blob: 7e117a231af10313f1b9bd963bf404eecaf94c9e [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>
17#include <linux/string.h>
18#include <linux/ctype.h>
19#include <linux/delay.h>
20#include <linux/utsname.h>
21#include <linux/ioport.h>
22#include <linux/init.h>
23#include <linux/smp_lock.h>
24#include <linux/initrd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/bootmem.h>
26#include <linux/tty.h>
27#include <linux/gfp.h>
28#include <linux/percpu.h>
29#include <linux/kmod.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070030#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/kernel_stat.h>
Rusty Russelld7cd5612006-12-07 02:14:08 +010032#include <linux/start_kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/security.h>
Jens Axboe3d442232008-06-26 11:21:34 +020034#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/workqueue.h>
36#include <linux/profile.h>
37#include <linux/rcupdate.h>
38#include <linux/moduleparam.h>
39#include <linux/kallsyms.h>
40#include <linux/writeback.h>
41#include <linux/cpu.h>
42#include <linux/cpuset.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070043#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/efi.h>
Thomas Gleixner906568c2007-02-16 01:28:01 -080045#include <linux/tick.h>
Andrew Morton6168a702007-02-17 21:22:39 -080046#include <linux/interrupt.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070047#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070048#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/unistd.h>
50#include <linux/rmap.h>
51#include <linux/mempolicy.h>
52#include <linux/key.h>
Jan Beulich4552d5d2006-06-26 13:57:28 +020053#include <linux/unwind.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 Molnarfbb9ce92006-07-03 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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67#include <asm/io.h>
68#include <asm/bugs.h>
69#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070070#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080071#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#ifdef CONFIG_X86_LOCAL_APIC
74#include <asm/smp.h>
75#endif
76
77/*
Coywolf Qi Hunt69c99ac2006-01-14 13:21:11 -080078 * This is one of the first .c files built. Error out early if we have compiler
79 * trouble.
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
Andrew Mortonf1a60db2006-12-06 20:39:59 -080082#if __GNUC__ == 4 && __GNUC_MINOR__ == 1 && __GNUC_PATCHLEVEL__ == 0
83#warning gcc-4.1.0 is known to miscompile the kernel. A different compiler version is recommended.
84#endif
85
Sam Ravnborgaae5f662007-02-26 16:45:41 +010086static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089extern void fork_init(unsigned long);
90extern void mca_init(void);
91extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092extern void prio_tree_init(void);
93extern void radix_tree_init(void);
94extern void free_initmem(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095#ifdef CONFIG_ACPI
96extern void acpi_early_init(void);
97#else
98static inline void acpi_early_init(void) { }
99#endif
Arjan van de Ven37b73c82006-01-06 00:12:01 -0800100#ifndef CONFIG_DEBUG_RODATA
101static inline void mark_rodata_ro(void) { }
102#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
104#ifdef CONFIG_TC
105extern void tc_init(void);
106#endif
107
108enum system_states system_state;
109EXPORT_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. */
119void (*late_time_init)(void);
120extern 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#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100134unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700135
136/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 * Setup routine for controlling SMP activation
138 *
139 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
140 * activation entirely (the MPS table probe still happens, though).
141 *
142 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
143 * greater than 0, limits the maximum number of CPUs activated in
144 * SMP mode to <NUM>.
145 */
Len Brown61ec7562007-08-16 03:34:22 -0400146#ifndef CONFIG_X86_IO_APIC
147static inline void disable_ioapic_setup(void) {};
148#endif
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150static int __init nosmp(char *str)
151{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100152 setup_max_cpus = 0;
Len Brown61ec7562007-08-16 03:34:22 -0400153 disable_ioapic_setup();
Jan Beulich8b3b2952007-07-15 23:41:07 -0700154 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155}
156
Jan Beulich8b3b2952007-07-15 23:41:07 -0700157early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
159static int __init maxcpus(char *str)
160{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100161 get_option(&str, &setup_max_cpus);
162 if (setup_max_cpus == 0)
Len Brown61ec7562007-08-16 03:34:22 -0400163 disable_ioapic_setup();
164
165 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167
Hugh Dickins81340972007-08-27 16:02:12 +0100168early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700169#else
Andi Kleenca74a6f2008-01-30 13:33:17 +0100170#define setup_max_cpus NR_CPUS
Jan Beulich8b3b2952007-07-15 23:41:07 -0700171#endif
172
173/*
174 * If set, this is an indication to the drivers that reset the underlying
175 * device before going ahead with the initialization otherwise driver might
176 * rely on the BIOS and skip the reset operation.
