blob: 7917695bf71eeec3ba1fba336fd7631d8c02a233 [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>
21#include <linux/utsname.h>
22#include <linux/ioport.h>
23#include <linux/init.h>
24#include <linux/smp_lock.h>
25#include <linux/initrd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/bootmem.h>
27#include <linux/tty.h>
28#include <linux/gfp.h>
29#include <linux/percpu.h>
30#include <linux/kmod.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070031#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/kernel_stat.h>
Rusty Russelld7cd5612006-12-07 02:14:08 +010033#include <linux/start_kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/security.h>
Jens Axboe3d442232008-06-26 11:21:34 +020035#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/workqueue.h>
37#include <linux/profile.h>
38#include <linux/rcupdate.h>
39#include <linux/moduleparam.h>
40#include <linux/kallsyms.h>
41#include <linux/writeback.h>
42#include <linux/cpu.h>
43#include <linux/cpuset.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070044#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/efi.h>
Thomas Gleixner906568c2007-02-16 01:28:01 -080046#include <linux/tick.h>
Andrew Morton6168a702007-02-17 21:22:39 -080047#include <linux/interrupt.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070048#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070049#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/unistd.h>
51#include <linux/rmap.h>
52#include <linux/mempolicy.h>
53#include <linux/key.h>
Adrian Bunkb6cd0b72006-06-27 02:53:54 -070054#include <linux/buffer_head.h>
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -070055#include <linux/page_cgroup.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070056#include <linux/debug_locks.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070057#include <linux/debugobjects.h>
Ingo Molnarfbb9ce952006-07-03 00:24:50 -070058#include <linux/lockdep.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080059#include <linux/pid_namespace.h>
Adrian Bunk1f217822006-12-19 13:01:28 -080060#include <linux/device.h>
Eric W. Biederman73c27992007-05-09 02:34:32 -070061#include <linux/kthread.h>
Adrian Bunke6fe6642007-11-09 22:39:39 +010062#include <linux/sched.h>
Adrian Bunka1c9eea2008-02-06 01:36:44 -080063#include <linux/signal.h>
Akinobu Mita199f0ca2008-04-29 01:03:13 -070064#include <linux/idr.h>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040065#include <linux/ftrace.h>
Arjan van de Ven22a9d642009-01-07 08:45:46 -080066#include <linux/async.h>
Zhaolei02af61b2009-04-10 14:26:18 +080067#include <linux/kmemtrace.h>
Frederic Weisbecker3f5ec132008-11-11 23:21:31 +010068#include <trace/boot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70#include <asm/io.h>
71#include <asm/bugs.h>
72#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070073#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080074#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
Linus Torvalds1da177e2005-04-16 15:20:36 -070076#ifdef CONFIG_X86_LOCAL_APIC
77#include <asm/smp.h>
78#endif
79
Sam Ravnborgaae5f662007-02-26 16:45:41 +010080static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083extern void fork_init(unsigned long);
84extern void mca_init(void);
85extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086extern void prio_tree_init(void);
87extern void radix_tree_init(void);
88extern void free_initmem(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089#ifdef CONFIG_ACPI
90extern void acpi_early_init(void);
91#else
92static inline void acpi_early_init(void) { }
93#endif
Arjan van de Ven37b73c82006-01-06 00:12:01 -080094#ifndef CONFIG_DEBUG_RODATA
95static inline void mark_rodata_ro(void) { }
96#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98#ifdef CONFIG_TC
99extern void tc_init(void);
100#endif
101
Paul E. McKenneya6826042009-02-25 18:03:42 -0800102enum system_states system_state __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103EXPORT_SYMBOL(system_state);
104
105/*
106 * Boot command-line arguments
107 */
108#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
109#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
110
111extern void time_init(void);
112/* Default late time init is NULL. archs can override this later. */
Jan Beulichd2e31922009-01-06 14:41:10 -0800113void (*__initdata late_time_init)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114extern void softirq_init(void);
115
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800116/* Untouched command line saved by arch-specific code. */
117char __initdata boot_command_line[COMMAND_LINE_SIZE];
118/* Untouched saved command line (eg. for /proc) */
119char *saved_command_line;
120/* Command line for parameter parsing */
121static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
123static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700124static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Jan Beulich8b3b2952007-07-15 23:41:07 -0700126#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100128unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700129
130/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 * Setup routine for controlling SMP activation
132 *
133 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
134 * activation entirely (the MPS table probe still happens, though).
