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/utsname.h>
22 #include <linux/ioport.h>
23 #include <linux/init.h>
24 #include <linux/smp_lock.h>
25 #include <linux/initrd.h>
26 #include <linux/bootmem.h>
27 #include <linux/acpi.h>
28 #include <linux/tty.h>
29 #include <linux/gfp.h>
30 #include <linux/percpu.h>
31 #include <linux/kmod.h>
32 #include <linux/vmalloc.h>
33 #include <linux/kernel_stat.h>
34 #include <linux/start_kernel.h>
35 #include <linux/security.h>
36 #include <linux/smp.h>
37 #include <linux/workqueue.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_cgroup.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/ftrace.h>
68 #include <linux/async.h>
69 #include <linux/kmemcheck.h>
70 #include <linux/kmemtrace.h>
71 #include <linux/shmem_fs.h>
72 #include <trace/boot.h>
76 #include <asm/setup.h>
77 #include <asm/sections.h>
78 #include <asm/cacheflush.h>
80 #ifdef CONFIG_X86_LOCAL_APIC
84 static int kernel_init(void *);
86 extern void init_IRQ(void);
87 extern void fork_init(unsigned long);
88 extern void mca_init(void);
89 extern void sbus_init(void);
90 extern void prio_tree_init(void);
91 extern void radix_tree_init(void);
92 extern void free_initmem(void);
93 #ifndef CONFIG_DEBUG_RODATA
94 static inline void mark_rodata_ro(void) { }
98 extern void tc_init(void);
101 enum system_states system_state __read_mostly;
102 EXPORT_SYMBOL(system_state);
105 * Boot command-line arguments
107 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
108 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
110 extern void time_init(void);
111 /* Default late time init is NULL. archs can override this later. */
112 void (*__initdata late_time_init)(void);
113 extern void softirq_init(void);
115 /* Untouched command line saved by arch-specific code. */
116 char __initdata boot_command_line[COMMAND_LINE_SIZE];
117 /* Untouched saved command line (eg. for /proc) */
118 char *saved_command_line;
119 /* Command line for parameter parsing */
120 static char *static_command_line;
122 static char *execute_command;
123 static char *ramdisk_execute_command;
126 /* Setup configured maximum number of CPUs to activate */
127 unsigned int __initdata setup_max_cpus = NR_CPUS;
130 * Setup routine for controlling SMP activation
132 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
133 * activation entirely (the MPS table probe still happens, though).
135 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
136 * greater than 0, limits the maximum number of CPUs activated in
140 void __weak arch_disable_smp_support(void) { }
142 static int __init nosmp(char *str)
145 arch_disable_smp_support();
150 early_param("nosmp", nosmp);
152 static int __init maxcpus(char *str)
154 get_option(&str, &setup_max_cpus);
155 if (setup_max_cpus == 0)
156 arch_disable_smp_support();
161 early_param("maxcpus", maxcpus);
163 const unsigned int setup_max_cpus = NR_CPUS;
167 * If set, this is an indication to the drivers that reset the underlying
168 * device before going ahead with the initialization otherwise driver might
169 * rely on the BIOS and skip the reset operation.
171 * This is useful if kernel is booting in an unreliable environment.
172 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
173 * skipped and devices will be in unknown state.
175 unsigned int reset_devices;
176 EXPORT_SYMBOL(reset_devices);
178 static int __init set_reset_devices(char *str)
184 __setup("reset_devices", set_reset_devices);
186 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
187 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
188 static const char *panic_later, *panic_param;
190 extern struct obs_kernel_param __setup_start[], __setup_end[];
192 static int __init obsolete_checksetup(char *line)
194 struct obs_kernel_param *p;
195 int had_early_param = 0;
199 int n = strlen(p->str);
200 if (!strncmp(line, p->str, n)) {
202 /* Already done in parse_early_param?
203 * (Needs exact match on param part).
