blob: 61eb661593915cbe9a6e85a1481c3b5b149dcad7 [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>
30#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/workqueue.h>
35#include <linux/profile.h>
36#include <linux/rcupdate.h>
37#include <linux/moduleparam.h>
38#include <linux/kallsyms.h>
39#include <linux/writeback.h>
40#include <linux/cpu.h>
41#include <linux/cpuset.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070042#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/efi.h>
Thomas Gleixner906568c2007-02-16 01:28:01 -080044#include <linux/tick.h>
Andrew Morton6168a702007-02-17 21:22:39 -080045#include <linux/interrupt.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070046#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070047#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/unistd.h>
49#include <linux/rmap.h>
50#include <linux/mempolicy.h>
51#include <linux/key.h>
Jan Beulich4552d5d2006-06-26 13:57:28 +020052#include <linux/unwind.h>
Adrian Bunkb6cd0b72006-06-27 02:53:54 -070053#include <linux/buffer_head.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070054#include <linux/debug_locks.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070055#include <linux/debugobjects.h>
Ingo Molnarfbb9ce92006-07-03 00:24:50 -070056#include <linux/lockdep.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>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040063#include <linux/ftrace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064
65#include <asm/io.h>
66#include <asm/bugs.h>
67#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070068#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080069#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#ifdef CONFIG_X86_LOCAL_APIC
72#include <asm/smp.h>
73#endif
74
75/*
Coywolf Qi Hunt69c99ac2006-01-14 13:21:11 -080076 * This is one of the first .c files built. Error out early if we have compiler
77 * trouble.
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
Andrew Mortonf1a60db2006-12-06 20:39:59 -080080#if __GNUC__ == 4 && __GNUC_MINOR__ == 1 && __GNUC_PATCHLEVEL__ == 0
81#warning gcc-4.1.0 is known to miscompile the kernel. A different compiler version is recommended.
82#endif
83
Sam Ravnborgaae5f662007-02-26 16:45:41 +010084static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
86extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087extern void fork_init(unsigned long);
88extern void mca_init(void);
89extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090extern void prio_tree_init(void);
91extern void radix_tree_init(void);
92extern void free_initmem(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093#ifdef CONFIG_ACPI
94extern void acpi_early_init(void);
95#else
96static inline void acpi_early_init(void) { }
97#endif
Arjan van de Ven37b73c82006-01-06 00:12:01 -080098#ifndef CONFIG_DEBUG_RODATA
99static inline void mark_rodata_ro(void) { }
100#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101
102#ifdef CONFIG_TC
103extern void tc_init(void);
104#endif
105
106enum system_states system_state;
107EXPORT_SYMBOL(system_state);
108
109/*
110 * Boot command-line arguments
111 */
112#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
113#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
114
115extern void time_init(void);
116/* Default late time init is NULL. archs can override this later. */
117void (*late_time_init)(void);
118extern void softirq_init(void);
119
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800120/* Untouched command line saved by arch-specific code. */
121char __initdata boot_command_line[COMMAND_LINE_SIZE];
122/* Untouched saved command line (eg. for /proc) */
123char *saved_command_line;
124/* Command line for parameter parsing */
125static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700128static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
Jan Beulich8b3b2952007-07-15 23:41:07 -0700130#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100132unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700133
134/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 * Setup routine for controlling SMP activation
136 *
137 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
138 * activation entirely (the MPS table probe still happens, though).
139 *
140 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
141 * greater than 0, limits the maximum number of CPUs activated in
142 * SMP mode to <NUM>.
