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
9 * Simplified starting of init: Michael A. Griffith <grif@acm.org>
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_ext.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>
80 #include <linux/list.h>
81 #include <linux/integrity.h>
82 #include <linux/proc_ns.h>
87 #include <asm/setup.h>
88 #include <asm/sections.h>
89 #include <asm/cacheflush.h>
91 #include <asm/kaiser.h>
94 static int kernel_init(void *);
96 extern void init_IRQ(void);
97 extern void fork_init(void);
98 extern void radix_tree_init(void);
99 #ifndef CONFIG_DEBUG_RODATA
100 static inline void mark_rodata_ro(void) { }
104 * Debug helper: via this flag we know that we are in 'early bootup code'
105 * where only the boot processor is running with IRQ disabled. This means
106 * two things - IRQ must not be enabled before the flag is cleared and some
107 * operations which are not allowed with IRQ disabled are allowed while the
110 bool early_boot_irqs_disabled __read_mostly;
112 enum system_states system_state __read_mostly;
113 EXPORT_SYMBOL(system_state);
116 * Boot command-line arguments
118 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
119 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
121 extern void time_init(void);
122 /* Default late time init is NULL. archs can override this later. */
123 void (*__initdata late_time_init)(void);
125 /* Untouched command line saved by arch-specific code. */
126 char __initdata boot_command_line[COMMAND_LINE_SIZE];
127 /* Untouched saved command line (eg. for /proc) */
128 char *saved_command_line;
129 /* Command line for parameter parsing */
130 static char *static_command_line;
131 /* Command line for per-initcall parameter parsing */
132 static char *initcall_command_line;
134 static char *execute_command;
135 static char *ramdisk_execute_command;
138 * Used to generate warnings if static_key manipulation functions are used
139 * before jump_label_init is called.
141 bool static_key_initialized __read_mostly;
142 EXPORT_SYMBOL_GPL(static_key_initialized);
145 * If set, this is an indication to the drivers that reset the underlying
146 * device before going ahead with the initialization otherwise driver might
147 * rely on the BIOS and skip the reset operation.
149 * This is useful if kernel is booting in an unreliable environment.
150 * For ex. kdump situation where previous kernel has crashed, BIOS has been
151 * skipped and devices will be in unknown state.
153 unsigned int reset_devices;
154 EXPORT_SYMBOL(reset_devices);
156 static int __init set_reset_devices(char *str)
162 __setup("reset_devices", set_reset_devices);
164 static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
165 const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
166 static const char *panic_later, *panic_param;
168 extern const struct obs_kernel_param __setup_start[], __setup_end[];
170 static int __init obsolete_checksetup(char *line)
172 const struct obs_kernel_param *p;
173 int had_early_param = 0;
177 int n = strlen(p->str);
178 if (parameqn(line, p->str, n)) {
180 /* Already done in parse_early_param?
181 * (Needs exact match on param part).
182 * Keep iterating, as we can have early
183 * params and __setups of same names 8( */
184 if (line[n] == '\0' || line[n] == '=')
186 } else if (!p->setup_func) {
187 pr_warn("Parameter %s is obsolete, ignored\n",
190 } else if (p->setup_func(line + n))
194 } while (p < __setup_end);
196 return had_early_param;
200 * This should be approx 2 Bo*oMips to start (note initial shift), and will
201 * still work even if initially too large, it will just take slightly longer
203 unsigned long loops_per_jiffy = (1<<12);
204 EXPORT_SYMBOL(loops_per_jiffy);
206 static int __init debug_kernel(char *str)
208 console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
212 static int __init quiet_kernel(char *str)
214 console_loglevel = CONSOLE_LOGLEVEL_QUIET;
218 early_param("debug", debug_kernel);
219 early_param("quiet", quiet_kernel);
221 static int __init loglevel(char *str)
226 * Only update loglevel value when a correct setting was passed,
227 * to prevent blind crashes (when loglevel being set to 0) that
228 * are quite hard to debug
230 if (get_option(&str, &newlevel)) {
231 console_loglevel = newlevel;
238 early_param("loglevel", loglevel);
240 /* Change NUL term back to "=", to make "param" the whole string. */
241 static int __init repair_env_string(char *param, char *val,
242 const char *unused, void *arg)
245 /* param=val or param="val"? */
246 if (val == param+strlen(param)+1)
248 else if (val == param+strlen(param)+2) {
250 memmove(val-1, val, strlen(val)+1);
258 /* Anything after -- gets handed straight to init. */
259 static int __init set_init_arg(char *param, char *val,
260 const char *unused, void *arg)
267 repair_env_string(param, val, unused, NULL);
269 for (i = 0; argv_init[i]; i++) {
270 if (i == MAX_INIT_ARGS) {
271 panic_later = "init";
276 argv_init[i] = param;
281 * Unknown boot options get handed to init, unless they look like
282 * unused parameters (modprobe will find them in /proc/cmdline).
