After Upgrading to 8.0.9 Attached Screen Loses Signal (nvidia)

[FlashPan]

Hello,

I initially setup my PVE with 8.0.4 and everything worked fine. The attached screen would display the pre and full boot sequence (going past the blue grub screen).

Since I've updated to 8.0.9, after the blue grub screen the monitor goes into standby mode.

I know the card and monitor is ok as they both work before the PVE boot sequence. I can enter the bios etc and have also tried another screen and the same occurs past the blue grub screen.

Running lspci| grep -i vga I see:

02:00.0 VGA compatible controller: NVIDIA Corporation NV44 [Quadro NVS 285] (rev a1)

Am not using this gpu for passthrough, just there if I need to get screen access directly on the PVE.

All my VMs are displaying fine in their console mode.

Have read about setting nomodeset but this advice is working from the angle of the initial setup/install of PVE.

Do you think this is the value I need? If so how can I set this please?

Thank you in advance and cheers.

 

[Philipp Hufnagl]

Hello

Can you issue commands on the PVE? Can you give me the output of journalctl -b?

 

[FlashPan]

Hello Philipp,

Thank you for your response.

Here is the output you wanted.

https://pastebin.com/G8uLgP4A

I can access the webgui fine and everything seems to be working as normal. Just teh external attached screen output on the PVE host.

Thanks

 

[Philipp Hufnagl]

Hmmm... there are a lot of hardware errors in there. Have you checked if there is a BIOS update available for you?

 

[FlashPan]

Hi Philipp, I just noticed a lot of errors when runnign the journal command. When I set this box up some months ago I was on the latest bios. Found now a newer/latest bios version now and installed it.

Still getting the same screen issue. Once I get past the blue grub screen, the monitor I can see will go blank when it is say it is loading the ram disk.

Here is another output of the journal command.

https://pastebin.com/EtECpz1n

Am still seeing errors in this outpout but overall everthing seems to work fine. CPU is what came with the Z640. The only extras I've added is the raid and nvidia card.

Thanks and cheers

 

[Philipp Hufnagl]

That is strange. Once your kernel boots, it reports that it does not find the monitor.

Since you have still installed older kernel, can you try if the problem still occurs with an older one?

 

[FlashPan]

Sure, I'll give that a go. Will disconnect all my drives, boot and storage and then install my original iso I downloaded some time ago to another drive. Will take me a little time so I may not be able to come back to report today but will try.

I do have a thought. Long shot but I did run this box headless for a while and installed some updates from the webgui. Could that have had an impact?

Thanks for your help and attention I do appreciate it. (and appreciate the possible timezone difference between us both).

 

[Philipp Hufnagl]

I forgott before: Could you send me the output of dmidecode -t processor bios system.

I do have a thought. Long shot but I did run this box headless for a while and installed some updates from the webgui. Could that have had an impact?

I don't think so. This seems more like your system does not know how to speak with the monitor. It is more likely to be the kernel, the graphics driver or faulty hardware.

Sure, I'll give that a go. Will disconnect all my drives, boot and storage and then install my original iso I downloaded some time ago to another drive. Will take me a little time so I may not be able to come back to report today but will try.

I meant more, something like selecting another kernel in the blue grub menu or installing an older kernel. Reinstalling Proxmox totally sounds like a lot of work. But if you choose to do this, make sure to back up important data, specially in /home and /etc before!

 

[FlashPan]

Ahh I see, Sorry this is still quite a new realm for me.

In the grub menu I could change to 6.2.16-3.pve (changing from 6.2.16-19-pve)

Screen displayed little further getting past the ram disk message but only a few lines and then the screen went blank again.

I've got quite a bit of kit to move to be able to extract the HP Z640. So was thinking to remove the raid card and drives, then attempt to install my original iso on a single drive to test. Keeping my existing raid/installation intact to possibly revert to.

I may be able to dig out another nvidia card (think the same type) to test as well.

Being on 6.2.16-3.pve here is the output for dmidecode -t processor bios system

https://pastebin.com/Hzza05dy

Thank you

 

[FlashPan]

Positive-ish progress

For giggles. I connected a slightly newer 3rd monitor and it is displaying fine. I can see the logon prompt (this was after reverting to 6.2.16-19-pve).

Am thinking about refresh rates now on the monitor(s) but having 2 not work I think is a bit rare. And then also working fine until pve tries to boot.

 

[FlashPan]

Just to make it more confusing. Connect to the 3rd monitor, boot pve and everything loads/displays fine. Disconnect the monitor cable from that, connect to the "problem" 2nd monitor and I can see the logon screen fine. Reboot PVE and the "problem" monitor is blank again as before (still seeing the preboot screens as before). I see no refresh rates/settings on the OSD menu on the problem monitor.