143 */
Len Brown61ec7562007-08-16 03:34:22 -0400144#ifndef CONFIG_X86_IO_APIC
145static inline void disable_ioapic_setup(void) {};
146#endif
147
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148static int __init nosmp(char *str)
149{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100150 setup_max_cpus = 0;
Len Brown61ec7562007-08-16 03:34:22 -0400151 disable_ioapic_setup();
Jan Beulich8b3b2952007-07-15 23:41:07 -0700152 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153}
154
Jan Beulich8b3b2952007-07-15 23:41:07 -0700155early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156
157static int __init maxcpus(char *str)
158{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100159 get_option(&str, &setup_max_cpus);
160 if (setup_max_cpus == 0)
Len Brown61ec7562007-08-16 03:34:22 -0400161 disable_ioapic_setup();
162
163 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164}
165
Hugh Dickins81340972007-08-27 16:02:12 +0100166early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700167#else
Andi Kleenca74a6f2008-01-30 13:33:17 +0100168#define setup_max_cpus NR_CPUS
Jan Beulich8b3b2952007-07-15 23:41:07 -0700169#endif
170
171/*
172 * If set, this is an indication to the drivers that reset the underlying
173 * device before going ahead with the initialization otherwise driver might
174 * rely on the BIOS and skip the reset operation.
175 *
176 * This is useful if kernel is booting in an unreliable environment.
177 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
178 * skipped and devices will be in unknown state.
179 */
180unsigned int reset_devices;
181EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Vivek Goyal7e962872006-09-27 01:50:44 -0700183static int __init set_reset_devices(char *str)
184{
185 reset_devices = 1;
186 return 1;
187}
188
189__setup("reset_devices", set_reset_devices);
190
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
192char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
193static const char *panic_later, *panic_param;
194
195extern struct obs_kernel_param __setup_start[], __setup_end[];
196
197static int __init obsolete_checksetup(char *line)
198{
199 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200200 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
202 p = __setup_start;
203 do {
204 int n = strlen(p->str);
205 if (!strncmp(line, p->str, n)) {
206 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200207 /* Already done in parse_early_param?
208 * (Needs exact match on param part).
209 * Keep iterating, as we can have early
210 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200212 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 } else if (!p->setup_func) {
214 printk(KERN_WARNING "Parameter %s is obsolete,"
215 " ignored\n", p->str);
216 return 1;
217 } else if (p->setup_func(line + n))
218 return 1;
219 }
220 p++;
221 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200222
223 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
226/*
227 * This should be approx 2 Bo*oMips to start (note initial shift), and will
228 * still work even if initially too large, it will just take slightly longer
229 */
230unsigned long loops_per_jiffy = (1<<12);
231
232EXPORT_SYMBOL(loops_per_jiffy);
233
234static int __init debug_kernel(char *str)
235{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800237 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
240static int __init quiet_kernel(char *str)
241{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800243 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244}
245
Yinghai Luf6f21c812008-02-08 04:21:58 -0800246early_param("debug", debug_kernel);
247early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
249static int __init loglevel(char *str)
250{
251 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800252 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253}
254
Yinghai Luf6f21c812008-02-08 04:21:58 -0800255early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257/*
258 * Unknown boot options get handed to init, unless they look like
259 * failed parameters
260 */
261static int __init unknown_bootoption(char *param, char *val)
262{
263 /* Change NUL term back to "=", to make "param" the whole string. */
264 if (val) {
265 /* param=val or param="val"? */
266 if (val == param+strlen(param)+1)
267 val[-1] = '=';
268 else if (val == param+strlen(param)+2) {
269 val[-2] = '=';
270 memmove(val-1, val, strlen(val)+1);
271 val--;
272 } else
273 BUG();
274 }
275
276 /* Handle obsolete-style parameters */
277 if (obsolete_checksetup(param))
278 return 0;
279
280 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200281 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 * line work?"
283 */
284 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
285 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
286 return 0;
287 }
288
289 if (panic_later)
290 return 0;
291
292 if (val) {
293 /* Environment option */
294 unsigned int i;
295 for (i = 0; envp_init[i]; i++) {
296 if (i == MAX_INIT_ENVS) {
297 panic_later = "Too many boot env vars at `%s'";
298 panic_param = param;
299 }
300 if (!strncmp(param, envp_init[i], val - param))
301 break;
302 }
303 envp_init[i] = param;
304 } else {
305 /* Command line option */
306 unsigned int i;
307 for (i = 0; argv_init[i]; i++) {
308 if (i == MAX_INIT_ARGS) {
309 panic_later = "Too many boot init vars at `%s'";
310 panic_param = param;
311 }
312 }
313 argv_init[i] = param;
314 }
315 return 0;
316}
317
Ingo Molnar12d6f212008-01-30 13:33:58 +0100318#ifdef CONFIG_DEBUG_PAGEALLOC
319int __read_mostly debug_pagealloc_enabled = 0;
320#endif
321
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322static int __init init_setup(char *str)
323{
324 unsigned int i;
325
326 execute_command = str;
327 /*
328 * In case LILO is going to boot us with default command line,
329 * it prepends "auto" before the whole cmdline which makes
330 * the shell think it should execute a script with such name.
