On Thu, Jul 31, 2025 at 09:17:51AM -0700, Jacob Keller wrote:
>
>
> On 7/31/2025 8:56 AM, Michal Kubiak wrote:
> > On Mon, Jul 07, 2025 at 03:48:36PM -0700, Jacob Keller wrote:
> >>
> >>
> >> On 7/4/2025 9:18 AM, Michal Kubiak wrote:
> >>> @@ -1100,14 +994,10 @@ static void ice_put_rx_mbuf(struct ice_rx_ring
> >>> *rx_ring, struct xdp_buff *xdp,
> >>> for (i = 0; i < post_xdp_frags; i++) {
> >>> buf = &rx_ring->rx_buf[idx];
> >>>
> >>> - if (verdict & (ICE_XDP_TX | ICE_XDP_REDIR)) {
> >>> - ice_rx_buf_adjust_pg_offset(buf, xdp->frame_sz);
> >>> + if (verdict & (ICE_XDP_TX | ICE_XDP_REDIR))
> >>> *xdp_xmit |= verdict;
> >>> - } else if (verdict & ICE_XDP_CONSUMED) {
> >>> + else if (verdict & ICE_XDP_CONSUMED)
> >>> buf->pagecnt_bias++;
> >>
> >> Why do we still keep pagecnt_bias++ here?
> >
> > My mistake. You're right - as I checked, we never use pagecnt_bias after
> > applying this patch.
> > I will remove pagecnt_bias completely in v2.
> >
> > Thanks,
> > Michal
> >
>
> I think Olek was also aware of this, but please make sure the v2 has
> fixed the errors with 9K MTU as well :D
Sure! I've already synced with Olek about the large MTU regression.
BTW, thanks for your report and detailed testing. TBH, I didn't run the
iperf with 9K MTU before. :)
Before sending the v2, I'll definitely test it the way you suggested.
