OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
omarchy-theming critical rare DISPUTE

Stop the whole desktop hard-freezing when the Walker launcher renders

omarchyomarchy-3amdhyprlandwaylanddesktop
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

Verified against upstream that the Omarchy 3 mechanism this record describes is accurate. bin/omarchy-launch-walker on the master branch of omacom/omarchy sets GSK_RENDERER=cairo only inside the if ! pgrep -f "walker --gapplication-service" branch, default/walker/walker.desktop carries a bare Exec=walker --gapplication-service with no override, and bin/omarchy-restart-walker restarts app-walker@autostart.service, which confirms the generated unit name the record uses. The cited issue 6443 exists, is titled exactly as the record's cause describes, and its body supports every step of the chain including the 30ddfeda commit that moved the variable out of default/hypr/envs.conf, so the source audit was right. The record's failure is that it was filed as an Omarchy record when it is an Omarchy 3 record. On this Omarchy 4.0.2-1 workstation pacman -Q walker fails, pacman -Ql omarchy lists no walker path, /usr/share/omarchy/bin contains no omarchy-launch-walker or omarchy-refresh-walker, ~/.config/autostart/ holds only limine-snapper-notify.desktop, org.fcitx.Fcitx5.desktop and print-applet.desktop, and the quattro tree contains no path matching walker. dhh closed issue 6443 with "Walker is gone on quattro." I did not reject it because the kernel and GTK4 halves are unchanged and the same oops can hit another GTK4 application on amdgpu, and because Walker is still installable from the AUR at 2.17.0-1. The second real defect is the fix's fallback: ~/.config/hypr/envs.conf with env = GSK_RENDERER,cairo is hyprlang syntax, that file does not exist on Omarchy 4, and Omarchy 4's Hyprland config is Lua using hl.env, so the record as written pointed an Omarchy 4 reader at a file nothing reads. I read /usr/share/uwsm/env.d/10-omarchy, which names ~/.config/uwsm/env.d/* as the preferred user override, and /usr/share/doc/uwsm/README.md lines 112 and 634 to 654, which confirm uwsm sources uwsm/env.d/* from each config dir and that ~/.profile is only sourced when the session did not start through uwsm start. ~/.config/environment.d/ is also live on this box, holding a working omarchy-firefox-wayland.conf. On the GTK claim, I read gsk/gskrenderer.c at tag 4.22.4: get_renderer_for_name still accepts cairo at line 496, and the renderer_possibilities table still tries Vulkan before the OpenGL renderer on Wayland, so the record's "default Vulkan" claim holds, but its "Vulkan/ngl" phrasing is stale, because line 501 now warns "The new GL renderer has been renamed to gl" and the help text records that the old GL renderer was removed in GTK 4.18. Arch's gtk4 1:4.22.4-1 depends on vulkan-icd-loader and libgtk-4.so.1 links libvulkan.so.1, so Vulkan is genuinely compiled in here. I lowered frequency from occasional to rare because the distribution no longer ships the trigger, and kept severity critical because an unrecoverable freeze is still the consequence for anyone affected. Not exercised: this workstation has an NVIDIA card, not amdgpu, so I could not reproduce the TTM oops or confirm whether kernel 7.1.9-arch1-2 still carries the underlying amdgpu defect, and I installed nothing.

The cause above was rewritten on 2026-09-13 to match this note; the fix was corrected by the audit itself.
RISK

Until this is applied, the failure mode is an unrecoverable freeze requiring a hard reset, which risks filesystem damage on btrfs or ext4 with unflushed writes. On Omarchy 3, ~/.config/autostart/walker.desktop is rewritten by bin/omarchy-refresh-walker, by install/config/walker-elephant.sh and by migrations, so re-check the Exec= line after any omarchy update. GSK_RENDERER=cairo disables GPU acceleration for that application's rendering, so expect slower redraws and higher CPU use. Set it as narrowly as the problem allows rather than session-wide.

Symptom

The entire desktop freezes. The display is frozen and keyboard and mouse are dead, but audio keeps playing, because PipeWire never touches the GPU. Only a hard reset recovers. journalctl -k from the previous boot shows BUG: kernel NULL pointer dereference at RIP: ttm_lru_bulk_move_pos_tail+0x4f/0xb0 [ttm] with Comm: walker, preceded by WARNING: drivers/gpu/drm/ttm/ttm_resource.c:235 at ttm_resource_add_bulk_move, then watchdog: BUG: soft lockup - CPU#16 stuck for 495s! [kworker/u97:3] inside amdgpu_dm_atomic_commit_tail.