331 * So we ignore all arguments entered _before_ init=... [MJ]
332 */
333 for (i = 1; i < MAX_INIT_ARGS; i++)
334 argv_init[i] = NULL;
335 return 1;
336}
337__setup("init=", init_setup);
338
Olof Johanssonffdfc402005-09-06 15:17:19 -0700339static int __init rdinit_setup(char *str)
340{
341 unsigned int i;
342
343 ramdisk_execute_command = str;
344 /* See "auto" comment in init_setup */
345 for (i = 1; i < MAX_INIT_ARGS; i++)
346 argv_init[i] = NULL;
347 return 1;
348}
349__setup("rdinit=", rdinit_setup);
350
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351#ifndef CONFIG_SMP
352
353#ifdef CONFIG_X86_LOCAL_APIC
354static void __init smp_init(void)
355{
356 APIC_init_uniprocessor();
357}
358#else
359#define smp_init() do { } while (0)
360#endif
361
362static inline void setup_per_cpu_areas(void) { }
Mike Travise0982e92008-03-26 14:23:48 -0700363static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364static inline void smp_prepare_cpus(unsigned int maxcpus) { }
365
366#else
367
Mike Travis321a8e92008-04-04 18:11:02 -0700368#if NR_CPUS > BITS_PER_LONG
369cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
370EXPORT_SYMBOL(cpu_mask_all);
371#endif
372
Mike Travise0982e92008-03-26 14:23:48 -0700373/* Setup number of possible processor ids */
374int nr_cpu_ids __read_mostly = NR_CPUS;
375EXPORT_SYMBOL(nr_cpu_ids);
376
377/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
378static void __init setup_nr_cpu_ids(void)
379{
380 int cpu, highest_cpu = 0;
381
382 for_each_possible_cpu(cpu)
383 highest_cpu = cpu;
384
385 nr_cpu_ids = highest_cpu + 1;
386}
387
Mike Travisdd5af902008-01-30 13:33:32 +0100388#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800389unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
391EXPORT_SYMBOL(__per_cpu_offset);
392
393static void __init setup_per_cpu_areas(void)
394{
395 unsigned long size, i;
396 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800397 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200400 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
401 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800403 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 __per_cpu_offset[i] = ptr - __per_cpu_start;
405 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800406 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 }
408}
Mike Travisdd5af902008-01-30 13:33:32 +0100409#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411/* Called by boot processor to activate the rest. */
412static void __init smp_init(void)
413{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800414 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
Max Krasnyanskye761b772008-07-15 04:43:49 -0700416 /*
417 * Set up the current CPU as possible to migrate to.
418 * The other ones will be done by cpu_up/cpu_down()
419 */
420 cpu = smp_processor_id();
421 cpu_set(cpu, cpu_active_map);
422
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800424 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100425 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800427 if (!cpu_online(cpu))
428 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 }
430
431 /* Any cleanup work */
432 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100433 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434}
435
436#endif
437
438/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800439 * We need to store the untouched command line for future reference.
440 * We also need to store the touched command line since the parameter
441 * parsing is performed in place, and we should allow a component to
442 * store reference of name/value for future reference.
443 */
444static void __init setup_command_line(char *command_line)
445{
446 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
447 static_command_line = alloc_bootmem(strlen (command_line)+1);
448 strcpy (saved_command_line, boot_command_line);
449 strcpy (static_command_line, command_line);
450}
451
452/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 * We need to finalize in a non-__init function or else race conditions
454 * between the root thread and the init thread may cause start_kernel to
455 * be reaped by free_initmem before the root thread has proceeded to
456 * cpu_idle.
