On Fri, 18 Mar 2022 20:19:29 +0100 Lorenzo Bianconi wrote:
> diff --git a/net/core/dev.c b/net/core/dev.c
> index ba69ddf85af6..92d560e648ab 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -4737,7 +4737,7 @@ static u32 netif_receive_generic_xdp(struct sk_buff *skb,
> * native XDP provides, thus we need to do it here as well.
> */
> if (skb_cloned(skb) || skb_is_nonlinear(skb) ||
> - skb_headroom(skb) < XDP_PACKET_HEADROOM) {
> + skb_headroom(skb) < XDP_PACKET_HEADROOM_MIN) {
> int hroom = XDP_PACKET_HEADROOM - skb_headroom(skb);
> int troom = skb->tail + skb->data_len - skb->end;
>
IIUC the initial purpose of SKB mode was to be able to test or
experiment with XDP "until drivers add support". If that's still
the case the semantics of XDP SKB should be as close to ideal
XDP implementation as possible.
We had a knob for specifying needed headroom, is that thing not
working / not a potentially cleaner direction?
