[#875] kvm: how to enable "features: hidden state=on" ?

[Denny Fuchs]

hi,

since ProxMox uses not *.xml: how can I activate:

<domain type='kvm>
...
<features>
...
<kvm>
<hidden state='on'/>
</kvm>
</features>

I know that it is possible to use "args" in *.conf ... so I tried:

args: '-features hidden state=on'

but:

kvm: -features hidden state=on: invalid option

What I have done wrong?

cu denny

 

[wbumiller]

As far as I can tell this translates to passing 'kvm=off' to the -cpu parameter after the cpu type. Since we currently generate this parameter only from a list of valid cpu types and have no option to customize this particular parameter you cannot do this yet.
I added an entry to our tracker for this request: https://bugzilla.proxmox.com/show_bug.cgi?id=875

 

[Denny Fuchs]

hi Wolfgang,

thank you. We need it to get a software working for tests. Otherwise it fails, because of the hypervisor things.

One question: we need it only for one VM and one boot. Is it possible to create a *.xml and fire up the VM?

Update:

it seems, that I've got it working. I've just copied the whole CMD from qm showcmd <vmid>, and added "...-cpu ...., kvm=off". It seems, that the software doesn't recognize the hypervisor anymore

 

[futiless]

Hi can you get this to work temporarily ? I can't even try Proxmox with passthrough, regardless of making the settings stick. In ESXi this was the reason I couldn't get it to function. Does it make sense that I could somehow get my geforce 1080 to passthrough finally?

 

[Denny Fuchs]

hi futiless,

it works only, if you start the VM with the script. Also after every change on the VM, you have to use qm showcmd ... and change it again.

cu denny

 

[wbumiller]

I somehow missed your original bump in bugzilla, updated it now.

 

[futiless]

hi futiless,

it works only, if you start the VM with the script. Also after every change on the VM, you have to use qm showcmd ... and change it again.

cu denny

I am lost what qm smhowcmd is exactly and when to apply it?

 

[Denny Fuchs]

I am lost what qm smhowcmd is exactly and when to apply it?

qm showcmd just shows you, with which parameters the VM is configured.

 

[gerfloim]

@futiless
did you get this to work finally? I'm stuck at the same place: successfully managed to passthrough a Nvidia Quadro P400 to my Win10VM, but Driver hangs with Error 43

 

[futiless]

@futiless
did you get this to work finally? I'm stuck at the same place: successfully managed to passthrough a Nvidia Quadro P400 to my Win10VM, but Driver hangs with Error 43

Good news I think you are looking for the following entry ;

edit

/etc/pve/nodes/SS1/qemu-server/ (x).conf

where x is your vmid

add

args: -cpu host,kvm=off

this will kill internal virtualization function calls for making VM's within the vm. This to my knowledge fixes the glitch but introduces the limitation. Which annoys me because I know AMD products can with consumer grade and well ... you are apparently even having problems with a Quadro ?!

oh and disable power saving (ACPI) .. maybe you won't have the problem with a Quadro (fingers crossed for you) but most people can't use and power saving features if you have had to use this flag. Sort of a mixed blessing.

Anyways let me know if it works so that others can see the solution.

 

[gerfloim]

OK - I tried this, but without luck - still get error code 43 in device manager.

Here is my config:

xyz@pve:~# cat /etc/pve/qemu-server/100.conf
agent: 1
args: -cpu host,kvm=off
balloon: 2048
bios: ovmf
bootdisk: scsi0
cores: 4
cpu: host
efidisk0: local-zfs-vms:vm-100-disk-1,size=128K
hostpci0: 21:00,pcie=1
ide0: local-zfs-iso:iso/Windows_Server_2016_64b_deutsch.iso,media=cdrom,size=5782784K
ide2: local-zfs-iso:iso/virtio-win-0.1.141.iso,media=cdrom,size=309208K
kvm: 1
machine: q35
memory: 8096
name: NVTest
net0: virtio=CE:4A:22:D4:14:B5,bridge=vmbr0
numa: 1
ostype: win10
scsi0: local-zfs-ct:vm-100-disk-1,discard=on,iothread=1,size=32G
scsihw: virtio-scsi-pci
smbios1: uuid=c542b089-675c-4d9e-92b2-e5069bfdeada
sockets: 2

Any other ideas?
Maybe anybody has a patched NVIDIA driver to bypass NVIDIA kvm detection like described here?

 

[dcsapak]

if you also use the gpu to show the display output, you should set x-vga=on on the hostpci setting
this disables the virtual gpu, and sets the vendor and all necessary cpu flags (no need to specify them in args)

if you do not want this, you have to specify the all cpu flags yourself in the args

(just set the x-vga=on, look at the resulting command with qm showcmd <VMID> and copy the -cpu part to args)

 

[kariboo92]

Hi,
using Proxmox 7.1.12

I've configured the VM config file with:
cpu: host,hidden=1

The process is then run with "kvm=off":

-cpu host,hv_ipi,hv_relaxed,hv_reset,hv_runtime,hv_spinlocks=0x1fff,hv_stimer,hv_synic,hv_time,hv_vapic,hv_vpindex,kvm=off,+kvm_pv_eoi,+kvm_pv_unhalt

but Windows 10 is still reporting to running a virtual machine.

Edit: Found that using the following is working:
args: -cpu 'host,-hypervisor,kvm=off'