blob: 217ed23e94875d59ffd7b6a95562fcba36112e08 [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 Molnar9b5609f2008-02-14 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 Molnarfbb9ce952006-07-03 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);
Michal Schmidtb1e4d202011-10-10 00:03:37 +0200166 if (parameqn(line, p->str, n)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 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{
Alexander Sverdlin808bf292011-09-21 09:51:40 +0200212 int newlevel;
213
214 /*
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
218 */
219 if (get_option(&str, &newlevel)) {
220 console_loglevel = newlevel;
221 return 0;
222 }
223
224 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
Yinghai Luf6f21c812008-02-08 04:21:58 -0800227early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
229/*
230 * Unknown boot options get handed to init, unless they look like
Rusty Russellf066a4f2009-12-01 14:56:44 +1030231 * unused parameters (modprobe will find them in /proc/cmdline).
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 */
233static int __init unknown_bootoption(char *param, char *val)
234{
235 /* Change NUL term back to "=", to make "param" the whole string. */
236 if (val) {
237 /* param=val or param="val"? */
238 if (val == param+strlen(param)+1)
239 val[-1] = '=';
240 else if (val == param+strlen(param)+2) {
241 val[-2] = '=';
242 memmove(val-1, val, strlen(val)+1);
243 val--;
244 } else
245 BUG();
246 }
247
248 /* Handle obsolete-style parameters */
249 if (obsolete_checksetup(param))
250 return 0;
251
Rusty Russellf066a4f2009-12-01 14:56:44 +1030252 /* Unused module parameter. */
253 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255
256 if (panic_later)
257 return 0;
258
259 if (val) {
260 /* Environment option */
261 unsigned int i;
262 for (i = 0; envp_init[i]; i++) {
263 if (i == MAX_INIT_ENVS) {
264 panic_later = "Too many boot env vars at `%s'";
265 panic_param = param;
266 }
267 if (!strncmp(param, envp_init[i], val - param))
268 break;
269 }
270 envp_init[i] = param;
271 } else {
272 /* Command line option */
273 unsigned int i;
274 for (i = 0; argv_init[i]; i++) {
275 if (i == MAX_INIT_ARGS) {
276 panic_later = "Too many boot init vars at `%s'";
277 panic_param = param;
278 }
279 }
280 argv_init[i] = param;
281 }
282 return 0;
283}
284
Ingo Molnar12d6f212008-01-30 13:33:58 +0100285#ifdef CONFIG_DEBUG_PAGEALLOC
286int __read_mostly debug_pagealloc_enabled = 0;
287#endif
288
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289static int __init init_setup(char *str)
290{
291 unsigned int i;
292
293 execute_command = str;
294 /*
295 * In case LILO is going to boot us with default command line,
296 * it prepends "auto" before the whole cmdline which makes
297 * the shell think it should execute a script with such name.
298 * So we ignore all arguments entered _before_ init=... [MJ]
299 */
300 for (i = 1; i < MAX_INIT_ARGS; i++)
301 argv_init[i] = NULL;
302 return 1;
303}
304__setup("init=", init_setup);
305
Olof Johanssonffdfc402005-09-06 15:17:19 -0700306static int __init rdinit_setup(char *str)
307{
308 unsigned int i;
309
310 ramdisk_execute_command = str;
311 /* See "auto" comment in init_setup */
312 for (i = 1; i < MAX_INIT_ARGS; i++)
313 argv_init[i] = NULL;
314 return 1;
315}
316__setup("rdinit=", rdinit_setup);
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318#ifndef CONFIG_SMP
Amerigo Wang34db18a02011-03-22 16:34:06 -0700319static const unsigned int setup_max_cpus = NR_CPUS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320#ifdef CONFIG_X86_LOCAL_APIC
321static void __init smp_init(void)
322{
323 APIC_init_uniprocessor();
324}
325#else
326#define smp_init() do { } while (0)
327#endif
328
Mike Travise0982e92008-03-26 14:23:48 -0700329static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330static inline void smp_prepare_cpus(unsigned int maxcpus) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331#endif
332
333/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800334 * We need to store the untouched command line for future reference.
335 * We also need to store the touched command line since the parameter
336 * parsing is performed in place, and we should allow a component to
337 * store reference of name/value for future reference.
338 */
339static void __init setup_command_line(char *command_line)
340{
341 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
342 static_command_line = alloc_bootmem(strlen (command_line)+1);
343 strcpy (saved_command_line, boot_command_line);
344 strcpy (static_command_line, command_line);
345}
346
347/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 * We need to finalize in a non-__init function or else race conditions
349 * between the root thread and the init thread may cause start_kernel to
350 * be reaped by free_initmem before the root thread has proceeded to
351 * cpu_idle.
352 *
353 * gcc-3.4 accidentally inlines this function, so use noinline.
354 */
355
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200356static __initdata DECLARE_COMPLETION(kthreadd_done);
357
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800358static noinline void __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700360 int pid;
361
Paul E. McKenney7db905e2009-09-02 14:01:24 -0700362 rcu_scheduler_starting();
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200363 /*
Peter Zijlstra97158562010-06-30 10:37:11 +0200364 * We need to spawn init first so that it obtains pid 1, however
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200365 * the init task will end up wanting to create kthreads, which, if
366 * we schedule it before we create kthreadd, will OOPS.
367 */
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100368 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700370 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800371 rcu_read_lock();
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700372 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800373 rcu_read_unlock();
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200374 complete(&kthreadd_done);
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200375
376 /*
377 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200378 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200379 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200380 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800381 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200382 schedule();
Linus Torvalds288d5ab2011-08-03 22:03:29 -1000383
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800384 /* Call into cpu_idle with preempt disabled */
Linus Torvalds288d5ab2011-08-03 22:03:29 -1000385 preempt_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200387}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389/* Check for early params. */
390static int __init do_early_param(char *param, char *val)
391{
Rusty Russell914dcaa2010-08-11 23:04:18 -0600392 const struct obs_kernel_param *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394 for (p = __setup_start; p < __setup_end; p++) {
Michal Schmidtb1e4d202011-10-10 00:03:37 +0200395 if ((p->early && parameq(param, p->str)) ||
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700396 (strcmp(param, "console") == 0 &&
397 strcmp(p->str, "earlycon") == 0)
398 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 if (p->setup_func(val) != 0)
400 printk(KERN_WARNING
401 "Malformed early option '%s'\n", param);
402 }
403 }
404 /* We accept everything at this stage. */
405 return 0;
406}
407
Magnus Damm13977092009-03-30 14:37:25 -0700408void __init parse_early_options(char *cmdline)
409{
410 parse_args("early options", cmdline, NULL, 0, do_early_param);
411}
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413/* Arch code calls this early on, or if not, just before other parsing. */
414void __init parse_early_param(void)
415{
416 static __initdata int done = 0;
417 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
418
419 if (done)
420 return;
421
422 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800423 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Magnus Damm13977092009-03-30 14:37:25 -0700424 parse_early_options(tmp_cmdline);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 done = 1;
426}
427
428/*
429 * Activate the first processor.
430 */
431
Stas Sergeev44fd2292006-03-23 02:59:44 -0800432static void __init boot_cpu_init(void)
433{
434 int cpu = smp_processor_id();
435 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030436 set_cpu_online(cpu, true);
Peter Zijlstra933b0612009-12-16 18:04:31 +0100437 set_cpu_active(cpu, true);
Rusty Russell915441b2009-01-01 10:12:15 +1030438 set_cpu_present(cpu, true);
439 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800440}
441
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000442void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700443{
444}
445
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000446void __init __weak thread_info_cache_init(void)
447{
448}
449
Pekka Enberg444f4782009-06-11 18:29:06 +0300450/*
451 * Set up kernel memory allocators
452 */
453static void __init mm_init(void)
454{
KAMEZAWA Hiroyukica371c02009-06-12 10:33:53 +0300455 /*
456 * page_cgroup requires countinous pages as memmap
457 * and it's bigger than MAX_ORDER unless SPARSEMEM.
458 */
459 page_cgroup_init_flatmem();
Pekka Enberg444f4782009-06-11 18:29:06 +0300460 mem_init();
461 kmem_cache_init();
Tejun Heo099a19d2010-06-27 18:50:00 +0200462 percpu_init_late();
Benjamin Herrenschmidtc868d552009-06-17 13:48:39 +1000463 pgtable_cache_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300464 vmalloc_init();
465}
466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467asmlinkage void __init start_kernel(void)
468{
469 char * command_line;
Rusty Russell914dcaa2010-08-11 23:04:18 -0600470 extern const struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700471
472 smp_setup_processor_id();
473
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700474 /*
475 * Need to run as early as possible, to initialize the
476 * lockdep hash:
477 */
478 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700479 debug_objects_early_init();
Ingo Molnar42059422008-02-14 09:44:08 +0100480
481 /*
482 * Set up the the initial canary ASAP:
483 */
484 boot_init_stack_canary();
485
Paul Menageddbcc7e2007-10-18 23:39:30 -0700486 cgroup_init_early();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700487
488 local_irq_disable();
Tejun Heo2ce802f2011-01-20 12:06:35 +0100489 early_boot_irqs_disabled = true;
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491/*
492 * Interrupts are still disabled. Do necessary setups, then
493 * enable them
494 */
Thomas Gleixner906568c2007-02-16 01:28:01 -0800495 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800496 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 page_address_init();
Alex Riesen657cafa2009-05-24 15:30:48 +0200498 printk(KERN_NOTICE "%s", linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700500 mm_init_owner(&init_mm, &init_task);
Linus Torvalds6345d242011-05-29 11:32:28 -0700501 mm_init_cpumask(&init_mm);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800502 setup_command_line(command_line);
Mike Travise0982e92008-03-26 14:23:48 -0700503 setup_nr_cpu_ids();
Tejun Heod6647bd2009-07-21 17:11:50 +0900504 setup_per_cpu_areas();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800505 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506
Haicheng Li1f522502010-05-24 14:32:51 -0700507 build_all_zonelists(NULL);
Pekka Enberg83b519e2009-06-10 19:40:04 +0300508 page_alloc_init();
509
510 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
511 parse_early_param();
512 parse_args("Booting kernel", static_command_line, __start___param,
513 __stop___param - __start___param,
514 &unknown_bootoption);
Jeremy Fitzhardinge97ce2c82011-10-12 16:17:54 -0700515
516 jump_label_init();
517
Pekka Enberg83b519e2009-06-10 19:40:04 +0300518 /*
519 * These use large bootmem allocations and must precede
520 * kmem_cache_init()
521 */
Mike Travis162a7e72011-05-24 17:13:20 -0700522 setup_log_buf(0);
Pekka Enberg83b519e2009-06-10 19:40:04 +0300523 pidhash_init();
Pekka Enberg83b519e2009-06-10 19:40:04 +0300524 vfs_caches_init_early();
525 sort_main_extable();
526 trap_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300527 mm_init();
KOSAKI Motohirode03c722011-05-24 17:12:15 -0700528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 /*
530 * Set up the scheduler prior starting any interrupts (such as the
531 * timer interrupt). Full topology setup happens at smp_init()
532 * time - but meanwhile we still have a functioning scheduler.
533 */
534 sched_init();
535 /*
536 * Disable preemption - early bootup scheduling is extremely
537 * fragile until we cpu_idle() for the first time.
538 */
539 preempt_disable();
Ard van Breemenc4a68302007-01-05 16:36:19 -0800540 if (!irqs_disabled()) {
541 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
542 "enabled *very* early, fixing it\n");
543 local_irq_disable();
544 }
Peter Zijlstra9f58a202010-11-17 23:17:35 +0100545 idr_init_cache();
Peter Zijlstra24a24bb2010-11-17 23:17:33 +0100546 perf_event_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 rcu_init();
Yinghai Lu773e3eb2010-02-10 01:20:33 -0800548 radix_tree_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800549 /* init some links before init_ISA_irqs() */
550 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 init_IRQ();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100552 prio_tree_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800554 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700556 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700557 time_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700558 profile_init();
Takao Indohd8ad7d12011-03-29 12:35:04 -0400559 call_function_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700560 if (!irqs_disabled())
Ron Lee24d431d2008-11-27 02:31:57 +1030561 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
562 "enabled early\n");
Tejun Heo2ce802f2011-01-20 12:06:35 +0100563 early_boot_irqs_disabled = false;
Heiko Carstens93e02812006-07-03 00:24:24 -0700564 local_irq_enable();
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000565
566 /* Interrupts are enabled now so all GFP allocations are safe. */
Rafael J. Wysocki452aa692010-03-05 13:42:13 -0800567 gfp_allowed_mask = __GFP_BITS_MASK;
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000568
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300569 kmem_cache_init_late();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
571 /*
572 * HACK ALERT! This is early. We're enabling the console before
573 * we've done PCI setups etc, and console_init() must be aware of
574 * this. But we do want output early, in case something goes wrong.
