Unrecoverable failure in required component org.gnome.Shell.desktop

[coxackie]

Trying booster in Arch with zen kernel - no config files. The boot process hangs when trying to enter gnome shell. I get

gnome-session-binary[527]: Unrecoverable failure in required component org.gnome.Shell.desktop

in the journal. It does not happen when I use images from mkinitcpio. I can provide further info upon request.

EDIT: I realised that this is caused by not loading the amdgpu module. Same happens with mkinitcpio. However, if I include amdgpu to booster, it just does not boot.

[anatol]

It is really strange that amdgpu is required for booster to mount the root filesystem. I definitely interested to learn what happens there. Do you have any extra logs that shed additional light on this amdgpu issue? Maybe this driver needs extra firmware blobs?

[anatol]

One thing you can try is to add amdgpu firmware files manually. At /etc/booster.yaml add extra_files: /usr/lib/firmware/amdgpu/.

[anatol]

Filed a separate ticket to track the firmware files feature #48

[coxackie]

One thing I certainly noticed is that, when adding amdgpu at mkinitcpio module, the size of the image increased quite a lot more than if I add it to booster. Maybe it is worth checking what mkinitcpio does...

[anatol]

mkinitcpio does add modules firmware. The #48 proposes adding this feature to booster. But before going forward with it I want to check that is the real culprit of your issue. Please let me know if extra_files: option from above makes any difference for you.

[coxackie]

sure - just to make sure: you want me to put both modules: amdgpu and extra_files: /usr/lib/firmware/amdgpu/ in /etc/booster.yaml?

[anatol]

sure - just to make sure: you want me to put both modules: amdgpu and extra_files: /usr/lib/firmware/amdgpu/ in /etc/booster.yaml?

yes. And if it does not boot then please provide boot output with booster.debug=1 kernel option.

[oech3]

If Early KMS(including gpu ko in initramfs) solve it,gnome may running before gpu is initialized.see
https://wiki.archlinux.org/index.php/NVIDIA/Troubleshooting#Xorg_fails_during_boot,_but_otherwise_starts_fine
https://superuser.com/questions/1469766/make-systemd-service-wait-for-a-device-generically

[anatol]

thanks @oech3 I've see at wiki that some folks recommend adding the gpu drivers to initramfs but it was not clear for me why.

Anyway has anyone tried adding the firmware files to config? Or optionally try booster from master branch (booster-git AUR package) - it should add the dependent firmware files automatically. Let me know if it works or does not work for you.

[coxackie]

I added a couple of days ago modules: amdgpu and extra_files: /usr/lib/firmware/amdgpu/, but the result was the same, unfortunately: hang. You mean me to put booster.debug=1 in the options part of the conf file (using systemd-boot). If yes, how do I capture the boot output later? Sorry for the noob questions, but at least I am willing to help... :P

[anatol]

but the result was the same, unfortunately: hang

bummer.

Try to add debug booster.debug=1 boot parameters. Is there anything interesting in the logs?

If yes, how do I capture the boot output later?

Using your phone camera is probably the easiest way.

Is there a way to reproduce this problem with Nvidia GPU? So I can try to do play with it locally.

[oech3]

@coxackie can you boot if you disable gdm.service ?
If it boot,does systemctl start gdm work after waiting 10s?

[coxackie]

@oech3 I will check, but I am pretty sure this is not the issue (when I use amdgpu). It hangs from the very beginning.

To be honest, I would not like to use the amdgpu module. If I do not, for some reason org.gnome.Shell.desktop has a problem, but if I change tty with Ctrl-Alt-F2 and then go back immediately with Ctrl-Alt-F1, the gnome login screen appears. It is ridiculous, but in all honesty I prefer this than putting all the amd files in the image, as it increases the size a lot, and there is not so much space in the partition (dual booting).

[anatol]

I added modules: nvidia,nvidia_modeset,nvidia_uvm,nvidia_drm to my config and tried to boot with this image. I found that umount(/sys) fails with EBUSY that means some driver (nvidia?) still uses it. I unmount it lazily and now my problem is gone. Now I can boot my machines with nvidia drivers added to the image.

I pushed 82feb91 to master. Is there any chance you can try to boot with it?

[coxackie]

Sorry for being delayed in answering - I will try it again when a new version comes out, and then let you know here what is going on.

[anatol]

Thanks!

Try master branch - it contains a change that adds firmware files automatically. It might help in your situation.

[CyberX5]

Hi,

I dont think this is fixed, specifically adding amdgpu to modules: or modules_force_load: does not boot.

Using -universal doesn't appear to change anything, adding extra_files: /usr/lib/firmware/amdgpu/ doesn't ether, im using booster.disable_concurrent_module_loading.

The output with booster.debug just freezes at loading module amdgpu.

Heres the output of booster -kernelVersion 5.12.12-arch1-1 -debug -force -universal -output /boot/booster-linux.img file.txt

If i can help somehow feel free to tell me 🙂.

[coxackie]

Sorry for the huge delay between my responses. Turns out that org.gnome.Shell.desktop is now perfectly fine without specifying the amdgpu module, so I do not have to deal with it anymore, and I am happy - size becomes quite smaller.

[anatol]

There are some fixes at wip branch that might help with this issue. If you have a chance please pull the wip branch and check whether this issue is fixed or now.

[CyberX5]

Its fixed, adding amdgpu to modules: or modules_force_load: works. Didnt need to do anything else.

Thank you 🙂.