Suspend, sleep and resume on Omarchy

Suspend is enabled by default on Omarchy 4.x. Hibernation is not: you opt into it. This page covers what the shipped system actually does around sleep, which failures are real and current on 4.0.4, and the order to try fixes in. Everything below was checked against the v4.0.4 source tree and against the open sleep issues, most of them filed on 4.0.x. The manual chapter is System sleep.

Status on 4.0.4

Suspend is on by default and the manual’s advice is to try it and hide the menu entry if it misbehaves. There are 460 issues whose text touches sleep, and the ones filed on 4.0.x split cleanly in two. The hard failures, where the machine never resumes, hangs on the way into sleep, or comes back with a corrupted panel, cluster on NVIDIA and hybrid NVIDIA laptops, Intel Meteor Lake panels that freeze for about a minute on resume, Intel Macs, and whatever the current kernel has just broken. Those live in the kernel, the GPU driver, or firmware, which is why the first fix below is a kernel swap rather than a config edit. The softer failures, where the machine is fine but something around the lock screen is wrong afterwards, are Omarchy’s own: a wake script that skips a dark monitor, a lock timer that never blanks again, a fingerprint retry loop, a lid race with logind.

Hibernation is a different story. It is opt-in, it requires Limine, and on NVIDIA it is still openly unfinished. Treat working hibernation as a nice surprise rather than a baseline.

What Omarchy does automatically

Locking before sleep is the part Omarchy owns. A user unit, omarchy-sleep-lock.service, runs omarchy-system-sleep-monitor, which holds a systemd-inhibit --what=sleep --mode=delay lock and watches logind’s PrepareForSleep signal. When the signal arrives it runs omarchy-system-sleep-lock, which asks Quickshell to lock and polls until the session reports secure. The catch with a delay inhibitor is that logind only waits so long: when the window expires it suspends whether or not the lock landed. Omarchy ships /etc/systemd/logind.conf.d/20-inhibit-delay.conf raising InhibitDelayMaxSec from the 5 second default to 15, and the lock script reads that value back from logind at runtime and keeps a fifth of it in reserve, capped at 12 seconds. If it loses the race you get a critical notification saying the session was left unlocked.

Closing the lid is bound in Hyprland, not just in logind. default/hypr/bindings/utilities.lua binds switch:on:Lid Switch to omarchy-system-lid-close, which locks immediately, unless external monitors are connected, and then reconciles clamshell state. The point of locking on the lid event instead of on PrepareForSleep is timing: the shell is usually already secure by the time the sleep monitor asks. Omarchy ships no logind lid drop-in, so logind’s compiled default of HandleLidSwitch=suspend is still what actually triggers the suspend, and both paths fire on the same close.

Three system-sleep hooks ship in default/systemd/system-sleep/:

• unmount-fuse lazily unmounts gvfsd-fuse mounts before sleep and restarts gvfs afterwards. A FUSE daemon stuck in uninterruptible sleep used to time out the 20 second process freeze, so suspend silently failed and the machine stayed fully awake with the lid shut, draining the battery. Added by PR #4940, merged 2026-03-10 and first shipped in v3.5.0.
• keyboard-backlight turns the keyboard backlight off before hibernate, because some ASUS LED controllers block S4.
• force-igpu uses supergfxctl to park a discrete GPU in Vfio before hibernate and restore Integrated mode afterwards, since the NVIDIA driver cannot freeze a powered-off dGPU.

Apple hardware gets two install-time quirks. install/hardware/apple/fix-suspend-nvme.sh installs a service that clears d3cold_allowed on the NVMe controller for MacBook8,1, 9,1, 10,1 and MacBookPro13,x and 14,x. fix-t2.sh adds pm_async=off mem_sleep_default=deep to the Limine cmdline on T2 Macs, which v4.0.0 listed as fixing T2 suspend and fan defaults.

Hibernation setup is a single command, omarchy hibernation setup. Nothing in the 4.0.4 installer calls it, so a fresh machine has no swapfile and no Hibernate entry until you do. It creates a /swap btrfs subvolume with a swapfile the size of your RAM, marks it nodatacow, adds the fstab entry, installs the keyboard-backlight hook, adds the resume mkinitcpio hook, writes resume= and resume_offset= into a Limine drop-in, adds rtc_cmos.use_acpi_alarm=1 on s2idle systems, and rebuilds the UKI. omarchy hibernation remove takes the swap and the mkinitcpio hook back out, but as #12096 shows it leaves the Limine drop-ins behind.

