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/extable.h>
16 #include <linux/module.h>
17 #include <linux/proc_fs.h>
18 #include <linux/kernel.h>
19 #include <linux/syscalls.h>
20 #include <linux/stackprotector.h>
21 #include <linux/string.h>
22 #include <linux/ctype.h>
23 #include <linux/delay.h>
24 #include <linux/ioport.h>
25 #include <linux/init.h>
26 #include <linux/initrd.h>
27 #include <linux/bootmem.h>
28 #include <linux/acpi.h>
29 #include <linux/tty.h>
30 #include <linux/nmi.h>
31 #include <linux/percpu.h>
32 #include <linux/kmod.h>
33 #include <linux/vmalloc.h>
34 #include <linux/kernel_stat.h>
35 #include <linux/start_kernel.h>
36 #include <linux/security.h>
37 #include <linux/smp.h>
38 #include <linux/profile.h>
39 #include <linux/rcupdate.h>
40 #include <linux/moduleparam.h>
41 #include <linux/kallsyms.h>
42 #include <linux/writeback.h>
43 #include <linux/cpu.h>
44 #include <linux/cpuset.h>
45 #include <linux/cgroup.h>
46 #include <linux/efi.h>
47 #include <linux/tick.h>
48 #include <linux/interrupt.h>
49 #include <linux/taskstats_kern.h>
50 #include <linux/delayacct.h>
51 #include <linux/unistd.h>
52 #include <linux/rmap.h>
53 #include <linux/mempolicy.h>
54 #include <linux/key.h>
55 #include <linux/buffer_head.h>
56 #include <linux/page_ext.h>
57 #include <linux/debug_locks.h>
58 #include <linux/debugobjects.h>
59 #include <linux/lockdep.h>
60 #include <linux/kmemleak.h>
61 #include <linux/pid_namespace.h>
62 #include <linux/device.h>
63 #include <linux/kthread.h>
64 #include <linux/sched.h>
65 #include <linux/signal.h>
66 #include <linux/idr.h>
67 #include <linux/kgdb.h>
68 #include <linux/ftrace.h>
69 #include <linux/async.h>
70 #include <linux/kmemcheck.h>
71 #include <linux/sfi.h>
72 #include <linux/shmem_fs.h>
73 #include <linux/slab.h>
74 #include <linux/perf_event.h>
75 #include <linux/ptrace.h>
76 #include <linux/blkdev.h>
77 #include <linux/elevator.h>
78 #include <linux/sched_clock.h>
79 #include <linux/sched/task.h>
80 #include <linux/context_tracking.h>
81 #include <linux/random.h>
82 #include <linux/list.h>
83 #include <linux/integrity.h>
84 #include <linux/proc_ns.h>
86 #include <linux/cache.h>
87 #include <linux/rodata_test.h>
91 #include <asm/setup.h>
92 #include <asm/sections.h>
93 #include <asm/cacheflush.h>
95 static int kernel_init(void *);
97 extern void init_IRQ(void);
98 extern void fork_init(void);
99 extern void radix_tree_init(void);
102 * Debug helper: via this flag we know that we are in 'early bootup code'
103 * where only the boot processor is running with IRQ disabled. This means
104 * two things - IRQ must not be enabled before the flag is cleared and some
105 * operations which are not allowed with IRQ disabled are allowed while the
108 bool early_boot_irqs_disabled __read_mostly;
110 enum system_states system_state __read_mostly;
111 EXPORT_SYMBOL(system_state);
114 * Boot command-line arguments
116 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
117 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
119 extern void time_init(void);
120 /* Default late time init is NULL. archs can override this later. */
121 void (*__initdata late_time_init)(void);
123 /* Untouched command line saved by arch-specific code. */
124 char __initdata boot_command_line[COMMAND_LINE_SIZE];
125 /* Untouched saved command line (eg. for /proc) */
126 char *saved_command_line;
127 /* Command line for parameter parsing */
128 static char *static_command_line;
129 /* Command line for per-initcall parameter parsing */
130 static char *initcall_command_line;
132 static char *execute_command;
133 static char *ramdisk_execute_command;
136 * Used to generate warnings if static_key manipulation functions are used
137 * before jump_label_init is called.
139 bool static_key_initialized __read_mostly;
140 EXPORT_SYMBOL_GPL(static_key_initialized);
143 * If set, this is an indication to the drivers that reset the underlying
144 * device before going ahead with the initialization otherwise driver might
145 * rely on the BIOS and skip the reset operation.
147 * This is useful if kernel is booting in an unreliable environment.
