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/smp_lock.h>
24 #include <linux/initrd.h>
25 #include <linux/bootmem.h>
26 #include <linux/acpi.h>
27 #include <linux/tty.h>
28 #include <linux/percpu.h>
29 #include <linux/kmod.h>
30 #include <linux/vmalloc.h>
31 #include <linux/kernel_stat.h>
32 #include <linux/start_kernel.h>
33 #include <linux/security.h>
34 #include <linux/smp.h>
35 #include <linux/workqueue.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_cgroup.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/kmemtrace.h>
70 #include <linux/sfi.h>
71 #include <linux/shmem_fs.h>
72 #include <linux/slab.h>
73 #include <trace/boot.h>
77 #include <asm/setup.h>
78 #include <asm/sections.h>
79 #include <asm/cacheflush.h>
81 #ifdef CONFIG_X86_LOCAL_APIC
85 static int kernel_init(void *);
87 extern void init_IRQ(void);
88 extern void fork_init(unsigned long);
89 extern void mca_init(void);
90 extern void sbus_init(void);
91 extern void prio_tree_init(void);
92 extern void radix_tree_init(void);
93 extern void free_initmem(void);
94 #ifndef CONFIG_DEBUG_RODATA
95 static inline void mark_rodata_ro(void) { }
99 extern void tc_init(void);
102 enum system_states system_state __read_mostly;
103 EXPORT_SYMBOL(system_state);
106 * Boot command-line arguments
108 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
109 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
111 extern void time_init(void);
112 /* Default late time init is NULL. archs can override this later. */
113 void (*__initdata late_time_init)(void);
114 extern void softirq_init(void);
116 /* Untouched command line saved by arch-specific code. */
117 char __initdata boot_command_line[COMMAND_LINE_SIZE];
118 /* Untouched saved command line (eg. for /proc) */
119 char *saved_command_line;
120 /* Command line for parameter parsing */
121 static char *static_command_line;
123 static char *execute_command;
124 static char *ramdisk_execute_command;
127 /* Setup configured maximum number of CPUs to activate */
128 unsigned int __initdata setup_max_cpus = NR_CPUS;
131 * Setup routine for controlling SMP activation
133 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
134 * activation entirely (the MPS table probe still happens, though).
136 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
137 * greater than 0, limits the maximum number of CPUs activated in
141 void __weak arch_disable_smp_support(void) { }
143 static int __init nosmp(char *str)
146 arch_disable_smp_support();
151 early_param("nosmp", nosmp);
153 /* this is hard limit */
154 static int __init nrcpus(char *str)
158 get_option(&str, &nr_cpus);
159 if (nr_cpus > 0 && nr_cpus < nr_cpu_ids)
160 nr_cpu_ids = nr_cpus;
165 early_param("nr_cpus", nrcpus);
167 static int __init maxcpus(char *str)
169 get_option(&str, &setup_max_cpus);
170 if (setup_max_cpus == 0)
171 arch_disable_smp_support();
176 early_param("maxcpus", maxcpus);
178 static const unsigned int setup_max_cpus = NR_CPUS;
182 * If set, this is an indication to the drivers that reset the underlying
183 * device before going ahead with the initialization otherwise driver might
184 * rely on the BIOS and skip the reset operation.
186 * This is useful if kernel is booting in an unreliable environment.
187 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
188 * skipped and devices will be in unknown state.
190 unsigned int reset_devices;
191 EXPORT_SYMBOL(reset_devices);
193 static int __init set_reset_devices(char *str)
199 __setup("reset_devices", set_reset_devices);
201 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
202 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
203 static const char *panic_later, *panic_param;
205 extern struct obs_kernel_param __setup_start[], __setup_end[];
207 static int __init obsolete_checksetup(char *line)
209 struct obs_kernel_param *p;
210 int had_early_param = 0;
214 int n = strlen(p->str);
215 if (!strncmp(line, p->str, n)) {
217 /* Already done in parse_early_param?
218 * (Needs exact match on param part).
