Kernel Core Internals

The glue that holds every subsystem together: boot, logging, parameters, panics, and module init

What is "kernel core"?

The kernel core is the collection of infrastructure that does not belong to any single subsystem but that every subsystem depends on. It lives primarily in init/, kernel/, and arch/x86/kernel/ (and the equivalent arch directories), and it answers the question: how does a compressed binary become a running operating system that can schedule tasks, respond to interrupts, and load drivers?

Topics covered here include:

• The C entry point (start_kernel()) and the ordered sequence of subsystem initialization calls it drives
• How the kernel receives and parses boot parameters from the bootloader
• How modules and built-in subsystems register their initialization functions through the initcall mechanism
• What happens when something goes catastrophically wrong — oops, panic, and crash dump collection
• The printk logging infrastructure that underpins every diagnostic message in the kernel

This is not memory management, not the scheduler, not networking — it is the machinery that sets up all of those before handing control off to them.

---

Pages in this section

Page	What it covers
printk and Kernel Logging	Log levels, ring buffer, rate limiting, dynamic debug, dmesg
Early Boot and start_kernel()	C entry point, subsystem init order, initcalls, __init sections
Kernel Boot Parameters	Command-line parsing, __setup, early_param, module params
Kernel Panic and Oops	Oops handling, panic(), stack unwinding, kdump, KASAN/KFENCE
Module Init and Initcalls	module_init/module_exit, struct module, loading flow, signing
Kernel Core War Stories	Real incidents: init ordering bugs, __init misuse, panic loops

---

Suggested reading order

If you are new to kernel internals, read these in order:

1. Early Boot and start_kernel() — understand the init sequence before anything else
2. Kernel Boot Parameters — how the kernel is configured before any code runs
3. Module Init and Initcalls — how subsystems and drivers hook into the boot sequence
4. printk and Kernel Logging — observability infrastructure used by everything else
5. Kernel Panic and Oops — what happens when something goes wrong
6. Kernel Core War Stories — real-world examples that tie the concepts together

---

Kernel boot flow

Power on / bootloader (GRUB/EFI)
        |
        | loads compressed kernel image (bzImage / Image)
        v
arch entry point (arch/x86/kernel/head_64.S)
        |
        | sets up early page tables, GDT, stack
        v
startup_64 / __startup_64
        |
        | decompresses kernel, relocates if KASLR
        v
x86_64_start_kernel()   (arch/x86/kernel/head64.c)
        |
        | early CR0/CR4/EFER setup, early IDT
        v
start_kernel()          (init/main.c)          <--- C entry point
        |
        |-- set_task_stack_end_magic(&init_task)
        |-- smp_setup_processor_id()
        |-- boot_cpu_init()
        |-- setup_arch()         (memory map, page tables, ACPI)
        |-- setup_per_cpu_areas()
        |-- sched_init()
        |-- rcu_init()
        |-- init_IRQ()
        |-- timekeeping_init()
        |-- kmem_cache_init()
        |-- console_init()       (printk now goes to screen)
        |
        v
rest_init()
        |
        |-- kernel_thread(kernel_init, ...)  --> PID 1
        |-- kernel_thread(kthreadd, ...)     --> PID 2
        |
        v
cpu_startup_entry(CPUHP_ONLINE)   <-- CPU 0 becomes idle thread
        |
        v (PID 1 runs)
kernel_init()
        |
        |-- do_initcalls()        (runs all initcall levels)
        |-- prepare_namespace()   (mount root filesystem)
        |
        v
exec /sbin/init  (or init= boot parameter)
        |
        v
userspace init system (systemd / sysvinit)

---

Quick reference: key kernel globals

Symbol	Header	Description
init_task	<linux/init_task.h>	The static task_struct for PID 0 (idle thread); the root of all process lineage
jiffies	<linux/jiffies.h>	Monotonically increasing tick counter; wraps after ~497 days on 32-bit
HZ	<asm/param.h>	Timer interrupt frequency in ticks per second; typically 100, 250, or 1000
nr_cpu_ids	<linux/threads.h>	Number of possible CPUs (set by setup_nr_cpu_ids() during early boot)
NUMA_NO_NODE	<linux/numa.h>	Sentinel value (-1) meaning "no specific NUMA node"; used in allocator APIs
boot_command_line[]	<linux/init.h>	Copy of the command line received from the bootloader
system_state	<linux/kernel.h>	Current boot phase: SYSTEM_BOOTING, SYSTEM_RUNNING, SYSTEM_HALT, etc.

---

Further reading

• Kernel Modules — writing and loading out-of-tree modules
• printk — logging in depth
• Kernel Debugging — KGDB, kdump, oops analysis
• init/main.c — the canonical source for start_kernel() and rest_init()
• Documentation/admin-guide/kernel-parameters.txt — exhaustive list of boot parameters