177 *
178 * This is useful if kernel is booting in an unreliable environment.
179 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
180 * skipped and devices will be in unknown state.
181 */
182unsigned int reset_devices;
183EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Vivek Goyal7e962872006-09-27 01:50:44 -0700185static int __init set_reset_devices(char *str)
186{
187 reset_devices = 1;
188 return 1;
189}
190
191__setup("reset_devices", set_reset_devices);
192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
194char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
195static const char *panic_later, *panic_param;
196
197extern struct obs_kernel_param __setup_start[], __setup_end[];
198
199static int __init obsolete_checksetup(char *line)
200{
201 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200202 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204 p = __setup_start;
205 do {
206 int n = strlen(p->str);
207 if (!strncmp(line, p->str, n)) {
208 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200209 /* Already done in parse_early_param?
210 * (Needs exact match on param part).
211 * Keep iterating, as we can have early
212 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200214 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 } else if (!p->setup_func) {
216 printk(KERN_WARNING "Parameter %s is obsolete,"
217 " ignored\n", p->str);
218 return 1;
219 } else if (p->setup_func(line + n))
220 return 1;
221 }
222 p++;
223 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200224
225 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226}
227
228/*
229 * This should be approx 2 Bo*oMips to start (note initial shift), and will
230 * still work even if initially too large, it will just take slightly longer
231 */
232unsigned long loops_per_jiffy = (1<<12);
233
234EXPORT_SYMBOL(loops_per_jiffy);
235
236static int __init debug_kernel(char *str)
237{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800239 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240}
241
242static int __init quiet_kernel(char *str)
243{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800245 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
Yinghai Luf6f21c812008-02-08 04:21:58 -0800248early_param("debug", debug_kernel);
249early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
251static int __init loglevel(char *str)
252{
253 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800254 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255}
256
Yinghai Luf6f21c812008-02-08 04:21:58 -0800257early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259/*
260 * Unknown boot options get handed to init, unless they look like
261 * failed parameters
262 */
263static int __init unknown_bootoption(char *param, char *val)
264{
265 /* Change NUL term back to "=", to make "param" the whole string. */
266 if (val) {
267 /* param=val or param="val"? */
268 if (val == param+strlen(param)+1)
269 val[-1] = '=';
270 else if (val == param+strlen(param)+2) {
271 val[-2] = '=';
272 memmove(val-1, val, strlen(val)+1);
273 val--;
274 } else
275 BUG();
276 }
277
278 /* Handle obsolete-style parameters */
279 if (obsolete_checksetup(param))
280 return 0;
281
282 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200283 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 * line work?"
285 */
286 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
287 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
288 return 0;
289 }
290
291 if (panic_later)
292 return 0;
293
294 if (val) {
295 /* Environment option */
296 unsigned int i;
297 for (i = 0; envp_init[i]; i++) {
298 if (i == MAX_INIT_ENVS) {
299 panic_later = "Too many boot env vars at `%s'";
300 panic_param = param;
301 }
302 if (!strncmp(param, envp_init[i], val - param))
303 break;
304 }
305 envp_init[i] = param;
306 } else {
307 /* Command line option */
308 unsigned int i;
309 for (i = 0; argv_init[i]; i++) {
310 if (i == MAX_INIT_ARGS) {
311 panic_later = "Too many boot init vars at `%s'";
312 panic_param = param;
313 }
314 }
315 argv_init[i] = param;
316 }
317 return 0;
318}
319
Ingo Molnar12d6f212008-01-30 13:33:58 +0100320#ifdef CONFIG_DEBUG_PAGEALLOC
321int __read_mostly debug_pagealloc_enabled = 0;
322#endif
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324static int __init init_setup(char *str)
325{
326 unsigned int i;
327
328 execute_command = str;
329 /*
330 * In case LILO is going to boot us with default command line,
331 * it prepends "auto" before the whole cmdline which makes
332 * the shell think it should execute a script with such name.
