"Unknown video mode" on fresh Proxmox 7 install

[kyriazis]

(for @t.lamprecht)

Installed Proxmox 7 on two separate machines (one Haswell-based server platform [upgrade], and the one Broadwell-based server platform [new install]), both machines connected to a Raritan KVM with VGA dongle.

Install was uneventful in both instances, boot up sequence shows no errors, but when I reach the login screen, I get an "Unknown video mode" message on the screen.

No suspicious messages in dmesg, /var/log/syslog, /var/log/debug. One thing that *could* cause as issue is that Raritan VGA dongles seem to have a bad EDID programmed in them, but that never caused an issue with previous Proxmox versions, or any other OS for that matter.

syslog excerpt showing bad edid:

Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410227] mgag200 0000:07:00.0: VGA-1: EDID is invalid:
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410230]       [00] BAD  00 ff ff ff ff ff ff 00 ff ff ff ff ff ff ff ff
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410230]       [00] BAD  ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410231]       [00] BAD  ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410231]       [00] BAD  ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410232]       [00] BAD  ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410232]       [00] BAD  ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410233]       [00] BAD  ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.410233]       [00] BAD  ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff ff
Jul 14 10:04:21 vis-hsw-00 kernel: [   13.411291] fbcon: mgag200drmfb (fb0) is primary device

Attaching output of lspci -knn for both machines.

Thank you!

George

 

[t.lamprecht]

No suspicious messages in dmesg, /var/log/syslog, /var/log/debug. One thing that *could* cause as issue is that Raritan VGA dongles seem to have a bad EDID programmed in them, but that never caused an issue with previous Proxmox versions, or any other OS for that matter.

That really seems suspicious. Maybe the nerwer module of the newer kernel got a bit more strict here?
What other distros/OS did you try?
And, IIRC, you mentioned that you used the Proxmox VE 7.0 ISO to install this one, and it worked just fine?

 

[kyriazis]

On same type of machines, I've installed CentOS 6/7/8, Ubuntu 14/16/18, Windows 10, and Proxmox starting from 5.x upgraded regularly.

Proxmox 7 had the same behavior no matter if it was installed via ISO or update.

I've also checked all the relevant /sys/class/{drm,graphics,vtconsole} files and compared between a working Proxmox 6 and the above Proxmox 7 systems and they show the same information. Specifically, /sys/class/graphics/fb0/modes and /sys/class/drm/card0-VGA-1/modes show the same modes. From the first file is looks like it programmed 1024x768:

# more /sys/class/graphics/fb0/modes
U:1024x768p-0

Again, the contents of this file are the same between Proxmox 6 & 7.

We have a cluster of 27 systems (and growing), so at this point I am stopping the upgrade until I find a solution to this. Half of the systems are using the abovementioned Raritan VGA dongles, and the other half have no monitor connected and managed via web BMC console. I will also try to upgrade one with a BMC console and see how it behaves.

I will be fine if there is a "workaround", ie. change a configuration file to force a particular video mode.

Thank you!

George

 

[t.lamprecht]

On same type of machines, I've installed CentOS 6/7/8, Ubuntu 14/16/18, Windows 10, and Proxmox starting from 5.x upgraded regularly.

So all rather "old" stuff, relatively speaking. Could be interesting to see if the same happens with a plain Debian 11 Bullseye installation (that release uses the 5.10 kernel, so only a single release behind PVE 7.0's 5.11 one) - that could help to rule a few things out.

I will be fine if there is a "workaround", ie. change a configuration file to force a particular video mode.

Did you try things like adding "nomodeset" to the kernel parameter?

The mgag200 module your server are using did not see much change between the 5.4 Kernel of Proxmox VE 6.4 and the 5.11 kernel of Proxmox VE 5.11. Albeit, they transformed it to also accept the older desktop Matrox graphic cards in 5.8, so maybe this is some fallout from there, but as mga g200 usage isn't exactly rare, it would surprise me a bit if it went broken unoticed for that long, maybe really just in combination with that KVM dongle.

Can you by chance also test a "normal" monitor to see if the mgag200 module/HW itself would work?

 

[fuzzyduck]

i have this as well. I dist upgraded to 7 and then got a failed rpool import. Then i manually added and booted. Blank screen
then after many retries i booted the fresh iso and installed 7. Then after grub again a blank screen.
i even switched monitors on both occasions....