148 * For ex. kdump situation where previous kernel has crashed, BIOS has been
149 * skipped and devices will be in unknown state.
151 unsigned int reset_devices;
152 EXPORT_SYMBOL(reset_devices);
154 static int __init set_reset_devices(char *str)
160 __setup("reset_devices", set_reset_devices);
162 static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
163 const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
164 static const char *panic_later, *panic_param;
166 extern const struct obs_kernel_param __setup_start[], __setup_end[];
168 static bool __init obsolete_checksetup(char *line)
170 const struct obs_kernel_param *p;
171 bool had_early_param = false;
175 int n = strlen(p->str);
176 if (parameqn(line, p->str, n)) {
178 /* Already done in parse_early_param?
179 * (Needs exact match on param part).
180 * Keep iterating, as we can have early
181 * params and __setups of same names 8( */
182 if (line[n] == '\0' || line[n] == '=')
183 had_early_param = true;
184 } else if (!p->setup_func) {
185 pr_warn("Parameter %s is obsolete, ignored\n",
188 } else if (p->setup_func(line + n))
192 } while (p < __setup_end);
194 return had_early_param;
198 * This should be approx 2 Bo*oMips to start (note initial shift), and will
199 * still work even if initially too large, it will just take slightly longer
201 unsigned long loops_per_jiffy = (1<<12);
202 EXPORT_SYMBOL(loops_per_jiffy);
204 static int __init debug_kernel(char *str)
206 console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
210 static int __init quiet_kernel(char *str)
212 console_loglevel = CONSOLE_LOGLEVEL_QUIET;
216 early_param("debug", debug_kernel);
217 early_param("quiet", quiet_kernel);
219 static int __init loglevel(char *str)
224 * Only update loglevel value when a correct setting was passed,
225 * to prevent blind crashes (when loglevel being set to 0) that
226 * are quite hard to debug
228 if (get_option(&str, &newlevel)) {
229 console_loglevel = newlevel;
236 early_param("loglevel", loglevel);
238 /* Change NUL term back to "=", to make "param" the whole string. */
239 static int __init repair_env_string(char *param, char *val,
240 const char *unused, void *arg)
243 /* param=val or param="val"? */
244 if (val == param+strlen(param)+1)
246 else if (val == param+strlen(param)+2) {
248 memmove(val-1, val, strlen(val)+1);
256 /* Anything after -- gets handed straight to init. */
257 static int __init set_init_arg(char *param, char *val,
258 const char *unused, void *arg)
265 repair_env_string(param, val, unused, NULL);
267 for (i = 0; argv_init[i]; i++) {
268 if (i == MAX_INIT_ARGS) {
269 panic_later = "init";
274 argv_init[i] = param;
279 * Unknown boot options get handed to init, unless they look like
280 * unused parameters (modprobe will find them in /proc/cmdline).
282 static int __init unknown_bootoption(char *param, char *val,
283 const char *unused, void *arg)
285 repair_env_string(param, val, unused, NULL);
287 /* Handle obsolete-style parameters */
288 if (obsolete_checksetup(param))
291 /* Unused module parameter. */
292 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
299 /* Environment option */
301 for (i = 0; envp_init[i]; i++) {
302 if (i == MAX_INIT_ENVS) {
306 if (!strncmp(param, envp_init[i], val - param))
309 envp_init[i] = param;
311 /* Command line option */
313 for (i = 0; argv_init[i]; i++) {
314 if (i == MAX_INIT_ARGS) {
315 panic_later = "init";
319 argv_init[i] = param;
324 static int __init init_setup(char *str)
328 execute_command = str;
330 * In case LILO is going to boot us with default command line,
331 * it prepends "auto" before the whole cmdline which makes
332 * the shell think it should execute a script with such name.
333 * So we ignore all arguments entered _before_ init=... [MJ]
335 for (i = 1; i < MAX_INIT_ARGS; i++)
339 __setup("init=", init_setup);
341 static int __init rdinit_setup(char *str)
345 ramdisk_execute_command = str;
346 /* See "auto" comment in init_setup */
347 for (i = 1; i < MAX_INIT_ARGS; i++)
351 __setup("rdinit=", rdinit_setup);
354 static const unsigned int setup_max_cpus = NR_CPUS;
355 static inline void setup_nr_cpu_ids(void) { }
356 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
360 * We need to store the untouched command line for future reference.
361 * We also need to store the touched command line since the parameter
362 * parsing is performed in place, and we should allow a component to
363 * store reference of name/value for future reference.
