Fixes to weston startup by marmarek · Pull Request #156 · rhinstaller/initial-setup

marmarek · 2024-12-17T12:38:45Z

This PR fixes two aspect of weston startup:

Switching VT - necessary with libseat >= 0.9.0
Handling multiple GPUs

See individual commit messages for details

Conan-Kudo · 2025-04-20T16:35:02Z

FYI @AdamWill @nirik: This pull request should also resolve RH#2358688.

M4rtinK · 2025-04-22T16:40:45Z

So I've tried the fix by installing a Rawhide VM & installing initial-setup-gui.

First I tried to run the Initial Setup GUI without the changes - set graphical.target target as the default systemd target & enable initial-setup.service. Indeed, in this case the GUI does not run & you end up on the login prompt.

Then I edited the /usr/libexec/initial-setup/run-gui-backend file with the changes from this PR, set graphical.target target as the default systemd target & enabled initial-setup.service - this fixes the issue with the GUI not showing up - it is now visible!

But unfortunately the fix does not appear to be complete - although the GUI does show up, when you press the FINISH CONFIGURATION button, something gets stuck indefinitely & the login prompt is not reached:

It is possible to switch to a different TTY, but the one in which Initial Setup was running is stuck.

Journal dump from this happening:
journal_stuck_dump.txt

marmarek · 2025-04-23T17:20:00Z

the login prompt is not reached:

Was there supposed to be a login prompt? initial-setup uses tty7, but I think normally getty is started only on tty1-tty6.

AdamWill · 2025-04-23T18:58:02Z

after initial-setup runs, either a login prompt or a logged-in session must be presented next. this is in the release criteria, and it's how it has always worked previously.

marmarek · 2025-04-24T13:29:38Z

Lightdm (and probably others too) will switch to appropriate VT on its own. That isn't the case for the text console. I guess I can add chvt back to tty1 (or saved fgconsole if that works?), will that be enough?

Conan-Kudo · 2025-04-24T13:44:14Z

It should indeed switch back to what it started with before. So basically capture the existing vt, switch, then switch back.

marmarek · 2025-04-24T14:06:26Z

This version should (hopefully) do that now.

Conan-Kudo · 2025-04-24T14:08:30Z

kbd should be specified as a runtime dependency of the subpackage containing this script.

Conan-Kudo · 2025-04-24T14:09:25Z

Needs to be added for this subpackage block: https://github.com/rhinstaller/initial-setup/blob/master/initial-setup.spec#L93-L111

marmarek · 2025-04-24T14:12:52Z

CI is unhappy, not sure why:

2025-04-24 14:07:18.247 logging.py        INFO   Creating tar archive
2025-04-24 14:07:18.324 logging.py        INFO   removing 'initial_setup-0.3.101' (and everything under it)
2025-04-24 14:07:18.350 logging.py        INFO   The archive is in initial-setup-0.3.101.tar.gz
2025-04-24 14:07:18.351 commands.py       DEBUG  Command: bash -c cp dist/*.tar.gz .
2025-04-24 14:07:18.354 commands.py       DEBUG  Command: bash -c ls -1 initial-setup-*.tar.gz
2025-04-24 14:07:18.357 logging.py        INFO   ls: cannot access 'initial-setup-*.tar.gz': No such file or directory
2025-04-24 14:07:18.358 commands.py       ERROR  Command 'bash -c ls -1 initial-setup-*.tar.gz' failed.
2025-04-24 14:07:18.358 commands.py       DEBUG  Command stderr: ls: cannot access 'initial-setup-*.tar.gz': No such file or directory

... and restore original VT before exiting. libseat >= 0.9.0 doesn't do that anymore, and it's necessary to get access to devices. More details at https://lists.sr.ht/~kennylevinsen/seatd-devel/%3CZy_-FRQnBTeNPXVj@mail-itl%3E QubesOS/qubes-issues#9568

Normally weston uses card0 only. In dual-GPU setup, it may not be the one with monitor connected, and when it doesn't find any connector monitor, it fails to startup. Look also at outputs connected to other cards too. There doesn't seems to be any option to check all connected devices, so enumerate them via sysfs manually.

