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 #define DEBUG /* Enable initcall_debug */
14 #include <linux/types.h>
15 #include <linux/module.h>
16 #include <linux/proc_fs.h>
17 #include <linux/kernel.h>
18 #include <linux/syscalls.h>
19 #include <linux/stackprotector.h>
20 #include <linux/string.h>
21 #include <linux/ctype.h>
22 #include <linux/delay.h>
23 #include <linux/ioport.h>
24 #include <linux/init.h>
25 #include <linux/initrd.h>
26 #include <linux/bootmem.h>
27 #include <linux/acpi.h>
28 #include <linux/tty.h>
29 #include <linux/percpu.h>
30 #include <linux/kmod.h>
31 #include <linux/vmalloc.h>
32 #include <linux/kernel_stat.h>
33 #include <linux/start_kernel.h>
34 #include <linux/security.h>
35 #include <linux/smp.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/sfi.h>
70 #include <linux/shmem_fs.h>
71 #include <linux/slab.h>
72 #include <linux/perf_event.h>
73 #include <linux/file.h>
74 #include <linux/ptrace.h>
75 #include <linux/blkdev.h>
76 #include <linux/elevator.h>
77 #include <linux/sched_clock.h>
78 #include <linux/context_tracking.h>
79 #include <linux/random.h>
80 #include <linux/list.h>
84 #include <asm/setup.h>
85 #include <asm/sections.h>
86 #include <asm/cacheflush.h>
88 #ifdef CONFIG_X86_LOCAL_APIC
92 static int kernel_init(void *);
94 extern void init_IRQ(void);
95 extern void fork_init(unsigned long);
96 extern void radix_tree_init(void);
97 #ifndef CONFIG_DEBUG_RODATA
98 static inline void mark_rodata_ro(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
= false;
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 situaiton 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 int __init
obsolete_checksetup(char *line
)
170 const struct obs_kernel_param
*p
;
171 int had_early_param
= 0;
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
] == '=')
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);
203 EXPORT_SYMBOL(loops_per_jiffy
);
205 static int __init
debug_kernel(char *str
)
207 console_loglevel
= CONSOLE_LOGLEVEL_DEBUG
;
211 static int __init
quiet_kernel(char *str
)
213 console_loglevel
= CONSOLE_LOGLEVEL_QUIET
;
217 early_param("debug", debug_kernel
);
218 early_param("quiet", quiet_kernel
);
220 static int __init
loglevel(char *str
)
225 * Only update loglevel value when a correct setting was passed,
226 * to prevent blind crashes (when loglevel being set to 0) that
227 * are quite hard to debug
229 if (get_option(&str
, &newlevel
)) {
230 console_loglevel
= newlevel
;
237 early_param("loglevel", loglevel
);
239 /* Change NUL term back to "=", to make "param" the whole string. */
240 static int __init
repair_env_string(char *param
, char *val
, const char *unused
)
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);
257 * Unknown boot options get handed to init, unless they look like
258 * unused parameters (modprobe will find them in /proc/cmdline).
260 static int __init
unknown_bootoption(char *param
, char *val
, const char *unused
)
262 repair_env_string(param
, val
, unused
);
264 /* Handle obsolete-style parameters */
265 if (obsolete_checksetup(param
))
268 /* Unused module parameter. */
269 if (strchr(param
, '.') && (!val
|| strchr(param
, '.') < val
))
276 /* Environment option */
278 for (i
= 0; envp_init
[i
]; i
++) {
279 if (i
== MAX_INIT_ENVS
) {
283 if (!strncmp(param
, envp_init
[i
], val
- param
))
286 envp_init
[i
] = param
;
288 /* Command line option */
290 for (i
= 0; argv_init
[i
]; i
++) {
291 if (i
== MAX_INIT_ARGS
) {
292 panic_later
= "init";
296 argv_init
[i
] = param
;
301 static int __init
init_setup(char *str
)
305 execute_command
= str
;
307 * In case LILO is going to boot us with default command line,
308 * it prepends "auto" before the whole cmdline which makes
309 * the shell think it should execute a script with such name.