You think we put this down to a problem monitor?

 

[Philipp Hufnagl]

The line that strikes me the most in your journal is the following:

s1pve01 kernel: nouveau 0000:02:00.0: [drm] Cannot find any crtc or sizes

It appears several times. It's basically the graphics driver telling you the kernel told it, it does not understand how to talk with the display.

 

[FlashPan]

Thank you Philipp,

So am still trying ot get my head around if this is a monitor problem, graphics card, driver or kernel problem.

The monitor works fine on other OSs/machines I connect it to. I even connected the PVE to a KVM switch with other machines, they picked up the monitor fine but the PVE still not past the grub screen.

Am not 100% sure it is a graphics card issue.

Is their a mechanism to reinstall the nouveau driver? Or is that not worth it if the kernel cannot see the monitor? As I take it the kernel sees the monitor then passes that to the driver?

So then it could be a kernel issue where it just does not like/see my monitor? (so for the time being am stuffed).

I think if we cannto move much more on this, then time for a different monitor. As my tests show another monitor working.

Thanks and cheers

 

[Philipp Hufnagl]

When you google the issue, you can find that the exact same message is passed along by completely other graphic drivers (for example, amdgpu). This does not strictly rue out the graphic driver, but I think a kernel bug is more likely. Also, the way you described the error, it occurs the exact moment when your System is booting the kernel.

Have you tried it with different Kernels?

Is their a mechanism to reinstall the nouveau driver? Or is that not worth it if the kernel cannot see the monitor? As I take it the kernel sees the monitor then passes that to the driver?

Apt has a parameter called "reinstall" as in apt reinstall .... While this should not break anything, I would recommend making a backup before anyway.

 

[FlashPan]

Hello, yep I tried it with the previous kernel 6.2.16-3.pve and still the same. Day job getting in the way now so not been able to test with a fresh install on an earlier version yet.

Will look/research at the apt reinstall option (thanks for that).

 

[wings]

I'm tearing my hair out because I just ran into this same issue I think.

I've got the following specs:
* Ryzen 7800X3D
* NVIDIA RTX 3070 Ti
* 165Hz ultrawide 3440x1440 monitor.

I've tried connecting an HDMI 4K monitor (it's all I have other than this one ) and no dice there either... basically whatever I do I get stuck at "Loading initial ramdisk ..."

I believe the system is working and booting, but no real way to tell... originally I tried doing this by installing Debian Bookworm and then installing Proxmox on top per the wiki, with the same results, and journalctl indicated exactly this - that it was booting but couldn't see a display.

A fresh install of 8.1 has resulted in the same issue. Any clues would be very appreciated!

EDIT: I managed to get into the machine by connecting directly to onboard video and forcing the BIOS to output to integrated graphics. A bit of hacking later I should have SSH working and a static IP set, and will be able to boot the machine "normally" and start debugging...

 

[FlashPan]

Glad you got it working

This morning I had some updates to install and that included 6.5.11-4 kernal. Now on PVE 8.1.3 and still the same issue for me.

I think I read somewhere that PVE does not like RTX cards?

I can still access via the webgui.

For my box I do not have have an integrated GPU so I have to use a card.

Now that you changed to internal gpu is interesting but frustrating for me as it works fine on the other monitor I have.

 

[wings]

Glad you got it working

This morning I had some updates to install and that included 6.5.11-4 kernal. Now on PVE 8.1.3 and still the same issue for me.

I think I read somewhere that PVE does not like RTX cards?

I can still access via the webgui.

For my box I do not have have an integrated GPU so I have to use a card.

Now that you changed to internal gpu is interesting but frustrating for me as it works fine on the other monitor I have.

Try this (ASSUMING you are running the proprietary NVIDIA driver here, btw)

wings@blackberry:/etc/modprobe.d$ sudo cat blacklist-nvidia-nouveau.conf

blacklist nouveau
blacklist amdgpu
options nouveau modeset=0

wings@blackberry:/etc/modprobe.d$ sudo cat nvidia_drm.conf
options nvidia_drm modeset=1

Put something like that in modprobe and try rebooting.

You'll also need to do `update-initramfs -u -k all` before you reboot, I think.

 

[Reymouth]

@wings I encountered the same malfunction with my KVM and I fixed it temporarily by switching back to proxmox-kernel-helper in 8.0.4, this malfunction appeared in 8.0.5

 

[Philipp Hufnagl]

Hello

This morning I had some updates to install and that included 6.5.11-4 kernal. Now on PVE 8.1.3 and still the same issue for me.

Can you please give me a new journal anyway in case something changed?
journalctl --since '2023-11-25' > $(hostname)-journal.txt