This is an Omarchy 3 failure on AMD graphics. Omarchy 4 (Quattro) does not ship Walker, so the Omarchy-specific trigger is gone there. The kernel and GTK4 halves still apply to any GTK4 application on amdgpu, so the same oops can appear with a different Comm:.

Cause

On Omarchy 3 the GSK_RENDERER=cairo workaround is applied only inside bin/omarchy-launch-walker, and only in a branch that does not normally run. The resident walker --gapplication-service process is started by the XDG autostart entry ~/.config/autostart/walker.desktop, whose Exec=walker --gapplication-service carries no environment override. systemd-xdg-autostart-generator turns that entry into app-walker@autostart.service and starts it at session start, before any keybind is pressed, so the guard if ! pgrep -f "walker --gapplication-service" in the launch script is dead code and the variable is never applied.

Walker therefore runs with GTK4's default renderer. On Wayland that is the Vulkan renderer: gsk_renderer_new_for_surface_full walks the display override, then GSK_RENDERER, then the backend, then Vulkan, and only then the OpenGL renderer, and Arch's gtk4 is built against vulkan-icd-loader. On amdgpu, Walker page-faults on a GEM buffer, the kernel takes a NULL dereference inside TTM while holding the LRU spinlock, and every subsequent display atomic commit deadlocks on that lock, which is why input dies while audio survives. The underlying defect is kernel-side TTM and amdgpu. The missing environment variable is what exposes it.

The workaround used to be session-wide in Omarchy 3's default/hypr/envs.conf and was narrowed to the launcher, which is how the gap opened. Omarchy 4 ships neither Walker nor that file. Its Hyprland config is Lua and its session environment comes from /usr/share/uwsm/env.d/10-omarchy.

Fix

**On Omarchy 3**, confirm the variable is genuinely absent from the running service:


pgrep -a walker
tr '\0' '\n' < /proc/$(pgrep -x walker)/environ | grep GSK_RENDERER

Force the Cairo renderer on the autostart entry:


sed -i 's|^Exec=walker --gapplication-service|Exec=env GSK_RENDERER=cairo walker --gapplication-service|' \
  ~/.config/autostart/walker.desktop
systemctl --user daemon-reload
systemctl --user restart app-walker@autostart.service

Or set it session-wide in ~/.config/hypr/envs.conf, which is Omarchy 3's hyprlang config:


env = GSK_RENDERER,cairo

then log out and back in.

**On Omarchy 4 (Quattro)** neither Walker nor ~/.config/hypr/envs.conf exists, so neither step above applies. If you need the Cairo renderer for some other GTK4 application that is oopsing on amdgpu, set it where Omarchy 4 actually reads session environment. /usr/share/uwsm/env.d/10-omarchy names ~/.config/uwsm/env.d/* as the preferred user override, and uwsm sources those files as shell:


mkdir -p ~/.config/uwsm/env.d
echo 'export GSK_RENDERER=cairo' > ~/.config/uwsm/env.d/50-gsk-renderer

~/.config/environment.d/*.conf also reaches the systemd user manager and the graphical session:


GSK_RENDERER=cairo

Log out and back in for either file to take effect. Do not put it in ~/.profile: uwsm sources a POSIX profile only when the session was not started through uwsm start, so on Omarchy 4 that path is unreliable and the variable may never reach the session.

To scope the change to one application rather than the whole session, wrap that application's Exec= line with env GSK_RENDERER=cairo instead.

Check the value you set is still accepted by your gtk4:


GSK_RENDERER=help gtk4-demo 2>&1 | head -20

cairo remains valid on gtk4 4.22.4. ngl does not: GTK renamed that renderer to gl, and GSK_RENDERER=ngl now logs The new GL renderer has been renamed to gl and selects the OpenGL renderer anyway.

Verify

On Omarchy 3, tr '\0' '\n' < /proc/$(pgrep -x walker)/environ | grep GSK_RENDERER prints GSK_RENDERER=cairo, and no new Comm: walker TTM oopses appear in journalctl -k -b -1 over subsequent days.

On Omarchy 4, confirm the variable reached the session rather than only the shell:


systemctl --user show-environment | grep GSK_RENDERER

Then confirm the application picked it up, substituting its process name:


tr '\0' '\n' < /proc/$(pgrep -x <app>)/environ | grep GSK_RENDERER

Sources

walker-gtk4-vulkan-renderer-amdgpu-hard-freeze