Can't get LibVF to pick up after reboot

[arc-compute]

Can't get LibVF to pick up after reboot

[arthurrasmusson]

Hey @jon-bit. Thanks for pointing this out - I'll try my best to help you with whatever is going on here.

Can you start by running ./scripts/generate-debug-information.sh then dump the resulting ./logs/debug.log for me?

Thanks.

[jon-bit]

I ran it again because I rebooted and saw this

Failed to disable unit: Unit file nvidia-vgpud.service does not exist.
Failed to stop nvidia-vgpud.service: Unit nvidia-vgpud.service not loaded.

but if that log is needed Here it is

https://pastebin.com/PChqHd3u

[arthurrasmusson]

Delete your debug.log file then run sudo systemctl restart gvm-post.service after reboot. If you can re-run the debug log script and post it after doing that I'll take a look.

I suspect this is your issue:
Arc-Compute/GVM-user#3

You can also reach me in the Open-IOV Community Group Chat - I'll probably be able to help you more quickly in there:
https://discord.gg/Rb9K9DYxKK

[jon-bit]

I ran the restart on gvm and got

Failed to restart gvm-post.servic.service: Unit gvm-post.servic.service not found.

Here is the output of debug.log

https://pastebin.com/tw3fsWa8

[mbuchel]

you typed in the restart command wrong, try just:

sudo systemctl restart gvm-post

it should autocomplete for you

[mbuchel]

also you have a 3070, the support for ampere is still super experimental

[jon-bit]

OK I just saw this in the logs

Device Check Test succeeded
Device Major Number Identifier Null succeeded
Device Major Number Identifier succeeded
DeviceFileMode Nvidia Check succeeded
Open /dev/nvidiactl File succeeded
Open /dev/nvidia0 File succeeded
All tests succeeded
# GVM-user TEST-DEVICE END
# GVM-user TEST-NVIDIA-API START
RM Version Check Ok: Ensure your version check is correctly: 510.47.03
RM Version Check Incorrect: Invalid version is reported correctly, please check your driver version.
RM Version Check Ignore succeeded
RM Alloc Root Ok succeeded
RM Alloc Root Fail (Invalid FD) succeeded
RM Alloc Root Fail (Bad FD) succeeded
RM Alloc Root Fail (Bad Argument) succeeded
RM Free Root Ok succeeded
RM Free Root Fail (Invalid FD) succeeded
RM Free Root Fail (Double Deallocate) succeeded
Get Probed Ids succeeded
2/11 tests failed
# GVM-user TEST-NVIDIA-API END
# GVM-user TEST-NVIDIA-MANAGER START
Created gpu: 0x00000100 (0x10DE, 0x2482, 0x10DE, 0x5052)
Failed RM Control Mechanism:
	client: 0xC1D00046
	object: 0xD0014603
	cmd: 0xA0810101
	flags: 0x00000000
	params: 0x7fff9a52c380
	size: 0x000011C0
	status: 0x0000003A
Destroyed gpu: 0x00000100 (0x10DE, 0x2482, 0x10DE, 0x5052)
Create MDevs succeeded
All tests succeeded
# GVM-user TEST-NVIDIA-MANAGER END
# UNAME START
Linux fedora 6.0.18-300.fc37.x86_64 #1 SMP PREEMPT_DYNAMIC Sat Jan 7 17:10:00 UTC 2023 x86_64 GNU/Linux
# UNAME END
# CPUINFO START

[mbuchel]

you are using 525.60, the libvf.io scripts may not have the update to support this yet, but it is kindof moot as vgpu unlock does not support 3070, the gvm suite will but it is not yet there. feel free to join the open-iov discord.

[jon-bit]

Sorry but I don't have discord. Regardless how do we fix this?

[mbuchel]

you can install gvm-user utils and select nvidia/525.60 branch to compile it

https://github.com/Open-IOV/GVM-user/tree/nvidia/525

the issue will be handling the vgpu unlock for 3070 series, which is much more complex

[arthurrasmusson]

@jon-bit See GPU Support on Open-IOV:
https://open-iov.org/index.php/GPU_Support

[arthurrasmusson]

@jon-bit as this issue appears to originate from an unsupported device I'm going close this thread for now.

[jon-bit]

I'll respect the fact that this is not a supported card but do you know of any other software that could help with GPU virtualization?