204 * Keep iterating, as we can have early
205 * params and __setups of same names 8( */
206 if (line[n] == '\0' || line[n] == '=')
208 } else if (!p->setup_func) {
209 printk(KERN_WARNING "Parameter %s is obsolete,"
210 " ignored\n", p->str);
212 } else if (p->setup_func(line + n))
216 } while (p < __setup_end);
218 return had_early_param;
222 * This should be approx 2 Bo*oMips to start (note initial shift), and will
223 * still work even if initially too large, it will just take slightly longer
225 unsigned long loops_per_jiffy = (1<<12);
227 EXPORT_SYMBOL(loops_per_jiffy);
229 static int __init debug_kernel(char *str)
231 console_loglevel = 10;
235 static int __init quiet_kernel(char *str)
237 console_loglevel = 4;
241 early_param("debug", debug_kernel);
242 early_param("quiet", quiet_kernel);
244 static int __init loglevel(char *str)
246 get_option(&str, &console_loglevel);
250 early_param("loglevel", loglevel);
253 * Unknown boot options get handed to init, unless they look like
256 static int __init unknown_bootoption(char *param, char *val)
258 /* Change NUL term back to "=", to make "param" the whole string. */
260 /* param=val or param="val"? */
261 if (val == param+strlen(param)+1)
263 else if (val == param+strlen(param)+2) {
265 memmove(val-1, val, strlen(val)+1);
271 /* Handle obsolete-style parameters */
272 if (obsolete_checksetup(param))
276 * Preemptive maintenance for "why didn't my misspelled command
279 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
280 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
288 /* Environment option */
290 for (i = 0; envp_init[i]; i++) {
291 if (i == MAX_INIT_ENVS) {
292 panic_later = "Too many boot env vars at `%s'";
295 if (!strncmp(param, envp_init[i], val - param))
298 envp_init[i] = param;
300 /* Command line option */
302 for (i = 0; argv_init[i]; i++) {
303 if (i == MAX_INIT_ARGS) {
304 panic_later = "Too many boot init vars at `%s'";
308 argv_init[i] = param;
313 #ifdef CONFIG_DEBUG_PAGEALLOC
314 int __read_mostly debug_pagealloc_enabled = 0;
317 static int __init init_setup(char *str)
321 execute_command = str;
323 * In case LILO is going to boot us with default command line,
324 * it prepends "auto" before the whole cmdline which makes
325 * the shell think it should execute a script with such name.
326 * So we ignore all arguments entered _before_ init=... [MJ]
328 for (i = 1; i < MAX_INIT_ARGS; i++)
332 __setup("init=", init_setup);
334 static int __init rdinit_setup(char *str)
338 ramdisk_execute_command = str;
339 /* See "auto" comment in init_setup */
340 for (i = 1; i < MAX_INIT_ARGS; i++)
344 __setup("rdinit=", rdinit_setup);
348 #ifdef CONFIG_X86_LOCAL_APIC
349 static void __init smp_init(void)
351 APIC_init_uniprocessor();
354 #define smp_init() do { } while (0)
357 static inline void setup_nr_cpu_ids(void) { }
358 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
362 #if NR_CPUS > BITS_PER_LONG
363 cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
364 EXPORT_SYMBOL(cpu_mask_all);
367 /* Setup number of possible processor ids */
368 int nr_cpu_ids __read_mostly = NR_CPUS;
369 EXPORT_SYMBOL(nr_cpu_ids);
371 /* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
372 static void __init setup_nr_cpu_ids(void)
374 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
377 /* Called by boot processor to activate the rest. */
378 static void __init smp_init(void)
383 * Set up the current CPU as possible to migrate to.
384 * The other ones will be done by cpu_up/cpu_down()
386 set_cpu_active(smp_processor_id(), true);
388 /* FIXME: This should be done in userspace --RR */
389 for_each_present_cpu(cpu) {
390 if (num_online_cpus() >= setup_max_cpus)
392 if (!cpu_online(cpu))
396 /* Any cleanup work */
397 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
398 smp_cpus_done(setup_max_cpus);
404 * We need to store the untouched command line for future reference.
405 * We also need to store the touched command line since the parameter
406 * parsing is performed in place, and we should allow a component to
407 * store reference of name/value for future reference.
409 static void __init setup_command_line(char *command_line)
411 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
412 static_command_line = alloc_bootmem(strlen (command_line)+1);
413 strcpy (saved_command_line, boot_command_line);
414 strcpy (static_command_line, command_line);
418 * We need to finalize in a non-__init function or else race conditions
419 * between the root thread and the init thread may cause start_kernel to
420 * be reaped by free_initmem before the root thread has proceeded to
423 * gcc-3.4 accidentally inlines this function, so use noinline.