Known problems

Issue	Models	Status	Fixed in
#12190 suspend entry hangs on linux-omarchy 7.2.5-3, stock linux 7.2.3 fine	Dell XPS 13 DX13260, Intel Wildcat Lake	open	not fixed
#12129 black screen, no wake	ASUS TUF F17, RTX 4070 Mobile	open	not fixed
#12041 rainbow corruption after lid resume, survives shell restart	Lenovo Slim Pro 7 14ARP8, AMD iGPU plus RTX 3050	open	not fixed
#5695 i915 commit timeouts, roughly 60s freeze on resume	ThinkPad P1 Gen 7, Alienware m16 R2, Meteor Lake	open	not fixed
#12194 Thunderbolt PME wakes the machine every 45s with the lid shut	MacBookPro14,2 and Alpine Ridge siblings	open	not fixed
#12193 lid close then quick reopen leaves a dark screen	any laptop, reported on MacBookPro14,2	open, PR #12210	not fixed
#12147 one monitor stays dark after wake while Hyprland reports DPMS on	LG 4K over DisplayPort in a four monitor hybrid Intel plus RTX 5070 setup	open	not fixed
#12140 lock screen never blanks again after a resume	reported on an AMD RX 9060 XT desktop, cause is in the shell’s lock timer	open	not fixed
#5554 hibernate resume aborts to a fresh boot, or the machine never powers off, or the display comes back corrupted	NVIDIA desktops and laptops, results differ between hybrid and dGPU-only	open, tracking issue	not fixed
#12096 hibernation remove leaves stale resume= in the UKI	any	open	not fixed
#12095 shipped usbcore autosuspend drop-in is a no-op, BLE mice fail to reconnect	Intel Bluetooth, AX201 and similar	open	not fixed
#11412 fingerprint unlock dead after suspend, fprintd stuck busy	Framework Laptop 13 and ThinkPad T14 Gen 7, Goodix sensor	open	not fixed
#2635 monitors get no signal after wake, machine still running	NVIDIA 3070 first, then AMD RX 9070 XT, RX 580, Beelink SER5 and SER9	open	not fixed
#4184 no wake plus heavy battery drain during sleep	Framework 13 AMD	open, FUSE freeze cause fixed	v3.5.0 for the FUSE freeze via PR #4940

Two patterns are worth naming. First, several of these reproduce outside Omarchy. In #5695 a reporter hit the same i915 timeouts on Ubuntu with GNOME on the same ThinkPad, and the same reporter measured zero events on a 6.17 kernel against 17 on 7.0. Second, #12190 is a clean kernel bisect by package: 13 suspends out of 13 on stock linux 7.2.3, one successful suspend entry out of four and zero resumes on linux-omarchy 7.2.5-3, which is the kernel v4.0.4 shipped to everyone.

Fixes that work

Work in this order.

1. Update, then reboot. omarchy update gets you to 4.0.4 and the current shell.
2. Read the journal from the boot after the failure, not the failing one. Once userspace freezes, journald stops, so a hard power-off loses whatever the kernel printed but had not flushed. journalctl -b -1 -k | grep -E "PM: suspend (entry|exit)" tells you whether the kernel entered sleep, resumed, or never came back.
3. Swap kernels. If suspend entry hangs or resume never happens, boot stock linux, or linux-lts for Meteor Lake panel freezes, and test the same cycle. It is the step with the clearest evidence behind it right now: #12190 on the shipped kernel and #5695 on 7.x both went away with a different kernel package.
4. On NVIDIA, enable nvidia-suspend.service, nvidia-hibernate.service and nvidia-resume.service, then check cat /proc/driver/nvidia/params after a reboot. Omarchy ships no NVIDIA sleep parameters of its own. On the 610 driver 4.x installs, UseKernelSuspendNotifiers is the parameter that matters, and the finding that repeats through #5554 is that PreserveVideoMemoryAllocations=1, which gpu-screen-recorder’s gsr-nvidia.conf sets behind your back, makes hibernate resume abort with nv_pmops_freeze returns -5 when the driver is loaded from the initramfs. Reporters there got resume working by setting it to 0 or by disabling the early KMS drop-in /etc/mkinitcpio.conf.d/nvidia.conf, mostly on hybrid laptops; dGPU-only desktops had mixed results. Rerun limine-update after any change. The services alone did not help the #12129 reporter.
5. If only the screen is dead and the machine is alive over SSH, force a real DPMS transition rather than a redundant enable: hyprctl dispatch 'hl.dsp.dpms({ action = "disable" })' then the same with enable. A plain omarchy system wake can no-op when Hyprland already believes the panel is lit.
6. If the machine wakes itself, look at /proc/acpi/wakeup for Thunderbolt root ports, xHCI controllers and the Wi-Fi PME, and disable the ones you do not need to wake on.
7. If nothing helps, omarchy toggle suspend removes Suspend from the system menu so you stop triggering a known-bad path.

For hibernation specifically: it needs Limine, free space equal to your RAM, and a correct resume_offset. Verify with btrfs inspect-internal map-swapfile -r /swap/swapfile against the value in /etc/limine-entry-tool.d/resume.conf. Use systemctl start systemd-hibernate.service when testing from a script, because systemctl hibernate hands the request to logind and returns before anything has happened.

Report it

Sleep bugs are only actionable with the boot after the failure attached. Run omarchy debug, pick Upload log, and paste the logs.omarchy.org URL into the issue. For lock and idle problems specifically, omarchy debug idle dumps the shell’s idle state, the sleep-lock unit status, and current idle inhibitors. Include your exact kernel package and version, since linux-omarchy and stock linux behave differently, plus cat /sys/power/mem_sleep so it is clear whether you are on s2idle or deep. If you can, report the result on both kernels. That side by side count of triggered, entered and resumed cycles per kernel is what makes #12190 a clean report.

Related

Component pages: NVIDIA, hybrid GPU laptops, multi-monitor, battery and power, T2 Macs, Bluetooth.

Fix pages: suspend will not resume on s2idle, hibernate fails or hangs, Bluetooth stops after resume, black screen after login, lock screen will not unlock, battery drains fast.