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 #include <linux/types.h>
13 #include <linux/module.h>
14 #include <linux/proc_fs.h>
15 #include <linux/kernel.h>
16 #include <linux/syscalls.h>
17 #include <linux/stackprotector.h>
18 #include <linux/string.h>
19 #include <linux/ctype.h>
20 #include <linux/delay.h>
21 #include <linux/ioport.h>
22 #include <linux/init.h>
23 #include <linux/initrd.h>
24 #include <linux/bootmem.h>
25 #include <linux/acpi.h>
26 #include <linux/tty.h>
27 #include <linux/percpu.h>
28 #include <linux/kmod.h>
29 #include <linux/vmalloc.h>
30 #include <linux/kernel_stat.h>
31 #include <linux/start_kernel.h>
32 #include <linux/security.h>
33 #include <linux/smp.h>
34 #include <linux/profile.h>
35 #include <linux/rcupdate.h>
36 #include <linux/moduleparam.h>
37 #include <linux/kallsyms.h>
38 #include <linux/writeback.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/cgroup.h>
42 #include <linux/efi.h>
43 #include <linux/tick.h>
44 #include <linux/interrupt.h>
45 #include <linux/taskstats_kern.h>
46 #include <linux/delayacct.h>
47 #include <linux/unistd.h>
48 #include <linux/rmap.h>
49 #include <linux/mempolicy.h>
50 #include <linux/key.h>
51 #include <linux/buffer_head.h>
52 #include <linux/page_cgroup.h>
53 #include <linux/debug_locks.h>
54 #include <linux/debugobjects.h>
55 #include <linux/lockdep.h>
56 #include <linux/kmemleak.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/device.h>
59 #include <linux/kthread.h>
60 #include <linux/sched.h>
61 #include <linux/signal.h>
62 #include <linux/idr.h>
63 #include <linux/kgdb.h>
64 #include <linux/ftrace.h>
65 #include <linux/async.h>
66 #include <linux/kmemcheck.h>
67 #include <linux/sfi.h>
68 #include <linux/shmem_fs.h>
69 #include <linux/slab.h>
70 #include <linux/perf_event.h>
74 #include <asm/setup.h>
75 #include <asm/sections.h>
76 #include <asm/cacheflush.h>
78 #ifdef CONFIG_X86_LOCAL_APIC
82 static int kernel_init(void *);
84 extern void init_IRQ(void);
85 extern void fork_init(unsigned long);
86 extern void mca_init(void);
87 extern void sbus_init(void);
88 extern void prio_tree_init(void);
89 extern void radix_tree_init(void);
90 extern void free_initmem(void);
91 #ifndef CONFIG_DEBUG_RODATA
92 static inline void mark_rodata_ro(void) { }
96 extern void tc_init(void);
100 * Debug helper: via this flag we know that we are in 'early bootup code'
101 * where only the boot processor is running with IRQ disabled. This means
102 * two things - IRQ must not be enabled before the flag is cleared and some
103 * operations which are not allowed with IRQ disabled are allowed while the
106 bool early_boot_irqs_disabled __read_mostly;
108 enum system_states system_state __read_mostly;
109 EXPORT_SYMBOL(system_state);
112 * Boot command-line arguments
114 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
115 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
116 #ifdef CONFIG_INIT_PASS_ALL_PARAMS
117 #define INIT_PASS_FUNCTION pass_all_bootoptions
119 #define INIT_PASS_FUNCTION pass_unknown_bootoptions
122 extern void time_init(void);
123 /* Default late time init is NULL. archs can override this later. */
124 void (*__initdata late_time_init)(void);
125 extern void softirq_init(void);
127 /* Untouched command line saved by arch-specific code. */
128 char __initdata boot_command_line[COMMAND_LINE_SIZE];
129 /* Untouched saved command line (eg. for /proc) */
130 char *saved_command_line;
131 /* Command line for parameter parsing */
132 static char *static_command_line;
134 static char *execute_command;
135 static char *ramdisk_execute_command;
138 /* Setup configured maximum number of CPUs to activate */
139 unsigned int setup_max_cpus = NR_CPUS;
140 EXPORT_SYMBOL(setup_max_cpus);
144 * Setup routine for controlling SMP activation
146 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
147 * activation entirely (the MPS table probe still happens, though).