365 static void __init setup_command_line(char *command_line)
368 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
369 initcall_command_line =
370 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
371 static_command_line = memblock_virt_alloc(strlen(command_line) + 1, 0);
372 strcpy(saved_command_line, boot_command_line);
373 strcpy(static_command_line, command_line);
377 * We need to finalize in a non-__init function or else race conditions
378 * between the root thread and the init thread may cause start_kernel to
379 * be reaped by free_initmem before the root thread has proceeded to
382 * gcc-3.4 accidentally inlines this function, so use noinline.
385 static __initdata DECLARE_COMPLETION(kthreadd_done);
387 static noinline void __ref rest_init(void)
391 rcu_scheduler_starting();
393 * We need to spawn init first so that it obtains pid 1, however
394 * the init task will end up wanting to create kthreads, which, if
395 * we schedule it before we create kthreadd, will OOPS.
397 kernel_thread(kernel_init, NULL, CLONE_FS);
398 numa_default_policy();
399 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
401 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
403 complete(&kthreadd_done);
406 * The boot idle thread must execute schedule()
407 * at least once to get things moving:
409 init_idle_bootup_task(current);
410 schedule_preempt_disabled();
411 /* Call into cpu_idle with preempt disabled */
412 cpu_startup_entry(CPUHP_ONLINE);
415 /* Check for early params. */
416 static int __init do_early_param(char *param, char *val,
417 const char *unused, void *arg)
419 const struct obs_kernel_param *p;
421 for (p = __setup_start; p < __setup_end; p++) {
422 if ((p->early && parameq(param, p->str)) ||
423 (strcmp(param, "console") == 0 &&
424 strcmp(p->str, "earlycon") == 0)
426 if (p->setup_func(val) != 0)
427 pr_warn("Malformed early option '%s'\n", param);
430 /* We accept everything at this stage. */
434 void __init parse_early_options(char *cmdline)
436 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
440 /* Arch code calls this early on, or if not, just before other parsing. */
441 void __init parse_early_param(void)
443 static int done __initdata;
444 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
449 /* All fall through to do_early_param. */
450 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
451 parse_early_options(tmp_cmdline);
455 void __init __weak arch_post_acpi_subsys_init(void) { }
457 void __init __weak smp_setup_processor_id(void)
461 # if THREAD_SIZE >= PAGE_SIZE
462 void __init __weak thread_stack_cache_init(void)
468 * Set up kernel memory allocators
470 static void __init mm_init(void)
473 * page_ext requires contiguous pages,
474 * bigger than MAX_ORDER unless SPARSEMEM.
476 page_ext_init_flatmem();
485 asmlinkage __visible void __init start_kernel(void)
490 set_task_stack_end_magic(&init_task);
491 smp_setup_processor_id();
492 debug_objects_early_init();
495 * Set up the the initial canary ASAP:
497 boot_init_stack_canary();
502 early_boot_irqs_disabled = true;
505 * Interrupts are still disabled. Do necessary setups, then
510 pr_notice("%s", linux_banner);
511 setup_arch(&command_line);
512 mm_init_cpumask(&init_mm);
513 setup_command_line(command_line);
515 setup_per_cpu_areas();
516 boot_cpu_state_init();
517 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
519 build_all_zonelists(NULL, NULL);
522 pr_notice("Kernel command line: %s\n", boot_command_line);
524 after_dashes = parse_args("Booting kernel",
525 static_command_line, __start___param,
526 __stop___param - __start___param,
527 -1, -1, NULL, &unknown_bootoption);
528 if (!IS_ERR_OR_NULL(after_dashes))
529 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
535 * These use large bootmem allocations and must precede
540 vfs_caches_init_early();
546 * Set up the scheduler prior starting any interrupts (such as the
547 * timer interrupt). Full topology setup happens at smp_init()
548 * time - but meanwhile we still have a functioning scheduler.
552 * Disable preemption - early bootup scheduling is extremely
553 * fragile until we cpu_idle() for the first time.
556 if (WARN(!irqs_disabled(),
557 "Interrupts were enabled *very* early, fixing it\n"))
562 * Allow workqueue creation and work item queueing/cancelling
563 * early. Work item execution depends on kthreads and starts after
566 workqueue_init_early();
570 /* trace_printk() and trace points may be used after this */
573 context_tracking_init();
574 /* init some links before init_ISA_irqs() */
584 sched_clock_postinit();
588 call_function_init();
589 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
590 early_boot_irqs_disabled = false;
593 kmem_cache_init_late();
596 * HACK ALERT! This is early. We're enabling the console before
597 * we've done PCI setups etc, and console_init() must be aware of
598 * this. But we do want output early, in case something goes wrong.