219 * Keep iterating, as we can have early
220 * params and __setups of same names 8( */
221 if (line[n] == '\0' || line[n] == '=')
223 } else if (!p->setup_func) {
224 printk(KERN_WARNING "Parameter %s is obsolete,"
225 " ignored\n", p->str);
227 } else if (p->setup_func(line + n))
231 } while (p < __setup_end);
233 return had_early_param;
237 * This should be approx 2 Bo*oMips to start (note initial shift), and will
238 * still work even if initially too large, it will just take slightly longer
240 unsigned long loops_per_jiffy = (1<<12);
242 EXPORT_SYMBOL(loops_per_jiffy);
244 static int __init debug_kernel(char *str)
246 console_loglevel = 10;
250 static int __init quiet_kernel(char *str)
252 console_loglevel = 4;
256 early_param("debug", debug_kernel);
257 early_param("quiet", quiet_kernel);
259 static int __init loglevel(char *str)
261 get_option(&str, &console_loglevel);
265 early_param("loglevel", loglevel);
268 * Unknown boot options get handed to init, unless they look like
269 * unused parameters (modprobe will find them in /proc/cmdline).
271 static int __init unknown_bootoption(char *param, char *val)
273 /* Change NUL term back to "=", to make "param" the whole string. */
275 /* param=val or param="val"? */
276 if (val == param+strlen(param)+1)
278 else if (val == param+strlen(param)+2) {
280 memmove(val-1, val, strlen(val)+1);
286 /* Handle obsolete-style parameters */
287 if (obsolete_checksetup(param))
290 /* Unused module parameter. */
291 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
298 /* Environment option */
300 for (i = 0; envp_init[i]; i++) {
301 if (i == MAX_INIT_ENVS) {
302 panic_later = "Too many boot env vars at `%s'";
305 if (!strncmp(param, envp_init[i], val - param))
308 envp_init[i] = param;
310 /* Command line option */
312 for (i = 0; argv_init[i]; i++) {
313 if (i == MAX_INIT_ARGS) {
314 panic_later = "Too many boot init vars at `%s'";
318 argv_init[i] = param;
323 #ifdef CONFIG_DEBUG_PAGEALLOC
324 int __read_mostly debug_pagealloc_enabled = 0;
327 static int __init init_setup(char *str)
331 execute_command = str;
333 * In case LILO is going to boot us with default command line,
334 * it prepends "auto" before the whole cmdline which makes
335 * the shell think it should execute a script with such name.
336 * So we ignore all arguments entered _before_ init=... [MJ]
338 for (i = 1; i < MAX_INIT_ARGS; i++)
342 __setup("init=", init_setup);
344 static int __init rdinit_setup(char *str)
348 ramdisk_execute_command = str;
349 /* See "auto" comment in init_setup */
350 for (i = 1; i < MAX_INIT_ARGS; i++)
354 __setup("rdinit=", rdinit_setup);
358 #ifdef CONFIG_X86_LOCAL_APIC
359 static void __init smp_init(void)
361 APIC_init_uniprocessor();
364 #define smp_init() do { } while (0)
367 static inline void setup_nr_cpu_ids(void) { }
368 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
372 /* Setup number of possible processor ids */
373 int nr_cpu_ids __read_mostly = NR_CPUS;
374 EXPORT_SYMBOL(nr_cpu_ids);
376 /* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
377 static void __init setup_nr_cpu_ids(void)
379 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
382 /* Called by boot processor to activate the rest. */
383 static void __init smp_init(void)
387 /* FIXME: This should be done in userspace --RR */
388 for_each_present_cpu(cpu) {
389 if (num_online_cpus() >= setup_max_cpus)
391 if (!cpu_online(cpu))
395 /* Any cleanup work */
396 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
397 smp_cpus_done(setup_max_cpus);
403 * We need to store the untouched command line for future reference.
404 * We also need to store the touched command line since the parameter
405 * parsing is performed in place, and we should allow a component to
406 * store reference of name/value for future reference.
408 static void __init setup_command_line(char *command_line)
410 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
411 static_command_line = alloc_bootmem(strlen (command_line)+1);
412 strcpy (saved_command_line, boot_command_line);
413 strcpy (static_command_line, command_line);
417 * We need to finalize in a non-__init function or else race conditions
418 * between the root thread and the init thread may cause start_kernel to
419 * be reaped by free_initmem before the root thread has proceeded to
422 * gcc-3.4 accidentally inlines this function, so use noinline.
425 static __initdata DECLARE_COMPLETION(kthreadd_done);
427 static noinline void __init_refok rest_init(void)
428 __releases(kernel_lock)
432 rcu_scheduler_starting();
434 * We need to spawn init first so that it obtains pid-1, however
435 * the init task will end up wanting to create kthreads, which, if
436 * we schedule it before we create kthreadd, will OOPS.