149 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
150 * greater than 0, limits the maximum number of CPUs activated in
154 void __weak arch_disable_smp_support(void) { }
156 static int __init nosmp(char *str)
159 arch_disable_smp_support();
164 early_param("nosmp", nosmp);
166 /* this is hard limit */
167 static int __init nrcpus(char *str)
171 get_option(&str, &nr_cpus);
172 if (nr_cpus > 0 && nr_cpus < nr_cpu_ids)
173 nr_cpu_ids = nr_cpus;
178 early_param("nr_cpus", nrcpus);
180 static int __init maxcpus(char *str)
182 get_option(&str, &setup_max_cpus);
183 if (setup_max_cpus == 0)
184 arch_disable_smp_support();
189 early_param("maxcpus", maxcpus);
191 static const unsigned int setup_max_cpus = NR_CPUS;
195 * If set, this is an indication to the drivers that reset the underlying
196 * device before going ahead with the initialization otherwise driver might
197 * rely on the BIOS and skip the reset operation.
199 * This is useful if kernel is booting in an unreliable environment.
200 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
201 * skipped and devices will be in unknown state.
203 unsigned int reset_devices;
204 EXPORT_SYMBOL(reset_devices);
206 static int __init set_reset_devices(char *str)
212 __setup("reset_devices", set_reset_devices);
214 static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
215 const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
216 static const char *panic_later, *panic_param;
218 extern const struct obs_kernel_param __setup_start[], __setup_end[];
220 static int __init obsolete_checksetup(char *line)
222 const struct obs_kernel_param *p;
223 int had_early_param = 0;
227 int n = strlen(p->str);
228 if (!strncmp(line, p->str, n)) {
230 /* Already done in parse_early_param?
231 * (Needs exact match on param part).
232 * Keep iterating, as we can have early
233 * params and __setups of same names 8( */
234 if (line[n] == '\0' || line[n] == '=')
236 } else if (!p->setup_func) {
237 printk(KERN_WARNING "Parameter %s is obsolete,"
238 " ignored\n", p->str);
240 } else if (p->setup_func(line + n))
244 } while (p < __setup_end);
246 return had_early_param;
250 * This should be approx 2 Bo*oMips to start (note initial shift), and will
251 * still work even if initially too large, it will just take slightly longer
253 unsigned long loops_per_jiffy = (1<<12);
255 EXPORT_SYMBOL(loops_per_jiffy);
257 static int __init debug_kernel(char *str)
259 console_loglevel = 10;
263 static int __init quiet_kernel(char *str)
265 console_loglevel = 2;
269 early_param("debug", debug_kernel);
270 early_param("quiet", quiet_kernel);
272 static int __init loglevel(char *str)
274 get_option(&str, &console_loglevel);
278 early_param("loglevel", loglevel);
281 * Select boot options to hand to init. If all is set hand off them all
282 * otherwise only hand off unused ones which do not apply to modules
283 * (modprobe will find them in /proc/cmdline).
285 static int __init pass_bootoption(char *param, char *val, int all)
287 /* Change NUL term back to "=", to make "param" the whole string. */
289 /* param=val or param="val"? */
290 if (val == param+strlen(param)+1)
292 else if (val == param+strlen(param)+2) {
294 memmove(val-1, val, strlen(val)+1);
300 /* Handle obsolete-style parameters */
301 if (obsolete_checksetup(param) && !all)
304 /* Unused module parameter. */
305 if (!all && strchr(param, '.') && (!val || strchr(param, '.') < val))
312 /* Environment option */
314 for (i = 0; envp_init[i]; i++) {
315 if (i == MAX_INIT_ENVS) {
316 panic_later = "Too many boot env vars at `%s'";
319 if (!strncmp(param, envp_init[i], val - param))
322 envp_init[i] = param;
324 /* Command line option */
326 for (i = 0; argv_init[i]; i++) {
327 if (i == MAX_INIT_ARGS) {
328 panic_later = "Too many boot init vars at `%s'";
332 argv_init[i] = param;
336 static int __init pass_unknown_bootoptions(char *param, char *val, int known)
340 return pass_bootoption(param, val, 0);
342 static int __init pass_all_bootoptions(char *param, char *val, int known)
344 return pass_bootoption(param, val, 1);
347 #ifdef CONFIG_DEBUG_PAGEALLOC
348 int __read_mostly debug_pagealloc_enabled = 0;
351 static int __init init_setup(char *str)
355 execute_command = str;
357 * In case LILO is going to boot us with default command line,
358 * it prepends "auto" before the whole cmdline which makes
359 * the shell think it should execute a script with such name.