 

[fuzzyduck]

t. lamprecht, you where helping me on the rpool import in another topic but i think this is it! exit initramfs after the import gave me a blank screen. another monitor was telling me wrong timings from the vga and couldnt show a image.

 

[t.lamprecht]

Are you sure that your host also uses the mgag200 module and even a Raritan KVM dongle?

 

[fuzzyduck]

no its just de VGA onboard on an Intel S2400SC. super basic thing.

Someone from the discord said:

​

`I installed 7 and got output on a 1024x768 monitor after install (during I had to use a 1280x1024)
had to restart after switching monitor ofc.`

 

[fuzzyduck]

im booting 6.4.1 iso to see if i got a screen after installing.
here i am stuck in limbo. freshly formatted the raid1 install 7 fresh.... no screen

ill report here to see if i got screen after reverting to a fresh 6.4.1

i bet my whole import rpool topic can go to the bin after this. this is so gonna be an vga issue

 

[fuzzyduck]

i installed 6.4.1 and i got prompt. So its something in 7 that doesnt give me vga after a succesful install of 7.
well im cured for now. gonna redo my whole setup. ill keep following this and my topic though for updates and i case you need me

 

[t.lamprecht]

Hmm, thanks for reporting back.

What you could try is installing the opt-in 5.11 kernel for 6.4 to see if that's all what's needed to break the console.

apt install pve-kernel-5.11

If so you can chose the default 5.4 one on next boot and deinstall the 5.11 kernel againapt purge pve-kernel-5.11; apt autoremove , if not we know at least that it isn't the kernel.

 

[fuzzyduck]

im not going to install a kernel when i dont have a prompt to go back to a previous one? Im absolutely bummed out i have to redo EVERYTHING

 

[t.lamprecht]

im not going to install a kernel when i dont have a prompt to go back to a previous one? Im absolutely bummed out i have to redo EVERYTHING

As said, the initial boot menu (either the black or blue thing) allows to choose the kernel, the newest will be selected by default, but you can go back to the older one. That prompt is before any kernel even can run, so it really should not be affected.

 

[kyriazis]

setting "nomodeset", or for that matter setting "vga=0x0317" work around the issue. Thanks for the hint.

I will plugin monitors next time I get in the lab. I agree with you that mga is quite common, but so are Raritan VGA dongles in a server / lab environment.

So, if someone does not have a login prompt, you can go to the grub menu, edit the boot command line, and add one of the above (temporarily, until you can do it in a more permanent way in /etc/default/grub).

Thanks,

george

 

[t.lamprecht]

setting "nomodeset", or for that matter setting "vga=0x0317" work around the issue. Thanks for the hint.

Ok, thanks for confirming that.

So, if someone does not have a login prompt, you can go to the grub menu, edit the boot command line, and add one of the above (temporarily, until you can do it in a more permanent way in /etc/default/grub).

For others reading this: the permanent way may be different if systemd-boot is used, both ways are documented at:
https://pve.proxmox.com/pve-docs/chapter-sysadmin.html#sysboot_edit_kernel_cmdline

 

[kyriazis]

@t.lamprecht , do you know if changes in /etc/default/grub will disappear after a package update?

Thanks!

george

 

[t.lamprecht]

Nope, most files in /etc are seen as "configuration from the user" and if a package changes a file in /etc it ships, and an admin made any changes to that file, then the user is asked if they want to install the updated version or keep the locally modified one, with the latter being the default.

Especially for such a central file like /etc/default/grub a change is normally only done on major upgrades, if at all.

 

[fuzzyduck]

So when reinstalling 6.4 I now choose the EFI boot method for the iso installer on the usb stick and after installing I now longer get the blue grub menu, But I think I saw text on a black green at boot?. All is running well now.

Question: in EFIboot and a possible upgrade to 7 in the future, will I still run into trouble booting into a blank screen with no VGA output?

 

[t.lamprecht]

Question: in EFIboot and a possible upgrade to 7 in the future, will I still run into trouble booting into a blank screen with no VGA output?

EFI or not is not really related to what happens to the graphic stack of the kernel...

Rather try nomodeset as that is what helped the actual thread OP, if that does not help you then this thread is unrelated to your issues...

 

[kyriazis]

@t.lamprecht , is this going to be fixed in an update?