438 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
439 numa_default_policy();
440 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
442 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
444 complete(&kthreadd_done);
448 * The boot idle thread must execute schedule()
449 * at least once to get things moving:
451 init_idle_bootup_task(current);
452 preempt_enable_no_resched();
456 /* Call into cpu_idle with preempt disabled */
460 /* Check for early params. */
461 static int __init do_early_param(char *param, char *val)
463 struct obs_kernel_param *p;
465 for (p = __setup_start; p < __setup_end; p++) {
466 if ((p->early && strcmp(param, p->str) == 0) ||
467 (strcmp(param, "console") == 0 &&
468 strcmp(p->str, "earlycon") == 0)
470 if (p->setup_func(val) != 0)
472 "Malformed early option '%s'\n", param);
475 /* We accept everything at this stage. */
479 void __init parse_early_options(char *cmdline)
481 parse_args("early options", cmdline, NULL, 0, do_early_param);
484 /* Arch code calls this early on, or if not, just before other parsing. */
485 void __init parse_early_param(void)
487 static __initdata int done = 0;
488 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
493 /* All fall through to do_early_param. */
494 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
495 parse_early_options(tmp_cmdline);
500 * Activate the first processor.
503 static void __init boot_cpu_init(void)
505 int cpu = smp_processor_id();
506 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
507 set_cpu_online(cpu, true);
508 set_cpu_active(cpu, true);
509 set_cpu_present(cpu, true);
510 set_cpu_possible(cpu, true);
513 void __init __weak smp_setup_processor_id(void)
517 void __init __weak thread_info_cache_init(void)
522 * Set up kernel memory allocators
524 static void __init mm_init(void)
527 * page_cgroup requires countinous pages as memmap
528 * and it's bigger than MAX_ORDER unless SPARSEMEM.
530 page_cgroup_init_flatmem();
533 pgtable_cache_init();
537 asmlinkage void __init start_kernel(void)
540 extern struct kernel_param __start___param[], __stop___param[];
542 smp_setup_processor_id();
545 * Need to run as early as possible, to initialize the
549 debug_objects_early_init();
552 * Set up the the initial canary ASAP:
554 boot_init_stack_canary();
559 early_boot_irqs_off();
560 early_init_irq_lock_class();
563 * Interrupts are still disabled. Do necessary setups, then
570 printk(KERN_NOTICE "%s", linux_banner);
571 setup_arch(&command_line);
572 mm_init_owner(&init_mm, &init_task);
573 setup_command_line(command_line);
575 setup_per_cpu_areas();
576 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
578 build_all_zonelists(NULL);
581 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
583 parse_args("Booting kernel", static_command_line, __start___param,
584 __stop___param - __start___param,
585 &unknown_bootoption);
587 * These use large bootmem allocations and must precede
591 vfs_caches_init_early();
596 * Set up the scheduler prior starting any interrupts (such as the
597 * timer interrupt). Full topology setup happens at smp_init()
598 * time - but meanwhile we still have a functioning scheduler.
602 * Disable preemption - early bootup scheduling is extremely
603 * fragile until we cpu_idle() for the first time.
606 if (!irqs_disabled()) {
607 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
608 "enabled *very* early, fixing it\n");
613 /* init some links before init_ISA_irqs() */
623 if (!irqs_disabled())
624 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
626 early_boot_irqs_on();
629 /* Interrupts are enabled now so all GFP allocations are safe. */
630 gfp_allowed_mask = __GFP_BITS_MASK;
632 kmem_cache_init_late();
635 * HACK ALERT! This is early. We're enabling the console before
636 * we've done PCI setups etc, and console_init() must be aware of
637 * this. But we do want output early, in case something goes wrong.
