ESXi Host Reboots When Powering On VM with PCI Passthrough (Marvell/QLogic FC HBA)

Products

• VMware vSphere ESXi

Issue/Introduction

• The ESXi host unexpectedly reboot / PSOD when powering on a virtual machine configured with a Marvell/QLogic Fibre Channel (FC) HBA attached via PCI passthrough (VMDirectPath I/O).
  
  • Disruption of running workloads
    
    
• Potential service downtime during VM lifecycle operations

  • VM power-on may appear to hang (e.g., stall at ~88% progress in vCenter) immediately before the host reboots.

• Affected FC adapters: Marvell/QLogic 16/32/64Gb FC HBAs (PCI Vendor ID 1077).
  
  
• Observed on platforms such as HPE ProLiant DL380 Gen10 Plus.
  
  
• vmkernel.log located at ESXi /var/run/log/ shows the device failing to recover after a PCIe reset, e.g.:
  • WARNING: PCI: 743: Dev <SBDF> is unresponsive after reset
  • PCIPassthru: 5852: <SBDF> :Reset for device failed with Failure
  • PCI: 1404: Skipping device reset on <SBDF> because PCIe link to the device is down

Environment

• VMware vSphere ESXi 7.x
• VMware vSphere ESXi 8.x
• VMware vSphere ESXi 9.0.x
• VMware vSphere ESXi 9.1.1

Cause

These Marvell/QLogic FC adapters do not support Function Level Reset (FLR). By default, ESXi falls back to resetting such devices via a secondary bus (bridge) reset when the device is claimed for passthrough. On some platforms, this bridge reset is not handled cleanly by the device/platform, leaving the PCIe device unresponsive and, in this case, causing a host PSOD/reboot during passthrough device initialisation.

These adapters do support the D3D0 power-state reset method (D0 → D3hot → D0), which works reliably as an alternative to bridge reset.

Resolution

Broadcom Engineering is aware of the issue and is working on a fix in an upcoming release

 

Workaround

Configure ESXi to use the D3D0 reset method instead of the default bridge reset for the affected device IDs, via /etc/vmware/passthru.map.

1. Identify the PCI vendor/device ID of the passthrough adapter:
      esxcli hardware pci list
      Look for vendorID and deviceID for all the Marvel/QLogic FC HBAs.
   
   
2. Edit /etc/vmware/passthru.map and add entries for the affected devices
    
   Note: Qlogic/Marvell 16/32/64Gb FC card cannot be shared

   1077  2281  d3d0  false
1077  2261  d3d0  false

3. (Vendor ID 1077 = Marvell/QLogic; device IDs listed are the confirmed affected 16/32/64Gb FC HBA models. If a different device ID is seen for the same adapter family based on esxcli command output from step 1, add it in the same <vendor> <device> d3d0 false format.)
   
   
4. Reboot the host for the change to take effect.

Persistence note: /etc/vmware/passthru.map is a sticky-bit configuration file — changes persist across reboots, but only if the host is shut down cleanly, or backup.sh 0 has run before the next reboot (it runs periodically on its own).

Verification:

• After reboot, confirm the reset method for the device:
      esxcli hardware pci list 
  
  
• Check the Reset Method field reflects the change, then power on the VM with the passthrough device attached and confirm the host no longer reboots.