457 *
458 * gcc-3.4 accidentally inlines this function, so use noinline.
459 */
460
Sam Ravnborg92080302007-05-17 20:43:54 +0200461static void noinline __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 __releases(kernel_lock)
463{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700464 int pid;
465
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100466 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700468 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700469 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200471
472 /*
473 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200474 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200475 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200476 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800477 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200478 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800479 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200480
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800481 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200483}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
485/* Check for early params. */
486static int __init do_early_param(char *param, char *val)
487{
488 struct obs_kernel_param *p;
489
490 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700491 if ((p->early && strcmp(param, p->str) == 0) ||
492 (strcmp(param, "console") == 0 &&
493 strcmp(p->str, "earlycon") == 0)
494 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 if (p->setup_func(val) != 0)
496 printk(KERN_WARNING
497 "Malformed early option '%s'\n", param);
498 }
499 }
500 /* We accept everything at this stage. */
501 return 0;
502}
503
504/* Arch code calls this early on, or if not, just before other parsing. */
505void __init parse_early_param(void)
506{
507 static __initdata int done = 0;
508 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
509
510 if (done)
511 return;
512
513 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800514 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
516 done = 1;
517}
518
519/*
520 * Activate the first processor.
521 */
522
Stas Sergeev44fd2292006-03-23 02:59:44 -0800523static void __init boot_cpu_init(void)
524{
525 int cpu = smp_processor_id();
526 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
527 cpu_set(cpu, cpu_online_map);
528 cpu_set(cpu, cpu_present_map);
529 cpu_set(cpu, cpu_possible_map);
530}
531
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000532void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700533{
534}
535
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000536void __init __weak thread_info_cache_init(void)
537{
538}
539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540asmlinkage void __init start_kernel(void)
541{
542 char * command_line;
543 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700544
545 smp_setup_processor_id();
546
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700547 /*
548 * Need to run as early as possible, to initialize the
549 * lockdep hash:
550 */
Andi Kleenc9538ed2006-09-26 10:52:34 +0200551 unwind_init();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700552 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700553 debug_objects_early_init();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700554 cgroup_init_early();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700555
556 local_irq_disable();
557 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700558 early_init_irq_lock_class();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560/*
561 * Interrupts are still disabled. Do necessary setups, then
562 * enable them
563 */
564 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800565 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800566 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 page_address_init();
568 printk(KERN_NOTICE);
Linus Torvalds89937802006-12-11 09:28:46 -0800569 printk(linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700571 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800572 setup_command_line(command_line);
Jan Beulich690a9732006-10-21 18:37:01 +0200573 unwind_setup();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700575 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800576 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 /*
579 * Set up the scheduler prior starting any interrupts (such as the
580 * timer interrupt). Full topology setup happens at smp_init()
581 * time - but meanwhile we still have a functioning scheduler.
582 */
583 sched_init();
584 /*
585 * Disable preemption - early bootup scheduling is extremely
586 * fragile until we cpu_idle() for the first time.
587 */
588 preempt_disable();
589 build_all_zonelists();
590 page_alloc_init();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800591 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 parse_early_param();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800593 parse_args("Booting kernel", static_command_line, __start___param,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 __stop___param - __start___param,
595 &unknown_bootoption);
Ard van Breemenc4a68302007-01-05 16:36:19 -0800596 if (!irqs_disabled()) {
597 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
598 "enabled *very* early, fixing it\n");
599 local_irq_disable();
600 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 sort_main_extable();
602 trap_init();
603 rcu_init();
604 init_IRQ();
605 pidhash_init();
606 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800607 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700609 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700610 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200611 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700612 profile_init();
613 if (!irqs_disabled())
614 printk("start_kernel(): bug: interrupts were enabled early\n");
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700615 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700616 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617
618 /*
619 * HACK ALERT! This is early. We're enabling the console before
620 * we've done PCI setups etc, and console_init() must be aware of
621 * this. But we do want output early, in case something goes wrong.
