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

Stop AMD GPU freezes and reboots when idle or waking from sleep

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

The single cited source, https://wiki.archlinux.org/title/AMDGPU, does support the symptom and the general approach. I fetched its raw wikitext and the section "System freezes or reboots when idle" says PowerPlay features such as GFXOFF can cause frequent unrecoverable driver crashes that coincide with idle GPU usage on a multi-monitor setup and especially waking from sleep, which is the record's symptom almost word for word, and it names amdgpu.ppfeaturemask=0xffff7fff plus the 0xfffd7fff, 0xfffd3fff and 0 alternatives. So the source stands and I removed nothing. The recommended mask is still wrong, and the source is why: the wiki itself says 0xffff7fff "leaves all other (implemented and unimplemented) PowerPlay features enabled", and the record rewrote that into "disables only PP_GFXOFF_MASK and leaves everything else enabled", which reads as a minimal change when it is not. I checked the driver at tag v7.1: drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c sets uint amdgpu_pp_feature_mask = 0xfff7bfff; under the comment "OverDrive(bit 14) disabled by default / GFX DCS(bit 19) disabled by default", and drivers/gpu/drm/amd/include/amd_shared.h gives PP_OVERDRIVE_MASK = 0x4000, PP_GFXOFF_MASK = 0x8000, PP_STUTTER_MODE = 0x20000, PP_GFX_DCS_MASK = 0x80000. So 0xffff7fff re-enables Overdrive and GFX Async DCS, and the same file's amdgpu_init calls add_taint(TAINT_CPU_OUT_OF_SPEC, ...) and pr_crit("Overdrive is enabled ...") when the Overdrive bit is set. The correct minimal value on kernel 7.1 is the driver default with GFXOFF cleared, 0xfff73fff, which the wiki also lists, and I replaced the copied ladder with the computed form the wiki uses elsewhere for the Overdrive case. The bigger defect is the same one the last thirty-seven re-audits found. The fix told an Omarchy reader to edit /etc/default/grub or the cmdline: line in /boot/limine.conf. I confirmed on this workstation (omarchy 4.0.2-1, kernel 7.1.9-arch1-2) that /etc/default/grub does not exist and no grub package is installed, and the installed /etc/limine-entry-tool.conf template documents the real path: drop-ins under /etc/limine-entry-tool.d/*.conf, += to append and = to replace. That agrees with the Limine Arch wiki page and with the existing corpus record limine-kernel-parameters-not-applying-omarchy, which I read so this correction does not contradict it. I read /etc/limine-entry-tool.d/omarchy-defaults.conf locally and fetched the same file at upstream tag v4.0.3 (published 2026-09-08, the newest tag) and they are identical, so this holds on the current release. I rewrote the danger because the old one said only that disabling PowerPlay costs power and that a bootloader reinstall can drop the parameter, and it missed that a copied mask can itself produce flicker and broken resume, and that a bare = in the drop-in silently deletes initramfs_async=0 and can leave an encrypted machine at a text prompt with no desktop. NOT EXERCISED: this workstation has an NVIDIA GeForce RTX 3090 and no amdgpu module is loaded (/sys/module/amdgpu does not exist), so no AMD behaviour, no mask and no freeze was reproduced here. Every mask claim comes from the kernel source at v7.1 and the wiki, and I ran nothing that writes to the bootloader or the initramfs.

The cause above was not rewritten and may still contain the error described. The fix below is the corrected version.
RISK

A wrong mask is the danger here, not a typo in the parameter name. amdgpu.ppfeaturemask is a raw bitmask with no validation, so a copied value silently enables features the kernel keeps off. 0xffff7fff, the value most forum posts quote, turns on Overdrive and GFX Async DCS on kernel 7.1. Overdrive taints the kernel with TAINT_CPU_OUT_OF_SPEC and makes any bug report unwelcome upstream, and the Arch wiki warns that enabling unstable PowerPlay bits can cause screen flicker or broken resume from suspend, which is the failure you were trying to fix. amdgpu.ppfeaturemask=0 disables DPM as well, so the card can sit at boot clocks with poor performance and higher idle power. Disabling GFXOFF at all increases idle power draw and heat, which matters on a laptop.

