vCenter UI hangs during VMFS extension on hosts utilizing Marvell/QLogic qed driver

Products

• VMware vSphere ESXi

Issue/Introduction

 

• The vCenter Server user interface (UI) becomes unresponsive or "stuck" when attempting to open the "Increase Datastore Capacity" or "Extend VMFS" wizard.

• Storage inventory tasks, such as "Retrieve disk information" or "Query storage provider," fail to complete or remain in a "Running" state indefinitely.

• The ESXi host vmkernel.log or vmkalert.log contains the following error: ALERT: [qed_l12_start_xmit:532(qedfc0)]Failed to prepare tx packet with status = Busy

• Direct management of the storage stack via esxcli storage commands may result in a timeout or hang.

 

Environment

ESXi 8.0.3

Cause

This issue is caused by a heap memory exhaustion within the Marvell/QLogic qed driver suite. When the driver's internal transmit (Tx) buffers or heap memory reaches capacity, the HBA is unable to process further I/O requests or SCSI management commands. This state prevents the ESXi host's storage management service (hostd) from returning storage metadata to vCenter Server, leading to a UI deadlock in the vSphere Client.

Resolution

To permanently resolve this issue, perform the following:

1. Schedule Maintenance: Evacuate all Virtual Machines from the affected ESXi host using vMotion.

2. Reboot: Perform a cold boot of the ESXi host to clear the driver heap state.

3. Update Driver/Firmware: Ensure the Marvell/QLogic driver and firmware are updated to the latest version listed on the VMware Compatibility Guide (VCG) and aligned with the VMware Cloud Foundation (VCF) Bill of Materials (BOM) if applicable. Specifically, ensure the qed driver is at a version containing fixes for memory management and Tx queue handling.

Workaround If an immediate extension of the VMFS datastore is required before a reboot can be scheduled:

1. Bypass the vCenter Server inventory aggregation by logging directly into the ESXi Host Client (https://<ESXi_IP>/ui).

2. Navigate to Storage > Datastores.

3. Select the target datastore and choose Increase Capacity.

4. Proceed with the extension via the local host interface. Note: This may still fail if the driver stack is completely unresponsive, in which case a reboot is the only viable path.