Recover from "device '' not found" dropping to an emergency shell after the Quattro upgrade
Checked every claim against local files on this omarchy 4.0.2-1 workstation and against omacom/omarchy#6894 and #6951 read in full today. The failure itself cannot be exercised here: this machine boots, it is already pinned, I have no sudo, and nothing was rebuilt, no UKI written and no VM used. So the mechanism is confirmed from source and from config files, and the recovery steps are confirmed as correct commands on Omarchy 4 rather than as a completed recovery.
Confirmed on this machine. /etc/limine-entry-tool.d/omarchy-defaults.conf does use KERNEL_CMDLINE[default]+= for quiet splash and for initramfs_async=0, and /etc/default/limine does carry a pinned cryptdevice= and root= line, exactly as the record says. /proc/cmdline shows all of those parameters on the running kernel, which is the strongest check available here: that file and that += syntax demonstrably produce a working cmdline. The += mechanism is documented by the package, not only asserted by a collaborator: /etc/limine-entry-tool.conf and line 380 of /usr/share/doc/limine-entry-tool/README.md both say += "Ignores /etc/kernel/cmdline and /proc/cmdline", and load_config in /usr/lib/limine/limine-common-functions loads /etc/default/limine last of four layers. preserve_kernel_cmdline_root is at line 520 of /usr/share/omarchy/bin/omarchy-upgrade-to-quattro and is called at line 2362, after install_omarchy_quattro_packages at 2359, so the record's window claim holds on the installed version, and it still holds at tag v4.0.3 (lines 2360 and 2363), which I fetched because 4.0.2 is no longer the newest release. #6951 is still open. #6894's collaborator comment supports the cause in detail and recommends the same recovery, appending to /etc/default/limine with += and running limine-mkinitcpio, so the citation genuinely supports the claim. Issue #6894 is closed as completed, which the record did not say.
limine-mkinitcpio over the preset form is correct, which I was asked to confirm. /etc/mkinitcpio.d/ is empty here, and /usr/bin/mkinitcpio line 986 dies with No presets found in %s when it is, so the record's quoted message is right. The underlying reason is firmer than the record gave: limine-mkinitcpio-hook ships /etc/pacman.d/hooks/90-mkinitcpio-install.hook, which overrides Arch's hook of the same name and calls limine-mkinitcpio-install instead of the script that generates presets, and mkinitcpio on PATH resolves to /usr/local/bin/mkinitcpio, a wrapper from that same package which warns "This does not update Limine boot entries" and offers limine-mkinitcpio. I also checked that the quattro upgrade never touches /etc/mkinitcpio.d/, so an upgraded machine can still hold a stale preset and mkinitcpio -P can run there and leave the boot entries untouched. That is the stronger reason to avoid it, so I put it first. I checked sudo pacman -S linux against /usr/bin/omarchy-update-pacman-guard: the guard aborts only when both -S and -u are present, so that command is not blocked and the record is not proposing a blocked path.
What was wrong. The fix's one worked example is root=/dev/mapper/root, and the Omarchy 4 ISO opens LUKS as omarchy_root (configurator on the quattro branch of omacom/omarchy-iso), so /dev/mapper/omarchy_root. On a boot fix, one example with no way to tell which name applies is a defect, so I rewrote the fix to derive the name from ls -l /dev/mapper/ and to give the encrypted and unencrypted forms separately, with cryptdevice= and root= agreeing on the mapper name. Second, the record tells the reader to get real values off the machine but never warns that the values on hand come from a snapshot boot: /proc/cmdline there carries the snapshot's rootflags=subvol= and not @, and copying it reproduces the unbootable state. The package warns about this itself in /etc/limine-entry-tool.conf, and it now appears in both the fix and the danger. Everything else in the record held and I left it as written.
