4 * Copyright (C) 1991, 1992 Linus Torvalds
6 * GK 2/5/95 - Changed to support mounting root fs via NFS
7 * Added initrd & change_root: Werner Almesberger & Hans Lermen, Feb '96
8 * Moan early if gcc is old, avoiding bogus kernels - Paul Gortmaker, May '96
12 #define DEBUG /* Enable initcall_debug */
14 #include <linux/types.h>
15 #include <linux/module.h>
16 #include <linux/proc_fs.h>
17 #include <linux/kernel.h>
18 #include <linux/syscalls.h>
19 #include <linux/stackprotector.h>
20 #include <linux/string.h>
21 #include <linux/ctype.h>
22 #include <linux/delay.h>
23 #include <linux/ioport.h>
24 #include <linux/init.h>
25 #include <linux/initrd.h>
26 #include <linux/bootmem.h>
27 #include <linux/acpi.h>
28 #include <linux/tty.h>
29 #include <linux/percpu.h>
30 #include <linux/kmod.h>
31 #include <linux/vmalloc.h>
32 #include <linux/kernel_stat.h>
33 #include <linux/start_kernel.h>
34 #include <linux/security.h>
35 #include <linux/smp.h>
36 #include <linux/profile.h>
37 #include <linux/rcupdate.h>
38 #include <linux/moduleparam.h>
39 #include <linux/kallsyms.h>
40 #include <linux/writeback.h>
41 #include <linux/cpu.h>
42 #include <linux/cpuset.h>
43 #include <linux/cgroup.h>
44 #include <linux/efi.h>
45 #include <linux/tick.h>
46 #include <linux/interrupt.h>
47 #include <linux/taskstats_kern.h>
48 #include <linux/delayacct.h>
49 #include <linux/unistd.h>
50 #include <linux/rmap.h>
51 #include <linux/mempolicy.h>
52 #include <linux/key.h>
53 #include <linux/buffer_head.h>
54 #include <linux/page_cgroup.h>
55 #include <linux/debug_locks.h>
56 #include <linux/debugobjects.h>
57 #include <linux/lockdep.h>
58 #include <linux/kmemleak.h>
59 #include <linux/pid_namespace.h>
60 #include <linux/device.h>
61 #include <linux/kthread.h>
62 #include <linux/sched.h>
63 #include <linux/signal.h>
64 #include <linux/idr.h>
65 #include <linux/kgdb.h>
66 #include <linux/ftrace.h>
67 #include <linux/async.h>
68 #include <linux/kmemcheck.h>
69 #include <linux/sfi.h>
70 #include <linux/shmem_fs.h>
71 #include <linux/slab.h>
72 #include <linux/perf_event.h>
73 #include <linux/file.h>
74 #include <linux/ptrace.h>
75 #include <linux/blkdev.h>
76 #include <linux/elevator.h>
77 #include <linux/sched_clock.h>
78 #include <linux/context_tracking.h>
79 #include <linux/random.h>
83 #include <asm/setup.h>
84 #include <asm/sections.h>
85 #include <asm/cacheflush.h>
87 #ifdef CONFIG_X86_LOCAL_APIC
91 static int kernel_init(void *);
93 extern void init_IRQ(void);
94 extern void fork_init(unsigned long);
95 extern void mca_init(void);
96 extern void sbus_init(void);
97 extern void radix_tree_init(void);
98 #ifndef CONFIG_DEBUG_RODATA
99 static inline void mark_rodata_ro(void) { }
103 extern void tc_init(void);
107 * Debug helper: via this flag we know that we are in 'early bootup code'
108 * where only the boot processor is running with IRQ disabled. This means
109 * two things - IRQ must not be enabled before the flag is cleared and some
110 * operations which are not allowed with IRQ disabled are allowed while the
113 bool early_boot_irqs_disabled __read_mostly;
115 enum system_states system_state __read_mostly;
116 EXPORT_SYMBOL(system_state);
119 * Boot command-line arguments
121 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
122 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
124 extern void time_init(void);
125 /* Default late time init is NULL. archs can override this later. */
126 void (*__initdata late_time_init)(void);
128 /* Untouched command line saved by arch-specific code. */
129 char __initdata boot_command_line[COMMAND_LINE_SIZE];
130 /* Untouched saved command line (eg. for /proc) */
131 char *saved_command_line;
132 /* Command line for parameter parsing */
133 static char *static_command_line;
135 static char *execute_command;
136 static char *ramdisk_execute_command;
139 * Used to generate warnings if static_key manipulation functions are used
140 * before jump_label_init is called.