310 * So we ignore all arguments entered _before_ init=... [MJ]
312 for (i
= 1; i
< MAX_INIT_ARGS
; i
++)
316 __setup("init=", init_setup
);
318 static int __init
rdinit_setup(char *str
)
322 ramdisk_execute_command
= str
;
323 /* See "auto" comment in init_setup */
324 for (i
= 1; i
< MAX_INIT_ARGS
; i
++)
328 __setup("rdinit=", rdinit_setup
);
331 static const unsigned int setup_max_cpus
= NR_CPUS
;
332 #ifdef CONFIG_X86_LOCAL_APIC
333 static void __init
smp_init(void)
335 APIC_init_uniprocessor();
338 #define smp_init() do { } while (0)
341 static inline void setup_nr_cpu_ids(void) { }
342 static inline void smp_prepare_cpus(unsigned int maxcpus
) { }
346 * We need to store the untouched command line for future reference.
347 * We also need to store the touched command line since the parameter
348 * parsing is performed in place, and we should allow a component to
349 * store reference of name/value for future reference.
351 static void __init
setup_command_line(char *command_line
)
354 memblock_virt_alloc(strlen(boot_command_line
) + 1, 0);
355 initcall_command_line
=
356 memblock_virt_alloc(strlen(boot_command_line
) + 1, 0);
357 static_command_line
= memblock_virt_alloc(strlen(command_line
) + 1, 0);
358 strcpy (saved_command_line
, boot_command_line
);
359 strcpy (static_command_line
, command_line
);
363 * We need to finalize in a non-__init function or else race conditions
364 * between the root thread and the init thread may cause start_kernel to
365 * be reaped by free_initmem before the root thread has proceeded to
368 * gcc-3.4 accidentally inlines this function, so use noinline.
371 static __initdata
DECLARE_COMPLETION(kthreadd_done
);
373 static noinline
void __init_refok
rest_init(void)
377 rcu_scheduler_starting();
379 * We need to spawn init first so that it obtains pid 1, however
380 * the init task will end up wanting to create kthreads, which, if
381 * we schedule it before we create kthreadd, will OOPS.
383 kernel_thread(kernel_init
, NULL
, CLONE_FS
);
384 numa_default_policy();
385 pid
= kernel_thread(kthreadd
, NULL
, CLONE_FS
| CLONE_FILES
);
387 kthreadd_task
= find_task_by_pid_ns(pid
, &init_pid_ns
);
389 complete(&kthreadd_done
);
392 * The boot idle thread must execute schedule()
393 * at least once to get things moving:
395 init_idle_bootup_task(current
);
396 schedule_preempt_disabled();
397 /* Call into cpu_idle with preempt disabled */
398 cpu_startup_entry(CPUHP_ONLINE
);
401 /* Check for early params. */
402 static int __init
do_early_param(char *param
, char *val
, const char *unused
)
404 const struct obs_kernel_param
*p
;
406 for (p
= __setup_start
; p
< __setup_end
; p
++) {
407 if ((p
->early
&& parameq(param
, p
->str
)) ||
408 (strcmp(param
, "console") == 0 &&
409 strcmp(p
->str
, "earlycon") == 0)
411 if (p
->setup_func(val
) != 0)
412 pr_warn("Malformed early option '%s'\n", param
);
415 /* We accept everything at this stage. */
419 void __init
parse_early_options(char *cmdline
)
421 parse_args("early options", cmdline
, NULL
, 0, 0, 0, do_early_param
);
424 /* Arch code calls this early on, or if not, just before other parsing. */
425 void __init
parse_early_param(void)
427 static __initdata
int done
= 0;
428 static __initdata
char tmp_cmdline
[COMMAND_LINE_SIZE
];
433 /* All fall through to do_early_param. */
434 strlcpy(tmp_cmdline
, boot_command_line
, COMMAND_LINE_SIZE
);
435 parse_early_options(tmp_cmdline
);
440 * Activate the first processor.
