On Thu, Mar 28, 2024 at 5:40 AM [email protected] <[email protected]> wrote:
>
> I validated this modification on my x710 network card, but I found that
> the outer UDP checksum of the transmitted packets is incorrect, leading
> to communication abnormalities. I think it's necessary to disable the outer
> UDP checksum because although the capability reported by DPDK
> indicates support, in reality, the hardware doesn't actually support
> offloading,
> resulting in outer UDP checksum errors.
>
> tx_geneve_tso_offload="false", tx_ip_csum_offload="true",
> tx_out_ip_csum_offload="true",
> tx_out_udp_csum_offload="true", tx_sctp_csum_offload="true",
> tx_tcp_csum_offload="true",
> tx_tcp_seg_offload="false", tx_udp_csum_offload="true",
> tx_vxlan_tso_offload="false"
Well, good timing, thanks for the report.
I was testing ipv6 in ipv4 (which seemed to work) and I realised
something is wrong at the outer -> inner conversion by looking at the
ol_flags in my debug prints.
Now, trying the opposite (ipv4 in ipv6), I think I reproduce your
issue with a E810 nic:
04:50:46.211854 50:7c:6f:3c:0c:26 > 50:7c:6f:3c:10:5a, ethertype IPv6
(0x86dd), length 168: (hlim 64, next-header UDP (17) payload length:
114) 2001:4e48::2.39854 > 2001::1.geneve: [bad udp cksum 0x89ad ->
0xfadd!] Geneve, Flags [none], vni 0x0, proto TEB (0x6558)
52:54:00:00:11:01 > 4e:a9:1d:ce:85:4a, ethertype IPv4 (0x0800),
length 98: (tos 0x0, ttl 64, id 9408, offset 0, flags [DF], proto ICMP
(1), length 84)
172.31.22.2 > 172.31.22.1: ICMP echo request, id 1442, seq 9, length 64
Please have a try with the v2 (I'll post soon).
If it still fails, can you provide a reproducer (ideally without OVN
to reduce the scope)?
--
David Marchand
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