575 */
576 console_init();
577 if (panic_later)
578 panic(panic_later, panic_param);
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700579
580 lockdep_info();
581
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700582 /*
583 * Need to run this when irqs are enabled, because it wants
584 * to self-test [hard/soft]-irqs on/off lock inversion bugs
585 * too:
586 */
587 locking_selftest();
588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589#ifdef CONFIG_BLK_DEV_INITRD
590 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700591 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200593 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700594 page_to_pfn(virt_to_page((void *)initrd_start)),
595 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 initrd_start = 0;
597 }
598#endif
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700599 page_cgroup_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100600 enable_debug_pagealloc();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700601 debug_objects_mem_init();
Catalin Marinas9b090f22011-05-19 16:25:30 +0100602 kmemleak_init();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700603 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 numa_policy_init();
605 if (late_time_init)
606 late_time_init();
Thomas Gleixnerfa84e9e2009-08-21 22:01:12 +0200607 sched_clock_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 calibrate_delay();
609 pidmap_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 anon_vma_init();
611#ifdef CONFIG_X86
612 if (efi_enabled)
613 efi_enter_virtual_mode();
614#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000615 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100616 cred_init();
Jan Beulich44813742009-09-21 17:03:05 -0700617 fork_init(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 proc_caches_init();
619 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 key_init();
621 security_init();
Jason Wessel0b4b3822010-05-20 21:04:29 -0500622 dbg_late_init();
Jan Beulich44813742009-09-21 17:03:05 -0700623 vfs_caches_init(totalram_pages);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 signals_init();
625 /* rootfs populating might need page-writeback */
626 page_writeback_init();
627#ifdef CONFIG_PROC_FS
628 proc_root_init();
629#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700630 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700632 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700633 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634
635 check_bugs();
636
637 acpi_early_init(); /* before LAPIC and SMP init */
Feng Tang6ae69962009-08-14 15:13:46 -0400638 sfi_init_late();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400640 ftrace_init();
641
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 /* Do the rest non-__init'ed, we're now alive */
643 rest_init();
644}
645
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700646/* Call all constructor functions linked into the kernel. */
647static void __init do_ctors(void)
648{
649#ifdef CONFIG_CONSTRUCTORS
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800650 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700651
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800652 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
653 (*fn)();
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700654#endif
655}
656
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800657int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500658core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
Ingo Molnar4a683bf2009-08-21 12:53:36 +0200660static char msgbuf[64];
Ingo Molnar4a683bf2009-08-21 12:53:36 +0200661
Kevin Winchestere4461272010-08-09 17:20:32 -0700662static int __init_or_module do_one_initcall_debug(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700663{
Frederic Weisbecker74239072008-11-11 23:24:42 +0100664 ktime_t calltime, delta, rettime;
Américo Wang30dbb20e2010-05-26 18:57:53 +0800665 unsigned long long duration;
666 int ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700667
Kevin Winchester22c5c032010-08-09 17:20:32 -0700668 printk(KERN_DEBUG "calling %pF @ %i\n", fn, task_pid_nr(current));
669 calltime = ktime_get();
Américo Wang30dbb20e2010-05-26 18:57:53 +0800670 ret = fn();
Kevin Winchester22c5c032010-08-09 17:20:32 -0700671 rettime = ktime_get();
672 delta = ktime_sub(rettime, calltime);
673 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
674 printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n", fn,
675 ret, duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700676
Kevin Winchester22c5c032010-08-09 17:20:32 -0700677 return ret;
678}
679
Kevin Winchestere4461272010-08-09 17:20:32 -0700680int __init_or_module do_one_initcall(initcall_t fn)
Kevin Winchester22c5c032010-08-09 17:20:32 -0700681{
682 int count = preempt_count();
683 int ret;
684
685 if (initcall_debug)
686 ret = do_one_initcall_debug(fn);
687 else
688 ret = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700689
690 msgbuf[0] = 0;
691
Américo Wang30dbb20e2010-05-26 18:57:53 +0800692 if (ret && ret != -ENODEV && initcall_debug)
693 sprintf(msgbuf, "error code %d ", ret);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700694
695 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700696 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700697 preempt_count() = count;
698 }
699 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700700 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700701 local_irq_enable();
702 }
703 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700704 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700705 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700706
Américo Wang30dbb20e2010-05-26 18:57:53 +0800707 return ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700708}
709
710
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700711extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712
713static void __init do_initcalls(void)
714{
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800715 initcall_t *fn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800717 for (fn = __early_initcall_end; fn < __initcall_end; fn++)
718 do_one_initcall(*fn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719}
720
721/*
722 * Ok, the machine is now initialized. None of the devices
723 * have been touched yet, but the CPU subsystem is up and
724 * running, and memory and process management works.