135 *
136 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
137 * greater than 0, limits the maximum number of CPUs activated in
138 * SMP mode to <NUM>.
139 */
Ingo Molnar65a4e572009-01-31 03:36:17 +0100140
141void __weak arch_disable_smp_support(void) { }
Len Brown61ec7562007-08-16 03:34:22 -0400142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static int __init nosmp(char *str)
144{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100145 setup_max_cpus = 0;
Ingo Molnar65a4e572009-01-31 03:36:17 +0100146 arch_disable_smp_support();
147
Jan Beulich8b3b2952007-07-15 23:41:07 -0700148 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
Jan Beulich8b3b2952007-07-15 23:41:07 -0700151early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
153static int __init maxcpus(char *str)
154{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100155 get_option(&str, &setup_max_cpus);
156 if (setup_max_cpus == 0)
Ingo Molnar65a4e572009-01-31 03:36:17 +0100157 arch_disable_smp_support();
Len Brown61ec7562007-08-16 03:34:22 -0400158
159 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160}
161
Hugh Dickins81340972007-08-27 16:02:12 +0100162early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700163#else
Ingo Molnar65a4e572009-01-31 03:36:17 +0100164const unsigned int setup_max_cpus = NR_CPUS;
Jan Beulich8b3b2952007-07-15 23:41:07 -0700165#endif
166
167/*
168 * If set, this is an indication to the drivers that reset the underlying
169 * device before going ahead with the initialization otherwise driver might
170 * rely on the BIOS and skip the reset operation.
171 *
172 * This is useful if kernel is booting in an unreliable environment.
173 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
174 * skipped and devices will be in unknown state.
175 */
176unsigned int reset_devices;
177EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Vivek Goyal7e962872006-09-27 01:50:44 -0700179static int __init set_reset_devices(char *str)
180{
181 reset_devices = 1;
182 return 1;
183}
184
185__setup("reset_devices", set_reset_devices);
186
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
188char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
189static const char *panic_later, *panic_param;
190
191extern struct obs_kernel_param __setup_start[], __setup_end[];
192
193static int __init obsolete_checksetup(char *line)
194{
195 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200196 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
198 p = __setup_start;
199 do {
200 int n = strlen(p->str);
201 if (!strncmp(line, p->str, n)) {
202 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200203 /* Already done in parse_early_param?
204 * (Needs exact match on param part).
205 * Keep iterating, as we can have early
206 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200208 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 } else if (!p->setup_func) {
210 printk(KERN_WARNING "Parameter %s is obsolete,"
211 " ignored\n", p->str);
212 return 1;
213 } else if (p->setup_func(line + n))
214 return 1;
215 }
216 p++;
217 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200218
219 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220}
221
222/*
223 * This should be approx 2 Bo*oMips to start (note initial shift), and will
224 * still work even if initially too large, it will just take slightly longer
225 */
226unsigned long loops_per_jiffy = (1<<12);
227
228EXPORT_SYMBOL(loops_per_jiffy);
229
230static int __init debug_kernel(char *str)
231{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800233 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234}
235
236static int __init quiet_kernel(char *str)
237{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800239 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240}
241
Yinghai Luf6f21c812008-02-08 04:21:58 -0800242early_param("debug", debug_kernel);
243early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245static int __init loglevel(char *str)
246{
247 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800248 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249}
250
Yinghai Luf6f21c812008-02-08 04:21:58 -0800251early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252
253/*
254 * Unknown boot options get handed to init, unless they look like
255 * failed parameters
256 */
257static int __init unknown_bootoption(char *param, char *val)
258{
259 /* Change NUL term back to "=", to make "param" the whole string. */
260 if (val) {
261 /* param=val or param="val"? */
262 if (val == param+strlen(param)+1)
263 val[-1] = '=';
264 else if (val == param+strlen(param)+2) {
265 val[-2] = '=';
266 memmove(val-1, val, strlen(val)+1);
267 val--;
268 } else
269 BUG();
270 }
271
272 /* Handle obsolete-style parameters */
273 if (obsolete_checksetup(param))
274 return 0;
275
276 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200277 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 * line work?"