333 * So we ignore all arguments entered _before_ init=... [MJ]
334 */
335 for (i = 1; i < MAX_INIT_ARGS; i++)
336 argv_init[i] = NULL;
337 return 1;
338}
339__setup("init=", init_setup);
340
Olof Johanssonffdfc402005-09-06 15:17:19 -0700341static int __init rdinit_setup(char *str)
342{
343 unsigned int i;
344
345 ramdisk_execute_command = str;
346 /* See "auto" comment in init_setup */
347 for (i = 1; i < MAX_INIT_ARGS; i++)
348 argv_init[i] = NULL;
349 return 1;
350}
351__setup("rdinit=", rdinit_setup);
352
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353#ifndef CONFIG_SMP
354
355#ifdef CONFIG_X86_LOCAL_APIC
356static void __init smp_init(void)
357{
358 APIC_init_uniprocessor();
359}
360#else
361#define smp_init() do { } while (0)
362#endif
363
364static inline void setup_per_cpu_areas(void) { }
Mike Travise0982e92008-03-26 14:23:48 -0700365static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366static inline void smp_prepare_cpus(unsigned int maxcpus) { }
367
368#else
369
Mike Travis321a8e92008-04-04 18:11:02 -0700370#if NR_CPUS > BITS_PER_LONG
371cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
372EXPORT_SYMBOL(cpu_mask_all);
373#endif
374
Mike Travise0982e92008-03-26 14:23:48 -0700375/* Setup number of possible processor ids */
376int nr_cpu_ids __read_mostly = NR_CPUS;
377EXPORT_SYMBOL(nr_cpu_ids);
378
379/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
380static void __init setup_nr_cpu_ids(void)
381{
382 int cpu, highest_cpu = 0;
383
384 for_each_possible_cpu(cpu)
385 highest_cpu = cpu;
386
387 nr_cpu_ids = highest_cpu + 1;
388}
389
Mike Travisdd5af902008-01-30 13:33:32 +0100390#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800391unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
393EXPORT_SYMBOL(__per_cpu_offset);
394
395static void __init setup_per_cpu_areas(void)
396{
397 unsigned long size, i;
398 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800399 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
401 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200402 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
403 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800405 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 __per_cpu_offset[i] = ptr - __per_cpu_start;
407 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800408 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 }
410}
Mike Travisdd5af902008-01-30 13:33:32 +0100411#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413/* Called by boot processor to activate the rest. */
414static void __init smp_init(void)
415{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800416 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417
Max Krasnyanskye761b772008-07-15 04:43:49 -0700418 /*
419 * Set up the current CPU as possible to migrate to.
420 * The other ones will be done by cpu_up/cpu_down()
421 */
422 cpu = smp_processor_id();
423 cpu_set(cpu, cpu_active_map);
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800426 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100427 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800429 if (!cpu_online(cpu))
430 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 }
432
433 /* Any cleanup work */
434 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100435 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
438#endif
439
440/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800441 * We need to store the untouched command line for future reference.
442 * We also need to store the touched command line since the parameter
443 * parsing is performed in place, and we should allow a component to
444 * store reference of name/value for future reference.
445 */
446static void __init setup_command_line(char *command_line)
447{
448 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
449 static_command_line = alloc_bootmem(strlen (command_line)+1);
450 strcpy (saved_command_line, boot_command_line);
451 strcpy (static_command_line, command_line);
452}
453
454/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 * We need to finalize in a non-__init function or else race conditions
456 * between the root thread and the init thread may cause start_kernel to
457 * be reaped by free_initmem before the root thread has proceeded to
458 * cpu_idle.
459 *
460 * gcc-3.4 accidentally inlines this function, so use noinline.
461 */
462
Sam Ravnborg92080302007-05-17 20:43:54 +0200463static void noinline __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 __releases(kernel_lock)
465{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700466 int pid;
467
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100468 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700470 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700471 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200473
474 /*
475 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200476 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200477 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200478 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800479 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200480 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800481 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200482
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800483 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200485}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
487/* Check for early params. */
488static int __init do_early_param(char *param, char *val)
489{
490 struct obs_kernel_param *p;
491
492 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700493 if ((p->early && strcmp(param, p->str) == 0) ||
494 (strcmp(param, "console") == 0 &&
495 strcmp(p->str, "earlycon") == 0)
496 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 if (p->setup_func(val) != 0)
498 printk(KERN_WARNING
499 "Malformed early option '%s'\n", param);
500 }
501 }
502 /* We accept everything at this stage. */
503 return 0;
504}
505
506/* Arch code calls this early on, or if not, just before other parsing. */
507void __init parse_early_param(void)
508{
509 static __initdata int done = 0;
510 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
511
512 if (done)
513 return;
514
515 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800516 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
518 done = 1;
519}
520
521/*
522 * Activate the first processor.