284 static int __init unknown_bootoption(char *param, char *val,
285 const char *unused, void *arg)
287 repair_env_string(param, val, unused, NULL);
289 /* Handle obsolete-style parameters */
290 if (obsolete_checksetup(param))
293 /* Unused module parameter. */
294 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
301 /* Environment option */
303 for (i = 0; envp_init[i]; i++) {
304 if (i == MAX_INIT_ENVS) {
308 if (!strncmp(param, envp_init[i], val - param))
311 envp_init[i] = param;
313 /* Command line option */
315 for (i = 0; argv_init[i]; i++) {
316 if (i == MAX_INIT_ARGS) {
317 panic_later = "init";
321 argv_init[i] = param;
326 static int __init init_setup(char *str)
330 execute_command = str;
332 * In case LILO is going to boot us with default command line,
333 * it prepends "auto" before the whole cmdline which makes
334 * the shell think it should execute a script with such name.
335 * So we ignore all arguments entered _before_ init=... [MJ]
337 for (i = 1; i < MAX_INIT_ARGS; i++)
341 __setup("init=", init_setup);
343 static int __init rdinit_setup(char *str)
347 ramdisk_execute_command = str;
348 /* See "auto" comment in init_setup */
349 for (i = 1; i < MAX_INIT_ARGS; i++)
353 __setup("rdinit=", rdinit_setup);
356 static const unsigned int setup_max_cpus = NR_CPUS;
357 static inline void setup_nr_cpu_ids(void) { }
358 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
362 * We need to store the untouched command line for future reference.
363 * We also need to store the touched command line since the parameter
364 * parsing is performed in place, and we should allow a component to
365 * store reference of name/value for future reference.
367 static void __init setup_command_line(char *command_line)
370 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
371 initcall_command_line =
372 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
373 static_command_line = memblock_virt_alloc(strlen(command_line) + 1, 0);
374 strcpy(saved_command_line, boot_command_line);
375 strcpy(static_command_line, command_line);
379 * We need to finalize in a non-__init function or else race conditions
380 * between the root thread and the init thread may cause start_kernel to
381 * be reaped by free_initmem before the root thread has proceeded to
384 * gcc-3.4 accidentally inlines this function, so use noinline.
387 static __initdata DECLARE_COMPLETION(kthreadd_done);
389 static noinline void __init_refok rest_init(void)
393 rcu_scheduler_starting();
394 smpboot_thread_init();
396 * We need to spawn init first so that it obtains pid 1, however
397 * the init task will end up wanting to create kthreads, which, if
398 * we schedule it before we create kthreadd, will OOPS.
