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
117 extern void time_init(void);
118 /* Default late time init is NULL. archs can override this later. */
119 void (*__initdata late_time_init)(void);
120 extern void softirq_init(void);
122 /* Untouched command line saved by arch-specific code. */
123 char __initdata boot_command_line[COMMAND_LINE_SIZE];
124 /* Untouched saved command line (eg. for /proc) */
125 char *saved_command_line;
126 /* Command line for parameter parsing */
127 static char *static_command_line;
129 static char *execute_command;
130 static char *ramdisk_execute_command;
133 * If set, this is an indication to the drivers that reset the underlying
134 * device before going ahead with the initialization otherwise driver might
135 * rely on the BIOS and skip the reset operation.
137 * This is useful if kernel is booting in an unreliable environment.
138 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
139 * skipped and devices will be in unknown state.
141 unsigned int reset_devices;
142 EXPORT_SYMBOL(reset_devices);
144 static int __init set_reset_devices(char *str)
150 __setup("reset_devices", set_reset_devices);
152 static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
153 const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
154 static const char *panic_later, *panic_param;
156 extern const struct obs_kernel_param __setup_start[], __setup_end[];
158 static int __init obsolete_checksetup(char *line)
160 const struct obs_kernel_param *p;
161 int had_early_param = 0;
165 int n = strlen(p->str);
166 if (parameqn(line, p->str, n)) {
168 /* Already done in parse_early_param?
169 * (Needs exact match on param part).
170 * Keep iterating, as we can have early
171 * params and __setups of same names 8( */
172 if (line[n] == '\0' || line[n] == '=')
174 } else if (!p->setup_func) {
175 printk(KERN_WARNING "Parameter %s is obsolete,"
176 " ignored\n", p->str);
178 } else if (p->setup_func(line + n))
182 } while (p < __setup_end);
184 return had_early_param;
188 * This should be approx 2 Bo*oMips to start (note initial shift), and will
189 * still work even if initially too large, it will just take slightly longer
191 unsigned long loops_per_jiffy = (1<<12);
193 EXPORT_SYMBOL(loops_per_jiffy);
195 static int __init debug_kernel(char *str)
197 console_loglevel = 10;
201 static int __init quiet_kernel(char *str)
203 console_loglevel = 4;
207 early_param("debug", debug_kernel);
208 early_param("quiet", quiet_kernel);
210 static int __init loglevel(char *str)
215 * Only update loglevel value when a correct setting was passed,
216 * to prevent blind crashes (when loglevel being set to 0) that
217 * are quite hard to debug
219 if (get_option(&str, &newlevel)) {
220 console_loglevel = newlevel;
227 early_param("loglevel", loglevel);
230 * Unknown boot options get handed to init, unless they look like
231 * unused parameters (modprobe will find them in /proc/cmdline).
233 static int __init unknown_bootoption(char *param, char *val)
235 /* Change NUL term back to "=", to make "param" the whole string. */
237 /* param=val or param="val"? */
238 if (val == param+strlen(param)+1)
240 else if (val == param+strlen(param)+2) {
242 memmove(val-1, val, strlen(val)+1);
248 /* Handle obsolete-style parameters */
249 if (obsolete_checksetup(param))
252 /* Unused module parameter. */
253 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
260 /* Environment option */
262 for (i = 0; envp_init[i]; i++) {
263 if (i == MAX_INIT_ENVS) {
264 panic_later = "Too many boot env vars at `%s'";
267 if (!strncmp(param, envp_init[i], val - param))
270 envp_init[i] = param;
272 /* Command line option */
274 for (i = 0; argv_init[i]; i++) {
275 if (i == MAX_INIT_ARGS) {
276 panic_later = "Too many boot init vars at `%s'";
280 argv_init[i] = param;
285 static int __init init_setup(char *str)
289 execute_command = str;
291 * In case LILO is going to boot us with default command line,
292 * it prepends "auto" before the whole cmdline which makes
293 * the shell think it should execute a script with such name.
294 * So we ignore all arguments entered _before_ init=... [MJ]
296 for (i = 1; i < MAX_INIT_ARGS; i++)
300 __setup("init=", init_setup);
302 static int __init rdinit_setup(char *str)
306 ramdisk_execute_command = str;
307 /* See "auto" comment in init_setup */
308 for (i = 1; i < MAX_INIT_ARGS; i++)
312 __setup("rdinit=", rdinit_setup);
315 static const unsigned int setup_max_cpus = NR_CPUS;
316 #ifdef CONFIG_X86_LOCAL_APIC
317 static void __init smp_init(void)
319 APIC_init_uniprocessor();
322 #define smp_init() do { } while (0)
325 static inline void setup_nr_cpu_ids(void) { }
326 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
330 * We need to store the untouched command line for future reference.