523 */
524
Stas Sergeev44fd2292006-03-23 02:59:44 -0800525static void __init boot_cpu_init(void)
526{
527 int cpu = smp_processor_id();
528 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
529 cpu_set(cpu, cpu_online_map);
530 cpu_set(cpu, cpu_present_map);
531 cpu_set(cpu, cpu_possible_map);
532}
533
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000534void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700535{
536}
537
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000538void __init __weak thread_info_cache_init(void)
539{
540}
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542asmlinkage void __init start_kernel(void)
543{
544 char * command_line;
545 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700546
547 smp_setup_processor_id();
548
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700549 /*
550 * Need to run as early as possible, to initialize the
551 * lockdep hash:
552 */
Andi Kleenc9538ed2006-09-26 10:52:34 +0200553 unwind_init();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700554 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700555 debug_objects_early_init();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700556 cgroup_init_early();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700557
558 local_irq_disable();
559 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700560 early_init_irq_lock_class();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/*
563 * Interrupts are still disabled. Do necessary setups, then
564 * enable them
565 */
566 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800567 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800568 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 page_address_init();
570 printk(KERN_NOTICE);
Linus Torvalds89937802006-12-11 09:28:46 -0800571 printk(linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700573 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800574 setup_command_line(command_line);
Jan Beulich690a9732006-10-21 18:37:01 +0200575 unwind_setup();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700577 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800578 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579
580 /*
581 * Set up the scheduler prior starting any interrupts (such as the
582 * timer interrupt). Full topology setup happens at smp_init()
583 * time - but meanwhile we still have a functioning scheduler.
584 */
585 sched_init();
586 /*
587 * Disable preemption - early bootup scheduling is extremely
588 * fragile until we cpu_idle() for the first time.
589 */
590 preempt_disable();
591 build_all_zonelists();
592 page_alloc_init();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800593 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 parse_early_param();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800595 parse_args("Booting kernel", static_command_line, __start___param,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 __stop___param - __start___param,
597 &unknown_bootoption);
Ard van Breemenc4a68302007-01-05 16:36:19 -0800598 if (!irqs_disabled()) {
599 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
600 "enabled *very* early, fixing it\n");
601 local_irq_disable();
602 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 sort_main_extable();
604 trap_init();
605 rcu_init();
606 init_IRQ();
607 pidhash_init();
608 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800609 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700611 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700612 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200613 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700614 profile_init();
615 if (!irqs_disabled())
616 printk("start_kernel(): bug: interrupts were enabled early\n");
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700617 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700618 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619
620 /*
621 * HACK ALERT! This is early. We're enabling the console before
622 * we've done PCI setups etc, and console_init() must be aware of
623 * this. But we do want output early, in case something goes wrong.
624 */
625 console_init();
626 if (panic_later)
627 panic(panic_later, panic_param);
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700628
629 lockdep_info();
630
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700631 /*
632 * Need to run this when irqs are enabled, because it wants
633 * to self-test [hard/soft]-irqs on/off lock inversion bugs
634 * too:
635 */
636 locking_selftest();
637
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638#ifdef CONFIG_BLK_DEV_INITRD
639 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700640 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200642 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700643 page_to_pfn(virt_to_page((void *)initrd_start)),
644 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 initrd_start = 0;
646 }
647#endif
Nick Piggindb64fe02008-10-18 20:27:03 -0700648 vmalloc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 vfs_caches_init_early();
Paul Jacksonc417f022006-01-08 01:02:01 -0800650 cpuset_init_early();
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700651 page_cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 mem_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100653 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100654 cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 kmem_cache_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700656 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700657 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700658 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 numa_policy_init();
660 if (late_time_init)
661 late_time_init();
662 calibrate_delay();
663 pidmap_init();
664 pgtable_cache_init();
665 prio_tree_init();
666 anon_vma_init();
667#ifdef CONFIG_X86
668 if (efi_enabled)
669 efi_enter_virtual_mode();
670#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000671 thread_info_cache_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672 fork_init(num_physpages);
673 proc_caches_init();
674 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 key_init();
676 security_init();
677 vfs_caches_init(num_physpages);
678 radix_tree_init();
679 signals_init();
680 /* rootfs populating might need page-writeback */
681 page_writeback_init();
682#ifdef CONFIG_PROC_FS
683 proc_root_init();
684#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700685 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700687 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700688 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
690 check_bugs();
691
692 acpi_early_init(); /* before LAPIC and SMP init */
693
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400694 ftrace_init();
695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 /* Do the rest non-__init'ed, we're now alive */
697 rest_init();
698}
699
Arjan van de Ven59f94152008-07-30 12:49:02 -0700700static int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500701core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702