142 bool static_key_initialized __read_mostly = false;
143 EXPORT_SYMBOL_GPL(static_key_initialized);
146 * If set, this is an indication to the drivers that reset the underlying
147 * device before going ahead with the initialization otherwise driver might
148 * rely on the BIOS and skip the reset operation.
150 * This is useful if kernel is booting in an unreliable environment.
151 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
152 * skipped and devices will be in unknown state.
154 unsigned int reset_devices;
155 EXPORT_SYMBOL(reset_devices);
157 static int __init set_reset_devices(char *str)
163 __setup("reset_devices", set_reset_devices);
165 static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
166 const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
167 static const char *panic_later, *panic_param;
169 extern const struct obs_kernel_param __setup_start[], __setup_end[];
171 static int __init obsolete_checksetup(char *line)
173 const struct obs_kernel_param *p;
174 int had_early_param = 0;
178 int n = strlen(p->str);
179 if (parameqn(line, p->str, n)) {
181 /* Already done in parse_early_param?
182 * (Needs exact match on param part).
183 * Keep iterating, as we can have early
184 * params and __setups of same names 8( */
185 if (line[n] == '\0' || line[n] == '=')
187 } else if (!p->setup_func) {
188 pr_warn("Parameter %s is obsolete, ignored\n",
191 } else if (p->setup_func(line + n))
195 } while (p < __setup_end);
197 return had_early_param;
201 * This should be approx 2 Bo*oMips to start (note initial shift), and will
202 * still work even if initially too large, it will just take slightly longer
204 unsigned long loops_per_jiffy = (1<<12);
206 EXPORT_SYMBOL(loops_per_jiffy);
208 static int __init debug_kernel(char *str)
210 console_loglevel = 10;
214 static int __init quiet_kernel(char *str)
216 console_loglevel = 4;
220 early_param("debug", debug_kernel);
221 early_param("quiet", quiet_kernel);
223 static int __init loglevel(char *str)
228 * Only update loglevel value when a correct setting was passed,
229 * to prevent blind crashes (when loglevel being set to 0) that
230 * are quite hard to debug
232 if (get_option(&str, &newlevel)) {
233 console_loglevel = newlevel;
240 early_param("loglevel", loglevel);
242 /* Change NUL term back to "=", to make "param" the whole string. */
243 static int __init repair_env_string(char *param, char *val, const char *unused)
246 /* param=val or param="val"? */
247 if (val == param+strlen(param)+1)
249 else if (val == param+strlen(param)+2) {
251 memmove(val-1, val, strlen(val)+1);
260 * Unknown boot options get handed to init, unless they look like
261 * unused parameters (modprobe will find them in /proc/cmdline).
263 static int __init unknown_bootoption(char *param, char *val, const char *unused)
265 repair_env_string(param, val, unused);
267 /* Handle obsolete-style parameters */
268 if (obsolete_checksetup(param))
271 /* Unused module parameter. */
272 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
279 /* Environment option */
281 for (i = 0; envp_init[i]; i++) {
282 if (i == MAX_INIT_ENVS) {
283 panic_later = "Too many boot env vars at `%s'";
286 if (!strncmp(param, envp_init[i], val - param))
289 envp_init[i] = param;
291 /* Command line option */
293 for (i = 0; argv_init[i]; i++) {
294 if (i == MAX_INIT_ARGS) {
295 panic_later = "Too many boot init vars at `%s'";
299 argv_init[i] = param;
304 static int __init init_setup(char *str)
308 execute_command = str;
310 * In case LILO is going to boot us with default command line,
311 * it prepends "auto" before the whole cmdline which makes
312 * the shell think it should execute a script with such name.