443 static void __init
boot_cpu_init(void)
445 int cpu
= smp_processor_id();
446 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
447 set_cpu_online(cpu
, true);
448 set_cpu_active(cpu
, true);
449 set_cpu_present(cpu
, true);
450 set_cpu_possible(cpu
, true);
453 void __init __weak
smp_setup_processor_id(void)
457 # if THREAD_SIZE >= PAGE_SIZE
458 void __init __weak
thread_info_cache_init(void)
464 * Set up kernel memory allocators
466 static void __init
mm_init(void)
469 * page_cgroup requires contiguous pages,
470 * bigger than MAX_ORDER unless SPARSEMEM.
472 page_cgroup_init_flatmem();
480 asmlinkage __visible
void __init
start_kernel(void)
483 extern const struct kernel_param __start___param
[], __stop___param
[];
486 * Need to run as early as possible, to initialize the
490 smp_setup_processor_id();
491 debug_objects_early_init();
494 * Set up the the initial canary ASAP:
496 boot_init_stack_canary();
501 early_boot_irqs_disabled
= true;
504 * Interrupts are still disabled. Do necessary setups, then
509 pr_notice("%s", linux_banner
);
510 setup_arch(&command_line
);
511 mm_init_cpumask(&init_mm
);
512 setup_command_line(command_line
);
514 setup_per_cpu_areas();
515 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
517 build_all_zonelists(NULL
, NULL
);
520 pr_notice("Kernel command line: %s\n", boot_command_line
);
522 parse_args("Booting kernel", static_command_line
, __start___param
,
523 __stop___param
- __start___param
,
524 -1, -1, &unknown_bootoption
);
529 * These use large bootmem allocations and must precede
534 vfs_caches_init_early();
540 * Set up the scheduler prior starting any interrupts (such as the
541 * timer interrupt). Full topology setup happens at smp_init()
542 * time - but meanwhile we still have a functioning scheduler.
546 * Disable preemption - early bootup scheduling is extremely
547 * fragile until we cpu_idle() for the first time.
550 if (WARN(!irqs_disabled(), "Interrupts were enabled *very* early, fixing it\n"))
555 context_tracking_init();
557 /* init some links before init_ISA_irqs() */
566 sched_clock_postinit();
569 call_function_init();
570 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
571 early_boot_irqs_disabled
= false;
574 kmem_cache_init_late();
577 * HACK ALERT! This is early. We're enabling the console before
578 * we've done PCI setups etc, and console_init() must be aware of
579 * this. But we do want output early, in case something goes wrong.
583 panic("Too many boot %s vars at `%s'", panic_later
,
589 * Need to run this when irqs are enabled, because it wants
590 * to self-test [hard/soft]-irqs on/off lock inversion bugs
595 #ifdef CONFIG_BLK_DEV_INITRD
596 if (initrd_start
&& !initrd_below_start_ok
&&
597 page_to_pfn(virt_to_page((void *)initrd_start
)) < min_low_pfn
) {
598 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
599 page_to_pfn(virt_to_page((void *)initrd_start
)),
605 debug_objects_mem_init();
607 setup_per_cpu_pageset();
617 if (efi_enabled(EFI_RUNTIME_SERVICES
))
618 efi_enter_virtual_mode();
620 #ifdef CONFIG_X86_ESPFIX64
621 /* Should be run before the first non-init thread is created */
624 thread_info_cache_init();
626 fork_init(totalram_pages
);
632 vfs_caches_init(totalram_pages
);
634 /* rootfs populating might need page-writeback */
635 page_writeback_init();
639 taskstats_init_early();
646 if (efi_enabled(EFI_RUNTIME_SERVICES
)) {
648 efi_free_boot_services();
653 /* Do the rest non-__init'ed, we're now alive */
657 /* Call all constructor functions linked into the kernel. */
658 static void __init
do_ctors(void)
660 #ifdef CONFIG_CONSTRUCTORS
661 ctor_fn_t