Arjan van de Ven59f94152008-07-30 12:49:02 -0700703int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700704{
705 int count = preempt_count();
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200706 ktime_t delta;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700707 char msgbuf[64];
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100708 struct boot_trace it;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700709
710 if (initcall_debug) {
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100711 it.caller = task_pid_nr(current);
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100712 printk("calling %pF @ %i\n", fn, it.caller);
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200713 it.calltime = ktime_get();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700714 }
715
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200716 it.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700717
718 if (initcall_debug) {
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200719 it.rettime = ktime_get();
720 delta = ktime_sub(it.rettime, it.calltime);
Tim Birdca538f62008-10-09 15:23:05 -0700721 it.duration = (unsigned long long) delta.tv64 >> 10;
722 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200723 it.result, it.duration);
Frederic Weisbecker56010202008-10-02 13:26:05 +0200724 trace_boot(&it, fn);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700725 }
726
727 msgbuf[0] = 0;
728
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200729 if (it.result && it.result != -ENODEV && initcall_debug)
730 sprintf(msgbuf, "error code %d ", it.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700731
732 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700733 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700734 preempt_count() = count;
735 }
736 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700737 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700738 local_irq_enable();
739 }
740 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700741 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700742 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700743
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200744 return it.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700745}
746
747
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700748extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750static void __init do_initcalls(void)
751{
752 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700754 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700755 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 /* Make sure there is no pending stuff from the initcall sequence */
758 flush_scheduled_work();
759}
760
761/*
762 * Ok, the machine is now initialized. None of the devices
763 * have been touched yet, but the CPU subsystem is up and
764 * running, and memory and process management works.
765 *
766 * Now we can finally start doing some real work..
767 */
768static void __init do_basic_setup(void)
769{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200770 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700771 init_workqueues();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 usermodehelper_init();
773 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800774 init_irq_proc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 do_initcalls();
776}
777
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700778static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700779{
780 initcall_t *call;
781
782 for (call = __initcall_start; call < __early_initcall_end; call++)
783 do_one_initcall(*call);
784}
785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786static void run_init_process(char *init_filename)
787{
788 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700789 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100792/* This is a non __init function. Force it to be noinline otherwise gcc
793 * makes it inline to init() and it becomes part of init.text section
794 */
795static int noinline init_post(void)
796{
797 free_initmem();
798 unlock_kernel();
799 mark_rodata_ro();
800 system_state = SYSTEM_RUNNING;
801 numa_default_policy();
802
803 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
804 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
805
806 (void) sys_dup(0);
807 (void) sys_dup(0);
808
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700809 current->signal->flags |= SIGNAL_UNKILLABLE;
810
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100811 if (ramdisk_execute_command) {
812 run_init_process(ramdisk_execute_command);
813 printk(KERN_WARNING "Failed to execute %s\n",
814 ramdisk_execute_command);
815 }
816
817 /*
818 * We try each of these until one succeeds.
819 *
820 * The Bourne shell can be used instead of init if we are
821 * trying to recover a really broken machine.
822 */
823 if (execute_command) {
824 run_init_process(execute_command);
825 printk(KERN_WARNING "Failed to execute %s. Attempting "
826 "defaults...\n", execute_command);
827 }
828 run_init_process("/sbin/init");
829 run_init_process("/etc/init");
830 run_init_process("/bin/init");
831 run_init_process("/bin/sh");
832
833 panic("No init found. Try passing init= option to kernel.");
834}
835
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100836static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837{
838 lock_kernel();
839 /*
840 * init can run on any cpu.
841 */
Mike Travis321a8e92008-04-04 18:11:02 -0700842 set_cpus_allowed_ptr(current, CPU_MASK_ALL_PTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 /*
844 * Tell the world that we're going to be the grim
845 * reaper of innocent orphaned children.
846 *
847 * We don't want people to have to make incorrect
848 * assumptions about where in the task array this
849 * can be found.
850 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800851 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700853 cad_pid = task_pid(current);
854
Andi Kleenca74a6f2008-01-30 13:33:17 +0100855 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100858 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 smp_init();
861 sched_init_smp();
862
863 cpuset_init_smp();
864
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 do_basic_setup();
866
867 /*
868 * check if there is an early userspace init. If yes, let it do all
869 * the work
870 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700871
872 if (!ramdisk_execute_command)
873 ramdisk_execute_command = "/init";
874
875 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
876 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700878 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
880 /*
881 * Ok, we have completed the initial bootup, and
882 * we're essentially up and running. Get rid of the
883 * initmem segments and start the user-mode stuff..
884 */
Frederic Weisbecker097d0362008-10-03 15:39:21 +0200885 stop_boot_trace();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100886 init_post();
887 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888}