Re: [Xen-devel] [PATCH net v5] xen-netback: fix fragment detection in checksum setup

[Ian Campbell <Ian.Campbell@xxxxxxxxxx>]

On Tue, 2013-12-10 at 09:12 -0800, Eric Dumazet wrote:
> On Tue, 2013-12-10 at 16:24 +0000, Paul Durrant wrote:
> 
> > Yes, I guess that's true but if we decide to pull up at all then is it 
> > harmful to pull more than we absolutely need?
> 
> Pulling too many bytes might hurt aggregation.
> 
> When one skb is merged to another one (skb_try_coalesce()), if you have
> payload in skb->head, we cannot free the head, and need one additional
> 'frag'

Does that ever happen to skb's which are received by a NIC and passed up
to the network stack (including forwarding by bridging of netfilter)? I
thought this path only happened for locally generated traffic.

If it can happen for received skb's then that seems like a pretty solid
reason not to over estimate in the pull up phase.

Ian.


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel