Broadcom BCM57414 dual-port 25G OCP adapters on ESXi 8.0.x hosts experience continuous link flapping on active ports when the secondary port contains an un-cabled SFP transceiver.
Symptoms:
vmkernel.log shows the bnxtnet driver initializing and immediately dropping the link.vmkernel.log shows the following sequence: bnxtnet: bnxtnet_update_phy:3816: [vmnic3] link_info->auto_mode: 4, phy_cfg.flags: 0x00000029 - (AUTO)bnxtnet: bnxtnet_display_link:2985: [vmnic3] NIC Link is downbnxtnet: bnxtnet_uplink_quiesce_dev:2606: [vmnic3] Quiesce dev 0x4528003be000vobd.log confirms vob.net.vmnic.linkstate.down and up transitions.vobd.log confirms linkstate transitions: [vob.net.vmnic.linkstate.down] vmnic vmnic3 linkstate down[vob.net.vmnic.linkstate.up] vmnic vmnic3 linkstate upThe two ports on this network adapter share internal clock/timing hardware. When an unused port (Port 1) has an optical module (SFP) inserted but is not connected to a switch, the adapter continuously attempts to auto-detect a link speed, a process that never succeeds since there is no far-end device.
This repeated detection cycle periodically competes for the shared internal clock resource that the active port (Port 2) relies on to maintain its connection. When that happens, Port 2's link is briefly interrupted before recovering, and the cycle repeats, producing the periodic link flapping observed.
This is categorized as working as designed based on the adapter's shared hardware architecture. This timing issue only occurs when no link is established but modules are present. If a cable is inserted into Port 1 after Port 2 has already established a link, Port 2's link status will not be affected.
Follow either method below to stabilize the link:
Please also refer Broadcom KB article Broadcom BCM57414 Adapter FEC CL74 Setting Confirmation (VMware ESXi 8.0)