426 static noinline void __init_refok rest_init(void)
427 __releases(kernel_lock)
431 rcu_scheduler_starting();
432 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
433 numa_default_policy();
434 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
435 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
439 * The boot idle thread must execute schedule()
440 * at least once to get things moving:
442 init_idle_bootup_task(current);
443 preempt_enable_no_resched();
447 /* Call into cpu_idle with preempt disabled */
451 /* Check for early params. */
452 static int __init do_early_param(char *param, char *val)
454 struct obs_kernel_param *p;
456 for (p = __setup_start; p < __setup_end; p++) {
457 if ((p->early && strcmp(param, p->str) == 0) ||
458 (strcmp(param, "console") == 0 &&
459 strcmp(p->str, "earlycon") == 0)
461 if (p->setup_func(val) != 0)
463 "Malformed early option '%s'\n", param);
466 /* We accept everything at this stage. */
470 void __init parse_early_options(char *cmdline)
472 parse_args("early options", cmdline, NULL, 0, do_early_param);
475 /* Arch code calls this early on, or if not, just before other parsing. */
476 void __init parse_early_param(void)
478 static __initdata int done = 0;
479 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
484 /* All fall through to do_early_param. */
485 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
486 parse_early_options(tmp_cmdline);
491 * Activate the first processor.
494 static void __init boot_cpu_init(void)
496 int cpu = smp_processor_id();
497 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
498 set_cpu_online(cpu, true);
499 set_cpu_present(cpu, true);
500 set_cpu_possible(cpu, true);
503 void __init __weak smp_setup_processor_id(void)
507 void __init __weak thread_info_cache_init(void)
512 * Set up kernel memory allocators
514 static void __init mm_init(void)
517 * page_cgroup requires countinous pages as memmap
518 * and it's bigger than MAX_ORDER unless SPARSEMEM.
520 page_cgroup_init_flatmem();
523 pgtable_cache_init();
527 asmlinkage void __init start_kernel(void)
530 extern struct kernel_param __start___param[], __stop___param[];
532 smp_setup_processor_id();
535 * Need to run as early as possible, to initialize the
539 debug_objects_early_init();
542 * Set up the the initial canary ASAP:
544 boot_init_stack_canary();
549 early_boot_irqs_off();
550 early_init_irq_lock_class();
553 * Interrupts are still disabled. Do necessary setups, then
560 printk(KERN_NOTICE "%s", linux_banner);
561 setup_arch(&command_line);
562 mm_init_owner(&init_mm, &init_task);
563 setup_command_line(command_line);
565 setup_per_cpu_areas();
566 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
568 build_all_zonelists();
571 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
573 parse_args("Booting kernel", static_command_line, __start___param,
574 __stop___param - __start___param,
575 &unknown_bootoption);
577 * These use large bootmem allocations and must precede
581 vfs_caches_init_early();
586 * Set up the scheduler prior starting any interrupts (such as the
587 * timer interrupt). Full topology setup happens at smp_init()
588 * time - but meanwhile we still have a functioning scheduler.
592 * Disable preemption - early bootup scheduling is extremely
593 * fragile until we cpu_idle() for the first time.
596 if (!irqs_disabled()) {
597 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
598 "enabled *very* early, fixing it\n");
602 /* init some links before init_ISA_irqs() */
612 if (!irqs_disabled())
613 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
615 early_boot_irqs_on();
618 /* Interrupts are enabled now so all GFP allocations are safe. */
619 set_gfp_allowed_mask(__GFP_BITS_MASK);
621 kmem_cache_init_late();
624 * HACK ALERT! This is early. We're enabling the console before
625 * we've done PCI setups etc, and console_init() must be aware of
626 * this. But we do want output early, in case something goes wrong.