You are editing the kernel command line on a machine that already will not boot. Keep += rather than = in /etc/default/limine, because a bare = replaces Omarchy's own defaults, initramfs_async=0 among them. Read the device names and the decrypted root's mapper name off the machine with lsblk -f, blkid and ls -l /dev/mapper/ instead of copying the example: an ISO-installed machine is omarchy_root and an upgraded one is commonly root, and the wrong name is another failed boot. Do not copy rootflags= out of /proc/cmdline while you are booted from a snapshot, because it names the snapshot's subvolume rather than @, which is the same warning the package prints in /etc/limine-entry-tool.conf. If you rebuild while booted from a snapshot, restore that snapshot over @ before rebooting, or the new entry stays rooted on the snapshot.
Symptom
The machine does not boot after upgrading to Omarchy 4 and lands in an initramfs emergency shell. The device in the message is empty quotes, not a UUID:
:: running hook [keymap]
:: Loading keymap...done
Error: device '' not found. Skipping fsck
:: mounting '' on real root
mount: /new_root: wrong fs type, bad option, bad superblock on , missing codepage or helper program, or other error.
:: running emergency hook [plymouth]
ERROR: Failed to mount '' on real root
sh: can't access tty; job control turned off
[rootfs ~]#
The empty quotes are the diagnostic detail. They mean the initramfs parsed a kernel command line carrying no root= at all, rather than a root= it could not resolve. A missing hook or a stale UUID would have printed the device name it failed to find.
A second reporter whose upgrade was interrupted mid-transaction hit the same message with two additional symptoms: the active entry in /boot/limine.conf had its cmdline: line truncated to initramfs_async=0 alone, and the kernel image that entry pointed at was missing because the pacman transaction never finished.
Cause
Established by collaborator triage on omacom/omarchy#6894, which is now closed as completed. /etc/limine-entry-tool.d/omarchy-defaults.conf appends to KERNEL_CMDLINE[default] with +=, and += by design stops limine-entry-tool falling back to /etc/kernel/cmdline or /proc/cmdline. On an install that predates the ISO pinning root= in /etc/default/limine, the moment that drop-in lands any UKI rebuilt afterwards carries only the drop-in's own parameters, and root= is gone. omarchy-upgrade-to-quattro repairs this in preserve_kernel_cmdline_root, but that runs after the package transaction, so there is a window in which the machine will not boot.
Confirmed on an omarchy 4.0.2-1 install. /etc/limine-entry-tool.d/omarchy-defaults.conf uses KERNEL_CMDLINE[default]+= for the quiet splash and initramfs_async=0 parameters, and /etc/default/limine carries the pinned cryptdevice= and root= line that a repaired or newly installed machine has. /proc/cmdline on the running kernel shows those parameters, which is what proves that file and that syntax are the ones that take effect.
The += behaviour is documented by the package itself, in /etc/limine-entry-tool.conf and at line 380 of /usr/share/doc/limine-entry-tool/README.md: += "Ignores /etc/kernel/cmdline and /proc/cmdline". /etc/default/limine is the last of the four layers load_config reads in /usr/lib/limine/limine-common-functions, which is why pinning there wins and why the same file recommends += rather than = there. The installer refuses to finish an install whose assembled config has no root= (orchestrator/phases_impl.py:1288 on the quattro branch of omacom/omarchy-iso), and that is the pinning an upgraded machine never received.
omacom/omarchy#6951, which pins root= before the packages that can drop it, is still open on 2026-09-11, so the window is still there. It is still there in the newest release: at tag v4.0.3, published 2026-09-08, omarchy-upgrade-to-quattro still calls install_omarchy_quattro_packages at line 2360 and preserve_kernel_cmdline_root only at line 2363.
Fix
Boot a working kernel first. The Limine menu's btrfs snapshot entries point at complete kernel copies, so pick one of those rather than the broken linux entry.