M4rtinK · 2025-05-16T11:47:01Z

Any updates ? I assume its still broken without the fix. :P

marmarek · 2025-05-16T12:15:12Z

Let me know if I can help, I think I addressed all earlier comments. CI looks to be broken, but I don't think I can do anything about it...

M4rtinK · 2025-05-16T15:07:12Z

Let me know if I can help, I think I addressed all earlier comments. CI looks to be broken, but I don't think I can do anything about it...

OK, will check on ~Monday. :)

Conan-Kudo · 2025-07-24T22:13:18Z

Can someone please look at this? We should get this fixed and released to F42 and Rawhide.

supakeen · 2026-01-07T08:51:17Z

I've been tinkering with some Fedora variants that could make use of this.

AdamWill · 2026-04-08T19:50:58Z

Ping @M4rtinK ? We still have complaints about this downstream.

Conan-Kudo · 2026-04-15T05:40:22Z

Can we just apply these downstream in Fedora and see if it fixes things? Because this has been held up due to lack of review for almost two years...

M4rtinK · 2026-05-19T09:21:40Z

Sorry for the delays, looking into this during this week. :)

M4rtinK · 2026-06-02T09:20:24Z

So after testing this PR it seems to fix the issue with Initial Setup GUI not starting due to Weston issues on x86 (at least) on current Rawhide. I've tried if it helps also on ARM where I have seen a similar issues, but it is not sufficient there (and the issue here - Weston outright crashing - might be different from what happens on x86). So lets go with this for x86 & fix the ARM situation separately. :)

M4rtinK

Looks good to me, thanks! :)

rvykydal

Looks good to me.

Copilot

Pull request overview

This PR updates the Weston-based GUI backend startup flow to better support modern seat handling (libseat >= 0.9.0) and multi-GPU systems by switching to the correct VT before launching Weston and by attempting to explicitly enumerate DRM devices.

Changes:

Switch to the session VT prior to launching Weston (and switch back afterward).
Enumerate /sys/class/drm/card* to select a primary DRM device and pass additional devices to Weston.
Add an RPM runtime dependency on kbd (for chvt/fgconsole) in the Wayland/Weston subpackage.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
scripts/run-gui-backend.guiweston	Adds VT switching and DRM device enumeration to improve Weston startup with libseat and multi-GPU setups.
initial-setup.spec	Adds `kbd` runtime dependency for the Wayland/Weston backend to support VT utilities used by the script.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

+original_vt=
+if [ -n "$XDG_VTNR" ]; then
+    original_vt=$(fgconsole)
+    # change to VT on which we run to get access to devices
+    chvt "${XDG_VTNR}"
+fi


KKoukiou · 2026-06-02T10:07:08Z

+cards=$(ls -d /sys/class/drm/card*| grep -v -- -| cut -d / -f 5)
+primary_card=$(echo "$cards" | head -1 )
+secondary_cards=$(echo "$cards"|
+                  tail -n +2|
+                  xargs printf "%s,")
+
+weston --config=${CONFIG_FILE} --socket=wl-firstboot-0 --drm-device="$primary_card" --additional-devices="$secondary_cards"


I think this i wrong. I have only one card and I don't get this:

kkoukiou@fedora:~/repos/anaconda-webui$ cards=$(ls -d /sys/class/drm/card*| grep -v -- -| cut -d / -f 5) kkoukiou@fedora:~/repos/anaconda-webui$ primary_card=$(echo "$cards" | head -1 ) kkoukiou@fedora:~/repos/anaconda-webui$ echo $primary_card card1 kkoukiou@fedora:~/repos/anaconda-webui$ secondary_cards=$(echo "$cards"| tail -n +2| xargs printf "%s,") kkoukiou@fedora:~/repos/anaconda-webui$ echo $secondary_card

