NSX-T Bare metal Edge node cannot claim datapath interfaces

Products

VMware NSX

Issue/Introduction

Expected fastpath interfaces (fp-ethx) are not visible using the get interfaces command.
In the /var/log/syslog file, you see entries similar to:

<30>1 YYYY-MM-DDTHH:MM:SS.204651+00:00 nsx-edge datapathd 9632 - - EAL:   probe driver: 8086:1528 rte_ixgbe_pmd
<4>1 YYYY-MM-DDTHH:MM:SS.204678+00:00 nsx-edge kernel - - - [ 1938.704675] vfio-pci 0000:04:00.0: Device is ineligible for IOMMU domain attach due to platform RMRR requirement. Contact your platform vendor.     <<<--------
<30>1 YYYY-MM-DDTHH:MM:SS.205030+00:00 nsx-edge datapathd 9632 - - EAL:   set IOMMU type 1 (Type 1) failed, error 1 (Operation not permitted)
<30>1 YYYY-MM-DDTHH:MM:SS.205075+00:00 nsx-edge datapathd 9632 - - EAL:   set IOMMU type 8 (No-IOMMU) failed, error 19 (No such device)
<30>1 YYYY-MM-DDTHH:MM:SS.205116+00:00 nsx-edge datapathd 9632 - - EAL:   0000:04:00.0 failed to select IOMMU type
<30>1 YYYY-MM-DDTHH:MM:SS.205157+00:00 nsx-edge datapathd 9632 - - EAL: Error - exiting with code: 1
<30>1 YYYY-MM-DDTHH:MM:SS.205197+00:00 nsx-edge datapathd 9632 - -    Cause: Requested device 0000:04:00.0 cannot be used
<31>1 YYYY-MM-DDTHH:MM:SS.205386+00:00 nsx-edge datapathd 9636 - - EAL: lcore 2 is ready (tid=6240a700;cpuset=[2])
<31>1 YYYY-MM-DDTHH:MM:SS.205500+00:00 nsx-edge datapathd 9636 - - EAL: lcore 7 is ready (tid=5fc05700;cpuset=[7])
<31>1 YYYY-MM-DDTHH:MM:SS.205605+00:00 nsx-edge datapathd 9636 - - EAL: lcore 4 is ready (tid=61408700;cpuset=[4])
<31>1 YYYY-MM-DDTHH:MM:SS.205710+00:00 nsx-edge datapathd 9636 - - EAL: PCI device 0000:04:00.0 on NUMA socket 0
<31>1 YYYY-MM-DDTHH:MM:SS.205812+00:00 nsx-edge datapathd 9636 - - EAL:   probe driver: 8086:1528 rte_ixgbe_pmd
<31>1 YYYY-MM-DDTHH:MM:SS.205913+00:00 nsx-edge datapathd 9636 - - EAL:   set IOMMU type 1 (Type 1) failed, error 1 (Operation not permitted)
<31>1 YYYY-MM-DDTHH:MM:SS.206015+00:00 nsx-edge datapathd 9636 - - EAL:   set IOMMU type 8 (No-IOMMU) failed, error 19 (No such device)
<27>1 YYYY-MM-DDTHH:MM:SS.206116+00:00 nsx-edge datapathd 9636 - - EAL:   0000:04:00.0 failed to select IOMMU type
<26>1 YYYY-MM-DDTHH:MM:SS.206258+00:00 nsx-edge datapathd 9636 - - EAL: Error - exiting with code: 1#012 Cause:
<26>1 YYYY-MM-DDTHH:MM:SS.206356+00:00 nsx-edge datapathd 9636 - - Requested device 0000:04:00.0 cannot be used
<29>1 YYYY-MM-DDTHH:MM:SS.207573+00:00 nsx-edge systemd 1 - - nsx-edge-datapath.service: Main process exited, code=exited, status=1/FAILURE
<30>1 YYYY-MM-DDTHH:MM:SS.227255+00:00 nsx-edge datapath-systemd-helper 9663 - - # Stopping at YYYY-MM-DDTHH:MM:SS.227098034Z

Note: The preceding log excerpts are only examples. Date, time, and environmental variables may vary depending on your environment.

Environment

VMware NSX

Cause

This issue occurs because some server vendors use Reserved Memory Range Reporting (RMRR) for health check and direct access to some hardware.
This prevents DPDK to claim the server physical NICs. This issue is known to affect at least the HP Proliant server series.

Resolution

Because RMRR is reliant on VT-d extensions being enabled, disabling VT-d in the BIOS allows the datapath to claim the expected interfaces normally. Disabling RMRR for specific interfaces differs depending on the platform and can be convoluted. Disabling VT-d is a simpler approach applicable to the NSX Edge.

To resolve this issue, disable VT-d in the BIOS.

Disclaimer: VMware is not responsible for the accuracy or content of third-party websites, and inclusion of such links does not imply endorsement. Please consult the vendor before making the changes below to ensure there is no impact on server functionality.

To disable on VT-D on HPE ProLiant GEN 10:

From the RBSU > System Utilties > System Configuration > BIOS/Platform Configuration (RBSU) > Virtualization Options.
Select Intel(R) VT-D to “Disabled”.

To Disable VT-d on an HPE ProLiant Gen 9 Server:

RBUS > System Options > Virtualization Options.
Select Intel(R) VT-d to “Disabled”.

Additional Information

Because RMRR is reliant on VT-d extensions being enabled, disabling VT-d in the BIOS allows the datapath to claim the expected interfaces normally. For more information, see Advisory: (Revision) Linux - KVM Virtual Machine Does Not Start and Displays "Device is ineligible for IOMMU domain attach due to platform RMRR requirement" Message on HPE ProLiant Systems with an Intel Processor.

This article describes the exact same behavior when KVM is used to try to enable PCI passthrough on a NIC on specific HPE ProLiant Servers. The article also provides information on how to disable RMRR for specific interfaces, but they differ for different platforms and can be convoluted. Disabling VT-d is a simpler approach applicable to the NSX Edge.