360 * So we ignore all arguments entered _before_ init=... [MJ]
362 for (i = 1; i < MAX_INIT_ARGS; i++)
366 __setup("init=", init_setup);
368 static int __init rdinit_setup(char *str)
372 ramdisk_execute_command = str;
373 /* See "auto" comment in init_setup */
374 for (i = 1; i < MAX_INIT_ARGS; i++)
378 __setup("rdinit=", rdinit_setup);
382 #ifdef CONFIG_X86_LOCAL_APIC
383 static void __init smp_init(void)
385 APIC_init_uniprocessor();
388 #define smp_init() do { } while (0)
391 static inline void setup_nr_cpu_ids(void) { }
392 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
396 /* Setup number of possible processor ids */
397 int nr_cpu_ids __read_mostly = NR_CPUS;
398 EXPORT_SYMBOL(nr_cpu_ids);
400 /* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
401 static void __init setup_nr_cpu_ids(void)
403 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
406 /* Called by boot processor to activate the rest. */
407 static void __init smp_init(void)
411 /* FIXME: This should be done in userspace --RR */
412 for_each_present_cpu(cpu) {
413 if (num_online_cpus() >= setup_max_cpus)
415 if (!cpu_online(cpu))
419 /* Any cleanup work */
420 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
421 smp_cpus_done(setup_max_cpus);
427 * We need to store the untouched command line for future reference.
428 * We also need to store the touched command line since the parameter
429 * parsing is performed in place, and we should allow a component to
430 * store reference of name/value for future reference.
432 static void __init setup_command_line(char *command_line)
434 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
435 static_command_line = alloc_bootmem(strlen (command_line)+1);
436 strcpy (saved_command_line, boot_command_line);
437 strcpy (static_command_line, command_line);
441 * We need to finalize in a non-__init function or else race conditions
442 * between the root thread and the init thread may cause start_kernel to
443 * be reaped by free_initmem before the root thread has proceeded to
446 * gcc-3.4 accidentally inlines this function, so use noinline.
449 static __initdata DECLARE_COMPLETION(kthreadd_done);
451 static noinline void __init_refok rest_init(void)
455 rcu_scheduler_starting();
457 * We need to spawn init first so that it obtains pid 1, however
458 * the init task will end up wanting to create kthreads, which, if
459 * we schedule it before we create kthreadd, will OOPS.
461 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
462 numa_default_policy();
463 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
465 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
467 complete(&kthreadd_done);
470 * The boot idle thread must execute schedule()
471 * at least once to get things moving:
473 init_idle_bootup_task(current);
474 preempt_enable_no_resched();
478 /* Call into cpu_idle with preempt disabled */
482 /* Check for early params. */
483 static int __init do_early_param(char *param, char *val, int known)
485 const struct obs_kernel_param *p;
490 for (p = __setup_start; p < __setup_end; p++) {
491 if ((p->early && strcmp(param, p->str) == 0) ||
492 (strcmp(param, "console") == 0 &&
493 strcmp(p->str, "earlycon") == 0)
495 if (p->setup_func(val) != 0)
497 "Malformed early option '%s'\n", param);
500 /* We accept everything at this stage. */
504 void __init parse_early_options(char *cmdline)
506 parse_args("early options", cmdline, NULL, 0, do_early_param);
509 /* Arch code calls this early on, or if not, just before other parsing. */
510 void __init parse_early_param(void)
512 static __initdata int done = 0;
513 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
518 /* All fall through to do_early_param. */
519 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
520 parse_early_options(tmp_cmdline);
525 * Activate the first processor.