725 *
726 * Now we can finally start doing some real work..
727 */
728static void __init do_basic_setup(void)
729{
Lai Jiangshan759ee092009-03-25 17:06:30 +0800730 cpuset_init_smp();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 usermodehelper_init();
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700732 shmem_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800734 init_irq_proc();
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700735 do_ctors();
Linus Torvaldsd5767c52011-09-28 10:23:44 -0700736 usermodehelper_enable();
wangyanqingb0f84372011-09-29 15:09:40 +0800737 do_initcalls();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738}
739
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700740static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700741{
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800742 initcall_t *fn;
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700743
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800744 for (fn = __initcall_start; fn < __early_initcall_end; fn++)
745 do_one_initcall(*fn);
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700746}
747
David Howellsd7627462010-08-17 23:52:56 +0100748static void run_init_process(const char *init_filename)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749{
750 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700751 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752}
753
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100754/* This is a non __init function. Force it to be noinline otherwise gcc
755 * makes it inline to init() and it becomes part of init.text section
756 */
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800757static noinline int init_post(void)
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100758{
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800759 /* need to finish all async __init code before freeing the memory */
760 async_synchronize_full();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100761 free_initmem();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100762 mark_rodata_ro();
763 system_state = SYSTEM_RUNNING;
764 numa_default_policy();
765
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100766
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700767 current->signal->flags |= SIGNAL_UNKILLABLE;
768
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100769 if (ramdisk_execute_command) {
770 run_init_process(ramdisk_execute_command);
771 printk(KERN_WARNING "Failed to execute %s\n",
772 ramdisk_execute_command);
773 }
774
775 /*
776 * We try each of these until one succeeds.
777 *
778 * The Bourne shell can be used instead of init if we are
779 * trying to recover a really broken machine.
780 */
781 if (execute_command) {
782 run_init_process(execute_command);
783 printk(KERN_WARNING "Failed to execute %s. Attempting "
784 "defaults...\n", execute_command);
785 }
786 run_init_process("/sbin/init");
787 run_init_process("/etc/init");
788 run_init_process("/bin/init");
789 run_init_process("/bin/sh");
790
Andreas Mohr9a85b8d2010-03-05 13:42:39 -0800791 panic("No init found. Try passing init= option to kernel. "
792 "See Linux Documentation/init.txt for guidance.");
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100793}
794
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100795static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796{
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200797 /*
798 * Wait until kthreadd is all set-up.
799 */
800 wait_for_completion(&kthreadd_done);
Miao Xie58568d22009-06-16 15:31:49 -0700801 /*
802 * init can allocate pages on any node
803 */
Miao Xie5ab116c2010-03-23 13:35:34 -0700804 set_mems_allowed(node_states[N_HIGH_MEMORY]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 /*
806 * init can run on any cpu.
807 */
Rusty Russell1a2142a2009-03-30 22:05:10 -0600808 set_cpus_allowed_ptr(current, cpu_all_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700810 cad_pid = task_pid(current);
811
Andi Kleenca74a6f2008-01-30 13:33:17 +0100812 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813
814 do_pre_smp_initcalls();
Peter Zijlstra004417a2010-11-25 18:38:29 +0100815 lockup_detector_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817 smp_init();
818 sched_init_smp();
819
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 do_basic_setup();
821
Eric W. Biederman2bd3a9972010-03-02 23:53:19 -0800822 /* Open the /dev/console on the rootfs, this should never fail */
823 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
824 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
825
826 (void) sys_dup(0);
827 (void) sys_dup(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 /*
829 * check if there is an early userspace init. If yes, let it do all
830 * the work
831 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700832
833 if (!ramdisk_execute_command)
834 ramdisk_execute_command = "/init";
835
836 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
837 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700838 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700839 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 /*
842 * Ok, we have completed the initial bootup, and
843 * we're essentially up and running. Get rid of the
844 * initmem segments and start the user-mode stuff..
845 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100846
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100847 init_post();
848 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849}