ESXi 7.0 U3L upgrade from Dell OEM ISO loads driver 4.1.9 driver but host disconnected after reboot

Products

• VMware vSphere ESXi

Issue/Introduction

Changes in the ntg3 driver version 4.1.9.0-4vmw might cause link issues for the fiber physical layer and connectivity on some NICs, such as Broadcom 1Gb, fails to come up.

Symptoms:
After upgrading the ntg3 driver to version 4.1.9.0-4vmw, Broadcom NICs with fiber physical connectivity might lose network. The vmkernel logs on the ESXi host will have traces as below

2023-04-21T12:45:18.689Z cpu44:2097903)Device: 507: ntg3:driver->ops.scanDevice:0 ms
2023-04-21T12:45:45.007Z cpu2:2097340)ntg3:0000:18:00.0:Ntg3UplinkAssociate:1131:Associated with vmnic0
2023-04-21T12:45:45.007Z cpu46:2097954)Uplink: 12244: Device vmnic0 not yet opened
2023-04-21T12:45:45.007Z cpu46:2097954)Device: 395: uplink_drv:driver->ops.attachDevice :0 ms
2023-04-21T12:45:45.007Z cpu46:2097954)Device: 400: Found driver uplink_drv for device 0x461943084701d68c
2023-04-21T12:45:45.007Z cpu46:2097954)Uplink: 14658: Opening device vmnic0
2023-04-21T12:45:45.007Z cpu46:2097954)Uplink: 12269: The default queue id for vmnic0 is 0x44000.
2023-04-21T12:45:45.007Z cpu46:2097954)VMKAPIMOD: 110: failed to query port alias for Not found:
2023-04-21T12:45:45.007Z cpu46:2097954)Uplink: 12282: enabled port 0x82000002 with mac XX:XX:XX:XX:XX:XX
2023-04-21T12:45:45.007Z cpu46:2097954)Uplink: 2750: vmnic0, fast switch on uplink(0x82000002) connected to PPS is not supported
2023-04-21T12:45:45.007Z cpu2:2097340)ntg3:vmnic0:Ntg3UplinkStateSet:614:Setting uplink state to 0x1d
2023-04-21T12:45:45.007Z cpu2:2097340)ntg3:vmnic0:Ntg3UplinkStartIO:672:Ntg3UplinkStartIO

Environment

VMware vSphere ESXi 7.0.3

Cause

Link down state condition for Broadcom 1 GB cards is observed when patching to VMware image "VMware-VMvisor-Installer-7.0U3l-21424296" with driver VMW_bootbank_ntg3_4.1.9.0-4vmw.

Resolution

• This issue is fixed in ESXi 7.0 Update 3m release.
• ESXi 7.0 Update 3m provides ntg3 driver version 4.1.9.0-5vmw. The fix also adds a module parameter, fifoElastic, which you can enable in case of jumbo frame drops in certain Dell switches. To enable the parameter, use the following command:
  esxcli system module parameters set -p 'fifoElastic=1' -m ntg3

Workaround:

• None

 

Additional Information

Impact/Risks:

• Upon reboot after patch, the host is in a disconnected state if Broadcom 1 GB cards are used for the management network