641 panic(panic_later, panic_param);
646 * Need to run this when irqs are enabled, because it wants
647 * to self-test [hard/soft]-irqs on/off lock inversion bugs
652 #ifdef CONFIG_BLK_DEV_INITRD
653 if (initrd_start && !initrd_below_start_ok &&
654 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
655 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
657 page_to_pfn(virt_to_page((void *)initrd_start)),
663 enable_debug_pagealloc();
666 debug_objects_mem_init();
668 setup_per_cpu_pageset();
678 efi_enter_virtual_mode();
680 thread_info_cache_init();
682 fork_init(totalram_pages);
688 vfs_caches_init(totalram_pages);
690 /* rootfs populating might need page-writeback */
691 page_writeback_init();
692 #ifdef CONFIG_PROC_FS
697 taskstats_init_early();
702 acpi_early_init(); /* before LAPIC and SMP init */
707 /* Do the rest non-__init'ed, we're now alive */
711 /* Call all constructor functions linked into the kernel. */
712 static void __init do_ctors(void)
714 #ifdef CONFIG_CONSTRUCTORS
715 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
717 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
723 core_param(initcall_debug, initcall_debug, bool, 0644);
725 static char msgbuf[64];
726 static struct boot_trace_call call;
727 static struct boot_trace_ret ret;
729 int do_one_initcall(initcall_t fn)
731 int count = preempt_count();
732 ktime_t calltime, delta, rettime;
734 if (initcall_debug) {
735 call.caller = task_pid_nr(current);
736 printk("calling %pF @ %i\n", fn, call.caller);
737 calltime = ktime_get();
738 trace_boot_call(&call, fn);
744 if (initcall_debug) {
745 disable_boot_trace();
746 rettime = ktime_get();
747 delta = ktime_sub(rettime, calltime);
748 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
749 trace_boot_ret(&ret, fn);
750 printk("initcall %pF returned %d after %Ld usecs\n", fn,
751 ret.result, ret.duration);
756 if (ret.result && ret.result != -ENODEV && initcall_debug)
757 sprintf(msgbuf, "error code %d ", ret.result);
759 if (preempt_count() != count) {
760 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
761 preempt_count() = count;
763 if (irqs_disabled()) {
764 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
768 printk("initcall %pF returned with %s\n", fn, msgbuf);
775 extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
777 static void __init do_initcalls(void)
781 for (fn = __early_initcall_end; fn < __initcall_end; fn++)
782 do_one_initcall(*fn);
784 /* Make sure there is no pending stuff from the initcall sequence */
785 flush_scheduled_work();
789 * Ok, the machine is now initialized. None of the devices
790 * have been touched yet, but the CPU subsystem is up and
791 * running, and memory and process management works.
793 * Now we can finally start doing some real work..
795 static void __init do_basic_setup(void)
799 usermodehelper_init();
807 static void __init do_pre_smp_initcalls(void)
811 for (fn = __initcall_start; fn < __early_initcall_end; fn++)
812 do_one_initcall(*fn);
815 static void run_init_process(char *init_filename)
817 argv_init[0] = init_filename;
818 kernel_execve(init_filename, argv_init, envp_init);
821 /* This is a non __init function. Force it to be noinline otherwise gcc
822 * makes it inline to init() and it becomes part of init.text section
824 static noinline int init_post(void)
825 __releases(kernel_lock)
827 /* need to finish all async __init code before freeing the memory */
828 async_synchronize_full();
832 system_state = SYSTEM_RUNNING;
833 numa_default_policy();
836 current->signal->flags |= SIGNAL_UNKILLABLE;
838 if (ramdisk_execute_command) {
839 run_init_process(ramdisk_execute_command);
840 printk(KERN_WARNING "Failed to execute %s\n",
841 ramdisk_execute_command);
845 * We try each of these until one succeeds.
847 * The Bourne shell can be used instead of init if we are
848 * trying to recover a really broken machine.
850 if (execute_command) {
851 run_init_process(execute_command);
852 printk(KERN_WARNING "Failed to execute %s. Attempting "
853 "defaults...\n", execute_command);
855 run_init_process("/sbin/init");
856 run_init_process("/etc/init");
857 run_init_process("/bin/init");
858 run_init_process("/bin/sh");
860 panic("No init found. Try passing init= option to kernel. "
861 "See Linux Documentation/init.txt for guidance.");
864 static int __init kernel_init(void * unused)
867 * Wait until kthreadd is all set-up.
869 wait_for_completion(&kthreadd_done);
873 * init can allocate pages on any node
875 set_mems_allowed(node_states[N_HIGH_MEMORY]);
877 * init can run on any cpu.
879 set_cpus_allowed_ptr(current, cpu_all_mask);
881 * Tell the world that we're going to be the grim
882 * reaper of innocent orphaned children.
884 * We don't want people to have to make incorrect
885 * assumptions about where in the task array this
888 init_pid_ns.child_reaper = current;
890 cad_pid = task_pid(current);
892 smp_prepare_cpus(setup_max_cpus);
894 do_pre_smp_initcalls();
902 /* Open the /dev/console on the rootfs, this should never fail */
903 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
904 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
909 * check if there is an early userspace init. If yes, let it do all
913 if (!ramdisk_execute_command)
914 ramdisk_execute_command = "/init";
916 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
917 ramdisk_execute_command = NULL;
922 * Ok, we have completed the initial bootup, and
923 * we're essentially up and running. Get rid of the
924 * initmem segments and start the user-mode stuff..