Re: [Xen-devel] [PATCHv1 net-next] xen-netback: remove unconditional pull_skb_tail in guest Tx path

[Ian Campbell <Ian.Campbell@citrix.com>]

On Wed, 2014-11-05 at 10:46 +0000, David Vrabel wrote:
> On 04/11/14 21:43, Eric Dumazet wrote:
> > On Tue, 2014-11-04 at 16:17 -0500, David Miller wrote:
> > 
> > 
> >>
> >> Every protocol demux starts with pskb_may_pull() to pull frag data
> >> into the linear area, if necessary, before looking at headers.
> > 
> > eth_get_headlen() might be relevant as well, to perform a single copy of
> > exactly all headers.
> 
> In netback's case we need an estimate of the header length before
> reading any of the packet, since peeking at any frag would prevent any
> TLB flush avoidance.
> 
> It might be useful to use eth_get_headlen() to adjust the estimate at
> runtime, but for now the fixed amount of 128 bytes is simple and seems
> good enough.

I think what Eric meant was that having done the 128 copy you could call
eth_get_headlen which in the common case should be a nop but would
ensure you always had the headers in the linear area for the uncommon
case.

It looks like the difference compared with skb_checksum_setup is that
eth_get_headlen deals with L4 too whereas skb_checksum_setup only goes
to L3 (and then only for some subset of protocols with checksums).

Ian.