OPINIONATED OMARCHY
TROUBLESHOOTING CORPUS
790
RECORDS
790
AUDITED
0
UNCHECKED
2388
SOURCES
LIVE
boot-kernel critical occasional DISPUTE

/boot (the ESP) was not mounted during the upgrade, so the new kernel went to the root filesystem

omarchyarchendeavouroscachyosmanjarouefiliminesystemd-bootgrubdesktoplaptop
CORRECTED An auditor found the fix wrong or incomplete and rewrote it.
AUDIT NOTE

Checked on this Omarchy 4 machine and against the mkinitcpio and limine-entry-tool sources. On Omarchy 4 /boot is the vfat ESP from fstab (fmask=0077,dmask=0077, confirmed with findmnt), the kernel package ships only /usr/lib/modules/<ver>/vmlinuz, and limine-mkinitcpio-hook overrides the stock hook with /etc/pacman.d/hooks/90-mkinitcpio-install.hook (same name, higher priority per alpm-hooks(5)), whose script runs initialize_header in /usr/lib/limine/limine-common-functions before touching anything. With the ESP unmounted that fails on check_boot_partition (Boot path '/boot' is not a mounted FAT32 boot partition.) or, without ESP_PATH, on find_boot_partition, and exits 1, so pacman prints error: command failed to execute correctly and nothing is written into the root-filesystem /boot directory. The record's symptom ("no errors at all") and cause (stray vmlinuz/initramfs on the root filesystem) therefore describe plain Arch, not Omarchy 4, though the consequence is the same: modules replaced, boot image not rebuilt, old kernel boots with no modules. Also wrong for Omarchy 4: mkinitcpio -P cannot serve as a fallback because /etc/mkinitcpio.d is empty here and mkinitcpio 41.1 dies with No presets found (line 986 of /usr/bin/mkinitcpio), and ls /boot needs sudo. What held: findmnt/lsblk diagnosis, the wiki ESP mount options and the vfat modules-load advice, daemon-reload before mount, pacman -S linux linux-firmware passing omarchy-update-pacman-guard (it aborts only when -S and -u are both present), the UKI path /boot/EFI/Linux/omarchy_linux.efi (CUSTOM_UKI_NAME=omarchy and the ${prefix}_${kernel}.efi pattern in limine-mkinitcpio-install, also used by omarchy-refresh-limine), limine-mkinitcpio and limine-update as real commands (limine-update runs limine-install --no-efi-register then limine-mkinitcpio), subvol=@ and the root mapper name matching this machine's cmdline, and the old mkinitcpio warning being absent from both the installed 41.1 script and master. bbs 194153 quotes that old warning and the chroot-and-reinstall cure. bbs 285144 is a uname/pacman -Q mismatch thread and only weakly supports the record. Not exercised: an actual unmounted-ESP upgrade, the chroot path, or GRUB and systemd-boot.

The cause above was rewritten on 2026-09-06 to match this note; the fix was corrected by the audit itself.
RISK

Never rm -rf /boot while unsure whether the ESP is mounted. With the ESP mounted you delete your only boot loader and kernel. Move it aside instead. On a dual-boot machine the ESP also holds Windows' boot loader, so do not reformat it. If you chroot from a live USB onto btrfs, get subvol=@ right or you will repair an empty top-level subvolume.

Symptom

The kernel upgrade ran, but the next boot fails: the boot loader cannot find the kernel, or the system boots and then everything modular breaks (no Wi-Fi, no GPU, modprobe: FATAL: Module ... not found in directory /lib/modules/7.0.3-arch1-2), or you land in an emergency shell. Once you get a shell, uname -r reports an older version than pacman -Q linux, and sudo ls /boot shows either an empty directory or kernels that do not match.

On plain Arch the upgrade printed no warning at all: current mkinitcpio no longer prints the old "/boot appears to be a separate partition but is not mounted" message. On Omarchy 4 the Limine hook did complain, and it is easy to scroll past inside omarchy update:


ERROR: Boot path '/boot' is not a mounted FAT32 boot partition.
error: command failed to execute correctly

The transaction still finished, because a failing PostTransaction hook does not abort pacman.

Cause

The EFI system partition was not mounted at /boot when the transaction ran, usually because the fstab entry is missing/wrong, because the mount silently failed earlier in the session (vfat module not loaded, dirty FAT flagged errors=remount-ro), or because the ESP is mounted somewhere else and you relied on systemd automount.

