blob: 3627bb37225c77552b83a8c9f6babc12bee2df2e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/init/main.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
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>
10 */
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/types.h>
13#include <linux/module.h>
14#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/kernel.h>
16#include <linux/syscalls.h>
Ingo Molnar9b5609fd2008-02-14 09:41:09 +010017#include <linux/stackprotector.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/string.h>
19#include <linux/ctype.h>
20#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/ioport.h>
22#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/initrd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/bootmem.h>
Len Brown4a7a16d2009-06-12 20:42:08 -040025#include <linux/acpi.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/tty.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/percpu.h>
28#include <linux/kmod.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070029#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/kernel_stat.h>
Rusty Russelld7cd5612006-12-07 02:14:08 +010031#include <linux/start_kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/security.h>
Jens Axboe3d442232008-06-26 11:21:34 +020033#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#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>
Paul Menageddbcc7e2007-10-18 23:39:30 -070041#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/efi.h>
Thomas Gleixner906568c2007-02-16 01:28:01 -080043#include <linux/tick.h>
Andrew Morton6168a702007-02-17 21:22:39 -080044#include <linux/interrupt.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070045#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070046#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/unistd.h>
48#include <linux/rmap.h>
49#include <linux/mempolicy.h>
50#include <linux/key.h>
Adrian Bunkb6cd0b72006-06-27 02:53:54 -070051#include <linux/buffer_head.h>
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -070052#include <linux/page_cgroup.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070053#include <linux/debug_locks.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070054#include <linux/debugobjects.h>
Ingo Molnarfbb9ce92006-07-03 00:24:50 -070055#include <linux/lockdep.h>
Catalin Marinas3c7b4e62009-06-11 13:22:39 +010056#include <linux/kmemleak.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080057#include <linux/pid_namespace.h>
Adrian Bunk1f217822006-12-19 13:01:28 -080058#include <linux/device.h>
Eric W. Biederman73c27992007-05-09 02:34:32 -070059#include <linux/kthread.h>
Adrian Bunke6fe6642007-11-09 22:39:39 +010060#include <linux/sched.h>
Adrian Bunka1c9eea2008-02-06 01:36:44 -080061#include <linux/signal.h>
Akinobu Mita199f0ca2008-04-29 01:03:13 -070062#include <linux/idr.h>
Jason Wessel0b4b3822010-05-20 21:04:29 -050063#include <linux/kgdb.h>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040064#include <linux/ftrace.h>
Arjan van de Ven22a9d642009-01-07 08:45:46 -080065#include <linux/async.h>
Vegard Nossumdfec0722008-04-04 00:51:41 +020066#include <linux/kmemcheck.h>
Feng Tang6ae69962009-08-14 15:13:46 -040067#include <linux/sfi.h>
Kay Sievers2b2af542009-04-30 15:23:42 +020068#include <linux/shmem_fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090069#include <linux/slab.h>
Peter Zijlstra24a24bb2010-11-17 23:17:33 +010070#include <linux/perf_event.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
72#include <asm/io.h>
73#include <asm/bugs.h>
74#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070075#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080076#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#ifdef CONFIG_X86_LOCAL_APIC
79#include <asm/smp.h>
80#endif
81
Sam Ravnborgaae5f662007-02-26 16:45:41 +010082static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
84extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085extern void fork_init(unsigned long);
86extern void mca_init(void);
87extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088extern void prio_tree_init(void);
89extern void radix_tree_init(void);
90extern void free_initmem(void);
Arjan van de Ven37b73c82006-01-06 00:12:01 -080091#ifndef CONFIG_DEBUG_RODATA
92static inline void mark_rodata_ro(void) { }
93#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95#ifdef CONFIG_TC
96extern void tc_init(void);
97#endif
98
Tejun Heo2ce802f2011-01-20 12:06:35 +010099/*
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
104 * flag is set.
105 */
106bool early_boot_irqs_disabled __read_mostly;
107
Paul E. McKenneya6826042009-02-25 18:03:42 -0800108enum system_states system_state __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109EXPORT_SYMBOL(system_state);
110
111/*
112 * Boot command-line arguments
113 */
114#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
115#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
116
117extern void time_init(void);
118/* Default late time init is NULL. archs can override this later. */
Jan Beulichd2e31922009-01-06 14:41:10 -0800119void (*__initdata late_time_init)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120extern void softirq_init(void);
121
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800122/* Untouched command line saved by arch-specific code. */
123char __initdata boot_command_line[COMMAND_LINE_SIZE];
124/* Untouched saved command line (eg. for /proc) */
125char *saved_command_line;
126/* Command line for parameter parsing */
127static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700130static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Jan Beulich8b3b2952007-07-15 23:41:07 -0700132/*
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.
136 *
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.
140 */
141unsigned int reset_devices;
142EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143
Vivek Goyal7e962872006-09-27 01:50:44 -0700144static int __init set_reset_devices(char *str)
145{
146 reset_devices = 1;
147 return 1;
148}
149
150__setup("reset_devices", set_reset_devices);
151
David Howellsd7627462010-08-17 23:52:56 +0100152static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
153const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154static const char *panic_later, *panic_param;
155
Rusty Russell914dcaa2010-08-11 23:04:18 -0600156extern const struct obs_kernel_param __setup_start[], __setup_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
158static int __init obsolete_checksetup(char *line)
159{
Rusty Russell914dcaa2010-08-11 23:04:18 -0600160 const struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200161 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
163 p = __setup_start;
164 do {
165 int n = strlen(p->str);
166 if (!strncmp(line, p->str, n)) {
167 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200168 /* 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( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200173 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 } else if (!p->setup_func) {
175 printk(KERN_WARNING "Parameter %s is obsolete,"
176 " ignored\n", p->str);
177 return 1;
178 } else if (p->setup_func(line + n))
179 return 1;
180 }
181 p++;
182 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200183
184 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185}
186
187/*
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
190 */
191unsigned long loops_per_jiffy = (1<<12);
192
193EXPORT_SYMBOL(loops_per_jiffy);
194
195static int __init debug_kernel(char *str)
196{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800198 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199}
200
201static int __init quiet_kernel(char *str)
202{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800204 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205}
206
Yinghai Luf6f21c812008-02-08 04:21:58 -0800207early_param("debug", debug_kernel);
208early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210static int __init loglevel(char *str)
211{
212 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800213 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214}
215
Yinghai Luf6f21c812008-02-08 04:21:58 -0800216early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218/*
219 * Unknown boot options get handed to init, unless they look like
Rusty Russellf066a4f2009-12-01 14:56:44 +1030220 * unused parameters (modprobe will find them in /proc/cmdline).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 */
222static int __init unknown_bootoption(char *param, char *val)
223{
224 /* Change NUL term back to "=", to make "param" the whole string. */
225 if (val) {
226 /* param=val or param="val"? */
227 if (val == param+strlen(param)+1)
228 val[-1] = '=';
229 else if (val == param+strlen(param)+2) {
230 val[-2] = '=';
231 memmove(val-1, val, strlen(val)+1);
232 val--;
233 } else
234 BUG();
235 }
236
237 /* Handle obsolete-style parameters */
238 if (obsolete_checksetup(param))
239 return 0;
240
Rusty Russellf066a4f2009-12-01 14:56:44 +1030241 /* Unused module parameter. */
242 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245 if (panic_later)
246 return 0;
247
248 if (val) {
249 /* Environment option */
250 unsigned int i;
251 for (i = 0; envp_init[i]; i++) {
252 if (i == MAX_INIT_ENVS) {
253 panic_later = "Too many boot env vars at `%s'";
254 panic_param = param;
255 }
256 if (!strncmp(param, envp_init[i], val - param))
257 break;
258 }
259 envp_init[i] = param;
260 } else {
261 /* Command line option */
262 unsigned int i;
263 for (i = 0; argv_init[i]; i++) {
264 if (i == MAX_INIT_ARGS) {
265 panic_later = "Too many boot init vars at `%s'";
266 panic_param = param;
267 }
268 }
269 argv_init[i] = param;
270 }
271 return 0;
272}
273
Ingo Molnar12d6f212008-01-30 13:33:58 +0100274#ifdef CONFIG_DEBUG_PAGEALLOC
275int __read_mostly debug_pagealloc_enabled = 0;
276#endif
277
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278static int __init init_setup(char *str)
279{
280 unsigned int i;
281
282 execute_command = str;
283 /*
284 * In case LILO is going to boot us with default command line,
285 * it prepends "auto" before the whole cmdline which makes
286 * the shell think it should execute a script with such name.
287 * So we ignore all arguments entered _before_ init=... [MJ]
288 */
289 for (i = 1; i < MAX_INIT_ARGS; i++)
290 argv_init[i] = NULL;
291 return 1;
292}
293__setup("init=", init_setup);
294
Olof Johanssonffdfc402005-09-06 15:17:19 -0700295static int __init rdinit_setup(char *str)
296{
297 unsigned int i;
298
299 ramdisk_execute_command = str;
300 /* See "auto" comment in init_setup */
301 for (i = 1; i < MAX_INIT_ARGS; i++)
302 argv_init[i] = NULL;
303 return 1;
304}
305__setup("rdinit=", rdinit_setup);
306
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307#ifndef CONFIG_SMP
Amerigo Wang34db18a02011-03-22 16:34:06 -0700308static const unsigned int setup_max_cpus = NR_CPUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309#ifdef CONFIG_X86_LOCAL_APIC
310static void __init smp_init(void)
311{
312 APIC_init_uniprocessor();
313}
314#else
315#define smp_init() do { } while (0)
316#endif
317
Mike Travise0982e92008-03-26 14:23:48 -0700318static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319static inline void smp_prepare_cpus(unsigned int maxcpus) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320#endif
321
322/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800323 * We need to store the untouched command line for future reference.
324 * We also need to store the touched command line since the parameter
325 * parsing is performed in place, and we should allow a component to
326 * store reference of name/value for future reference.
327 */
328static void __init setup_command_line(char *command_line)
329{
330 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
331 static_command_line = alloc_bootmem(strlen (command_line)+1);
332 strcpy (saved_command_line, boot_command_line);
333 strcpy (static_command_line, command_line);
334}
335
336/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 * We need to finalize in a non-__init function or else race conditions
338 * between the root thread and the init thread may cause start_kernel to
339 * be reaped by free_initmem before the root thread has proceeded to
340 * cpu_idle.
341 *
342 * gcc-3.4 accidentally inlines this function, so use noinline.
343 */
344
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200345static __initdata DECLARE_COMPLETION(kthreadd_done);
346
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800347static noinline void __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700349 int pid;
350
Paul E. McKenney7db905e2009-09-02 14:01:24 -0700351 rcu_scheduler_starting();
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200352 /*
Peter Zijlstra97158562010-06-30 10:37:11 +0200353 * We need to spawn init first so that it obtains pid 1, however
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200354 * the init task will end up wanting to create kthreads, which, if
355 * we schedule it before we create kthreadd, will OOPS.
356 */
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100357 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700359 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800360 rcu_read_lock();
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700361 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800362 rcu_read_unlock();
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200363 complete(&kthreadd_done);
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200364
365 /*
366 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200367 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200368 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200369 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800370 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200371 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800372 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200373
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800374 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200376}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
378/* Check for early params. */
379static int __init do_early_param(char *param, char *val)
380{
Rusty Russell914dcaa2010-08-11 23:04:18 -0600381 const struct obs_kernel_param *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382
383 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700384 if ((p->early && strcmp(param, p->str) == 0) ||
385 (strcmp(param, "console") == 0 &&
386 strcmp(p->str, "earlycon") == 0)
387 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 if (p->setup_func(val) != 0)
389 printk(KERN_WARNING
390 "Malformed early option '%s'\n", param);
391 }
392 }
393 /* We accept everything at this stage. */
394 return 0;
395}
396
Magnus Damm13977092009-03-30 14:37:25 -0700397void __init parse_early_options(char *cmdline)
398{
399 parse_args("early options", cmdline, NULL, 0, do_early_param);
400}
401
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402/* Arch code calls this early on, or if not, just before other parsing. */
403void __init parse_early_param(void)
404{
405 static __initdata int done = 0;
406 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
407
408 if (done)
409 return;
410
411 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800412 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Magnus Damm13977092009-03-30 14:37:25 -0700413 parse_early_options(tmp_cmdline);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 done = 1;
415}
416
417/*
418 * Activate the first processor.
419 */
420
Stas Sergeev44fd2292006-03-23 02:59:44 -0800421static void __init boot_cpu_init(void)
422{
423 int cpu = smp_processor_id();
424 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030425 set_cpu_online(cpu, true);
Peter Zijlstra933b0612009-12-16 18:04:31 +0100426 set_cpu_active(cpu, true);
Rusty Russell915441b2009-01-01 10:12:15 +1030427 set_cpu_present(cpu, true);
428 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800429}
430
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000431void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700432{
433}
434
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000435void __init __weak thread_info_cache_init(void)
436{
437}
438
Pekka Enberg444f4782009-06-11 18:29:06 +0300439/*
440 * Set up kernel memory allocators
441 */
442static void __init mm_init(void)
443{
KAMEZAWA Hiroyukica371c02009-06-12 10:33:53 +0300444 /*
445 * page_cgroup requires countinous pages as memmap
446 * and it's bigger than MAX_ORDER unless SPARSEMEM.
447 */
448 page_cgroup_init_flatmem();
Pekka Enberg444f4782009-06-11 18:29:06 +0300449 mem_init();
450 kmem_cache_init();
Tejun Heo099a19d2010-06-27 18:50:00 +0200451 percpu_init_late();
Benjamin Herrenschmidtc868d552009-06-17 13:48:39 +1000452 pgtable_cache_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300453 vmalloc_init();
454}
455
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456asmlinkage void __init start_kernel(void)
457{
458 char * command_line;
Rusty Russell914dcaa2010-08-11 23:04:18 -0600459 extern const struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700460
461 smp_setup_processor_id();
462
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700463 /*
464 * Need to run as early as possible, to initialize the
465 * lockdep hash:
466 */
467 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700468 debug_objects_early_init();
Ingo Molnar42059422008-02-14 09:44:08 +0100469
470 /*
471 * Set up the the initial canary ASAP:
472 */
473 boot_init_stack_canary();
474
Paul Menageddbcc7e2007-10-18 23:39:30 -0700475 cgroup_init_early();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700476
477 local_irq_disable();
Tejun Heo2ce802f2011-01-20 12:06:35 +0100478 early_boot_irqs_disabled = true;
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700479
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480/*
481 * Interrupts are still disabled. Do necessary setups, then
482 * enable them
483 */
Thomas Gleixner906568c2007-02-16 01:28:01 -0800484 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800485 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 page_address_init();
Alex Riesen657cafa2009-05-24 15:30:48 +0200487 printk(KERN_NOTICE "%s", linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700489 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800490 setup_command_line(command_line);
Mike Travise0982e92008-03-26 14:23:48 -0700491 setup_nr_cpu_ids();
Tejun Heod6647bd2009-07-21 17:11:50 +0900492 setup_per_cpu_areas();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800493 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
Haicheng Li1f522502010-05-24 14:32:51 -0700495 build_all_zonelists(NULL);
Pekka Enberg83b519e2009-06-10 19:40:04 +0300496 page_alloc_init();
497
498 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
499 parse_early_param();
500 parse_args("Booting kernel", static_command_line, __start___param,
501 __stop___param - __start___param,
502 &unknown_bootoption);
503 /*
504 * These use large bootmem allocations and must precede
505 * kmem_cache_init()
506 */
507 pidhash_init();
Pekka Enberg83b519e2009-06-10 19:40:04 +0300508 vfs_caches_init_early();
509 sort_main_extable();
510 trap_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300511 mm_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 /*
513 * Set up the scheduler prior starting any interrupts (such as the
514 * timer interrupt). Full topology setup happens at smp_init()
515 * time - but meanwhile we still have a functioning scheduler.
516 */
517 sched_init();
518 /*
519 * Disable preemption - early bootup scheduling is extremely
520 * fragile until we cpu_idle() for the first time.
521 */
522 preempt_disable();
Ard van Breemenc4a68302007-01-05 16:36:19 -0800523 if (!irqs_disabled()) {
524 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
525 "enabled *very* early, fixing it\n");
526 local_irq_disable();
527 }
Peter Zijlstra9f58a202010-11-17 23:17:35 +0100528 idr_init_cache();
Peter Zijlstra24a24bb2010-11-17 23:17:33 +0100529 perf_event_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 rcu_init();
Yinghai Lu773e3eb2010-02-10 01:20:33 -0800531 radix_tree_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800532 /* init some links before init_ISA_irqs() */
533 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 init_IRQ();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100535 prio_tree_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800537 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700539 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700540 time_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700541 profile_init();
542 if (!irqs_disabled())
Ron Lee24d431d2008-11-27 02:31:57 +1030543 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
544 "enabled early\n");
Tejun Heo2ce802f2011-01-20 12:06:35 +0100545 early_boot_irqs_disabled = false;
Heiko Carstens93e02812006-07-03 00:24:24 -0700546 local_irq_enable();
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000547
548 /* Interrupts are enabled now so all GFP allocations are safe. */
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800549 gfp_allowed_mask = __GFP_BITS_MASK;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000550
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300551 kmem_cache_init_late();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
553 /*
554 * HACK ALERT! This is early. We're enabling the console before
555 * we've done PCI setups etc, and console_init() must be aware of
556 * this. But we do want output early, in case something goes wrong.
557 */
558 console_init();
559 if (panic_later)
560 panic(panic_later, panic_param);
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700561
562 lockdep_info();
563
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700564 /*
565 * Need to run this when irqs are enabled, because it wants
566 * to self-test [hard/soft]-irqs on/off lock inversion bugs
567 * too:
568 */
569 locking_selftest();
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571#ifdef CONFIG_BLK_DEV_INITRD
572 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700573 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200575 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700576 page_to_pfn(virt_to_page((void *)initrd_start)),
577 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 initrd_start = 0;
579 }
580#endif
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700581 page_cgroup_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100582 enable_debug_pagealloc();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100583 kmemleak_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700584 debug_objects_mem_init();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700585 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 numa_policy_init();
587 if (late_time_init)
588 late_time_init();
Thomas Gleixnerfa84e9e2009-08-21 22:01:12 +0200589 sched_clock_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 calibrate_delay();
591 pidmap_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 anon_vma_init();
593#ifdef CONFIG_X86
594 if (efi_enabled)
595 efi_enter_virtual_mode();
596#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000597 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100598 cred_init();
Jan Beulich44813742009-09-21 17:03:05 -0700599 fork_init(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 proc_caches_init();
601 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 key_init();
603 security_init();
Jason Wessel0b4b3822010-05-20 21:04:29 -0500604 dbg_late_init();
Jan Beulich44813742009-09-21 17:03:05 -0700605 vfs_caches_init(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 signals_init();
607 /* rootfs populating might need page-writeback */
608 page_writeback_init();
609#ifdef CONFIG_PROC_FS
610 proc_root_init();
611#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700612 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700614 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700615 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616
617 check_bugs();
618
619 acpi_early_init(); /* before LAPIC and SMP init */
Feng Tang6ae69962009-08-14 15:13:46 -0400620 sfi_init_late();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400622 ftrace_init();
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 /* Do the rest non-__init'ed, we're now alive */
625 rest_init();
626}
627
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700628/* Call all constructor functions linked into the kernel. */
629static void __init do_ctors(void)
630{
631#ifdef CONFIG_CONSTRUCTORS
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800632 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700633
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800634 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
635 (*fn)();
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700636#endif
637}
638
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800639int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500640core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
Ingo Molnar4a683bf2009-08-21 12:53:36 +0200642static char msgbuf[64];
Ingo Molnar4a683bf2009-08-21 12:53:36 +0200643
Kevin Winchestere4461272010-08-09 17:20:32 -0700644static int __init_or_module do_one_initcall_debug(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700645{
Frederic Weisbecker74239072008-11-11 23:24:42 +0100646 ktime_t calltime, delta, rettime;
Américo Wang30dbb20e2010-05-26 18:57:53 +0800647 unsigned long long duration;
648 int ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700649
Kevin Winchester22c5c032010-08-09 17:20:32 -0700650 printk(KERN_DEBUG "calling %pF @ %i\n", fn, task_pid_nr(current));
651 calltime = ktime_get();
Américo Wang30dbb20e2010-05-26 18:57:53 +0800652 ret = fn();
Kevin Winchester22c5c032010-08-09 17:20:32 -0700653 rettime = ktime_get();
654 delta = ktime_sub(rettime, calltime);
655 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
656 printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n", fn,
657 ret, duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700658
Kevin Winchester22c5c032010-08-09 17:20:32 -0700659 return ret;
660}
661
Kevin Winchestere4461272010-08-09 17:20:32 -0700662int __init_or_module do_one_initcall(initcall_t fn)
Kevin Winchester22c5c032010-08-09 17:20:32 -0700663{
664 int count = preempt_count();
665 int ret;
666
667 if (initcall_debug)
668 ret = do_one_initcall_debug(fn);
669 else
670 ret = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700671
672 msgbuf[0] = 0;
673
Américo Wang30dbb20e2010-05-26 18:57:53 +0800674 if (ret && ret != -ENODEV && initcall_debug)
675 sprintf(msgbuf, "error code %d ", ret);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700676
677 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700678 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700679 preempt_count() = count;
680 }
681 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700682 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700683 local_irq_enable();
684 }
685 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700686 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700687 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700688
Américo Wang30dbb20e2010-05-26 18:57:53 +0800689 return ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700690}
691
692
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700693extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694
695static void __init do_initcalls(void)
696{
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800697 initcall_t *fn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800699 for (fn = __early_initcall_end; fn < __initcall_end; fn++)
700 do_one_initcall(*fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702
703/*
704 * Ok, the machine is now initialized. None of the devices
705 * have been touched yet, but the CPU subsystem is up and
706 * running, and memory and process management works.
