds padding here:
struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
				   unsigned int length, gfp_t gfp_mask)
{
	struct sk_buff *skb = NULL;
	unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
			      SKB_DATA_ALIGN(sizeof(struct skb_shared_info));