528 static void __init boot_cpu_init(void)
530 int cpu = smp_processor_id();
531 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
532 set_cpu_online(cpu, true);
533 set_cpu_active(cpu, true);
534 set_cpu_present(cpu, true);
535 set_cpu_possible(cpu, true);
538 void __init __weak smp_setup_processor_id(void)
542 void __init __weak thread_info_cache_init(void)
547 * Set up kernel memory allocators
549 static void __init mm_init(void)
552 * page_cgroup requires countinous pages as memmap
553 * and it's bigger than MAX_ORDER unless SPARSEMEM.
555 page_cgroup_init_flatmem();
559 pgtable_cache_init();
563 asmlinkage void __init start_kernel(void)
566 extern const struct kernel_param __start___param[], __stop___param[];
568 smp_setup_processor_id();
571 * Need to run as early as possible, to initialize the
575 debug_objects_early_init();
578 * Set up the the initial canary ASAP:
580 boot_init_stack_canary();
585 early_boot_irqs_disabled = true;
588 * Interrupts are still disabled. Do necessary setups, then
594 printk(KERN_NOTICE "%s", linux_banner);
595 setup_arch(&command_line);
596 mm_init_owner(&init_mm, &init_task);
597 setup_command_line(command_line);
599 setup_per_cpu_areas();
600 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
602 build_all_zonelists(NULL);
605 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
607 parse_args("Booting kernel", static_command_line, __start___param,
608 __stop___param - __start___param,
609 &INIT_PASS_FUNCTION);
611 * These use large bootmem allocations and must precede
615 vfs_caches_init_early();
620 * Set up the scheduler prior starting any interrupts (such as the
621 * timer interrupt). Full topology setup happens at smp_init()
622 * time - but meanwhile we still have a functioning scheduler.
626 * Disable preemption - early bootup scheduling is extremely
627 * fragile until we cpu_idle() for the first time.
630 if (!irqs_disabled()) {
631 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
632 "enabled *very* early, fixing it\n");
639 /* init some links before init_ISA_irqs() */
649 if (!irqs_disabled())
650 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
652 early_boot_irqs_disabled = false;
655 /* Interrupts are enabled now so all GFP allocations are safe. */
656 gfp_allowed_mask = __GFP_BITS_MASK;
658 kmem_cache_init_late();
661 * HACK ALERT! This is early. We're enabling the console before
662 * we've done PCI setups etc, and console_init() must be aware of
663 * this. But we do want output early, in case something goes wrong.
667 panic(panic_later, panic_param);
672 * Need to run this when irqs are enabled, because it wants
673 * to self-test [hard/soft]-irqs on/off lock inversion bugs
678 #ifdef CONFIG_BLK_DEV_INITRD
679 if (initrd_start && !initrd_below_start_ok &&
680 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
681 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
683 page_to_pfn(virt_to_page((void *)initrd_start)),
689 enable_debug_pagealloc();
690 debug_objects_mem_init();
692 setup_per_cpu_pageset();
702 efi_enter_virtual_mode();
704 thread_info_cache_init();
706 fork_init(totalram_pages);
712 vfs_caches_init(totalram_pages);
714 /* rootfs populating might need page-writeback */
715 page_writeback_init();
716 #ifdef CONFIG_PROC_FS
721 taskstats_init_early();
726 acpi_early_init(); /* before LAPIC and SMP init */
731 /* Do the rest non-__init'ed, we're now alive */
735 /* Call all constructor functions linked into the kernel. */
736 static void __init do_ctors(void)
738 #ifdef CONFIG_CONSTRUCTORS
739 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
741 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
747 core_param(initcall_debug, initcall_debug, bool, 0644);
749 static char msgbuf[64];
751 static int __init_or_module do_one_initcall_debug(initcall_t fn)
753 ktime_t calltime, delta, rettime;
754 unsigned long long duration;
757 printk(KERN_DEBUG "calling %pF @ %i\n", fn, task_pid_nr(current));
758 calltime = ktime_get();
760 rettime = ktime_get();
761 delta = ktime_sub(rettime, calltime);
762 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
763 printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n", fn,
769 int __init_or_module do_one_initcall(initcall_t fn)
771 int count = preempt_count();
775 ret = do_one_initcall_debug(fn);
781 if (ret && ret != -ENODEV && initcall_debug)
782 sprintf(msgbuf, "error code %d ", ret);
784 if (preempt_count() != count) {
785 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
786 preempt_count() = count;
788 if (irqs_disabled()) {
789 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
793 printk("initcall %pF returned with %s\n", fn, msgbuf);
800 extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
802 static void __init do_initcalls(void)
806 for (fn = __early_initcall_end; fn < __initcall_end; fn++)
807 do_one_initcall(*fn);
811 * Ok, the machine is now initialized. None of the devices
812 * have been touched yet, but the CPU subsystem is up and
813 * running, and memory and process management works.