630 panic(panic_later, panic_param);
635 * Need to run this when irqs are enabled, because it wants
636 * to self-test [hard/soft]-irqs on/off lock inversion bugs
641 #ifdef CONFIG_BLK_DEV_INITRD
642 if (initrd_start && !initrd_below_start_ok &&
643 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
644 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
646 page_to_pfn(virt_to_page((void *)initrd_start)),
652 enable_debug_pagealloc();
655 debug_objects_mem_init();
657 setup_per_cpu_pageset();
667 efi_enter_virtual_mode();
669 thread_info_cache_init();
671 fork_init(totalram_pages);
676 vfs_caches_init(totalram_pages);
679 /* rootfs populating might need page-writeback */
680 page_writeback_init();
681 #ifdef CONFIG_PROC_FS
686 taskstats_init_early();
691 acpi_early_init(); /* before LAPIC and SMP init */
695 /* Do the rest non-__init'ed, we're now alive */
699 /* Call all constructor functions linked into the kernel. */
700 static void __init do_ctors(void)
702 #ifdef CONFIG_CONSTRUCTORS
703 ctor_fn_t *call = (ctor_fn_t *) __ctors_start;
705 for (; call < (ctor_fn_t *) __ctors_end; call++)
711 core_param(initcall_debug, initcall_debug, bool, 0644);
713 static char msgbuf[64];
714 static struct boot_trace_call call;
715 static struct boot_trace_ret ret;
717 int do_one_initcall(initcall_t fn)
719 int count = preempt_count();
720 ktime_t calltime, delta, rettime;
722 if (initcall_debug) {
723 call.caller = task_pid_nr(current);
724 printk("calling %pF @ %i\n", fn, call.caller);
725 calltime = ktime_get();
726 trace_boot_call(&call, fn);
732 if (initcall_debug) {
733 disable_boot_trace();
734 rettime = ktime_get();
735 delta = ktime_sub(rettime, calltime);
736 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
737 trace_boot_ret(&ret, fn);
738 printk("initcall %pF returned %d after %Ld usecs\n", fn,
739 ret.result, ret.duration);
744 if (ret.result && ret.result != -ENODEV && initcall_debug)
745 sprintf(msgbuf, "error code %d ", ret.result);
747 if (preempt_count() != count) {
748 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
749 preempt_count() = count;
751 if (irqs_disabled()) {
752 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
756 printk("initcall %pF returned with %s\n", fn, msgbuf);
763 extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
765 static void __init do_initcalls(void)
769 for (call = __early_initcall_end; call < __initcall_end; call++)
770 do_one_initcall(*call);
772 /* Make sure there is no pending stuff from the initcall sequence */
773 flush_scheduled_work();
777 * Ok, the machine is now initialized. None of the devices
778 * have been touched yet, but the CPU subsystem is up and
779 * running, and memory and process management works.
781 * Now we can finally start doing some real work..
783 static void __init do_basic_setup(void)
785 rcu_init_sched(); /* needed by module_init stage. */
788 usermodehelper_init();
796 static void __init do_pre_smp_initcalls(void)
800 for (call = __initcall_start; call < __early_initcall_end; call++)
801 do_one_initcall(*call);
804 static void run_init_process(char *init_filename)
806 argv_init[0] = init_filename;
807 kernel_execve(init_filename, argv_init, envp_init);
810 /* This is a non __init function. Force it to be noinline otherwise gcc
811 * makes it inline to init() and it becomes part of init.text section
813 static noinline int init_post(void)
814 __releases(kernel_lock)
816 /* need to finish all async __init code before freeing the memory */
817 async_synchronize_full();
821 system_state = SYSTEM_RUNNING;
822 numa_default_policy();
824 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
825 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
830 current->signal->flags |= SIGNAL_UNKILLABLE;
832 if (ramdisk_execute_command) {
833 run_init_process(ramdisk_execute_command);
834 printk(KERN_WARNING "Failed to execute %s\n",
835 ramdisk_execute_command);
839 * We try each of these until one succeeds.
841 * The Bourne shell can be used instead of init if we are
842 * trying to recover a really broken machine.
844 if (execute_command) {
845 run_init_process(execute_command);
846 printk(KERN_WARNING "Failed to execute %s. Attempting "
847 "defaults...\n", execute_command);
849 run_init_process("/sbin/init");
850 run_init_process("/etc/init");
851 run_init_process("/bin/init");
852 run_init_process("/bin/sh");
854 panic("No init found. Try passing init= option to kernel.");
857 static int __init kernel_init(void * unused)
862 * init can allocate pages on any node
864 set_mems_allowed(node_possible_map);
866 * init can run on any cpu.
868 set_cpus_allowed_ptr(current, cpu_all_mask);
870 * Tell the world that we're going to be the grim
871 * reaper of innocent orphaned children.
873 * We don't want people to have to make incorrect
874 * assumptions about where in the task array this
877 init_pid_ns.child_reaper = current;
879 cad_pid = task_pid(current);
881 smp_prepare_cpus(setup_max_cpus);
883 do_pre_smp_initcalls();
892 * check if there is an early userspace init. If yes, let it do all
896 if (!ramdisk_execute_command)
897 ramdisk_execute_command = "/init";
899 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
900 ramdisk_execute_command = NULL;
905 * Ok, we have completed the initial bootup, and
906 * we're essentially up and running. Get rid of the
907 * initmem segments and start the user-mode stuff..