Memory Corruption by SRIOV VFs of Emulex OneConnect Network Adapters

Products

• VMware vSphere ESXi

Issue/Introduction

Symptoms:

SRIOV VFs (Virtual Function) of Emulex OneConnect Network Adapters may cause host memory corruption under the following conditions:

1. A Virtual Machine is using one or more SRIOV VFs of Emulex OneConnect Network Adapters, and

2. The Virtual Machine is using two or more VCPUs, and

3. The Virtual Machine is powered off abruptly, instead of shutdown gracefully using "guest shutdown" or equivalent.

The issue is observed for Emulex OCe12xxx and OCe14xxx series of Ethernet and Converged Network Adapters, including Emulex and OEM branded models. The memory corruptions occurs during the abrupt power-off of the Virtual Machine. The host could malfunction or crash as a result of the memory corruption. It affects various versions of ESXi and guest OS.

Environment

VMware vSphere ESXi 6.5
VMware ESXi 6.7.x
VMware vSphere ESXi 7.0

Cause

The issue is caused by bad DMA writes to host memory by the OneConnect network adapter.

Resolution

Currently there is no resolution. Future firmware updates of the OneConnect network adapters may resolve the issue.

Workaround:

The memory corruption can be prevented by removing any of the three necessary conditions mentioned above.