Fix hibernation that powers off but boots a fresh session instead of resuming
The underlying problem is real: a busybox initramfs with no resume hook, or with no resume=/resume_offset= on the kernel cmdline, will not resume. Four other claims are wrong on Omarchy 4 and one is wrong on plain Arch too. First, the cause's mechanism is false. /usr/lib/initcpio/install/filesystems (mkinitcpio 41.1-1) has only a build() function and there is no /usr/lib/initcpio/hooks/filesystems, and /usr/lib/initcpio/init runs run_hookfunctions 'run_hook' 'hook' $HOOKS and only then calls fsck_root and "$mount_handler" /new_root. Root is mounted after every run_hook, whatever the array position, so resume after filesystems is not too late. The Arch wiki page this record already cites agrees: its own example is HOOKS=(base udev autodetect microcode modconf kms keyboard keymap consolefont block filesystems resume fsck), with resume after filesystems, and it requires only that resume follow udev and follow encrypt or lvm2. Confirmed live on this workstation, where the effective array ends ... filesystems fsck btrfs-overlayfs resume and /sys/power/resume still reads 253:0, matching /dev/mapper/root, so the hook ran and resolved the device from last position. Second, sudo mkinitcpio -P fails outright on Omarchy 4: /etc/mkinitcpio.d/ is empty, and /usr/bin/mkinitcpio line 986 does [[ -e "${_optpreset[0]}" ]] || die 'No presets found in %s'. Presets are empty because limine-mkinitcpio-hook ships /etc/pacman.d/hooks/90-mkinitcpio-install.hook, which overrides the stock hook and execs limine-mkinitcpio-install instead of /usr/share/libalpm/scripts/mkinitcpio, and only the stock script's generate_presets writes /etc/mkinitcpio.d/<pkgbase>.preset. Third, the fix's sample HOOKS=(...) line is not Omarchy's array and editing /etc/mkinitcpio.conf on Omarchy has no effect, because /etc/mkinitcpio.conf.d/omarchy_hooks.conf (owned by omarchy-settings 4.0.2-1) assigns HOOKS wholesale afterwards. Fourth, the danger tells the reader to keep a fallback boot entry, and Omarchy 4 has none: MKINITCPIO_FALLBACK is set nowhere, and limine-mkinitcpio-install removes any fallback UKI when it is unset. The real recovery path is a Snapper snapshot entry, with snapper 0.13.1-3 and limine-snapper-sync 1.31.0-1 installed and BOOT_ORDER="*, *fallback, Snapshots" in /etc/limine-entry-tool.d/omarchy-defaults.conf. Fifth, the verify names /boot/initramfs-linux.img, which Omarchy 4 does not keep: ENABLE_UKI=yes and CUSTOM_UKI_NAME="omarchy" make limine-mkinitcpio-install build /boot/EFI/Linux/omarchy_linux.efi and delete the separate images, and /boot is a vfat ESP mounted dmask=0077, so ls /boot fails for a normal user. The cited issue does not support the record: omacom/omarchy 8471 is open with zero comments and its author writes that he has not been able to demonstrate a resume failure attributable to the ordering. Two sibling reports, 8888 (an open unmerged pull request) and 10375, repeat the same wrong mechanism and are also unconfirmed. I dropped the issue URL and added the quattro copy of omarchy-hibernation-setup, which is what actually writes the drop-ins. Severity lowered from critical to high because the consequence is a lost session on a machine that boots normally, not an unbootable system or data loss. NOT exercised: I did not run a hibernate and power-on cycle, did not rebuild an initramfs, and could not read /boot, which needs root.
Merged on 2026-09-13: omarchy-resume-hook-appended-after-filesystems was retired into this record and its page now 404s. That record asserted, in its slug and its title, that Omarchy's drop-in landing after filesystems breaks resume. Its 2026-09-12 re-audit returned reject: the mechanics were right, the defect was not real, and it was disproved live on the development workstation, where /sys/power/resume reads 253:0 against /dev/mapper/root, so the hook resolved the encrypted device from last position. It was first kept and rewritten to say so, then merged here because its slug could not be corrected and a reader arriving by that slug was being told the wrong thing before reading a word. Everything it held that was worth keeping is in this record's cause, which already stated the ordering rule, and in the danger above. Its full text and its verdict are in the git history and in raw/o3-power-suspend-audit.json.
A malformed HOOKS array produces an initramfs that cannot mount root, which is an unbootable system.
On Omarchy 4 there is no fallback boot entry to fall back to. MKINITCPIO_FALLBACK is set nowhere, and limine-mkinitcpio-install deletes any fallback UKI while it is unset, so the Limine menu carries one Omarchy entry plus Snapper snapshots. ENABLE_LIMINE_FALLBACK=yes in /etc/limine-entry-tool.d/omarchy-defaults.conf is about installing Limine on the UEFI fallback path and does not give you a fallback initramfs. Take a snapshot before rebuilding and know how to boot one from the Limine menu:
omarchy-snapshot create
On plain Arch, keep the fallback preset entry and confirm it boots before relying on it. Have installation media on hand either way.
