28allday
19fd794b6b
Single-script builder (nosignal.sh) that turns a stock Arch Linux ISO into a fully-offline installer for a themed Hyprland + caelestia (Quickshell) desktop: matching SDDM greeter, Btrfs/Limine bootable snapshots, chwd-style GPU detection, and a curated "os updates" layer (keybind cheatsheet, settings panels, system polish, on-box management skill). See README.md. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
68 lines
3 KiB
Markdown
Executable file
68 lines
3 KiB
Markdown
Executable file
# limine-uki-dead-entry
|
||
|
||
The repair script edits the bootloader, so run it with care. Workaround: pick the UKI entry.
|
||
|
||
## Symptom
|
||
After a `nosignal-update` that bumped the kernel (7.0.11→7.0.12), selecting
|
||
**"NoSignal OS"** in Limine drops to a **TTY**. Selecting **"Linux"** (the UKI
|
||
entry) boots normally.
|
||
|
||
## Root cause
|
||
- `/etc/default/limine` sets **`ENABLE_UKI=yes`**.
|
||
- Per `limine-entry-tool.conf:151`: *"Duplicate 'initramfs' and 'vmlinuz' files
|
||
are removed when 'limine-mkinitcpio' or 'limine-update' is run to generate a
|
||
UKI."* So the first kernel update **deletes** `/boot/vmlinuz-linux` and
|
||
`/boot/initramfs-linux.img`.
|
||
- But `limine.conf` carries an installer-seeded **first** entry:
|
||
```
|
||
/NoSignal (Arch Linux)
|
||
protocol: linux
|
||
path: boot():/vmlinuz-linux
|
||
module_path: boot():/initramfs-linux.img
|
||
…
|
||
```
|
||
pointing at the now-deleted files. With **no `default_entry`**, this dead
|
||
entry is the **auto-boot default** → TTY.
|
||
- The auto-generated UKI entry (`protocol: efi` → `/EFI/Linux/nosignal_linux.efi`)
|
||
is correct and bootable — that's the "Linux" entry that worked.
|
||
|
||
### Evidence
|
||
- `vmlinuz-linux` / `initramfs-linux.img`: **absent**.
|
||
- Dead entry present in both `limine.conf` and the pre-update `limine.conf.old`
|
||
(shipped from the ISO).
|
||
- UKI is **current**: its sha256 equals the freshly-built image — not stale
|
||
(the May-28 mtime is merely preserved). So the UKI is not the problem.
|
||
- GPU-agnostic; prior AMD rounds likely never ran a post-install kernel update.
|
||
|
||
### Secondary risk (NVIDIA-specific) — flag for the builder
|
||
The UKI is **139 MB** (NVIDIA driver + GSP firmware; `linux-firmware-nvidia` is
|
||
214 MB) on a **511 MB** ESP. limine.conf already shows the live UKI + a history
|
||
copy. As more kernel versions / snapshot UKIs accumulate, the ESP can fill, and
|
||
then UKI writes *would* genuinely fail (a real stale-UKI scenario). Recommend:
|
||
size the ESP ≥1–2 GB on NVIDIA, and/or trim the initramfs (don't embed the full
|
||
GSP firmware), and/or keep UKI snapshot history off the ESP.
|
||
|
||
## Fix
|
||
**Durable (builder, do this in the ISO):** don't seed a `protocol: linux` entry
|
||
when `ENABLE_UKI=yes`. Either
|
||
1. seed the OS entry as `protocol: efi` → `/EFI/Linux/<uki>_linux.efi`, or
|
||
2. drop the manual entry and let limine-entry-tool's auto UKI entry be the OS
|
||
entry (name it via `TARGET_OS_NAME`), and
|
||
3. set an explicit **`default_entry`** to the UKI so auto-boot never lands on a
|
||
bad entry.
|
||
|
||
**This component (repair an installed system):** `fix-limine-uki-entry.sh`
|
||
(root, idempotent, backs up limine.conf) rewrites the seeded entry's body from
|
||
`protocol: linux` + `vmlinuz-linux` to `protocol: efi` + the UKI, keeping the
|
||
label (stays first = default) and the cmdline. No-op if already fixed; refuses
|
||
if the UKI file is missing.
|
||
|
||
## Verify
|
||
```
|
||
grep -A3 'NoSignal (Arch Linux)' /boot/limine.conf # protocol: efi + UKI path
|
||
```
|
||
Then reboot and confirm the top/default entry boots to the desktop.
|
||
|
||
## Files
|
||
- `fix-limine-uki-entry.sh` — idempotent root repair.
|
||
- `migrations/1781434800-limine-uki-dead-entry.sh` — delegates to it.
|