On Omarchy 4 the second risk is the drop-in itself. Writing KERNEL_CMDLINE[default]= instead of +=, or naming the file so it sorts before omarchy-defaults.conf, throws Omarchy's own defaults away including initramfs_async=0. The packaged comment says that parameter is what stops plymouthd exiting on kernel 7.1, so an encrypted machine falls back to an unthemed text LUKS prompt or sits there with no graphical session. 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. Recovery is the Limine snapshot entry or the Limine boot menu editor, and Omarchy's Direct Boot bypasses that menu, so do not enable Direct Boot while you are still testing parameters. The parameter does survive kernel updates once it is in a drop-in, because the drop-in is what the UKI is rebuilt from.

Symptom

An AMD desktop or laptop locks up hard or spontaneously reboots when left idle, typically on a multi-monitor setup, and especially after resuming from sleep. Nothing is logged before the freeze.

Cause

AMD PowerPlay's GFXOFF feature (and, on some cards, stutter mode) puts the graphics engine into a power state it does not reliably recover from, producing unrecoverable driver crashes.

Fix

Disable GFXOFF with the PowerPlay feature mask, but compute the value from your own kernel's default instead of copying a number. Copied masks are the usual defect here, because they set bits the kernel deliberately leaves clear.


printf 'amdgpu.ppfeaturemask=0x%x\n' \
  "$(( $(cat /sys/module/amdgpu/parameters/ppfeaturemask) & ~0x8000 ))"

0x8000 is PP_GFXOFF_MASK in drivers/gpu/drm/amd/include/amd_shared.h. On kernel 7.1 the driver default is 0xfff7bfff, so the command above prints:


amdgpu.ppfeaturemask=0xfff73fff

**Do not use 0xffff7fff.** It sets every bit except GFXOFF, which switches on two features kernel 7.1 keeps off on purpose: PP_OVERDRIVE_MASK (0x4000) and PP_GFX_DCS_MASK (0x80000). Enabling Overdrive taints the kernel and logs Overdrive is enabled, please disable it before reporting any bugs unrelated to overdrive., and the Arch wiki warns that turning on undefined or unstable PowerPlay bits can itself cause screen flicker or broken resume from suspend.

If freezes persist, clear more bits from the same starting point:


0xfff53fff   GFXOFF plus PP_STUTTER_MODE (0x20000) cleared
0            every PowerPlay feature off, diagnostic only, and it costs power management

**Apply it on Omarchy 4 in a Limine drop-in.** There is no /etc/default/grub, and editing /boot/limine.conf does nothing, because Omarchy boots a Unified Kernel Image with the command line baked into the .efi and regenerates /boot/limine.conf on every kernel transaction. Use your own drop-in that sorts after omarchy-defaults.conf, with +=:


sudo tee /etc/limine-entry-tool.d/zz-local.conf >/dev/null <<'EOF'
# `+=` appends. A bare `=` REPLACES Omarchy's defaults and drops
# quiet/splash/initramfs_async=0.
KERNEL_CMDLINE[default]+=" amdgpu.ppfeaturemask=0xfff73fff"
EOF

sudo limine-mkinitcpio
sudo limine-update
sudo reboot

The full mechanism, including testing a parameter once from the Limine boot menu with nothing written to disk, is in the record limine-kernel-parameters-not-applying-omarchy.

**On other Arch-based systems:**

  • Limine: the same /etc/limine-entry-tool.d/*.conf drop-in, then sudo limine-update
  • systemd-boot: the options line in /boot/loader/entries/*.conf
  • GRUB: GRUB_CMDLINE_LINUX_DEFAULT in /etc/default/grub, then sudo grub-mkconfig -o /boot/grub/grub.cfg

Reboot.

Verify


cat /proc/cmdline                                  # shows amdgpu.ppfeaturemask=0xfff73fff
cat /sys/module/amdgpu/parameters/ppfeaturemask    # shows the same value

On Omarchy, /proc/cmdline must also still show Omarchy's own defaults (quiet splash loglevel=0 ... initramfs_async=0). If they are gone, your drop-in used a bare = or sorts before omarchy-defaults.conf.

Confirm no Overdrive taint was picked up:


sudo dmesg | grep -i overdrive     # must print nothing

Then leave the machine idle overnight with the monitors attached and check it is still responsive in the morning. As root, /sys/kernel/debug/dri/0/amdgpu_gfxoff_status reports the current GFXOFF state if debugfs is mounted.

Sources

amdgpu-idle-freeze-gfxoff-ppfeaturemask