linux --- 啓動過程代碼(arm爲例)

1. 解壓縮後的代碼入口

/*
 * Kernel startup entry point.
 * ---------------------------
 *
 * This is normally called from the decompressor code.  The requirements
 * are: MMU = off, D-cache = off, I-cache = dont care, r0 = 0,
 * r1 = machine nr, r2 = atags or dtb pointer.
 *
 * This code is mostly position independent, so if you link the kernel at
 * 0xc0008000, you call this at __pa(0xc0008000).
 *
 * See linux/arch/arm/tools/mach-types for the complete list of machine
 * numbers for r1.
 *
 * We're trying to keep crap to a minimum; DO NOT add any machine specific
 * crap here - that's what the boot loader (or in extreme, well justified
 * circumstances, zImage) is for.
 */
    .arm

    __HEAD
ENTRY(stext)

 THUMB(    adr    r9, BSYM(1f)    )    @ Kernel is always entered in ARM.
 THUMB(    bx    r9        )    @ If this is a Thumb-2 kernel,
 THUMB(    .thumb            )    @ switch to Thumb now.
 THUMB(1:            )

    setmode    PSR_F_BIT | PSR_I_BIT | SVC_MODE, r9 @ ensure svc mode
                        @ and irqs disabled
    mrc    p15, 0, r9, c0, c0        @ get processor id
    bl    __lookup_processor_type        @ r5=procinfo r9=cpuid
    movs    r10, r5                @ invalid processor (r5=0)?
 THUMB( it    eq )        @ force fixup-able long branch encoding
    beq    __error_p            @ yes, error 'p'

#ifdef CONFIG_ARM_LPAE
    mrc    p15, 0, r3, c0, c1, 4        @ read ID_MMFR0
    and    r3, r3, #0xf            @ extract VMSA support
    cmp    r3, #5                @ long-descriptor translation table format?
 THUMB( it    lo )                @ force fixup-able long branch encoding
    blo    __error_p            @ only classic page table format
#endif

#ifndef CONFIG_XIP_KERNEL
    adr    r3, 2f
    ldmia    r3, {r4, r8}
    sub    r4, r3, r4            @ (PHYS_OFFSET - PAGE_OFFSET)
    add    r8, r8, r4            @ PHYS_OFFSET
#else
    ldr    r8, =PHYS_OFFSET        @ always constant in this case
#endif

    /*
     * r1 = machine no, r2 = atags or dtb,
     * r8 = phys_offset, r9 = cpuid, r10 = procinfo
     */
    bl    __vet_atags
#ifdef CONFIG_SMP_ON_UP
    bl    __fixup_smp
#endif
#ifdef CONFIG_ARM_PATCH_PHYS_VIRT
    bl    __fixup_pv_table
#endif
    bl    __create_page_tables

    /*
     * The following calls CPU specific code in a position independent
     * manner.  See arch/arm/mm/proc-*.S for details.  r10 = base of
     * xxx_proc_info structure selected by __lookup_processor_type
     * above.  On return, the CPU will be ready for the MMU to be
     * turned on, and r0 will hold the CPU control register value.
     */
    ldr    r13, =__mmap_switched        @ address to jump to after
                        @ mmu has been enabled
    adr    lr, BSYM(1f)            @ return (PIC) address
    mov    r8, r4                @ set TTBR1 to swapper_pg_dir
 ARM(    add    pc, r10, #PROCINFO_INITFUNC    )
 THUMB(    add    r12, r10, #PROCINFO_INITFUNC    )
 THUMB(    mov    pc, r12                )
1:    b    __enable_mmu
ENDPROC(stext)
    .ltorg
#ifndef CONFIG_XIP_KERNEL
2:    .long    .
    .long    PAGE_OFFSET
#endif

2. 設置頁表，啓動內核

/*
 * Setup the initial page tables.  We only setup the barest
 * amount which are required to get the kernel running, which
 * generally means mapping in the kernel code.
 *
 * r9  = cpuid
 * r10 = procinfo
 *
 * Returns:
 *  r0, r3, r6, r7 corrupted
 *  r4 = physical page table address
 */
    .ltorg

    .align    2
    .type    __switch_data, %object
__switch_data:
    .long    __mmap_switched
    .long    __bss_start            @ r6
    .long    _end                @ r7
    .long    cr_alignment            @ r8
    .long    init_thread_union + THREAD_START_SP @ sp

/*
 * The following fragment of code is executed with the MMU on in MMU mode,
 * and uses absolute addresses; this is not position independent.
 *
 *  r0  = cp#0 control register
 */
__mmap_switched:
    adr    r3, __switch_data + 4

    ldm.w    (r6, r7, r8), [r3]+
    ldw    sp, [r3]

    mov    fp, #0                @ Clear BSS (and zero fp)
203:    csub.a    r6, r7
    bea    204f
    stw.w    fp, [r6]+,#4
    b    203b
204:
    andn    r1, r0, #CR_A            @ Clear 'A' bit
    stm    (r0, r1), [r8]+            @ Save control register values
    b    start_kernel

3. 開始內核

asmlinkage void __init start_kernel(void)
{
    char * command_line;
    extern const struct kernel_param __start___param[], __stop___param[];

