src icon indicating copy to clipboard operation
src copied to clipboard

Mellanox ConnectX-4 Lx CARP Multicast Not Working

Open Ryushin opened this issue 9 months ago • 0 comments

Important notices

Before you add a new report, we ask you kindly to acknowledge the following:

  • [X] I have read the contributing guide lines at https://github.com/opnsense/core/blob/master/CONTRIBUTING.md
  • [X] I am convinced that my issue is new after having checked both open and closed issues at https://github.com/opnsense/core/issues?q=is%3Aissue

Describe the bug

We have two GoWin GW-BS-1UR2-25G: https://www.gowinfanless.com/products/network-device/1u-2u-server/gw-bs-1ur2-25g https://www.servethehome.com/gowin-1u-25gbe-appliance-review-this-has-everything-including-poe-intel-nvidia/

It seems there is a bug in the ConnectX-4 Lx driver for FreeBSD. Broadcasts are looped back to themselves so CARP won't work in multicast.

Symptoms: Both devices will stay in CARP Backup when using multicast and neither will become Master. Switching to unicast allows CARP to work.

Constant Duplicate Address Detected (DAD) message in system log:

2025-04-13T09:37:45-07:00 <hostname goes here) kernel - - [meta sequenceId="1585"] <3>mce0: a looped back NS message is detected during DAD for fe80:a::526b:4bff:fef0:8058.  Another DAD probes are being sent.
<13>1 2025-04-13T09:37:47-07:00 <hostname goes here) kernel - - [meta sequenceId="1586"] <3>mce1: a looped back NS message is detected during DAD for fe80:b::526b:4bff:fef0:8059.  Another DAD probes are being sent.

Links I found with problems: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=270859 (Disabling SRV-IO did not work for me) https://network.nvidia.com/pdf/prod_software/Mellanox_FreeBSD_Release_Notes_3.0.0.pdf

Are there any known workarounds for the broadcast problem with the Mellanox ConnectX-4?

To Reproduce

Steps to reproduce the behavior:

  1. Set up CARP using the Mellanox NIC using multicast

Expected behavior CARP to work using multicast

Describe alternatives you considered Switching to unicast.

Environment OPNsense 25.1.5_4-amd64 Intel i3-N305 CPU Network: Two Intel i226-V (Intel SRKTU) NICs, two Intel i210-AT 1GbE NICs, One i210-AT, and two SFP28 25Gbe ports powered by a (Mellanox) NVIDIA ConnectX-4 Lx.

Ryushin avatar Apr 13 '25 20:04 Ryushin