*fn
= (ctor_fn_t
*) __ctors_start
;
663 for (; fn
< (ctor_fn_t
*) __ctors_end
; fn
++)
669 core_param(initcall_debug
, initcall_debug
, bool, 0644);
671 #ifdef CONFIG_KALLSYMS
672 struct blacklist_entry
{
673 struct list_head next
;
677 static __initdata_or_module
LIST_HEAD(blacklisted_initcalls
);
679 static int __init
initcall_blacklist(char *str
)
682 struct blacklist_entry
*entry
;
684 /* str argument is a comma-separated list of functions */
686 str_entry
= strsep(&str
, ",");
688 pr_debug("blacklisting initcall %s\n", str_entry
);
689 entry
= alloc_bootmem(sizeof(*entry
));
690 entry
->buf
= alloc_bootmem(strlen(str_entry
) + 1);
691 strcpy(entry
->buf
, str_entry
);
692 list_add(&entry
->next
, &blacklisted_initcalls
);
699 static bool __init_or_module
initcall_blacklisted(initcall_t fn
)
701 struct list_head
*tmp
;
702 struct blacklist_entry
*entry
;
705 fn_name
= kasprintf(GFP_KERNEL
, "%pf", fn
);
709 list_for_each(tmp
, &blacklisted_initcalls
) {
710 entry
= list_entry(tmp
, struct blacklist_entry
, next
);
711 if (!strcmp(fn_name
, entry
->buf
)) {
712 pr_debug("initcall %s blacklisted\n", fn_name
);
722 static int __init
initcall_blacklist(char *str
)
724 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
728 static bool __init_or_module
initcall_blacklisted(initcall_t fn
)
733 __setup("initcall_blacklist=", initcall_blacklist
);
735 static int __init_or_module
do_one_initcall_debug(initcall_t fn
)
737 ktime_t calltime
, delta
, rettime
;
738 unsigned long long duration
;
741 printk(KERN_DEBUG
"calling %pF @ %i\n", fn
, task_pid_nr(current
));
742 calltime
= ktime_get();
744 rettime
= ktime_get();
745 delta
= ktime_sub(rettime
, calltime
);
746 duration
= (unsigned long long) ktime_to_ns(delta
) >> 10;
747 printk(KERN_DEBUG
"initcall %pF returned %d after %lld usecs\n",
753 int __init_or_module
do_one_initcall(initcall_t fn
)
755 int count
= preempt_count();
759 if (initcall_blacklisted(fn
))
763 ret
= do_one_initcall_debug(fn
);
769 if (preempt_count() != count
) {
770 sprintf(msgbuf
, "preemption imbalance ");
771 preempt_count_set(count
);
773 if (irqs_disabled()) {
774 strlcat(msgbuf
, "disabled interrupts ", sizeof(msgbuf
));
777 WARN(msgbuf
[0], "initcall %pF returned with %s\n", fn
, msgbuf
);
783 extern initcall_t __initcall_start
[];
784 extern initcall_t __initcall0_start
[];
785 extern initcall_t __initcall1_start
[];
786 extern initcall_t __initcall2_start
[];
787 extern initcall_t __initcall3_start
[];
788 extern initcall_t __initcall4_start
[];
789 extern initcall_t __initcall5_start
[];
790 extern initcall_t __initcall6_start
[];
791 extern initcall_t __initcall7_start
[];
792 extern initcall_t __initcall_end
[];
794 static initcall_t
*initcall_levels
[] __initdata
= {
806 /* Keep these in sync with initcalls in include/linux/init.h */
807 static char *initcall_level_names
[] __initdata
= {
818 static void __init
do_initcall_level(int level
)
820 extern const struct kernel_param __start___param
[], __stop___param
[];
823 strcpy(initcall_command_line
, saved_command_line
);
824 parse_args(initcall_level_names
[level
],
825 initcall_command_line
, __start___param
,
826 __stop___param
- __start___param
,
830 for (fn
= initcall_levels
[level
]; fn
< initcall_levels
[level
+1]; fn
++)
831 do_one_initcall(*fn
);
834 static void __init
do_initcalls(void)
838 for (level
= 0; level
< ARRAY_SIZE(initcall_levels
) - 1; level
++)
839 do_initcall_level(level
);
843 * Ok, the machine is now initialized. None of the devices
844 * have been touched yet, but the CPU subsystem is up and
845 * running, and memory and process management works.