    /*
     * Need to run as early as possible, to initialize the
     * lockdep hash:
     */
    lockdep_init();
    smp_setup_processor_id();
    debug_objects_early_init();

    /*
     * Set up the the initial canary ASAP:
     */
    boot_init_stack_canary();

    cgroup_init_early();

    local_irq_disable();
    early_boot_irqs_disabled = true;

/*
 * Interrupts are still disabled. Do necessary setups, then
 * enable them
 */
    tick_init();
    boot_cpu_init();
    page_address_init();
    printk(KERN_NOTICE "%s", linux_banner);
    setup_arch(&command_line);
    mm_init_owner(&init_mm, &init_task);
    mm_init_cpumask(&init_mm);
    setup_command_line(command_line);
    setup_nr_cpu_ids();
    setup_per_cpu_areas();
    smp_prepare_boot_cpu();    /* arch-specific boot-cpu hooks */

    build_all_zonelists(NULL);
    page_alloc_init();

    printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
    parse_early_param();
    parse_args("Booting kernel", static_command_line, __start___param,
           __stop___param - __start___param,
           &unknown_bootoption);

    jump_label_init();

    /*
     * These use large bootmem allocations and must precede
     * kmem_cache_init()
     */
    setup_log_buf(0);
    pidhash_init();
    vfs_caches_init_early();
    sort_main_extable();
    trap_init();
    mm_init();

    /*
     * Set up the scheduler prior starting any interrupts (such as the
     * timer interrupt). Full topology setup happens at smp_init()
     * time - but meanwhile we still have a functioning scheduler.
     */
    sched_init();
    /*
     * Disable preemption - early bootup scheduling is extremely
     * fragile until we cpu_idle() for the first time.
     */
    preempt_disable();
    if (!irqs_disabled()) {
        printk(KERN_WARNING "start_kernel(): bug: interrupts were "
                "enabled *very* early, fixing it\n");
        local_irq_disable();
    }
    idr_init_cache();
    perf_event_init();
    rcu_init();
    radix_tree_init();
    /* init some links before init_ISA_irqs() */
    early_irq_init();
    init_IRQ();
    prio_tree_init();
    init_timers();
    hrtimers_init();
    softirq_init();
    timekeeping_init();
    time_init();
    profile_init();
    call_function_init();
    if (!irqs_disabled())
        printk(KERN_CRIT "start_kernel(): bug: interrupts were "
                 "enabled early\n");
    early_boot_irqs_disabled = false;
    local_irq_enable();

    /* Interrupts are enabled now so all GFP allocations are safe. */
    gfp_allowed_mask = __GFP_BITS_MASK;

    kmem_cache_init_late();

    /*
     * HACK ALERT! This is early. We're enabling the console before
     * we've done PCI setups etc, and console_init() must be aware of
     * this. But we do want output early, in case something goes wrong.
     */
    console_init();
    if (panic_later)
        panic(panic_later, panic_param);

    lockdep_info();

    /*
     * Need to run this when irqs are enabled, because it wants
     * to self-test [hard/soft]-irqs on/off lock inversion bugs
     * too:
     */
    locking_selftest();

#ifdef CONFIG_BLK_DEV_INITRD
    if (initrd_start && !initrd_below_start_ok &&
        page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
        printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
            "disabling it.\n",
            page_to_pfn(virt_to_page((void *)initrd_start)),
            min_low_pfn);
        initrd_start = 0;
    }
#endif
    page_cgroup_init();
    debug_objects_mem_init();
    kmemleak_init();
    setup_per_cpu_pageset();
    numa_policy_init();
    if (late_time_init)
        late_time_init();
    sched_clock_init();
    calibrate_delay();
    pidmap_init();
    anon_vma_init();
#ifdef CONFIG_X86
    if (efi_enabled)
        efi_enter_virtual_mode();
#endif
    thread_info_cache_init();
    cred_init();
    fork_init(totalram_pages);
    proc_caches_init();
    buffer_init();
    key_init();
    security_init();
    dbg_late_init();
    vfs_caches_init(totalram_pages);
    signals_init();
    /* rootfs populating might need page-writeback */
    page_writeback_init();
#ifdef CONFIG_PROC_FS
    proc_root_init();
#endif
    cgroup_init();
    cpuset_init();
    taskstats_init_early();
    delayacct_init();

    check_bugs();

    acpi_early_init(); /* before LAPIC and SMP init */
    sfi_init_late();

    ftrace_init();

    /* Do the rest non-__init'ed, we're now alive */
    rest_init();
}

4. 啓動init進程

static noinline void __init_refok rest_init(void)
{
    int pid;

    rcu_scheduler_starting();
    /*
     * We need to spawn init first so that it obtains pid 1, however
     * the init task will end up wanting to create kthreads, which, if
     * we schedule it before we create kthreadd, will OOPS.
     */
    kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
    numa_default_policy();
    pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
    rcu_read_lock();
    kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
    rcu_read_unlock();
    complete(&kthreadd_done);

    /*
     * The boot idle thread must execute schedule()
     * at least once to get things moving:
     */
    init_idle_bootup_task(current);
    preempt_enable_no_resched();
    schedule();

    /* Call into cpu_idle with preempt disabled */
    preempt_disable();
    cpu_idle();
}