313 * So we ignore all arguments entered _before_ init=... [MJ]
315 for (i = 1; i < MAX_INIT_ARGS; i++)
319 __setup("init=", init_setup);
321 static int __init rdinit_setup(char *str)
325 ramdisk_execute_command = str;
326 /* See "auto" comment in init_setup */
327 for (i = 1; i < MAX_INIT_ARGS; i++)
331 __setup("rdinit=", rdinit_setup);
334 static const unsigned int setup_max_cpus = NR_CPUS;
335 #ifdef CONFIG_X86_LOCAL_APIC
336 static void __init smp_init(void)
338 APIC_init_uniprocessor();
341 #define smp_init() do { } while (0)
344 static inline void setup_nr_cpu_ids(void) { }
345 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
349 * We need to store the untouched command line for future reference.
350 * We also need to store the touched command line since the parameter
351 * parsing is performed in place, and we should allow a component to
352 * store reference of name/value for future reference.
354 static void __init setup_command_line(char *command_line)
356 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
357 static_command_line = alloc_bootmem(strlen (command_line)+1);
358 strcpy (saved_command_line, boot_command_line);
359 strcpy (static_command_line, command_line);
363 * We need to finalize in a non-__init function or else race conditions
364 * between the root thread and the init thread may cause start_kernel to
365 * be reaped by free_initmem before the root thread has proceeded to
368 * gcc-3.4 accidentally inlines this function, so use noinline.
371 static __initdata DECLARE_COMPLETION(kthreadd_done);
373 static noinline void __init_refok rest_init(void)
377 rcu_scheduler_starting();
379 * We need to spawn init first so that it obtains pid 1, however
380 * the init task will end up wanting to create kthreads, which, if
381 * we schedule it before we create kthreadd, will OOPS.
383 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
384 numa_default_policy();
385 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
387 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
389 complete(&kthreadd_done);
392 * The boot idle thread must execute schedule()
393 * at least once to get things moving:
395 init_idle_bootup_task(current);
396 schedule_preempt_disabled();
397 /* Call into cpu_idle with preempt disabled */
398 cpu_startup_entry(CPUHP_ONLINE);
401 /* Check for early params. */
402 static int __init do_early_param(char *param, char *val, const char *unused)
404 const struct obs_kernel_param *p;
406 for (p = __setup_start; p < __setup_end; p++) {
407 if ((p->early && parameq(param, p->str)) ||
408 (strcmp(param, "console") == 0 &&
409 strcmp(p->str, "earlycon") == 0)
411 if (p->setup_func(val) != 0)
412 pr_warn("Malformed early option '%s'\n", param);
415 /* We accept everything at this stage. */
419 void __init parse_early_options(char *cmdline)
421 parse_args("early options", cmdline, NULL, 0, 0, 0, do_early_param);
424 /* Arch code calls this early on, or if not, just before other parsing. */
425 void __init parse_early_param(void)
427 static __initdata int done = 0;
428 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
433 /* All fall through to do_early_param. */
434 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
435 parse_early_options(tmp_cmdline);
440 * Activate the first processor.
443 static void __init boot_cpu_init(void)
445 int cpu = smp_processor_id();
446 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
447 set_cpu_online(cpu, true);
448 set_cpu_active(cpu, true);
449 set_cpu_present(cpu, true);
450 set_cpu_possible(cpu, true);
453 void __init __weak smp_setup_processor_id(void)
457 # if THREAD_SIZE >= PAGE_SIZE
458 void __init __weak thread_info_cache_init(void)
464 * Set up kernel memory allocators
466 static void __init mm_init(void)
469 * page_cgroup requires contiguous pages,
470 * bigger than MAX_ORDER unless SPARSEMEM.