279 */
280 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
281 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
282 return 0;
283 }
284
285 if (panic_later)
286 return 0;
287
288 if (val) {
289 /* Environment option */
290 unsigned int i;
291 for (i = 0; envp_init[i]; i++) {
292 if (i == MAX_INIT_ENVS) {
293 panic_later = "Too many boot env vars at `%s'";
294 panic_param = param;
295 }
296 if (!strncmp(param, envp_init[i], val - param))
297 break;
298 }
299 envp_init[i] = param;
300 } else {
301 /* Command line option */
302 unsigned int i;
303 for (i = 0; argv_init[i]; i++) {
304 if (i == MAX_INIT_ARGS) {
305 panic_later = "Too many boot init vars at `%s'";
306 panic_param = param;
307 }
308 }
309 argv_init[i] = param;
310 }
311 return 0;
312}
313
Ingo Molnar12d6f212008-01-30 13:33:58 +0100314#ifdef CONFIG_DEBUG_PAGEALLOC
315int __read_mostly debug_pagealloc_enabled = 0;
316#endif
317
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318static int __init init_setup(char *str)
319{
320 unsigned int i;
321
322 execute_command = str;
323 /*
324 * In case LILO is going to boot us with default command line,
325 * it prepends "auto" before the whole cmdline which makes
326 * the shell think it should execute a script with such name.
327 * So we ignore all arguments entered _before_ init=... [MJ]
328 */
329 for (i = 1; i < MAX_INIT_ARGS; i++)
330 argv_init[i] = NULL;
331 return 1;
332}
333__setup("init=", init_setup);
334
Olof Johanssonffdfc402005-09-06 15:17:19 -0700335static int __init rdinit_setup(char *str)
336{
337 unsigned int i;
338
339 ramdisk_execute_command = str;
340 /* See "auto" comment in init_setup */
341 for (i = 1; i < MAX_INIT_ARGS; i++)
342 argv_init[i] = NULL;
343 return 1;
344}
345__setup("rdinit=", rdinit_setup);
346
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347#ifndef CONFIG_SMP
348
349#ifdef CONFIG_X86_LOCAL_APIC
350static void __init smp_init(void)
351{
352 APIC_init_uniprocessor();
353}
354#else
355#define smp_init() do { } while (0)
356#endif
357
358static inline void setup_per_cpu_areas(void) { }
Mike Travise0982e92008-03-26 14:23:48 -0700359static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360static inline void smp_prepare_cpus(unsigned int maxcpus) { }
361
362#else
363
Mike Travis321a8e92008-04-04 18:11:02 -0700364#if NR_CPUS > BITS_PER_LONG
365cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
366EXPORT_SYMBOL(cpu_mask_all);
367#endif
368
Mike Travise0982e92008-03-26 14:23:48 -0700369/* Setup number of possible processor ids */
370int nr_cpu_ids __read_mostly = NR_CPUS;
371EXPORT_SYMBOL(nr_cpu_ids);
372
373/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
374static void __init setup_nr_cpu_ids(void)
375{
Rusty Russelle0c0ba72009-01-01 10:12:19 +1030376 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
Mike Travise0982e92008-03-26 14:23:48 -0700377}
378
Mike Travisdd5af902008-01-30 13:33:32 +0100379#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800380unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
382EXPORT_SYMBOL(__per_cpu_offset);
383
384static void __init setup_per_cpu_areas(void)
385{
386 unsigned long size, i;
387 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800388 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200391 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
392 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800394 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 __per_cpu_offset[i] = ptr - __per_cpu_start;
396 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800397 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398 }
399}
Mike Travisdd5af902008-01-30 13:33:32 +0100400#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
402/* Called by boot processor to activate the rest. */
403static void __init smp_init(void)
404{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800405 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Max Krasnyanskye761b772008-07-15 04:43:49 -0700407 /*
408 * Set up the current CPU as possible to migrate to.
409 * The other ones will be done by cpu_up/cpu_down()
410 */
Rusty Russell2b17fa52009-03-30 22:05:12 -0600411 set_cpu_active(smp_processor_id(), true);
Max Krasnyanskye761b772008-07-15 04:43:49 -0700412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800414 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100415 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800417 if (!cpu_online(cpu))
418 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 }
420
421 /* Any cleanup work */
422 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100423 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
426#endif
427
428/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800429 * We need to store the untouched command line for future reference.
