Re: [Intel-wired-lan] [PATCH v2 net] ixgbe: allow to increase MTU to some extent with XDP enabled

[Jason Xing]

On Tue, Jan 31, 2023 at 7:08 PM Alexander Lobakin
<alexandr.lobakin@xxxxxxxxx> wrote:
>
> From: Jason Xing <kerneljasonxing@xxxxxxxxx>
> Date: Tue, 31 Jan 2023 11:00:05 +0800
>
> > On Mon, Jan 30, 2023 at 11:09 PM Maciej Fijalkowski
> > <maciej.fijalkowski@xxxxxxxxx> wrote:
> >>
> >> On Fri, Jan 27, 2023 at 08:20:18PM +0800, Jason Xing wrote:
> >>> From: Jason Xing <kernelxing@xxxxxxxxxxx>
> >>>
> >>> I encountered one case where I cannot increase the MTU size directly
> >>> from 1500 to 2000 with XDP enabled if the server is equipped with
> >>> IXGBE card, which happened on thousands of servers in production
> >>> environment.
> >>
> >
> >> You said in this thread that you've done several tests - what were they?
> >
> > Tests against XDP are running on the server side when MTU varies from
> > 1500 to 3050 (not including ETH_HLEN, ETH_FCS_LEN and VLAN_HLEN) for a
>

> BTW, if ixgbe allows you to set MTU of 3050, it needs to be fixed. Intel
> drivers at some point didn't include the second VLAN tag into account,

Yes, I noticed that.

It should be like "int new_frame_size = new_mtu + ETH_HLEN +
ETH_FCS_LEN + (VLAN_HLEN * 2)" instead of only one VLAN_HLEN, which is
used to compute real size in ixgbe_change_mtu() function.
I'm wondering if I could submit another patch to fix the issue you
mentioned because the current patch tells a different issue. Does it
make sense?

If you're available, please help me review the v3 patch I've already
sent to the mailing-list. Thanks anyway.
The Link is https://lore.kernel.org/lkml/20230131032357.34029-1-kerneljasonxing@xxxxxxxxx/
.

Thanks,
Jason

> thus it was possible to trigger issues on Q-in-Q setups. AICS, not all
> of them were fixed.
>
> > few days.
> > I choose the iperf tool to test the maximum throughput and observe the
> > behavior when the machines are under greater pressure. Also, I use
> > netperf to send different size packets to the server side with
> > different modes (TCP_RR/_STREAM) applied.
> [...]
>
> Thanks,
> Olek