X window can't get display with some nvidia GPU driver errors from virtual machine boot

Graphics / Linux Linux
1

When I boot my QNX hypervisor ubuntu guest, nVidia GPU driver can’t display due to errors below:
NVRM: GPU 0000:00:03.0: RmInitAdapter failed! (0x26:0x65:1155)
NVRM: GPU 0000:00:03.0: rm_init_adapter failed, device minor number 0

ubuntu 18.04 with nvidia device driver has been installed in a hard disk, I can boot ubuntu with this hard disk as “Native” linux and got all display fine.
However if I boot the same hard disk’s ubuntu by a virtual machine loading from QNX hypervisor, I got the issue. If there are some special tips to use nvidia GPU driver from virtual machine?

After login to linux, I use SSH and run “nvidia-bug-report.sh” per link
https://devtalk.nvidia.com/default/topic/522835/linux/if-you-have-a-problem-please-read-this-first
and got result file “nvidia-bug-report.log.gz” . How to attached this result file with this report here?

2

You have to hide the hypervisor since running in a vm is not supported. Don’t know if QNX supports this.

3

So, nvidia GPU driver can’t be supported in any VM?

4

vmware esxi->supported with vgpu license, otherwise hypervisor can be hidden
xen->supported with vgpu license
kvm->hypervisor can be hidden
Edit: expanded this a bit.

5

Hi Generix,

“vmware esxi->supported with vgpu license, otherwise hypervisor can be hidden”
xen->supported with vgpu license
kvm->hypervisor can be hidden
Edit: expanded this a bit. "

—This means if this feature needs vgpu license to support it? Without the vgpu license, it does not work? What means “hypervisor can be hidden”?
To get it work, I have to contact to QNX or I have to contact to nvidia?

Thank you.

Ling

6

The nvidia supported way to pass through gpu to vms is vgpu. This works on Xen and vmware (kvm was in the works at RH, don’t know if that’s finished).
The unsupported way is to hide the hypervisor from the vm. This can be done with (at least) vmware and kvm.
Whether or not QNX can hide the hypervisor, I don’t know. A quick websearch points to ‘no’. You would have to ask QNX support for that.