Unicast Traffic Loss on Broadcom NetXtreme NICs (bnxtnet) while Broadcast/Multicast Traffic Continues

Products

• VMware vSphere ESXi

Issue/Introduction

• A physical network interface (vmnic) intermittently stops passing all unicast traffic.
• Broadcast and multicast traffic (e.g., ARP requests, CDP/LLDP) continue to be received and processed normally by the ESXi host.
• Packet captures at the UplinkSndKernel point show unicast traffic being handed off to the driver, but no corresponding traffic is observed at the physical switch port (egress failure).
• Inbound unicast traffic to Virtual Machines or VMkernel ports is not observed in software captures (ingress failure).
• The issue is only resolved by a full reboot of the ESXi host or a complete reload of the `bnxtnet` driver.

Environment

Vmware vSphere ESXi 

Cause

• This behavior is indicative of a Hardware MAC Filter Table Corruption or a state synchronization failure between the bnxtnet driver and the physical network adapter's firmware.
• When the ESXi hypervisor operates, it programs the NIC ASIC's internal hardware filtering table with the MAC addresses of all active Virtual Machines and VMkernel adapters. Due to a defect in certain versions of the `bnxtnet` driver or firmware, this table can become corrupted or cleared. Because the ASIC no longer recognizes the unicast MAC addresses as belonging to the host, it drops those frames at the silicon level before they reach the ESXi software. Broadcast and multicast frames bypass these filters by design, which is why they continue to function.

Resolution

To resolve this issue, the bnxtnet driver and NIC firmware must be updated to versions that include fixes for hardware state synchronization.

1. Check current versions using:
   • esxcli software vib list | grep bnxtnet
   • esxcli network nic get -n vmnicX
2. Upgrade: It is recommended to upgrade to at least bnxtnet driver version 234.0.159 and firmware version 234.1.128.0 (or later), as these releases contain specific fixes for hardware-initiated resets and synchronization errors.
3. Validate on VCG: Always ensure the chosen driver/firmware combination is supported for your specific hardware model on the [VMware Compatibility Guide](https://www.vmware.com/resources/compatibility/search.php).

Important: Hardware Vendor Engagement
You must contact your hardware vendor (e.g., Dell, HPE, Lenovo) to confirm these findings. Because this fault occurs within the physical NIC ASIC and involves the interaction between the vendor's firmware and the driver, the hardware vendor must:

• Confirm the presence of MAC table corruption or firmware "silent" resets in the hardware logs.
• Provide the specific, validated firmware package required for your server model.
• Assist in identifying if any physical degradation or specific Top-of-Rack (ToR) switch configurations (such as FEC mode mismatches) are contributing to the instability.