**Do not apply the "move resume before filesystems" fix that circulates for this.** Issues 8471, 8888 and 10375 on omacom/omarchy all assert that Omarchy's HOOKS+=(resume) drop-in landing last breaks resume. It does not: resume at the end still runs before root is mounted, and it still lands after encrypt, which is the only ordering that matters. The remedy those issues propose breaks a working machine in three ways. Deleting /etc/mkinitcpio.conf.d/omarchy_resume.conf breaks omarchy-hibernation-available and omarchy-hibernation-remove, which both test that exact path for ^HOOKS+=(resume)$, so the menu reports hibernation unavailable and it can never be removed cleanly, and re-running omarchy-hibernation-setup recreates the file so the array then carries resume twice. Copying the output of grep -h '^HOOKS' into a hand-written array reinstates kms, because the printed line always contains it while omarchy_hooks.conf strips it at build time on machines where NVIDIA drives every display, which drags nouveau and roughly 100 MB of GSP firmware back into the initramfs. And the array published in those issues omits plymouth, which Omarchy 4 ships and needs for the themed LUKS prompt.
Symptom
Hibernate works, the machine writes to disk and powers off cleanly, but on the next boot it just boots normally and my session is gone. No error message anywhere obvious.
Cause
The initramfs never attempted the resume. Two separate things have to be in place and either can be missing.
First, on a busybox-based initramfs the resume hook has to be in HOOKS. It has to come after the hooks that make the resume device available, which means after udev, and after encrypt or lvm2 when the swap sits on LUKS or LVM. Its position relative to filesystems and fsck does not matter. Neither of those hooks has a runtime component, and /usr/lib/initcpio/init runs every run_hook function before it fsck's or mounts the root filesystem, so a resume at the end of the array still runs before root is mounted.
Second, the kernel has to be told where the hibernation image lives, through resume= and, for a swapfile, resume_offset=. On systemd 255 and later the resume hook installs and calls /usr/lib/systemd/systemd-hibernate-resume, which can also read the HibernateLocation EFI variable, so a UEFI machine sometimes resumes without those parameters and a BIOS machine will not.
On a systemd-based initramfs, one using the systemd hook instead of udev, the resume mechanism is built in and a resume hook must not be added at all.
Fix
Print the array mkinitcpio will actually use. Drop-ins are concatenated onto /etc/mkinitcpio.conf in version-sorted basename order, so an assignment or an append in a later file wins:
bash -c 'source /etc/mkinitcpio.conf
for f in /etc/mkinitcpio.conf.d/*.conf; do source "$f"; done
printf "%s\n" "${HOOKS[@]}" | nl'
resume has to appear somewhere after udev, and after encrypt or lvm2 if the swap is on LUKS or LVM. It does not have to precede filesystems.
**Omarchy 4.** Do not edit /etc/mkinitcpio.conf and do not run mkinitcpio -P. /etc/mkinitcpio.conf is package-stock and HOOKS is assigned by /etc/mkinitcpio.conf.d/omarchy_hooks.conf, so an edit to the main file is overwritten at build time. /etc/mkinitcpio.d/ is empty on Omarchy, so mkinitcpio -P exits with No presets found in /etc/mkinitcpio.d. Let Omarchy configure it:
omarchy-hibernation-setup
That creates the /swap subvolume and /swap/swapfile, adds it to /etc/fstab, writes HOOKS+=(resume) into /etc/mkinitcpio.conf.d/omarchy_resume.conf, writes resume= and resume_offset= into /etc/limine-entry-tool.d/resume.conf, and rebuilds the UKI. Check what it produced:
cat /etc/mkinitcpio.conf.d/omarchy_resume.conf
cat /etc/limine-entry-tool.d/resume.conf
omarchy-hibernation-available && echo available
If only the initramfs and boot entries need rebuilding:
sudo limine-mkinitcpio
**Plain Arch, EndeavourOS, CachyOS, Manjaro.** Put resume in HOOKS in /etc/mkinitcpio.conf after udev, and after encrypt or lvm2 if they are present:
HOOKS=(base udev autodetect microcode modconf kms keyboard keymap consolefont block encrypt lvm2 filesystems resume fsck)
sudo mkinitcpio -P
**Both.** Confirm the kernel is told where the image is:
tr ' ' '\n' </proc/cmdline | grep -E '^resume'
For a swapfile you need resume= naming the device that holds it and resume_offset= for the file's physical offset. On btrfs:
sudo btrfs inspect-internal map-swapfile -r /swap/swapfile
On Omarchy set those in /etc/limine-entry-tool.d/resume.conf, not in /etc/default/limine, and rebuild with sudo limine-mkinitcpio.
**systemd-based initramfs.** If the array contains systemd rather than udev, resume is already provided. Do not add a resume hook. Check resume= and resume_offset= instead.
Verify
After a reboot with hibernation configured, the resume hook has run if /sys/power/resume names the resume device rather than 0:0:
cat /sys/power/resume # for example 253:0
cat /sys/power/resume_offset # non-zero when hibernating to a swapfile
ls -l /dev/mapper/root # major and minor must match /sys/power/resume
On Omarchy also:
omarchy-hibernation-available && echo available
Then hibernate and power the machine on. journalctl --list-boots shows no new boot, and the session is still the one you left.
Do not use lsinitcpio /boot/initramfs-linux.img on Omarchy 4. limine-mkinitcpio-install builds /boot/EFI/Linux/omarchy_linux.efi and deletes the separate initramfs and vmlinuz files, and /boot is a vfat ESP mounted dmask=0077, so it cannot be listed without root. On plain Arch, where the image is kept:
sudo lsinitcpio -l /boot/initramfs-linux.img | grep resume