Re: [PATCH bpf-next 1/3] xsk: avoid overwriting skb fields for multi-buffer traffic

[Stanislav Fomichev <stfomichev@gmail.com>]

On 09/22, Maciej Fijalkowski wrote:
> We are unnecessarily setting a bunch of skb fields per each processed
> descriptor, which is redundant for fragmented frames.
> 
> Let us set these respective members for first fragment only.
> 
> Signed-off-by: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
> ---
>  net/xdp/xsk.c | 10 +++++-----
>  1 file changed, 5 insertions(+), 5 deletions(-)
> 
> diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
> index 72e34bd2d925..72194f0a3fc0 100644
> --- a/net/xdp/xsk.c
> +++ b/net/xdp/xsk.c
> @@ -758,6 +758,10 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
>  				goto free_err;
>  
>  			xsk_set_destructor_arg(skb, desc->addr);
> +			skb->dev = dev;
> +			skb->priority = READ_ONCE(xs->sk.sk_priority);
> +			skb->mark = READ_ONCE(xs->sk.sk_mark);
> +			skb->destructor = xsk_destruct_skb;
>  		} else {
>  			int nr_frags = skb_shinfo(skb)->nr_frags;
>  			struct xsk_addr_node *xsk_addr;
> @@ -826,14 +830,10 @@ static struct sk_buff *xsk_build_skb(struct xdp_sock *xs,
>  
>  			if (meta->flags & XDP_TXMD_FLAGS_LAUNCH_TIME)
>  				skb->skb_mstamp_ns = meta->request.launch_time;
> +			xsk_tx_metadata_to_compl(meta, &skb_shinfo(skb)->xsk_meta);
>  		}
>  	}
>  
> -	skb->dev = dev;
> -	skb->priority = READ_ONCE(xs->sk.sk_priority);
> -	skb->mark = READ_ONCE(xs->sk.sk_mark);
> -	skb->destructor = xsk_destruct_skb;
> -	xsk_tx_metadata_to_compl(meta, &skb_shinfo(skb)->xsk_meta);
>  	xsk_inc_num_desc(skb);

What about IFF_TX_SKB_NO_LINEAR case? I'm not super familiar with
it, but I don't see priority/mark being set over there after this change.