815 * Now we can finally start doing some real work..
817 static void __init do_basic_setup(void)
820 usermodehelper_init();
828 static void __init do_pre_smp_initcalls(void)
832 for (fn = __initcall_start; fn < __early_initcall_end; fn++)
833 do_one_initcall(*fn);
836 static void run_init_process(const char *init_filename)
838 argv_init[0] = init_filename;
839 kernel_execve(init_filename, argv_init, envp_init);
842 /* This is a non __init function. Force it to be noinline otherwise gcc
843 * makes it inline to init() and it becomes part of init.text section
845 static noinline int init_post(void)
847 /* need to finish all async __init code before freeing the memory */
848 async_synchronize_full();
851 system_state = SYSTEM_RUNNING;
852 numa_default_policy();
855 current->signal->flags |= SIGNAL_UNKILLABLE;
857 if (ramdisk_execute_command) {
858 run_init_process(ramdisk_execute_command);
859 printk(KERN_WARNING "Failed to execute %s\n",
860 ramdisk_execute_command);
864 * We try each of these until one succeeds.
866 * The Bourne shell can be used instead of init if we are
867 * trying to recover a really broken machine.
869 if (execute_command) {
870 run_init_process(execute_command);
871 printk(KERN_WARNING "Failed to execute %s. Attempting "
872 "defaults...\n", execute_command);
874 run_init_process("/sbin/init");
875 run_init_process("/etc/init");
876 run_init_process("/bin/init");
877 run_init_process("/bin/sh");
879 panic("No init found. Try passing init= option to kernel. "
880 "See Linux Documentation/init.txt for guidance.");
883 static int __init kernel_init(void * unused)
886 * Wait until kthreadd is all set-up.
888 wait_for_completion(&kthreadd_done);
890 * init can allocate pages on any node
892 set_mems_allowed(node_states[N_HIGH_MEMORY]);
894 * init can run on any cpu.
896 set_cpus_allowed_ptr(current, cpu_all_mask);
898 * Tell the world that we're going to be the grim
899 * reaper of innocent orphaned children.
901 * We don't want people to have to make incorrect
902 * assumptions about where in the task array this
905 init_pid_ns.child_reaper = current;
907 cad_pid = task_pid(current);
909 smp_prepare_cpus(setup_max_cpus);
911 do_pre_smp_initcalls();
912 lockup_detector_init();
919 /* Open the /dev/console on the rootfs, this should never fail */
920 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
921 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
926 * We need to ensure that the filesystem is ready by this point, wait for
927 * async_populate_rootfs to complete.
929 populate_rootfs_wait();
932 * check if there is an early userspace init. If yes, let it do all
936 if (!ramdisk_execute_command)
937 ramdisk_execute_command = "/init";
939 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
940 ramdisk_execute_command = NULL;
945 * Ok, we have completed the initial bootup, and
946 * we're essentially up and running. Get rid of the
947 * initmem segments and start the user-mode stuff..