400 kernel_thread(kernel_init, NULL, CLONE_FS);
401 numa_default_policy();
402 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
404 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
406 complete(&kthreadd_done);
409 * The boot idle thread must execute schedule()
410 * at least once to get things moving:
412 init_idle_bootup_task(current);
413 schedule_preempt_disabled();
414 /* Call into cpu_idle with preempt disabled */
415 cpu_startup_entry(CPUHP_ONLINE);
418 /* Check for early params. */
419 static int __init do_early_param(char *param, char *val,
420 const char *unused, void *arg)
422 const struct obs_kernel_param *p;
424 for (p = __setup_start; p < __setup_end; p++) {
425 if ((p->early && parameq(param, p->str)) ||
426 (strcmp(param, "console") == 0 &&
427 strcmp(p->str, "earlycon") == 0)
429 if (p->setup_func(val) != 0)
430 pr_warn("Malformed early option '%s'\n", param);
433 /* We accept everything at this stage. */
437 void __init parse_early_options(char *cmdline)
439 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
443 /* Arch code calls this early on, or if not, just before other parsing. */
444 void __init parse_early_param(void)
446 static int done __initdata;
447 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
452 /* All fall through to do_early_param. */
453 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
454 parse_early_options(tmp_cmdline);
459 * Activate the first processor.
462 static void __init boot_cpu_init(void)
464 int cpu = smp_processor_id();
465 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
466 set_cpu_online(cpu, true);
467 set_cpu_active(cpu, true);
468 set_cpu_present(cpu, true);
469 set_cpu_possible(cpu, true);
472 void __init __weak smp_setup_processor_id(void)
476 # if THREAD_SIZE >= PAGE_SIZE
477 void __init __weak thread_info_cache_init(void)
483 * Set up kernel memory allocators
485 static void __init mm_init(void)
488 * page_ext requires contiguous pages,
489 * bigger than MAX_ORDER unless SPARSEMEM.
491 page_ext_init_flatmem();
503 asmlinkage __visible void __init start_kernel(void)
509 * Need to run as early as possible, to initialize the
513 set_task_stack_end_magic(&init_task);
514 smp_setup_processor_id();
515 debug_objects_early_init();
518 * Set up the the initial canary ASAP:
520 boot_init_stack_canary();
525 early_boot_irqs_disabled = true;
528 * Interrupts are still disabled. Do necessary setups, then
533 pr_notice("%s", linux_banner);
534 setup_arch(&command_line);
535 mm_init_cpumask(&init_mm);
536 setup_command_line(command_line);
538 setup_per_cpu_areas();
539 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
541 build_all_zonelists(NULL, NULL);
544 pr_notice("Kernel command line: %s\n", boot_command_line);
546 after_dashes = parse_args("Booting kernel",
547 static_command_line, __start___param,
548 __stop___param - __start___param,
549 -1, -1, NULL, &unknown_bootoption);
550 if (!IS_ERR_OR_NULL(after_dashes))
551 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
557 * These use large bootmem allocations and must precede
562 vfs_caches_init_early();
568 * Set up the scheduler prior starting any interrupts (such as the
569 * timer interrupt). Full topology setup happens at smp_init()
570 * time - but meanwhile we still have a functioning scheduler.
574 * Disable preemption - early bootup scheduling is extremely
575 * fragile until we cpu_idle() for the first time.
578 if (WARN(!irqs_disabled(),
579 "Interrupts were enabled *very* early, fixing it\n"))
584 /* trace_printk() and trace points may be used after this */
587 context_tracking_init();
589 /* init some links before init_ISA_irqs() */
599 sched_clock_postinit();
602 call_function_init();
603 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
604 early_boot_irqs_disabled = false;
607 kmem_cache_init_late();
610 * HACK ALERT! This is early. We're enabling the console before
611 * we've done PCI setups etc, and console_init() must be aware of
612 * this. But we do want output early, in case something goes wrong.