602 panic("Too many boot %s vars at `%s'", panic_later,
608 * Need to run this when irqs are enabled, because it wants
609 * to self-test [hard/soft]-irqs on/off lock inversion bugs
614 #ifdef CONFIG_BLK_DEV_INITRD
615 if (initrd_start && !initrd_below_start_ok &&
616 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
617 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
618 page_to_pfn(virt_to_page((void *)initrd_start)),
624 debug_objects_mem_init();
626 setup_per_cpu_pageset();
635 if (efi_enabled(EFI_RUNTIME_SERVICES))
636 efi_enter_virtual_mode();
638 #ifdef CONFIG_X86_ESPFIX64
639 /* Should be run before the first non-init thread is created */
642 thread_stack_cache_init();
657 taskstats_init_early();
662 acpi_subsystem_init();
663 arch_post_acpi_subsys_init();
666 if (efi_enabled(EFI_RUNTIME_SERVICES)) {
667 efi_free_boot_services();
672 /* Do the rest non-__init'ed, we're now alive */
676 /* Call all constructor functions linked into the kernel. */
677 static void __init do_ctors(void)
679 #ifdef CONFIG_CONSTRUCTORS
680 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
682 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
688 core_param(initcall_debug, initcall_debug, bool, 0644);
690 #ifdef CONFIG_KALLSYMS
691 struct blacklist_entry {
692 struct list_head next;
696 static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
698 static int __init initcall_blacklist(char *str)
701 struct blacklist_entry *entry;
703 /* str argument is a comma-separated list of functions */
705 str_entry = strsep(&str, ",");
707 pr_debug("blacklisting initcall %s\n", str_entry);
708 entry = alloc_bootmem(sizeof(*entry));
709 entry->buf = alloc_bootmem(strlen(str_entry) + 1);
710 strcpy(entry->buf, str_entry);
711 list_add(&entry->next, &blacklisted_initcalls);
718 static bool __init_or_module initcall_blacklisted(initcall_t fn)
720 struct blacklist_entry *entry;
721 char fn_name[KSYM_SYMBOL_LEN];
724 if (list_empty(&blacklisted_initcalls))
727 addr = (unsigned long) dereference_function_descriptor(fn);
728 sprint_symbol_no_offset(fn_name, addr);
731 * fn will be "function_name [module_name]" where [module_name] is not
732 * displayed for built-in init functions. Strip off the [module_name].
734 strreplace(fn_name, ' ', '\0');
736 list_for_each_entry(entry, &blacklisted_initcalls, next) {
737 if (!strcmp(fn_name, entry->buf)) {
738 pr_debug("initcall %s blacklisted\n", fn_name);
746 static int __init initcall_blacklist(char *str)
748 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
752 static bool __init_or_module initcall_blacklisted(initcall_t fn)
757 __setup("initcall_blacklist=", initcall_blacklist);
759 static int __init_or_module do_one_initcall_debug(initcall_t fn)
761 ktime_t calltime, delta, rettime;
762 unsigned long long duration;
765 printk(KERN_DEBUG "calling %pF @ %i\n", fn, task_pid_nr(current));
766 calltime = ktime_get();
768 rettime = ktime_get();
769 delta = ktime_sub(rettime, calltime);
770 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
771 printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n",
777 int __init_or_module do_one_initcall(initcall_t fn)
779 int count = preempt_count();
783 if (initcall_blacklisted(fn))
787 ret = do_one_initcall_debug(fn);
793 if (preempt_count() != count) {
794 sprintf(msgbuf, "preemption imbalance ");
795 preempt_count_set(count);
797 if (irqs_disabled()) {
798 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
801 WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
803 add_latent_entropy();
808 extern initcall_t __initcall_start[];
809 extern initcall_t __initcall0_start[];
810 extern initcall_t __initcall1_start[];
811 extern initcall_t __initcall2_start[];
812 extern initcall_t __initcall3_start[];
813 extern initcall_t __initcall4_start[];
814 extern initcall_t __initcall5_start[];
815 extern initcall_t __initcall6_start[];
816 extern initcall_t __initcall7_start[];
817 extern initcall_t __initcall_end[];
819 static initcall_t *initcall_levels[] __initdata = {
831 /* Keep these in sync with initcalls in include/linux/init.h */
832 static char *initcall_level_names[] __initdata = {
843 static void __init do_initcall_level(int level)
847 strcpy(initcall_command_line, saved_command_line);
848 parse_args(initcall_level_names[level],
849 initcall_command_line, __start___param,
850 __stop___param - __start___param,
852 NULL, &repair_env_string);
854 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
855 do_one_initcall(*fn);
858 static void __init do_initcalls(void)
862 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
863 do_initcall_level(level);
867 * Ok, the machine is now initialized. None of the devices
868 * have been touched yet, but the CPU subsystem is up and
869 * running, and memory and process management works.
