PCIe NIC passthrough

[brucexx]

I am trying to pass through a NIC to one of my VMs. I have one NIC with 4 ports and 1 nic with 2 port which I want to pass through to a VM.

DMAR: IOMMU enabled
DMAR-IR: Enabled IRQ remapping in x2apic mode
DMAR: Intel(R) Virtualization Technology for Directed I/O

I verified my hardware supports it, I have a dell server based on E5 Xeon v2, I also checked the I have the iommu groups isolated, meaning that the two ports card is in the same group (and nothing else is in this group).

I followed the instruction online listing the lspci and lspci -t and updated the VMs config accordingly.

machine: q35
hostpci0: 04:00.0,pcie=1
hostpci1: 04:00.1,pcie=1

where 04:00:0 and 04:00:1 are the PCIe ports of the same NIC.

I can see the card (with 2 ports) passed through to the VM when issued lspci at the VMs terminal:
01:00:0 - Ethernet Controller Broadcom Corporation BCM5720
02:00:0 - Ethernet Controller Broadcom Corporation BCM5720

..but when I try to bring the interfaces up and configure the network I get kernel panic with not syncing, fatal hardware error.

Does anybody have any suggestions of how to fix it or troubleshoot further ?

Thank you

 

[NIPE-SYSTEMS]

Try the following (the second thing in the section of the link):

https://pve.proxmox.com/wiki/Pci_pa...our_PCI_card_address.2C_and_configure_your_VM

This quote is taken from the wiki link. It describes how to define a PCI-device:

If you have a multi-function device (like a vga card with embedded audio chipset), you can pass all functions with

hostpci0: 01:00.0;01:00.1

According to this paragraph in the wiki, I would guess that this should also work for PCIe-devices. Furthermore you should include the argument

pcie=1

into the line in your config (to tell Proxmox that this device is a PCIe-device). In your example I think the following oneliner should work:

hostpci0: 04:00.0,04:00.1,pcie=1

Hope that helps.

 

[brucexx]

I tried that with no luck in. Thank you for suggestion tho... I went with a physical server for this machine, the amount of time troubleshooting this was not worth it. One day when the feature becomes easier to implement I will be glad to switch to VM again.