The Boot Chain

The whole handoff sequence in one diagram, with the exact artefact passed at each step and where each one lives on disk.

Debugging a boot that doesn't work means knowing exactly what got handed from one stage to the next, and where the artefact each stage is looking for actually lives on disk. Every stage in this chain does one job and then gets out of the way — it doesn't know what came before it beyond the one thing it was given, and it doesn't know what comes after beyond the one thing it produces. That narrowness is what makes the chain debuggable at all: a failure has a location, because every step either found its input or it didn't. This page is the map. The rest of this folder is the territory each stop on the map expands into.

The chain, end to end

The full boot chain, with the artefact handed over at each step.

Two topics sit beside this chain rather than inside it, because they aren't handoff steps: the kernel command line is text threaded through several of these arrows rather than a stage of its own, and Secure Boot is a verification gate that can sit between any two of the first three nodes, rejecting a handoff instead of performing one.

What is handed over at each step

Step	Artefact	Where it lives on disk	How to inspect it
Firmware picks a loader	a boot variable / boot order entry	firmware NVRAM — not on any filesystem	efibootmgr -v
Boot loader loads a kernel	\EFI\<vendor>\grubx64.efi (UEFI) or the MBR's first-stage code (legacy)	the EFI System Partition, or the disk's first sector	ls /boot/efi/EFI/*/
Kernel image	/boot/vmlinuz-*	/boot	ls -l /boot/vmlinuz-*
Initramfs	/boot/initrd.img-*	/boot	lsinitramfs /boot/initrd.img-$(uname -r) | head (Debian/Ubuntu)
Real root	whatever root= names — a device, a UUID, a LABEL	named on the kernel command line, not stored as a file	cat /proc/cmdline
Init	/sbin/init	the real root filesystem, reachable only after switch_root	ls -l /sbin/init

Where each step is covered

Step in the chain	Page
What firmware does before anything Linux exists	Firmware: BIOS and UEFI
The full chain, in one place	this page
Finding a kernel and initramfs, and handing off	Boot Loaders
What text actually gets passed to the kernel	The Kernel Command Line
Verifying every step before it's allowed to run	Secure Boot and Signed Kernels
Unpacking bzImage into a running kernel	Inside bzImage
CPU mode transitions before the first C function	Early Boot: Getting to C
start_kernel() and the initcall levels	start_kernel and the Initcall Order
Unpacking and using the initramfs	initramfs and Early User Space
Moving to the real root and becoming PID 1	switch_root and PID 1
How systemd turns units into a boot	systemd: The Model
Reading timestamps, diagnosing a broken boot	systemd in Practice, and Debugging a Broken Boot

Where it usually breaks

Symptom	Likely step	Covered in
Firmware finds nothing bootable	firmware has no usable boot-order entry (empty NVRAM, missing ESP)	Firmware: BIOS and UEFI
Loader menu appears, but no kernel is found	the loader's config points at a vmlinuz/initrd that isn't where it expects	Boot Loaders
Kernel panics with "unable to mount root fs"	the root= device named on the command line doesn't exist or isn't ready yet	switch_root and PID 1
Dropped to an (initramfs) shell prompt	early user space couldn't find or mount the real root itself	initramfs and Early User Space
Boots to emergency.target	systemd reached PID 1 fine, but a critical unit or mount failed after that	systemd in Practice, and Debugging a Broken Boot

Structure

/boot — where the kernel image, initramfs, and loader configuration live on most distributions

Pathfirmware → loader → kernel → initramfs → real root → PID 1

Observels -l /boot && cat /proc/cmdline

TrapEach stage only knows about the next one. A machine that reaches the loader menu has proved firmware and partitioning are fine, which eliminates half the chain before you start guessing.

References

• Boot Loader Specification — the standard /boot layout modern distributions are converging on, and the source for this page's artefact-location table.
• The kernel's admin guide to init — the kernel's own account of the "unable to mount root" class of failure, written for exactly this page's failure table.
• man 7 boot — the traditional boot sequence, stated compactly, and a useful cross-check against this page's diagram.