847 * Now we can finally start doing some real work..
849 static void __init
do_basic_setup(void)
852 usermodehelper_init();
857 usermodehelper_enable();
859 random_int_secret_init();
862 static void __init
do_pre_smp_initcalls(void)
866 for (fn
= __initcall_start
; fn
< __initcall0_start
; fn
++)
867 do_one_initcall(*fn
);
871 * This function requests modules which should be loaded by default and is
872 * called twice right after initrd is mounted and right before init is
873 * exec'd. If such modules are on either initrd or rootfs, they will be
874 * loaded before control is passed to userland.
876 void __init
load_default_modules(void)
878 load_default_elevator_module();
881 static int run_init_process(const char *init_filename
)
883 argv_init
[0] = init_filename
;
884 return do_execve(getname_kernel(init_filename
),
885 (const char __user
*const __user
*)argv_init
,
886 (const char __user
*const __user
*)envp_init
);
889 static int try_to_run_init_process(const char *init_filename
)
893 ret
= run_init_process(init_filename
);
895 if (ret
&& ret
!= -ENOENT
) {
896 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
903 static noinline
void __init
kernel_init_freeable(void);
905 static int __ref
kernel_init(void *unused
)
909 kernel_init_freeable();
910 /* need to finish all async __init code before freeing the memory */
911 async_synchronize_full();
914 system_state
= SYSTEM_RUNNING
;
915 numa_default_policy();
917 flush_delayed_fput();
919 if (ramdisk_execute_command
) {
920 ret
= run_init_process(ramdisk_execute_command
);
923 pr_err("Failed to execute %s (error %d)\n",
924 ramdisk_execute_command
, ret
);
928 * We try each of these until one succeeds.
930 * The Bourne shell can be used instead of init if we are
931 * trying to recover a really broken machine.
933 if (execute_command
) {
934 ret
= run_init_process(execute_command
);
937 pr_err("Failed to execute %s (error %d). Attempting defaults...\n",
938 execute_command
, ret
);
940 if (!try_to_run_init_process("/sbin/init") ||
941 !try_to_run_init_process("/etc/init") ||
942 !try_to_run_init_process("/bin/init") ||
943 !try_to_run_init_process("/bin/sh"))
946 panic("No working init found. Try passing init= option to kernel. "
947 "See Linux Documentation/init.txt for guidance.");
950 static noinline
void __init
kernel_init_freeable(void)
953 * Wait until kthreadd is all set-up.
955 wait_for_completion(&kthreadd_done
);
957 /* Now the scheduler is fully set up and can do blocking allocations */
958 gfp_allowed_mask
= __GFP_BITS_MASK
;
961 * init can allocate pages on any node
963 set_mems_allowed(node_states
[N_MEMORY
]);
965 * init can run on any cpu.
967 set_cpus_allowed_ptr(current
, cpu_all_mask
);
969 cad_pid
= task_pid(current
);
971 smp_prepare_cpus(setup_max_cpus
);
973 do_pre_smp_initcalls();
974 lockup_detector_init();
981 /* Open the /dev/console on the rootfs, this should never fail */
982 if (sys_open((const char __user
*) "/dev/console", O_RDWR
, 0) < 0)
983 pr_err("Warning: unable to open an initial console.\n");
988 * check if there is an early userspace init. If yes, let it do all
992 if (!ramdisk_execute_command
)
993 ramdisk_execute_command
= "/init";
995 if (sys_access((const char __user
*) ramdisk_execute_command
, 0) != 0) {
996 ramdisk_execute_command
= NULL
;
1001 * Ok, we have completed the initial bootup, and
1002 * we're essentially up and running. Get rid of the
1003 * initmem segments and start the user-mode stuff..
1006 /* rootfs is available now, try loading default modules */
1007 load_default_modules();