initramfs selection broken when kernel=kernel8.img

[gitbls]

Describe the bug

Something is broken in initramfs loading with kernel=kernel8.img on a Pi5, such that the wrong initramfs is loaded.

Steps to reproduce the behaviour

• Boot a Pi5 with the normal kernel. Add break=premount to /boot/firmware/cmdline.txt
• Reboot and the system will stop in initramfs
• Observe that the output from lsmod displays a few modules that have already been loaded
• exit initramfs and the system will continue booting
• Edit /boot/firmware/config.txt
  • Add kernel=kernel8.img
• Shutdown and reboot into initramfs
• Observe that the output from lsmod displays only the listing header and NO modules
• exit initramfs and the system will continue booting
• Edit /boot/firmware/config.txt
  • Comment out auto_initramfs=1
  • Add initramfs initramfs8 followkernel
• Shutdown and reboot init initramfs
• Observe that the output from lsmod works again, showing loaded modules

While this is not a problem in the normal use case, trying to use cryptsetup in initramfs fails due to the incorrect initramfs being loaded.

See https://gitlab.com/cryptsetup/cryptsetup/-/issues/940 for additional details and comments from the cryptsetup developer, Milan Broz.

Device (s)

Raspberry Pi 5

System

https://pastebin.com/UEh3iz2h

Logs

No response

Additional context

No response

[pelwell]

Thanks for reporting this - it's clearly a bug, and an easy one to fix. https://drive.google.com/file/d/1OxmVXmUHBH5woHupzSoWrubbax0SXfdg/view?usp=sharing contains an EEPROM image with a patch that should fix the problem. Once you've extracted the .bin file from the zip, install it with:

$ sudo rpi-eeprom-update -f pieeprom_is1965.bin

[gitbls]

Tested and confirmed fix.

Will it be in the next eeprom update?

Thx!

[pelwell]

The change hasn't been approved and merged yet, but I think it's unlikely we'll make a release before it gets merged, so yes.

[gitbls]

Perfect. Thanks again!

[mbroz]

Thanks! Maybe you can also update auto_initramfs documentation.

While debugging this issue, the description was confusing. It should probably say that it selects initramfs based on the autodetected kernel name. But how does it behave when an explicit kernel is selected (but no initramfs)?

[pelwell]

There's another auto_initramfs implementation that works for custom kernel names. The only requirements are that the kernel image name starts with kernel, the initramfs with initramfs, and that the rest matches (except that initramfs doesn't have .img on the end). Put another way, the names are of the form kernel<variant>.img and initramfs<variant>, where <variant> can be any reasonable string (don't go looking for exceptions).

Update: The .img suffix is optional - anything after the last . in the kernel name is removed (so don't go using . in the middle of a variant name without including a suffix...)

This scheme obviously works with the names currently used with RPiOS, but also allows for other variations.

The new image is available here: https://drive.google.com/file/d/1RL4HNb-5AyyIOc_JyNqqAumqc2SZxJwE/view?usp=drive_link