Plain Arch: mkinitcpio's pacman hook happily writes vmlinuz-linux and initramfs-linux.img into the plain /boot *directory* on the root filesystem, which the firmware cannot read.

Omarchy 4: limine-mkinitcpio-hook replaces that hook with /etc/pacman.d/hooks/90-mkinitcpio-install.hook. Its script checks that ESP_PATH (/boot, from /etc/default/limine) is a mounted vfat filesystem before building anything, fails with the error above, and writes nothing, so the UKI at /boot/EFI/Linux/omarchy_linux.efi on the unmounted ESP is simply left at the old version.

Either way the modules under /usr/lib/modules/<newver> are updated and the old ones removed, so the old kernel that the firmware does load has no matching modules.

Fix

First confirm the diagnosis:


findmnt /boot            # no output at all = the ESP is NOT mounted
lsblk -f                 # find the vfat/FAT32 partition
uname -r; pacman -Q linux
sudo ls -l /boot         # sudo: Omarchy mounts the ESP dmask=0077

If the system still boots (on the old kernel), repair it in place.

**Plain Arch only:** move the stray files off the root filesystem first, otherwise they stay there forever wasting space and confusing you later. On Omarchy 4 the hook wrote nothing, so the directory is empty and this step can be skipped (it is harmless if you do it anyway):


sudo mv /boot /boot.stray
sudo mkdir /boot

Add or fix the fstab entry, then mount:


sudo blkid -s UUID -o value /dev/nvme0n1p1     # your ESP

Omarchy's installer writes this shape (masks 0077, so only root can read the ESP):


# /etc/fstab
UUID=1234-ABCD  /boot  vfat  rw,relatime,fmask=0077,dmask=0077,codepage=437,iocharset=ascii,shortname=mixed,utf8,errors=remount-ro  0 2

The Arch wiki's fmask=0137,dmask=0027 variant is equally valid on plain Arch.


sudo systemctl daemon-reload
sudo mount /boot
findmnt /boot            # must now show the vfat device

Reinstall every kernel you have and regenerate the boot image. On Omarchy 4 the reinstall itself fires the Limine hook, which rebuilds the UKI, so the explicit command is belt and braces. pacman -S without -u is not blocked by Omarchy's update guard:


sudo pacman -S linux linux-firmware          # add linux-lts etc. if installed
sudo limine-mkinitcpio                        # Omarchy 4 / Limine UKI
# plain Arch instead:
sudo mkinitcpio -P

Do not use mkinitcpio -P on Omarchy 4: it has no presets in /etc/mkinitcpio.d/ and dies with No presets found, and the /usr/local/bin/mkinitcpio wrapper that Omarchy's hook package installs only warns and offers limine-mkinitcpio anyway.

Then reinstall the boot loader entries:


sudo limine-update            # Limine (also re-runs limine-mkinitcpio)
sudo bootctl update           # systemd-boot
sudo grub-mkconfig -o /boot/grub/grub.cfg   # GRUB

If it no longer boots at all, do the same from the Arch/Omarchy ISO, booted in UEFI mode (the Limine tools refuse to build a UKI unless /sys/firmware/efi exists inside the chroot):


lsblk -f
# LUKS first if encrypted:
cryptsetup open /dev/nvme0n1p2 root
mount -o subvol=@ /dev/mapper/root /mnt      # drop -o subvol=@ if not btrfs
mount --mkdir /dev/nvme0n1p1 /mnt/boot
arch-chroot /mnt
pacman -S linux linux-firmware
limine-mkinitcpio        # Omarchy 4 / Limine
mkinitcpio -P            # plain Arch instead
exit
umount -R /mnt

To stop it recurring when the ESP is not at /boot, preload the FAT modules so the mount never fails early:


# /etc/modules-load.d/vfat.conf
vfat
nls_cp437
nls_ascii

Once you are booting off the ESP again, delete /boot.stray if you created it.

Verify

findmnt /boot shows the vfat partition, ls /boot lists the vmlinuz/initramfs (or /boot/EFI/Linux/omarchy_linux.efi on Omarchy) with today's date, and after a reboot uname -r matches pacman -Q linux.

Sources

boot-esp-not-mounted-during-kernel-upgrade