Hi,
We use Mellanox CX-5 En SmartNICs for HW-offload in switchdev mode for
OVN L3 Gateway nodes, which handle cross-AZ traffic. While regular
routed traffic works well (gets offloaded into NIC in both directions
via TC), we observed, that flow, which involves connection tracking,
fails to be offloaded with next ovs-vswitchd log:
2025-05-16T10:53:47.604Z|25874|poll_loop(handler1)|DBG|wakeup due to
0-ms timeout at ofproto/ofproto-dpif-upcall.c:824 (1% CPU usage)
2025-05-16T10:53:47.606Z|25875|dpif_netlink(handler1)|ERR|failed to
offload flow: No such file or directory: ovn-node1-0
2025-05-16T10:53:47.606Z|25876|dpif_netlink(handler1)|DBG|system@ovs-system:
put[create] ufid:a9ea21cb-b110-4014-9893-212d3d7280b6
recirc_id(0x254a0),dp_hash(0x3e484431/0),skb_priority(0/0),tunnel(tun_id=0xff000e,src=10.0.1.10,dst=10.0.0.10,ttl=55/0,tp_src=10615/0,tp_dst=6081/0,geneve({class=0x102,type=0x80,len=4,0x5c0001}),flags(-df+csum+key)),in_port(2),skb_mark(0/0),ct_state(0xe1/0x2f),ct_zone(0x6/0),ct_mark(0/0),ct_label(0x2/0x3),ct_tuple4(src=172.21.4.9/0.0.0.0,dst=172.21.0.23/0.0.0.0,proto=6/0,tp_src=7359/0,tp_dst=6000/0),eth(src=00:02:33:e6:30:0e,dst=00:00:33:e6:30:0e),eth_type(0x0800),ipv4(src=172.21.4.9/252.0.0.0,dst=172.21.0.22,proto=6,tos=0/0x3,ttl=62,frag=no),tcp(src=7359/0,dst=6000/0),tcp_flags(0/0),
actions:ct_clear,set(tunnel(tun_id=0x104d,src=10.0.0.10,dst=10.0.0.20,ttl=64,tp_dst=6081,geneve({class=0x102,type=0x80,len=4,0x10012}),flags(df|csum|key))),set(eth(src=d0:fe:00:00:28:a2,dst=0a:00:23:a0:9a:e0)),set(ipv4(ttl=61)),2
DP port #2 is genev_sys_6081 is this example. I've enabled DBG logs for
all ovs-vswitched modules, but couldn't find any netlink messages, which
could bring more light on the problem...
OVS version is 3.3.1.
I've looked through the OVS code and couldn't find the place where on
netdev_tc_flow_put() could be returned ENOENT as an error. Maybe I'm on
the wrong way or missing some another important logic?
At the same time I see that Marcelo Ricardo Leitner says [1] that such
error was already seen in the setups with LAGs. This is what we have our
setup: 2 PFs are bound in LACP LAG and both ports are configured for HW
offloading. How to ensure that the problem is LAG or not?
Looking forward to get the direction for futher debugging. Thanks!
1: https://github.com/ovn-org/ovn/issues/160#issuecomment-1319291423
--
regards,
Vladislav Odintsov
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
