> From: Fijalkowski, Maciej <maciej.fijalkowski@xxxxxxxxx>
> Sent: Thursday, February 16, 2023 5:37 PM
[...]
> > > > + xdp_res = i40e_run_xdp(rx_ring, xdp, xdp_prog);
> > > > +
> > > > if (xdp_res) {
> > > > - if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
> > > > - xdp_xmit |= xdp_res;
> > > > + xdp_xmit |= xdp_res & (I40E_XDP_TX |
> > > I40E_XDP_REDIR);
> > >
> > > what was wrong with having above included in the
> > >
> > > } else if (xdp_res & (I40E_XDP_TX | I40E_XDP_REDIR)) {
> > >
> > > branch?
> > >
> >
> > For multi-buffer packets, only the first 'if' branch will be executed. We
> need to set
> > xdp_xmit for both single and multi-buffer packets.
>
> maybe pass xdp_xmit to i40e_process_rx_buffs and use it for buf_flip
> initialization? also you trimmed the code, but in there please don't
> define on-stack variables smaller than u32 (u16 next)
>
xdp_xmit may have been set by the xdp_res of a preceding packet in the current
napi call. We want to consume or flip the buffer based on only the xdp_res of current packet.
ACK for u16 -> u32.
