Re: [PATCH RFC net-next 4/4] veth: use generic skb XDP handling

["Björn Töpel" <bjorn@kernel.org>]

Maciej Fijalkowski <maciej.fijalkowski@intel.com> writes:

> On Tue, May 12, 2026 at 04:32:43PM +0200, Björn Töpel wrote:
>> Maciej Fijalkowski <maciej.fijalkowski@intel.com> writes:
>> 
>> > veth currently runs XDP for skb-backed packets by constructing an
>> > xdp_buff and then, for XDP_TX and XDP_REDIRECT, converting that
>> > skb-backed xdp_buff into an xdp_frame. The backing memory is still
>> > skb-owned, so veth has to pin the skb data and frags manually before
>> > consuming the skb.
>> >
>> > Use the generic skb XDP helper for skb-backed packets instead. This
>> > keeps skb-backed packets on the skb generic XDP path: XDP_REDIRECT uses
>> > xdp_do_generic_redirect() and XDP_TX uses generic_xdp_tx(). Packets that
>> > already arrive as xdp_frames keep using the existing veth xdp_frame path.
>> >
>> > veth still provides its own page_pool and xdp_rxq_info through struct
>> > xdp_generic_ctx. It also keeps using struct veth_xdp_buff storage so
>> > metadata kfuncs that need the skb continue to work after a possible skb
>> > COW.
>> 
>> 
>> Hmm, so now if veth with native XDP redirects to a native XDP NIC, we'll
>> end up in the generic path? We'll loose some bulking capabilities as
>> well, but maybe that's OK.
>
> Hi Bjorn - thanks a lot for taking a look at these changes!
>
> This touches a path where receiver has XDP prog loaded but sender used
> ndo_start_xmit() to produce skbs onto another end of veth pair; these skbs
> are then stored within ptr_ring where receiver picks them up. Native XDP
> still has bulking, however the skb path has a little bit of back-and-forth
> juggle between skb<->xdp_frame conversion. Now I got rid of these
> conversions but also of home-grown generic XDP path as well as bulking and
> the end result seems to land in the same performance results.
>
> Does it make sense to you?

Hmm! Thanks! ...but doesn't this change the semantics of redirect, for
skb-input/native case?

target of redirect    before (skb-input)                     after (skb-input)

real NIC via devmap   ndo_xdp_xmit(), bulked, XDP TX queue   netdev_start_xmit(), regular qdisc/TX path    
cpumap                xdp_frame on remote CPU's ptr_ring     skb on remote CPU's softnet backlog           
AF_XDP socket         native (potentially ZC)                xsk_generic_rcv (copy, generic)               
veth peer (devmap)    xdp_frame into peer's xdp_ring         skb into peer's xdp_ring (same ring, skb tag) 


(Let me re-read it!)


Björn