Chain loading kernel without a boot loader

Question

I am working on an embedded project that will do in-place upgrades. There may come a time when an upgrade will need to reinstall everything. Since this is a pretty risky thing, I would like to have some kind of failsafe fall-back mode to reattempt the upgrade.

I am thinking of adding a very basic kernel that has just enough to attempt another upgrade. I want to always boot into this basic kernel, check GPIOs, and then boot into the other kernel or continue booting depending on the state of the GPIOs. I'll install a button on the case to signal the appropriate GPIO.

I don't know how the boot loader works, and grub/syslinux isn't available AFAIK.

Is this sort of thing possible? Am I over-complicating the problem?

Answer 1

A Linux bootloader must, at a minimum, load the kernel and pass it a few parameters to tell the kernel where it's loaded. Many setups also require being able to pass a command line and an initial RAM drive, though on an embedded system with a known hardware setup this can be superfluous.

The best way to get a reliable bootloader is to use a working bootloader. Have a simple first-stage bootloader in a fixed configuration that chainloads a Linux bootloader (U-Boot or whatever your embedded system has), with a fallback mechanism. U-Boot itself can implement a fallback mechanism — see “Implementing Boot Image Fallback on U-Boot” on the U-Boot-Users mailing list, and Redundant Images on xpedite5370 for an example.