871 * Now we can finally start doing some real work..
873 static void __init do_basic_setup(void)
880 usermodehelper_enable();
882 random_int_secret_init();
885 static void __init do_pre_smp_initcalls(void)
889 for (fn = __initcall_start; fn < __initcall0_start; fn++)
890 do_one_initcall(*fn);
894 * This function requests modules which should be loaded by default and is
895 * called twice right after initrd is mounted and right before init is
896 * exec'd. If such modules are on either initrd or rootfs, they will be
897 * loaded before control is passed to userland.
899 void __init load_default_modules(void)
901 load_default_elevator_module();
904 static int run_init_process(const char *init_filename)
906 argv_init[0] = init_filename;
907 return do_execve(getname_kernel(init_filename),
908 (const char __user *const __user *)argv_init,
909 (const char __user *const __user *)envp_init);
912 static int try_to_run_init_process(const char *init_filename)
916 ret = run_init_process(init_filename);
918 if (ret && ret != -ENOENT) {
919 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
926 static noinline void __init kernel_init_freeable(void);
928 #if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
929 bool rodata_enabled __ro_after_init = true;
930 static int __init set_debug_rodata(char *str)
932 return strtobool(str, &rodata_enabled);
934 __setup("rodata=", set_debug_rodata);
937 #ifdef CONFIG_STRICT_KERNEL_RWX
938 static void mark_readonly(void)
940 if (rodata_enabled) {
944 pr_info("Kernel memory protection disabled.\n");
947 static inline void mark_readonly(void)
949 pr_warn("This architecture does not have kernel memory protection.\n");
953 static int __ref kernel_init(void *unused)
957 kernel_init_freeable();
958 /* need to finish all async __init code before freeing the memory */
959 async_synchronize_full();
962 system_state = SYSTEM_RUNNING;
963 numa_default_policy();
965 rcu_end_inkernel_boot();
967 if (ramdisk_execute_command) {
968 ret = run_init_process(ramdisk_execute_command);
971 pr_err("Failed to execute %s (error %d)\n",
972 ramdisk_execute_command, ret);
976 * We try each of these until one succeeds.
978 * The Bourne shell can be used instead of init if we are
979 * trying to recover a really broken machine.
981 if (execute_command) {
982 ret = run_init_process(execute_command);
985 panic("Requested init %s failed (error %d).",
986 execute_command, ret);
988 if (!try_to_run_init_process("/sbin/init") ||
989 !try_to_run_init_process("/etc/init") ||
990 !try_to_run_init_process("/bin/init") ||
991 !try_to_run_init_process("/bin/sh"))
994 panic("No working init found. Try passing init= option to kernel. "
995 "See Linux Documentation/admin-guide/init.rst for guidance.");
998 static noinline void __init kernel_init_freeable(void)
1001 * Wait until kthreadd is all set-up.
1003 wait_for_completion(&kthreadd_done);
1005 /* Now the scheduler is fully set up and can do blocking allocations */
1006 gfp_allowed_mask = __GFP_BITS_MASK;
1009 * init can allocate pages on any node
1011 set_mems_allowed(node_states[N_MEMORY]);
1013 * init can run on any cpu.
1015 set_cpus_allowed_ptr(current, cpu_all_mask);
1017 cad_pid = task_pid(current);
1019 smp_prepare_cpus(setup_max_cpus);
1023 do_pre_smp_initcalls();
1024 lockup_detector_init();
1029 page_alloc_init_late();
1033 /* Open the /dev/console on the rootfs, this should never fail */
1034 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
1035 pr_err("Warning: unable to open an initial console.\n");
1040 * check if there is an early userspace init. If yes, let it do all
1044 if (!ramdisk_execute_command)
1045 ramdisk_execute_command = "/init";
1047 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
1048 ramdisk_execute_command = NULL;
1049 prepare_namespace();
1053 * Ok, we have completed the initial bootup, and
1054 * we're essentially up and running. Get rid of the
1055 * initmem segments and start the user-mode stuff..
1057 * rootfs is available now, try loading the public keys
1058 * and default modules
1061 integrity_load_keys();
1062 load_default_modules();