331 * We also need to store the touched command line since the parameter
332 * parsing is performed in place, and we should allow a component to
333 * store reference of name/value for future reference.
335 static void __init setup_command_line(char *command_line)
337 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
338 static_command_line = alloc_bootmem(strlen (command_line)+1);
339 strcpy (saved_command_line, boot_command_line);
340 strcpy (static_command_line, command_line);
344 * We need to finalize in a non-__init function or else race conditions
345 * between the root thread and the init thread may cause start_kernel to
346 * be reaped by free_initmem before the root thread has proceeded to
349 * gcc-3.4 accidentally inlines this function, so use noinline.
352 static __initdata DECLARE_COMPLETION(kthreadd_done);
354 static noinline void __init_refok rest_init(void)
358 rcu_scheduler_starting();
360 * We need to spawn init first so that it obtains pid 1, however
361 * the init task will end up wanting to create kthreads, which, if
362 * we schedule it before we create kthreadd, will OOPS.
364 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
365 numa_default_policy();
366 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
368 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
370 complete(&kthreadd_done);
373 * The boot idle thread must execute schedule()
374 * at least once to get things moving:
376 init_idle_bootup_task(current);
377 schedule_preempt_disabled();
378 /* Call into cpu_idle with preempt disabled */
382 /* Check for early params. */
383 static int __init do_early_param(char *param, char *val)
385 const struct obs_kernel_param *p;
387 for (p = __setup_start; p < __setup_end; p++) {
388 if ((p->early && parameq(param, p->str)) ||
389 (strcmp(param, "console") == 0 &&
390 strcmp(p->str, "earlycon") == 0)
392 if (p->setup_func(val) != 0)
394 "Malformed early option '%s'\n", param);
397 /* We accept everything at this stage. */
401 void __init parse_early_options(char *cmdline)
403 parse_args("early options", cmdline, NULL, 0, do_early_param);
406 /* Arch code calls this early on, or if not, just before other parsing. */
407 void __init parse_early_param(void)
409 static __initdata int done = 0;
410 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
415 /* All fall through to do_early_param. */
416 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
417 parse_early_options(tmp_cmdline);
422 * Activate the first processor.
425 static void __init boot_cpu_init(void)
427 int cpu = smp_processor_id();
428 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
429 set_cpu_online(cpu, true);
430 set_cpu_active(cpu, true);
431 set_cpu_present(cpu, true);
432 set_cpu_possible(cpu, true);
435 void __init __weak smp_setup_processor_id(void)
439 void __init __weak thread_info_cache_init(void)
444 * Set up kernel memory allocators
446 static void __init mm_init(void)
449 * page_cgroup requires contiguous pages,
450 * bigger than MAX_ORDER unless SPARSEMEM.
452 page_cgroup_init_flatmem();
456 pgtable_cache_init();
460 asmlinkage void __init start_kernel(void)
463 extern const struct kernel_param __start___param[], __stop___param[];
466 * Need to run as early as possible, to initialize the
470 smp_setup_processor_id();
471 debug_objects_early_init();
474 * Set up the the initial canary ASAP:
476 boot_init_stack_canary();
481 early_boot_irqs_disabled = true;
484 * Interrupts are still disabled. Do necessary setups, then
490 printk(KERN_NOTICE "%s", linux_banner);
491 setup_arch(&command_line);
492 mm_init_owner(&init_mm, &init_task);
493 mm_init_cpumask(&init_mm);
494 setup_command_line(command_line);
496 setup_per_cpu_areas();
497 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
499 build_all_zonelists(NULL);
502 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
504 parse_args("Booting kernel", static_command_line, __start___param,
505 __stop___param - __start___param,
506 &unknown_bootoption);
511 * These use large bootmem allocations and must precede
516 vfs_caches_init_early();
522 * Set up the scheduler prior starting any interrupts (such as the
523 * timer interrupt). Full topology setup happens at smp_init()
524 * time - but meanwhile we still have a functioning scheduler.
528 * Disable preemption - early bootup scheduling is extremely
529 * fragile until we cpu_idle() for the first time.
532 if (!irqs_disabled()) {
533 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
534 "enabled *very* early, fixing it\n");
541 /* init some links before init_ISA_irqs() */
551 call_function_init();
552 if (!irqs_disabled())
553 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
555 early_boot_irqs_disabled = false;
558 /* Interrupts are enabled now so all GFP allocations are safe. */
559 gfp_allowed_mask = __GFP_BITS_MASK;
561 kmem_cache_init_late();
564 * HACK ALERT! This is early. We're enabling the console before
565 * we've done PCI setups etc, and console_init() must be aware of
566 * this. But we do want output early, in case something goes wrong.
