When connecting the CX8 NIC to the Q3400 switch, the subnet manager cannot be activated on either side

[mooyu]

Both ends display as follows:

(base) root@node-18:~# opensm

OpenSM 5.21.12.MLNX20250617.f74e01b8

Command Line Arguments:
Log File: /var/log/opensm.log

OpenSM 5.21.12.MLNX20250617.f74e01b8

Using default GUID 0x5000e6030005562a
Entering DISCOVERING state

Error from osm_opensm_bind (0x2A)
Perhaps another instance of OpenSM is already running
Exiting SM

(base) root@node-18:~# sminfo
ibwarn: [16405] mad_rpc_open_port: client_register for mgmt 1 failed
sminfo: iberror: failed: Failed to open ‘(null)’ port ‘0’

root@nvos:~# nv show system
operational applied

---

uptime 1 day, 16:28:08
hostname nvos
product-name nvos
product-release 25.02.5002
status System is ready
date-time
local-time 2025-09-11 06:31:25
timezone Etc/UTC Etc/UTC
health
status Not OK
version
product-release 25.02.5002
root@nvos:~# ib sm
-bash: ib: command not found
root@nvos:~# sminfo
ibwarn: [587839] _do_madrpc: recv failed: Connection timed out
ibwarn: [587839] mad_rpc: _do_madrpc failed; dport (DR path slid 0; dlid 0; 0)
sminfo: iberror: failed: query

[marlon1]

Hello,

This looks a known issue. Please update to newer version.

OpenSM 5.22.11.MLNX20250130.12243119 will fix the problem.

[marlon1]

Just to add to my last comment. SM is not supported on the XDR switch system.

[hedy]

In our case, the cleanest solution was aligning firmware + management stack rather than fighting OpenSM on the host. Once everything was on a supported combo, CX8 links came up immediately.

It may be worth checking with a vendor that’s already deployed Q3400 + CX8 fabrics—they usually know which SM mode actually works in practice.

For reference, we had decent experience working with NADDOD on a similar Q3400 + ConnectX-8 bring-up. They were already familiar with the SM behavior and firmware combinations, which saved a lot of trial-and-error on our side.