616 panic("Too many boot %s vars at `%s'", panic_later,
622 * Need to run this when irqs are enabled, because it wants
623 * to self-test [hard/soft]-irqs on/off lock inversion bugs
628 #ifdef CONFIG_BLK_DEV_INITRD
629 if (initrd_start && !initrd_below_start_ok &&
630 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
631 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
632 page_to_pfn(virt_to_page((void *)initrd_start)),
638 debug_objects_mem_init();
640 setup_per_cpu_pageset();
650 if (efi_enabled(EFI_RUNTIME_SERVICES))
651 efi_enter_virtual_mode();
653 #ifdef CONFIG_X86_ESPFIX64
654 /* Should be run before the first non-init thread is created */
657 thread_info_cache_init();
667 /* rootfs populating might need page-writeback */
668 page_writeback_init();
673 taskstats_init_early();
678 acpi_subsystem_init();
681 if (efi_enabled(EFI_RUNTIME_SERVICES)) {
683 efi_free_boot_services();
688 /* Do the rest non-__init'ed, we're now alive */
692 /* Call all constructor functions linked into the kernel. */
693 static void __init do_ctors(void)
695 #ifdef CONFIG_CONSTRUCTORS
696 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
698 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
704 core_param(initcall_debug, initcall_debug, bool, 0644);
706 #ifdef CONFIG_KALLSYMS
707 struct blacklist_entry {
708 struct list_head next;
712 static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
714 static int __init initcall_blacklist(char *str)
717 struct blacklist_entry *entry;
719 /* str argument is a comma-separated list of functions */
721 str_entry = strsep(&str, ",");
723 pr_debug("blacklisting initcall %s\n", str_entry);
724 entry = alloc_bootmem(sizeof(*entry));
725 entry->buf = alloc_bootmem(strlen(str_entry) + 1);
726 strcpy(entry->buf, str_entry);
727 list_add(&entry->next, &blacklisted_initcalls);
734 static bool __init_or_module initcall_blacklisted(initcall_t fn)
736 struct list_head *tmp;
737 struct blacklist_entry *entry;
740 fn_name = kasprintf(GFP_KERNEL, "%pf", fn);
744 list_for_each(tmp, &blacklisted_initcalls) {
745 entry = list_entry(tmp, struct blacklist_entry, next);
746 if (!strcmp(fn_name, entry->buf)) {
747 pr_debug("initcall %s blacklisted\n", fn_name);
757 static int __init initcall_blacklist(char *str)
759 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
763 static bool __init_or_module initcall_blacklisted(initcall_t fn)
768 __setup("initcall_blacklist=", initcall_blacklist);
770 static int __init_or_module do_one_initcall_debug(initcall_t fn)
772 ktime_t calltime, delta, rettime;
773 unsigned long long duration;
776 printk(KERN_DEBUG "calling %pF @ %i\n", fn, task_pid_nr(current));
777 calltime = ktime_get();
779 rettime = ktime_get();
780 delta = ktime_sub(rettime, calltime);
781 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
782 printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n",
788 int __init_or_module do_one_initcall(initcall_t fn)
790 int count = preempt_count();
794 if (initcall_blacklisted(fn))
798 ret = do_one_initcall_debug(fn);
804 if (preempt_count() != count) {
805 sprintf(msgbuf, "preemption imbalance ");
806 preempt_count_set(count);
808 if (irqs_disabled()) {
809 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
812 WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
818 extern initcall_t __initcall_start[];
819 extern initcall_t __initcall0_start[];
820 extern initcall_t __initcall1_start[];
821 extern initcall_t __initcall2_start[];
822 extern initcall_t __initcall3_start[];
823 extern initcall_t __initcall4_start[];
824 extern initcall_t __initcall5_start[];
825 extern initcall_t __initcall6_start[];
826 extern initcall_t __initcall7_start[];
827 extern initcall_t __initcall_end[];
829 static initcall_t *initcall_levels[] __initdata = {
841 /* Keep these in sync with initcalls in include/linux/init.h */
842 static char *initcall_level_names[] __initdata = {
853 static void __init do_initcall_level(int level)
857 strcpy(initcall_command_line, saved_command_line);
858 parse_args(initcall_level_names[level],
859 initcall_command_line, __start___param,
860 __stop___param - __start___param,
862 NULL, &repair_env_string);
864 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
865 do_one_initcall(*fn);
868 static void __init do_initcalls(void)
872 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
873 do_initcall_level(level);
877 * Ok, the machine is now initialized. None of the devices
878 * have been touched yet, but the CPU subsystem is up and
879 * running, and memory and process management works.