472 page_cgroup_init_flatmem();
476 pgtable_cache_init();
480 asmlinkage void __init start_kernel(void)
483 extern const struct kernel_param __start___param[], __stop___param[];
486 * Need to run as early as possible, to initialize the
490 smp_setup_processor_id();
491 debug_objects_early_init();
494 * Set up the the initial canary ASAP:
496 boot_init_stack_canary();
501 early_boot_irqs_disabled = true;
504 * Interrupts are still disabled. Do necessary setups, then
509 pr_notice("%s", linux_banner);
510 setup_arch(&command_line);
511 mm_init_owner(&init_mm, &init_task);
512 mm_init_cpumask(&init_mm);
513 setup_command_line(command_line);
515 setup_per_cpu_areas();
516 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
518 build_all_zonelists(NULL, NULL);
521 pr_notice("Kernel command line: %s\n", boot_command_line);
523 parse_args("Booting kernel", static_command_line, __start___param,
524 __stop___param - __start___param,
525 -1, -1, &unknown_bootoption);
530 * These use large bootmem allocations and must precede
535 vfs_caches_init_early();
541 * Set up the scheduler prior starting any interrupts (such as the
542 * timer interrupt). Full topology setup happens at smp_init()
543 * time - but meanwhile we still have a functioning scheduler.
547 * Disable preemption - early bootup scheduling is extremely
548 * fragile until we cpu_idle() for the first time.
551 if (WARN(!irqs_disabled(), "Interrupts were enabled *very* early, fixing it\n"))
556 context_tracking_init();
558 /* init some links before init_ISA_irqs() */
567 sched_clock_postinit();
570 call_function_init();
571 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
572 early_boot_irqs_disabled = false;
575 kmem_cache_init_late();
578 * HACK ALERT! This is early. We're enabling the console before
579 * we've done PCI setups etc, and console_init() must be aware of
580 * this. But we do want output early, in case something goes wrong.
584 panic(panic_later, panic_param);
589 * Need to run this when irqs are enabled, because it wants
590 * to self-test [hard/soft]-irqs on/off lock inversion bugs
595 #ifdef CONFIG_BLK_DEV_INITRD
596 if (initrd_start && !initrd_below_start_ok &&
597 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
598 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
599 page_to_pfn(virt_to_page((void *)initrd_start)),
605 debug_objects_mem_init();
607 setup_per_cpu_pageset();
616 if (efi_enabled(EFI_RUNTIME_SERVICES))
617 efi_enter_virtual_mode();
619 thread_info_cache_init();
621 fork_init(totalram_pages);
627 vfs_caches_init(totalram_pages);
629 /* rootfs populating might need page-writeback */
630 page_writeback_init();
631 #ifdef CONFIG_PROC_FS
636 taskstats_init_early();
641 acpi_early_init(); /* before LAPIC and SMP init */
644 if (efi_enabled(EFI_RUNTIME_SERVICES)) {
646 efi_free_boot_services();
651 /* Do the rest non-__init'ed, we're now alive */
655 /* Call all constructor functions linked into the kernel. */
656 static void __init do_ctors(void)
658 #ifdef CONFIG_CONSTRUCTORS
659 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
661 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
667 core_param(initcall_debug, initcall_debug, bool, 0644);
669 static int __init_or_module do_one_initcall_debug(initcall_t fn)
671 ktime_t calltime, delta, rettime;
672 unsigned long long duration;
675 pr_debug("calling %pF @ %i\n", fn, task_pid_nr(current));
676 calltime = ktime_get();
678 rettime = ktime_get();
679 delta = ktime_sub(rettime, calltime);
680 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
681 pr_debug("initcall %pF returned %d after %lld usecs\n",
687 int __init_or_module do_one_initcall(initcall_t fn)
689 int count = preempt_count();
694 ret = do_one_initcall_debug(fn);
700 if (preempt_count() != count) {
701 sprintf(msgbuf, "preemption imbalance ");
702 preempt_count_set(count);
704 if (irqs_disabled()) {
705 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
708 WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
714 extern initcall_t __initcall_start[];
715 extern initcall_t __initcall0_start[];
716 extern initcall_t __initcall1_start[];
717 extern initcall_t __initcall2_start[];
718 extern initcall_t __initcall3_start[];
719 extern initcall_t __initcall4_start[];
720 extern initcall_t __initcall5_start[];
721 extern initcall_t __initcall6_start[];
722 extern initcall_t __initcall7_start[];
723 extern initcall_t __initcall_end[];
725 static initcall_t *initcall_levels[] __initdata = {
737 /* Keep these in sync with initcalls in include/linux/init.h */
738 static char *initcall_level_names[] __initdata = {
749 static void __init do_initcall_level(int level)
751 extern const struct kernel_param __start___param[], __stop___param[];
754 strcpy(static_command_line, saved_command_line);
755 parse_args(initcall_level_names[level],
756 static_command_line, __start___param,
757 __stop___param - __start___param,
761 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
762 do_one_initcall(*fn);
765 static void __init do_initcalls(void)
769 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
770 do_initcall_level(level);
774 * Ok, the machine is now initialized. None of the devices
775 * have been touched yet, but the CPU subsystem is up and
776 * running, and memory and process management works.
