So do you see MTU size got decreased in this situation?
Just to make sure you know what we are doing here. I want to make sure your issue here is related to MACsec or not.
Or it issue is just simple case that even MACsec is not enabled, you still see mgbe_payload_cs_err got increased.
In this scenario, the MTU is not reduced anymore — that issue has been resolved, thank you.
There’s still one more issue: with MACsec disabled, we observe an increase in mgbe_payload_cs_err, while no such increase is seen when MACsec is enabled. This seems to be related to MACsec.
How frequently would that mgbe_payload_cs_err increase? If you run a iperf test for like 180 sec, then how would that counter increase?
I mean what is the scenario of the network connection here?
Also, what is the device on the other side? Does that one also have MACsec enabled?
After setting nvidia,macsec-enable = <0>, we observed one occurrence of mgbe_payload_cs_err after around 10 minutes of testing.
MTU Mismatches could lead to that increase. You should make sure every device on the network path all have the similar MTU setting.
The peer device is an x86 server, and its NIC does not have MACsec enabled. Both sides have MTU set to 9000.
If you use NV devkit in the same environment, then will you see this counter increase on NV devkit too?
Unfortunately, we cannot verify this scenario because our hardware design prevents two NV devkits from communicating with each other over XFI.