430 * We also need to store the touched command line since the parameter
431 * parsing is performed in place, and we should allow a component to
432 * store reference of name/value for future reference.
433 */
434static void __init setup_command_line(char *command_line)
435{
436 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
437 static_command_line = alloc_bootmem(strlen (command_line)+1);
438 strcpy (saved_command_line, boot_command_line);
439 strcpy (static_command_line, command_line);
440}
441
442/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 * We need to finalize in a non-__init function or else race conditions
444 * between the root thread and the init thread may cause start_kernel to
445 * be reaped by free_initmem before the root thread has proceeded to
446 * cpu_idle.
447 *
448 * gcc-3.4 accidentally inlines this function, so use noinline.
449 */
450
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800451static noinline void __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 __releases(kernel_lock)
453{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700454 int pid;
455
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100456 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700458 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700459 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200461
462 /*
463 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200464 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200465 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200466 init_idle_bootup_task(current);
Paul E. McKenneya6826042009-02-25 18:03:42 -0800467 rcu_scheduler_starting();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800468 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200469 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800470 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200471
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800472 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200474}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475
476/* Check for early params. */
477static int __init do_early_param(char *param, char *val)
478{
479 struct obs_kernel_param *p;
480
481 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700482 if ((p->early && strcmp(param, p->str) == 0) ||
483 (strcmp(param, "console") == 0 &&
484 strcmp(p->str, "earlycon") == 0)
485 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 if (p->setup_func(val) != 0)
487 printk(KERN_WARNING
488 "Malformed early option '%s'\n", param);
489 }
490 }
491 /* We accept everything at this stage. */
492 return 0;
493}
494
Magnus Damm13977092009-03-30 14:37:25 -0700495void __init parse_early_options(char *cmdline)
496{
497 parse_args("early options", cmdline, NULL, 0, do_early_param);
498}
499
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500/* Arch code calls this early on, or if not, just before other parsing. */
501void __init parse_early_param(void)
502{
503 static __initdata int done = 0;
504 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
505
506 if (done)
507 return;
508
509 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800510 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Magnus Damm13977092009-03-30 14:37:25 -0700511 parse_early_options(tmp_cmdline);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 done = 1;
513}
514
515/*
516 * Activate the first processor.
517 */
518
Stas Sergeev44fd2292006-03-23 02:59:44 -0800519static void __init boot_cpu_init(void)
520{
521 int cpu = smp_processor_id();
522 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030523 set_cpu_online(cpu, true);
524 set_cpu_present(cpu, true);
525 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800526}
527
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000528void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700529{
530}
531
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000532void __init __weak thread_info_cache_init(void)
533{
534}
535
Pekka Enberg444f4782009-06-11 18:29:06 +0300536/*
537 * Set up kernel memory allocators
538 */
539static void __init mm_init(void)
540{
541 mem_init();
542 kmem_cache_init();
543 vmalloc_init();
544}
545
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546asmlinkage void __init start_kernel(void)
547{
548 char * command_line;
549 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700550
551 smp_setup_processor_id();
552
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700553 /*
554 * Need to run as early as possible, to initialize the
555 * lockdep hash:
556 */
557 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700558 debug_objects_early_init();
Ingo Molnar42059422008-02-14 09:44:08 +0100559
560 /*
561 * Set up the the initial canary ASAP:
562 */
563 boot_init_stack_canary();
564
Paul Menageddbcc7e2007-10-18 23:39:30 -0700565 cgroup_init_early();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700566
567 local_irq_disable();
568 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700569 early_init_irq_lock_class();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571/*
572 * Interrupts are still disabled. Do necessary setups, then
573 * enable them
574 */
575 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800576 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800577 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 page_address_init();
Alex Riesen657cafa2009-05-24 15:30:48 +0200579 printk(KERN_NOTICE "%s", linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700581 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800582 setup_command_line(command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700584 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800585 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
Pekka Enberg83b519e2009-06-10 19:40:04 +0300587 build_all_zonelists();
588 page_alloc_init();
589
590 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
591 parse_early_param();
592 parse_args("Booting kernel", static_command_line, __start___param,
593 __stop___param - __start___param,
594 &unknown_bootoption);
595 /*
596 * These use large bootmem allocations and must precede
597 * kmem_cache_init()
598 */
599 pidhash_init();
Pekka Enberg83b519e2009-06-10 19:40:04 +0300600 vfs_caches_init_early();
601 sort_main_extable();
602 trap_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300603 mm_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 /*
605 * Set up the scheduler prior starting any interrupts (such as the
606 * timer interrupt). Full topology setup happens at smp_init()
607 * time - but meanwhile we still have a functioning scheduler.
608 */
609 sched_init();
610 /*
611 * Disable preemption - early bootup scheduling is extremely
612 * fragile until we cpu_idle() for the first time.
613 */
614 preempt_disable();
Ard van Breemenc4a68302007-01-05 16:36:19 -0800615 if (!irqs_disabled()) {
616 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
617 "enabled *very* early, fixing it\n");
618 local_irq_disable();
619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 rcu_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800621 /* init some links before init_ISA_irqs() */
622 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 init_IRQ();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800625 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700627 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700628 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200629 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700630 profile_init();
631 if (!irqs_disabled())
Ron Lee24d431d2008-11-27 02:31:57 +1030632 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
633 "enabled early\n");
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700634 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700635 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636
637 /*
638 * HACK ALERT! This is early. We're enabling the console before
639 * we've done PCI setups etc, and console_init() must be aware of
640 * this. But we do want output early, in case something goes wrong.
641 */
642 console_init();
643 if (panic_later)
644 panic(panic_later, panic_param);
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700645
646 lockdep_info();
647
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700648 /*
649 * Need to run this when irqs are enabled, because it wants
650 * to self-test [hard/soft]-irqs on/off lock inversion bugs
651 * too:
652 */
653 locking_selftest();
654
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655#ifdef CONFIG_BLK_DEV_INITRD
656 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700657 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200659 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700660 page_to_pfn(virt_to_page((void *)initrd_start)),
661 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 initrd_start = 0;
663 }
664#endif
Paul Jacksonc417f022006-01-08 01:02:01 -0800665 cpuset_init_early();
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700666 page_cgroup_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100667 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100668 cpu_hotplug_init();
Eduard - Gabriel Munteanub9ce08c2008-08-10 20:14:03 +0300669 kmemtrace_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700670 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700671 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700672 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 numa_policy_init();
674 if (late_time_init)
675 late_time_init();
676 calibrate_delay();
677 pidmap_init();
678 pgtable_cache_init();
679 prio_tree_init();
680 anon_vma_init();
681#ifdef CONFIG_X86
682 if (efi_enabled)
683 efi_enter_virtual_mode();
684#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000685 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100686 cred_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 fork_init(num_physpages);
688 proc_caches_init();
689 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 key_init();
691 security_init();
692 vfs_caches_init(num_physpages);
693 radix_tree_init();
694 signals_init();
695 /* rootfs populating might need page-writeback */
696 page_writeback_init();
697#ifdef CONFIG_PROC_FS
698 proc_root_init();
699#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700700 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700702 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700703 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
705 check_bugs();
706
707 acpi_early_init(); /* before LAPIC and SMP init */
708
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400709 ftrace_init();
710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 /* Do the rest non-__init'ed, we're now alive */
712 rest_init();
713}
714
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800715int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500716core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717
Arjan van de Ven59f94152008-07-30 12:49:02 -0700718int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700719{
720 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100721 ktime_t calltime, delta, rettime;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700722 char msgbuf[64];
Frederic Weisbecker74239072008-11-11 23:24:42 +0100723 struct boot_trace_call call;
724 struct boot_trace_ret ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700725
726 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100727 call.caller = task_pid_nr(current);
728 printk("calling %pF @ %i\n", fn, call.caller);
729 calltime = ktime_get();
730 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100731 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700732 }
733
Frederic Weisbecker74239072008-11-11 23:24:42 +0100734 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700735
736 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100737 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100738 rettime = ktime_get();
739 delta = ktime_sub(rettime, calltime);
Will Newton1d926f22008-11-21 14:08:59 -0800740 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Frederic Weisbecker74239072008-11-11 23:24:42 +0100741 trace_boot_ret(&ret, fn);
Tim Birdca538f62008-10-09 15:23:05 -0700742 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbecker74239072008-11-11 23:24:42 +0100743 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700744 }
745
746 msgbuf[0] = 0;
747
Frederic Weisbecker74239072008-11-11 23:24:42 +0100748 if (ret.result && ret.result != -ENODEV && initcall_debug)
749 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700750
751 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700752 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700753 preempt_count() = count;
754 }
755 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700756 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700757 local_irq_enable();
758 }
759 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700760 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700761 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700762
Frederic Weisbecker74239072008-11-11 23:24:42 +0100763 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700764}
765
766
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700767extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
769static void __init do_initcalls(void)
770{
771 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700773 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700774 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
776 /* Make sure there is no pending stuff from the initcall sequence */
777 flush_scheduled_work();
778}
779
780/*
781 * Ok, the machine is now initialized. None of the devices
782 * have been touched yet, but the CPU subsystem is up and
783 * running, and memory and process management works.