@KKoukiou I think the difference is you are echoing $secondary_card instead of $secondary_cards
My check:

[prnot5 ~/work/scratch]$ cards=$(ls -d /sys/class/drm/card*| grep -v -- -| cut -d / -f 5) [prnot5 ~/work/scratch]$ primary_card=$(echo "$cards" | head -1 ) [prnot5 ~/work/scratch]$ echo $primary_card card1 [prnot5 ~/work/scratch]$ secondary_cards=$(echo "$cards"| tail -n +2| xargs printf "%s,") [prnot5 ~/work/scratch]$ echo $secondary_cards ,

But as I said below I am not sure if comma is harmful for the command option.

--additional-devices=, seems to work just fine. weston uses strtok() to parse that option, it it skips empty names.

rvykydal

Oh right, I hit the issue with comma in #156 (comment) when playing with the script locally as well. Didn't run the weston command with the option though.

marmarek force-pushed the change-vt branch from 7d0349c to 136b8e2 Compare April 24, 2025 14:05

marmarek added 2 commits April 24, 2025 16:13

Switch VT before starting weston

e83afd2

... and restore original VT before exiting. libseat >= 0.9.0 doesn't do that anymore, and it's necessary to get access to devices. More details at https://lists.sr.ht/~kennylevinsen/seatd-devel/%3CZy_-FRQnBTeNPXVj@mail-itl%3E QubesOS/qubes-issues#9568

marmarek force-pushed the change-vt branch from 136b8e2 to 47ccca2 Compare April 24, 2025 14:13

M4rtinK approved these changes Jun 2, 2026

View reviewed changes

M4rtinK requested a review from rvykydal June 2, 2026 09:22

rvykydal approved these changes Jun 2, 2026

View reviewed changes

KKoukiou requested a review from Copilot June 2, 2026 10:01

Copilot started reviewing on behalf of KKoukiou June 2, 2026 10:01 View session

Copilot AI reviewed Jun 2, 2026

View reviewed changes

rvykydal reviewed Jun 3, 2026

View reviewed changes

rvykydal self-requested a review June 3, 2026 13:04

rvykydal approved these changes Jun 3, 2026

View reviewed changes

Conversation

marmarek commented Dec 17, 2024

Uh oh!

Conan-Kudo commented Apr 20, 2025

Uh oh!

M4rtinK commented Apr 22, 2025

Uh oh!

marmarek commented Apr 23, 2025

Uh oh!

AdamWill commented Apr 23, 2025

Uh oh!

marmarek commented Apr 24, 2025

Uh oh!

Conan-Kudo commented Apr 24, 2025

Uh oh!

marmarek commented Apr 24, 2025

Uh oh!

Conan-Kudo commented Apr 24, 2025

Uh oh!

Conan-Kudo commented Apr 24, 2025

Uh oh!

marmarek commented Apr 24, 2025

Uh oh!

M4rtinK commented May 16, 2025

Uh oh!

marmarek commented May 16, 2025

Uh oh!

M4rtinK commented May 16, 2025

Uh oh!

Conan-Kudo commented Jul 24, 2025

Uh oh!

supakeen commented Jan 7, 2026

Uh oh!

AdamWill commented Apr 8, 2026

Uh oh!

Conan-Kudo commented Apr 15, 2026

Uh oh!

M4rtinK commented May 19, 2026

Uh oh!

M4rtinK commented Jun 2, 2026

Uh oh!

M4rtinK left a comment

Choose a reason for hiding this comment

Uh oh!

rvykydal left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

KKoukiou Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rvykydal Jun 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

rvykydal Jun 3, 2026

Choose a reason for hiding this comment

Uh oh!

marmarek Jun 3, 2026

Choose a reason for hiding this comment

Uh oh!

rvykydal left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

8 participants

KKoukiou Jun 2, 2026 •

edited

Loading

rvykydal Jun 3, 2026 •

edited

Loading