Read the real values off the machine before you write anything:
lsblk -f # the LUKS container and the btrfs root
blkid # PARTUUIDs and UUIDs
ls -l /dev/mapper/ # the decrypted root's actual mapper name
cat /proc/cmdline # reference only, see the warning below
Two things that catch people here. The mapper name is not the same on every install: a machine installed by the Omarchy 4 ISO opens LUKS as omarchy_root, so /dev/mapper/omarchy_root, while a machine that came up through earlier releases is commonly /dev/mapper/root. Use what ls -l /dev/mapper/ shows. And while you are booted from a snapshot, /proc/cmdline carries the snapshot's rootflags=subvol= value, not @, so do not copy that parameter out of it. The package warns about exactly this in /etc/limine-entry-tool.conf.
Then pin the parameters the UKI is missing, in the file Omarchy reads, and rebuild. On an encrypted root, cryptdevice= names the LUKS partition and the mapper name to create, and root= names that mapper:
# add to /etc/default/limine, keeping += so Omarchy's own defaults survive
sudo tee -a /etc/default/limine >/dev/null <<'EOF'
KERNEL_CMDLINE[default]+=" cryptdevice=PARTUUID=<partuuid-of-the-luks-partition>:omarchy_root root=/dev/mapper/omarchy_root rootflags=subvol=@ rw rootfstype=btrfs"
EOF
sudo limine-mkinitcpio
On an unencrypted root, root= names the btrfs partition itself, by UUID rather than by a device node that can move:
sudo tee -a /etc/default/limine >/dev/null <<'EOF'
KERNEL_CMDLINE[default]+=" root=UUID=<uuid-of-the-btrfs-partition> rootflags=subvol=@ rw rootfstype=btrfs"
EOF
sudo limine-mkinitcpio
If the upgrade was interrupted and the kernel image itself is missing, finish the transaction from the snapshot before rebuilding:
sudo mount -o remount,rw /
sudo pacman -S linux
sudo limine-mkinitcpio
That pacman -S is not blocked by Omarchy's ALPM guard, which aborts only when both -S and -u are present. Do not turn it into pacman -Syu: that is the guard's case, and on a half-finished upgrade it is a partial-upgrade risk as well.
Use limine-mkinitcpio, not the mkinitcpio preset form. The preset form does not update the Limine boot entries or rebuild the UKI, which is the whole of what this machine needs. On a machine installed by the Omarchy 4 ISO it does not even run: limine-mkinitcpio-hook ships /etc/pacman.d/hooks/90-mkinitcpio-install.hook, which overrides Arch's own hook and never generates presets, so /etc/mkinitcpio.d/ is empty and mkinitcpio -P stops with No presets found in /etc/mkinitcpio.d. An upgraded machine may still have a preset file left from before, in which case mkinitcpio -P runs and silently leaves the boot entries stale. mkinitcpio on PATH is itself a wrapper from that package at /usr/local/bin/mkinitcpio, and it answers a preset run by warning that it did not update the Limine boot entries and offering to run limine-mkinitcpio for you.
One trap from the reporter who recovered this way. Running the rebuild while booted from a snapshot roots the new boot entry on the snapshot rather than on the real @ subvolume, and Omarchy notifies you to restore the snapshot before rebooting. Use the snapshot tool's restore menu to replace @ with the snapshot you are running from, then reboot into the normal entry.
Verify
cat /proc/cmdline # root= is present, with cryptdevice= on LUKS
mount | grep ' / ' # subvol=/@, not a snapshot path
omarchy-migrate --pending # empty, if the upgrade had stalled
Sources
- https://github.com/omacom/omarchy/issues/6894
- https://github.com/omacom/omarchy/pull/6951
- https://github.com/omacom/omarchy/blob/v4.0.3/bin/omarchy-upgrade-to-quattro
- https://github.com/omacom/omarchy-iso/blob/quattro/configs/airootfs/root/configurator
- https://github.com/omacom/omarchy-iso/blob/quattro/configs/airootfs/usr/share/omarchy-iso/orchestrator/phases_impl.py