Fighting error 43 – Nvidia GPU in a virtual machine

The infamous error 43 is triggered by the Nvidia Windows driver, in case it recognizes the GPU is used inside a virtual machine with GPU passthrough. This means the driver won’t load correctly during the system startup leaving you with cumbersome 800 by 600 display resolution.

An examination of the Windows Device Manager will unveil the culprit – Code 43

Windows has stopped this device because it has reported problems. (Code 43)

Luckily, KVM/QEMU is able to hide the fact that the Nvidia driver is running a virtual environment.

The tested system setup

This method has been tested for the following versions

Host:

• OS: Ubuntu 18.04, as well as a 16.04
• Kernel: 4.15 – 5.3.6
• Hypervisor: QEMU 2.12.0 up to 4.1
• Libvirt: version 4.6.0 up to 4.7.0

One can use virsh version in order to display the current version status.

Guest:

• OS: Windows 10 1803 – 1903
• VM chipset: i440FX, Q35
• Nvidia driver: 399.07 – 441.12
• GPU: Geforce GTX 970 and Geforce GTX 1060

Fixing error 43

In order to hide that the system is running on a hypervisor, KVM supports the “hidden” flag.

Depending on whether you call QEMU directly or call it via Libvirt the flag syntax differs.

Set KVM hidden via Libvirt xml

In order edit the virtual machines configuration use:

virsh edit your-windows-vm-name

Once your done editing, you can use CTRL+x CTRL+y to save the changes.

For QEMU 3.0 and later the required flags are “vendor_id” in the hyperv section, and “hidden” in the kvm section, of the features block, of the virtual machines xml definiton. I think it is recommended to use a 12 character value (see excerpt marked green).

Additionally, in case you are using QEMU 4.0 (or higher) in combination with a Q35 chip, the flag ‘ioapic driver='kvm'‘ needs to be added in the features section (see excerpt marked blue).

Excerpt from libvirt xml file

  ...
  <features>
    <acpi/>
    <apic/>
    <hyperv>
      <relaxed state='on'/>
      <vapic state='on'/>
      <spinlocks state='on' retries='8191'/>
      <vendor_id state='on' value='1234567890ab'/>
    </hyperv>
    <kvm>
      <hidden state='on'/>
    </kvm>
    <vmport state='off'/>
    <ioapic driver='kvm'/>
  </features>
  ...

Set KVM hidden for direct QEMU usage

In case you call QEMU directly you have to add the flags ‘kvm=off‘ and ‘vendor_id=null‘ on your cpu argument (see excerpt marked green).

Additionally, in case you are using QEMU 4.0 (or higher) in combination with a Q35 chip, use the machine argument ‘kernel_irqchip=on‘.

Example for virtual machine with host-passthrough set as CPU model and Q35 as machine type:

...
-cpu 'host,kvm=off,hv_vendor_id=null' \
-machine 'type=q35,kernel_irqchip=on' \
...

Using QEMU command arguments in the libvirt XML definition

Just for the sake of completeness (and historical compatibility of this article), there is a third way of enabling the hypervisor spoofing. Actually it is a combination of the two described earlier, as it is possible to add QEMU arguments to the libvirt xml defintion file.

These arguments are called after the definition has been parsed, and usually overwrite the settings made via XML structure. Thus it is not recommended to use both ways for one setting in the same definition file.

First of all find the very first line, which should read:

<domain type='kvm'>

and replace it with:

<domain type='kvm' xmlns:qemu='http://libvirt.org/schemas/domain/qemu/1.0'>

Find the line which ends with </devices>
and add the following block afterwards (or alter the block if it already exists):

   ...
   </devices>
   <qemu:commandline> 
     <qemu:arg value='-cpu'/> 
     <qemu:arg value='host,kvm=off,hv_vendor_id=null'/>
     <qemu:arg value='-machine'/> 
     <qemu:arg value='q35,kernel_irqchip=on'/> 
   </qemu:commandline>
</domain>

Attention: Make sure <devices> and <qemu:commandline> have the same indent.

Thats it – these settings work for me.

Sources

As usual, the glorious arch wiki has further information and help; i.e. it recommends using a 12-character alphanumeric string as vendor ID.

Post updates

21.09.2019 – added further information regarding qemu 4.0 and error 43

07.11.2019 – updated the article structure