570 panic(panic_later, panic_param);
575 * Need to run this when irqs are enabled, because it wants
576 * to self-test [hard/soft]-irqs on/off lock inversion bugs
581 #ifdef CONFIG_BLK_DEV_INITRD
582 if (initrd_start && !initrd_below_start_ok &&
583 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
584 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
586 page_to_pfn(virt_to_page((void *)initrd_start)),
592 debug_objects_mem_init();
594 setup_per_cpu_pageset();
604 efi_enter_virtual_mode();
606 thread_info_cache_init();
608 fork_init(totalram_pages);
614 vfs_caches_init(totalram_pages);
616 /* rootfs populating might need page-writeback */
617 page_writeback_init();
618 #ifdef CONFIG_PROC_FS
623 taskstats_init_early();
628 acpi_early_init(); /* before LAPIC and SMP init */
633 /* Do the rest non-__init'ed, we're now alive */
637 /* Call all constructor functions linked into the kernel. */
638 static void __init do_ctors(void)
640 #ifdef CONFIG_CONSTRUCTORS
641 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
643 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
649 core_param(initcall_debug, initcall_debug, bool, 0644);
651 static char msgbuf[64];
653 static int __init_or_module do_one_initcall_debug(initcall_t fn)
655 ktime_t calltime, delta, rettime;
656 unsigned long long duration;
659 printk(KERN_DEBUG "calling %pF @ %i\n", fn, task_pid_nr(current));
660 calltime = ktime_get();
662 rettime = ktime_get();
663 delta = ktime_sub(rettime, calltime);
664 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
665 printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n", fn,
671 int __init_or_module do_one_initcall(initcall_t fn)
673 int count = preempt_count();
677 ret = do_one_initcall_debug(fn);
683 if (ret && ret != -ENODEV && initcall_debug)
684 sprintf(msgbuf, "error code %d ", ret);
686 if (preempt_count() != count) {
687 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
688 preempt_count() = count;
690 if (irqs_disabled()) {
691 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
695 printk("initcall %pF returned with %s\n", fn, msgbuf);
702 extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
704 static void __init do_initcalls(void)
708 for (fn = __early_initcall_end; fn < __initcall_end; fn++)
709 do_one_initcall(*fn);
713 * Ok, the machine is now initialized. None of the devices
714 * have been touched yet, but the CPU subsystem is up and
715 * running, and memory and process management works.
717 * Now we can finally start doing some real work..
719 static void __init do_basic_setup(void)
722 usermodehelper_init();
727 usermodehelper_enable();
731 static void __init do_pre_smp_initcalls(void)
735 for (fn = __initcall_start; fn < __early_initcall_end; fn++)
736 do_one_initcall(*fn);
739 static void run_init_process(const char *init_filename)
741 argv_init[0] = init_filename;
742 kernel_execve(init_filename, argv_init, envp_init);
745 /* This is a non __init function. Force it to be noinline otherwise gcc
746 * makes it inline to init() and it becomes part of init.text section
748 static noinline int init_post(void)
750 /* need to finish all async __init code before freeing the memory */
751 async_synchronize_full();
754 system_state = SYSTEM_RUNNING;
755 numa_default_policy();
758 current->signal->flags |= SIGNAL_UNKILLABLE;
760 if (ramdisk_execute_command) {
761 run_init_process(ramdisk_execute_command);
762 printk(KERN_WARNING "Failed to execute %s\n",
763 ramdisk_execute_command);
767 * We try each of these until one succeeds.
769 * The Bourne shell can be used instead of init if we are
770 * trying to recover a really broken machine.
772 if (execute_command) {
773 run_init_process(execute_command);
774 printk(KERN_WARNING "Failed to execute %s. Attempting "
775 "defaults...\n", execute_command);
777 run_init_process("/sbin/init");
778 run_init_process("/etc/init");
779 run_init_process("/bin/init");
780 run_init_process("/bin/sh");
782 panic("No init found. Try passing init= option to kernel. "
783 "See Linux Documentation/init.txt for guidance.");
786 static int __init kernel_init(void * unused)
789 * Wait until kthreadd is all set-up.
791 wait_for_completion(&kthreadd_done);
793 * init can allocate pages on any node
795 set_mems_allowed(node_states[N_HIGH_MEMORY]);
797 * init can run on any cpu.
799 set_cpus_allowed_ptr(current, cpu_all_mask);
801 cad_pid = task_pid(current);
803 smp_prepare_cpus(setup_max_cpus);
805 do_pre_smp_initcalls();
806 lockup_detector_init();
813 /* Open the /dev/console on the rootfs, this should never fail */
814 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
815 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
820 * check if there is an early userspace init. If yes, let it do all
824 if (!ramdisk_execute_command)
825 ramdisk_execute_command = "/init";
827 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
828 ramdisk_execute_command = NULL;
833 * Ok, we have completed the initial bootup, and
834 * we're essentially up and running. Get rid of the
835 * initmem segments and start the user-mode stuff..