881 * Now we can finally start doing some real work..
883 static void __init do_basic_setup(void)
890 usermodehelper_enable();
892 random_int_secret_init();
895 static void __init do_pre_smp_initcalls(void)
899 for (fn = __initcall_start; fn < __initcall0_start; fn++)
900 do_one_initcall(*fn);
904 * This function requests modules which should be loaded by default and is
905 * called twice right after initrd is mounted and right before init is
906 * exec'd. If such modules are on either initrd or rootfs, they will be
907 * loaded before control is passed to userland.
909 void __init load_default_modules(void)
911 load_default_elevator_module();
914 static int run_init_process(const char *init_filename)
916 argv_init[0] = init_filename;
917 return do_execve(getname_kernel(init_filename),
918 (const char __user *const __user *)argv_init,
919 (const char __user *const __user *)envp_init);
922 static int try_to_run_init_process(const char *init_filename)
926 ret = run_init_process(init_filename);
928 if (ret && ret != -ENOENT) {
929 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
936 static noinline void __init kernel_init_freeable(void);
938 static int __ref kernel_init(void *unused)
942 kernel_init_freeable();
943 /* need to finish all async __init code before freeing the memory */
944 async_synchronize_full();
947 system_state = SYSTEM_RUNNING;
948 numa_default_policy();
950 flush_delayed_fput();
952 if (ramdisk_execute_command) {
953 ret = run_init_process(ramdisk_execute_command);
956 pr_err("Failed to execute %s (error %d)\n",
957 ramdisk_execute_command, ret);
961 * We try each of these until one succeeds.
963 * The Bourne shell can be used instead of init if we are
964 * trying to recover a really broken machine.
966 if (execute_command) {
967 ret = run_init_process(execute_command);
970 panic("Requested init %s failed (error %d).",
971 execute_command, ret);
973 if (!try_to_run_init_process("/sbin/init") ||
974 !try_to_run_init_process("/etc/init") ||
975 !try_to_run_init_process("/bin/init") ||
976 !try_to_run_init_process("/bin/sh"))
979 panic("No working init found. Try passing init= option to kernel. "
980 "See Linux Documentation/init.txt for guidance.");
983 static noinline void __init kernel_init_freeable(void)
986 * Wait until kthreadd is all set-up.
988 wait_for_completion(&kthreadd_done);
990 /* Now the scheduler is fully set up and can do blocking allocations */
991 gfp_allowed_mask = __GFP_BITS_MASK;
994 * init can allocate pages on any node
996 set_mems_allowed(node_states[N_MEMORY]);
998 * init can run on any cpu.
1000 set_cpus_allowed_ptr(current, cpu_all_mask);
1002 cad_pid = task_pid(current);
1004 smp_prepare_cpus(setup_max_cpus);
1006 do_pre_smp_initcalls();
1007 lockup_detector_init();
1012 page_alloc_init_late();
1016 /* Open the /dev/console on the rootfs, this should never fail */
1017 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
1018 pr_err("Warning: unable to open an initial console.\n");
1023 * check if there is an early userspace init. If yes, let it do all
1027 if (!ramdisk_execute_command)
1028 ramdisk_execute_command = "/init";
1030 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
1031 ramdisk_execute_command = NULL;
1032 prepare_namespace();
1036 * Ok, we have completed the initial bootup, and
1037 * we're essentially up and running. Get rid of the
1038 * initmem segments and start the user-mode stuff..
1040 * rootfs is available now, try loading the public keys
1041 * and default modules
1044 integrity_load_keys();
1045 load_default_modules();