OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
gpu-drivers high occasional DISPUTE

Fix a frozen AMD display with 'flip_done timed out' in the log

archomarchyendeavouroscachyosmanjaroamdlaptopdesktopwaylandhyprland
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

The single cited source resolves and does support the technical claim. I fetched https://wiki.archlinux.org/title/AMDGPU as raw wikitext and its section "Frozen or unresponsive display (flip_done timed out)" recommends exactly amdgpu.dcdebugmask=0x10 (Panel self refresh v1 and PSR-SU) or amdgpu.dcdebugmask=0x12 (Panel self refresh and memory stutter mode), so the record's wording is close to verbatim and the source stays. I checked the bit values against the kernel rather than trusting the wiki: enum DC_DEBUG_MASK in drivers/gpu/drm/amd/include/amd_shared.h gives DC_DISABLE_PSR = 0x10 and DC_DISABLE_STUTTER = 0x2, so 0x12 is both, confirmed. In drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c the parameter is module_param_named(dcdebugmask, amdgpu_dc_debug_mask, uint, 0444), which is read-only at runtime and confirms a reboot is genuinely required, and gives a better verify path than /proc/cmdline alone. What was wrong is the apply step. The record told an Omarchy reader to put the parameter in cmdline: in /boot/limine.conf, or in /etc/default/grub. Neither works on Omarchy 4, and the record carries omarchy in applies_to. On this workstation (omarchy 4.0.2-1, hyprland 0.56.2, kernel 7.1.9) there is no /etc/default/grub, the boot is a UKI, and the real drop-ins are /etc/limine-entry-tool.d/, which I read directly: omarchy-defaults.conf contains KERNEL_CMDLINE[default]+= lines for quiet splash and for initramfs_async=0, and omarchy-uki.conf sets ENABLE_UKI=yes. The packaged template /etc/limine-entry-tool.conf from limine-mkinitcpio-hook 1.37.1-1 states the operator rules, that += appends in /etc/limine-entry-tool.d/*.conf and a bare = replaces. So the fix, verify and danger are replaced to match the existing record limine-kernel-parameters-not-applying-omarchy rather than contradict it, keeping the systemd-boot and GRUB branches labelled for other distributions. The cause is also replaced, because the upstream report the wiki cites, https://gitlab.freedesktop.org/drm/amd/-/issues/4141, is worth scoping honestly: it was filed against a Ryzen 7840U with a Radeon 780M, is labelled PSR and Phoenix / Hawk Point, and was closed on 2026-08-12. Its comments need a login, so I could not read the closing note and cannot say whether a kernel fix landed. The wiki still recommends the workaround today. NOT EXERCISED: this workstation has an NVIDIA RTX 3090 and lsmod shows amdgpu is not loaded, so nothing about amdgpu behaviour was tested here. The Omarchy 4 boot mechanism was confirmed on this machine, the amdgpu facts come from the kernel source and the wiki only. Severity high and frequency occasional are left alone: a frozen display is serious, and the upstream report is confined to one APU family.

The cause above was rewritten on 2026-09-11 to match this note; the fix was corrected by the audit itself.
RISK

Disabling Panel Self Refresh costs battery life on a laptop, and 0x12 also disables memory stutter mode, which costs more. On Omarchy 4 the drop-in itself is the larger risk. Writing KERNEL_CMDLINE[default]= instead of +=, or naming your file so it sorts before omarchy-defaults.conf, silently drops Omarchy's initramfs_async=0, and the packaged comment in that file says that is what keeps Plymouth alive at the LUKS prompt, so an encrypted machine falls back to an unthemed text prompt or hangs. limine-mkinitcpio rewrites the UKI on the ESP, so run df -h /boot first, because a full ESP produces a truncated image that will not boot. A bad kernel parameter is a failure to boot. Recovery is the Limine boot menu editor or a Limine snapshot entry, and both are bypassed if you have enabled Omarchy's Direct Boot, in which case pick Limine from the firmware boot menu (usually F12, F8 or Esc). On a GRUB system the equivalent recovery is GRUB's own e editor at the menu.

Symptom

The screen stops updating entirely. The image is frozen but audio keeps playing and SSH still works. journalctl -k shows:


[drm:drm_atomic_helper_wait_for_flip_done] *ERROR* [CRTC:...] flip_done timed out

Cause

A bug in the amdgpu display code around Panel Self Refresh stops the atomic page flip from ever completing, so the compositor waits forever for a flip that never lands and the image freezes while the rest of the machine keeps running.

amdgpu.dcdebugmask is a bitmask over enum DC_DEBUG_MASK in the kernel. DC_DISABLE_PSR is 0x10 and turns off Panel Self Refresh v1 and PSR Selectively Updated. DC_DISABLE_STUTTER is 0x2 and turns off memory stutter mode, so 0x12 is both.

The upstream report the Arch wiki points at is https://gitlab.freedesktop.org/drm/amd/-/issues/4141, filed against a Ryzen 7840U with a Radeon 780M iGPU and labelled PSR and Phoenix / Hawk Point, so recent AMD APU laptops are where this is reported most. That report was closed on 2026-08-12 and its comments need a login to read, so whether a kernel fix has landed is not established here. The Arch wiki still recommended the workaround when this record was re-audited on 2026-09-11, so check your kernel version against the report before assuming you still need it.

Fix

dcdebugmask is a 0444 module parameter, readable but not writable at runtime, so it can only be set at boot. Start with:


amdgpu.dcdebugmask=0x10

which disables Panel Self Refresh v1 and PSR Selectively Updated. If that is not enough, change the same value to:


amdgpu.dcdebugmask=0x12

which disables Panel Self Refresh and memory stutter mode. Edit the value in place rather than adding a second parameter.

**Omarchy 4** has no /etc/default/grub, and /boot/limine.conf is regenerated on every kernel or limine transaction, so an edit there is discarded. Omarchy boots a Unified Kernel Image, which bakes the command line into the .efi image. Kernel parameters go in your own drop-in under /etc/limine-entry-tool.d/ whose filename sorts after Omarchy's omarchy-defaults.conf, and always with +=:


sudo tee -a /etc/limine-entry-tool.d/zz-local.conf >/dev/null <<'EOF'
# `+=` appends. A bare `=` would wipe Omarchy's own defaults.
KERNEL_CMDLINE[default]+=" amdgpu.dcdebugmask=0x10"
EOF
sudo limine-mkinitcpio && sudo limine-update && sudo reboot

Use tee -a rather than tee. A bare tee truncates the file and would drop any parameter you added earlier from another record. The full mechanism, including how to test a parameter from the Limine boot menu editor without writing anything to disk, is in the record limine-kernel-parameters-not-applying-omarchy.

**Other bootloaders:**

  • systemd-boot: append to options in /boot/loader/entries/*.conf
  • GRUB: GRUB_CMDLINE_LINUX_DEFAULT in /etc/default/grub, then sudo grub-mkconfig -o /boot/grub/grub.cfg

Reboot in every case.

Verify


cat /sys/module/amdgpu/parameters/dcdebugmask   # 16 for 0x10, 18 for 0x12
cat /proc/cmdline

/proc/cmdline contains amdgpu.dcdebugmask=0x10. On Omarchy 4 it must also still contain Omarchy's own defaults (quiet splash loglevel=0 systemd.show_status=false rd.udev.log_level=0 vt.global_cursor_default=0 and initramfs_async=0). If those have gone, your drop-in used = where it needed +=. Then journalctl -k -g flip_done stays empty after a full day of use and the display no longer freezes.

Sources

amdgpu-flip-done-timed-out