778 * Now we can finally start doing some real work..
780 static void __init do_basic_setup(void)
783 usermodehelper_init();
788 usermodehelper_enable();
790 random_int_secret_init();
793 static void __init do_pre_smp_initcalls(void)
797 for (fn = __initcall_start; fn < __initcall0_start; fn++)
798 do_one_initcall(*fn);
802 * This function requests modules which should be loaded by default and is
803 * called twice right after initrd is mounted and right before init is
804 * exec'd. If such modules are on either initrd or rootfs, they will be
805 * loaded before control is passed to userland.
807 void __init load_default_modules(void)
809 load_default_elevator_module();
812 static int run_init_process(const char *init_filename)
814 argv_init[0] = init_filename;
815 return do_execve(init_filename,
816 (const char __user *const __user *)argv_init,
817 (const char __user *const __user *)envp_init);
820 static int try_to_run_init_process(const char *init_filename)
824 ret = run_init_process(init_filename);
826 if (ret && ret != -ENOENT) {
827 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
834 static noinline void __init kernel_init_freeable(void);
836 static int __ref kernel_init(void *unused)
840 kernel_init_freeable();
841 /* need to finish all async __init code before freeing the memory */
842 async_synchronize_full();
845 system_state = SYSTEM_RUNNING;
846 numa_default_policy();
848 flush_delayed_fput();
850 if (ramdisk_execute_command) {
851 ret = run_init_process(ramdisk_execute_command);
854 pr_err("Failed to execute %s (error %d)\n",
855 ramdisk_execute_command, ret);
859 * We try each of these until one succeeds.
861 * The Bourne shell can be used instead of init if we are
862 * trying to recover a really broken machine.
864 if (execute_command) {
865 ret = run_init_process(execute_command);
868 pr_err("Failed to execute %s (error %d). Attempting defaults...\n",
869 execute_command, ret);
871 if (!try_to_run_init_process("/sbin/init") ||
872 !try_to_run_init_process("/etc/init") ||
873 !try_to_run_init_process("/bin/init") ||
874 !try_to_run_init_process("/bin/sh"))
877 panic("No working init found. Try passing init= option to kernel. "
878 "See Linux Documentation/init.txt for guidance.");
881 static noinline void __init kernel_init_freeable(void)
884 * Wait until kthreadd is all set-up.
886 wait_for_completion(&kthreadd_done);
888 /* Now the scheduler is fully set up and can do blocking allocations */
889 gfp_allowed_mask = __GFP_BITS_MASK;
892 * init can allocate pages on any node
894 set_mems_allowed(node_states[N_MEMORY]);
896 * init can run on any cpu.
898 set_cpus_allowed_ptr(current, cpu_all_mask);
900 cad_pid = task_pid(current);
902 smp_prepare_cpus(setup_max_cpus);
904 do_pre_smp_initcalls();
905 lockup_detector_init();
912 /* Open the /dev/console on the rootfs, this should never fail */
913 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
914 pr_err("Warning: unable to open an initial console.\n");
919 * check if there is an early userspace init. If yes, let it do all
923 if (!ramdisk_execute_command)
924 ramdisk_execute_command = "/init";
926 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
927 ramdisk_execute_command = NULL;
932 * Ok, we have completed the initial bootup, and
933 * we're essentially up and running. Get rid of the
934 * initmem segments and start the user-mode stuff..
937 /* rootfs is available now, try loading default modules */
938 load_default_modules();