784 *
785 * Now we can finally start doing some real work..
786 */
787static void __init do_basic_setup(void)
788{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200789 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700790 init_workqueues();
Lai Jiangshan759ee092009-03-25 17:06:30 +0800791 cpuset_init_smp();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 usermodehelper_init();
793 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800794 init_irq_proc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 do_initcalls();
796}
797
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700798static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700799{
800 initcall_t *call;
801
802 for (call = __initcall_start; call < __early_initcall_end; call++)
803 do_one_initcall(*call);
804}
805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806static void run_init_process(char *init_filename)
807{
808 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700809 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
811
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100812/* This is a non __init function. Force it to be noinline otherwise gcc
813 * makes it inline to init() and it becomes part of init.text section
814 */
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800815static noinline int init_post(void)
Hannes Ederacdd0522009-03-31 15:23:50 -0700816 __releases(kernel_lock)
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100817{
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800818 /* need to finish all async __init code before freeing the memory */
819 async_synchronize_full();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100820 free_initmem();
821 unlock_kernel();
822 mark_rodata_ro();
823 system_state = SYSTEM_RUNNING;
824 numa_default_policy();
825
826 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
827 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
828
829 (void) sys_dup(0);
830 (void) sys_dup(0);
831
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700832 current->signal->flags |= SIGNAL_UNKILLABLE;
833
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100834 if (ramdisk_execute_command) {
835 run_init_process(ramdisk_execute_command);
836 printk(KERN_WARNING "Failed to execute %s\n",
837 ramdisk_execute_command);
838 }
839
840 /*
841 * We try each of these until one succeeds.
842 *
843 * The Bourne shell can be used instead of init if we are
844 * trying to recover a really broken machine.
845 */
846 if (execute_command) {
847 run_init_process(execute_command);
848 printk(KERN_WARNING "Failed to execute %s. Attempting "
849 "defaults...\n", execute_command);
850 }
851 run_init_process("/sbin/init");
852 run_init_process("/etc/init");
853 run_init_process("/bin/init");
854 run_init_process("/bin/sh");
855
856 panic("No init found. Try passing init= option to kernel.");
857}
858
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100859static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
861 lock_kernel();
862 /*
863 * init can run on any cpu.
864 */
Rusty Russell1a2142a2009-03-30 22:05:10 -0600865 set_cpus_allowed_ptr(current, cpu_all_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 /*
867 * Tell the world that we're going to be the grim
868 * reaper of innocent orphaned children.
869 *
870 * We don't want people to have to make incorrect
871 * assumptions about where in the task array this
872 * can be found.
873 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800874 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700876 cad_pid = task_pid(current);
877
Andi Kleenca74a6f2008-01-30 13:33:17 +0100878 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100881 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 smp_init();
884 sched_init_smp();
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 do_basic_setup();
887
888 /*
889 * check if there is an early userspace init. If yes, let it do all
890 * the work
891 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700892
893 if (!ramdisk_execute_command)
894 ramdisk_execute_command = "/init";
895
896 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
897 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700899 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901 /*
902 * Ok, we have completed the initial bootup, and
903 * we're essentially up and running. Get rid of the
904 * initmem segments and start the user-mode stuff..
905 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100906
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100907 init_post();
908 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909}