Experimental Physics and
Industrial Control System

Hinko Kocevar via Tech-talk <[email protected]> · Thu, 9 Jan 2020 09:16:17 +0000

Hi Mark,

> However this command does work to pass all packets to the higher levels, even "bad" ones.
> sudo ethtool --features enp23s0f1 rx-all on
>
> When I did that the software worked fine on the 10 Gbit adapter!  It gets the expected response and is able to parse it correctly.

Nice!

> As the IOC is running the number of RX errors and frame errors reported by Ifconfig keeps increasing.  But the IOC does not report any errors.

Hmm, interesting..

> Now we need to figure out why it thinks those packets are bad, since the content I care about seems to be fine.

I'd fire up the wireshark / tcpdump and inspect the frame checksum sequence bytes (https://en.wikipedia.org/wiki/Ethernet_frame) on 10G. It might happen that wireshark is able to color code the frame CRC as valid/invalid on-the-fly. If not, the frame can be fed into a CRC32 calculator (https://github.com/jwbensley/Ethernet-CRC32 or similar) and the obtained CRC value compared to the frame CRC value.
If they are the same (and given the same behavior of dropping frames is observed in windows, too) it could be NIC firmware that messes up the CRC calculation for your SNAP packets (Ethernet ones seem to have good CRC, AFAICT from previous messages).
Otherwise, if CRCs are in fact different, something is mangling the 10G CRC along the path to your host. No idea what could do that, TBH.

Cheers,
Hinko

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System