707 *
708 * Now we can finally start doing some real work..
709 */
710static void __init do_basic_setup(void)
711{
Lai Jiangshan759ee092009-03-25 17:06:30 +0800712 cpuset_init_smp();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 usermodehelper_init();
Kay Sievers2b2af542009-04-30 15:23:42 +0200714 init_tmpfs();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800716 init_irq_proc();
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700717 do_ctors();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 do_initcalls();
719}
720
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700721static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700722{
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800723 initcall_t *fn;
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700724
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800725 for (fn = __initcall_start; fn < __early_initcall_end; fn++)
726 do_one_initcall(*fn);
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700727}
728
David Howellsd7627462010-08-17 23:52:56 +0100729static void run_init_process(const char *init_filename)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
731 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700732 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733}
734
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100735/* This is a non __init function. Force it to be noinline otherwise gcc
736 * makes it inline to init() and it becomes part of init.text section
737 */
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800738static noinline int init_post(void)
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100739{
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800740 /* need to finish all async __init code before freeing the memory */
741 async_synchronize_full();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100742 free_initmem();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100743 mark_rodata_ro();
744 system_state = SYSTEM_RUNNING;
745 numa_default_policy();
746
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100747
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700748 current->signal->flags |= SIGNAL_UNKILLABLE;
749
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100750 if (ramdisk_execute_command) {
751 run_init_process(ramdisk_execute_command);
752 printk(KERN_WARNING "Failed to execute %s\n",
753 ramdisk_execute_command);
754 }
755
756 /*
757 * We try each of these until one succeeds.
758 *
759 * The Bourne shell can be used instead of init if we are
760 * trying to recover a really broken machine.
761 */
762 if (execute_command) {
763 run_init_process(execute_command);
764 printk(KERN_WARNING "Failed to execute %s. Attempting "
765 "defaults...\n", execute_command);
766 }
767 run_init_process("/sbin/init");
768 run_init_process("/etc/init");
769 run_init_process("/bin/init");
770 run_init_process("/bin/sh");
771
Andreas Mohr9a85b8d2010-03-05 13:42:39 -0800772 panic("No init found. Try passing init= option to kernel. "
773 "See Linux Documentation/init.txt for guidance.");
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100774}
775
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100776static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200778 /*
779 * Wait until kthreadd is all set-up.
780 */
781 wait_for_completion(&kthreadd_done);
Miao Xie58568d22009-06-16 15:31:49 -0700782 /*
783 * init can allocate pages on any node
784 */
Miao Xie5ab116c2010-03-23 13:35:34 -0700785 set_mems_allowed(node_states[N_HIGH_MEMORY]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 /*
787 * init can run on any cpu.
788 */
Rusty Russell1a2142a2009-03-30 22:05:10 -0600789 set_cpus_allowed_ptr(current, cpu_all_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 /*
791 * Tell the world that we're going to be the grim
792 * reaper of innocent orphaned children.
793 *
794 * We don't want people to have to make incorrect
795 * assumptions about where in the task array this
796 * can be found.
797 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800798 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700800 cad_pid = task_pid(current);
801
Andi Kleenca74a6f2008-01-30 13:33:17 +0100802 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
804 do_pre_smp_initcalls();
Peter Zijlstra004417a2010-11-25 18:38:29 +0100805 lockup_detector_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 smp_init();
808 sched_init_smp();
809
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810 do_basic_setup();
811
Eric W. Biederman2bd3a9972010-03-02 23:53:19 -0800812 /* Open the /dev/console on the rootfs, this should never fail */
813 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
814 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
815
816 (void) sys_dup(0);
817 (void) sys_dup(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 /*
819 * check if there is an early userspace init. If yes, let it do all
820 * the work
821 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700822
823 if (!ramdisk_execute_command)
824 ramdisk_execute_command = "/init";
825
826 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
827 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700829 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830
831 /*
832 * Ok, we have completed the initial bootup, and
833 * we're essentially up and running. Get rid of the
834 * initmem segments and start the user-mode stuff..
835 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100836
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100837 init_post();
838 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839}