622 */
623 console_init();
624 if (panic_later)
625 panic(panic_later, panic_param);
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700626
627 lockdep_info();
628
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700629 /*
630 * Need to run this when irqs are enabled, because it wants
631 * to self-test [hard/soft]-irqs on/off lock inversion bugs
632 * too:
633 */
634 locking_selftest();
635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636#ifdef CONFIG_BLK_DEV_INITRD
637 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700638 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200640 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700641 page_to_pfn(virt_to_page((void *)initrd_start)),
642 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 initrd_start = 0;
644 }
645#endif
646 vfs_caches_init_early();
Paul Jacksonc417f022006-01-08 01:02:01 -0800647 cpuset_init_early();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 mem_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100649 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100650 cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 kmem_cache_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700652 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700653 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700654 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 numa_policy_init();
656 if (late_time_init)
657 late_time_init();
658 calibrate_delay();
659 pidmap_init();
660 pgtable_cache_init();
661 prio_tree_init();
662 anon_vma_init();
663#ifdef CONFIG_X86
664 if (efi_enabled)
665 efi_enter_virtual_mode();
666#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000667 thread_info_cache_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 fork_init(num_physpages);
669 proc_caches_init();
670 buffer_init();
671 unnamed_dev_init();
672 key_init();
673 security_init();
674 vfs_caches_init(num_physpages);
675 radix_tree_init();
676 signals_init();
677 /* rootfs populating might need page-writeback */
678 page_writeback_init();
679#ifdef CONFIG_PROC_FS
680 proc_root_init();
681#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700682 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700684 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700685 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686
687 check_bugs();
688
689 acpi_early_init(); /* before LAPIC and SMP init */
690
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400691 ftrace_init();
692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 /* Do the rest non-__init'ed, we're now alive */
694 rest_init();
695}
696
Arjan van de Ven59f94152008-07-30 12:49:02 -0700697static int initcall_debug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
699static int __init initcall_debug_setup(char *str)
700{
701 initcall_debug = 1;
702 return 1;
703}
704__setup("initcall_debug", initcall_debug_setup);
705
Arjan van de Ven59f94152008-07-30 12:49:02 -0700706int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700707{
708 int count = preempt_count();
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200709 ktime_t delta;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700710 char msgbuf[64];
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100711 struct boot_trace it;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700712
713 if (initcall_debug) {
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100714 it.caller = task_pid_nr(current);
715 it.func = fn;
716 printk("calling %pF @ %i\n", fn, it.caller);
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200717 it.calltime = ktime_get();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700718 }
719
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200720 it.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700721
722 if (initcall_debug) {
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200723 it.rettime = ktime_get();
724 delta = ktime_sub(it.rettime, it.calltime);
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100725 it.duration = (unsigned long long) delta.tv64 >> 20;
726 printk("initcall %pF returned %d after %Ld msecs\n", fn,
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200727 it.result, it.duration);
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100728 trace_boot(&it);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700729 }
730
731 msgbuf[0] = 0;
732
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200733 if (it.result && it.result != -ENODEV && initcall_debug)
734 sprintf(msgbuf, "error code %d ", it.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700735
736 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700737 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700738 preempt_count() = count;
739 }
740 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700741 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700742 local_irq_enable();
743 }
744 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700745 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700746 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700747
Frederic Weisbeckercb5ab742008-10-02 12:59:20 +0200748 return it.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700749}
750
751
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700752extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
754static void __init do_initcalls(void)
755{
756 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700758 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700759 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
761 /* Make sure there is no pending stuff from the initcall sequence */
762 flush_scheduled_work();
763}
764
765/*
766 * Ok, the machine is now initialized. None of the devices
767 * have been touched yet, but the CPU subsystem is up and
768 * running, and memory and process management works.
769 *
770 * Now we can finally start doing some real work..
771 */
772static void __init do_basic_setup(void)
773{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200774 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 /* drivers will send hotplug events */
776 init_workqueues();
777 usermodehelper_init();
778 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800779 init_irq_proc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 do_initcalls();
781}
782
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700783static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700784{
785 initcall_t *call;
786
787 for (call = __initcall_start; call < __early_initcall_end; call++)
788 do_one_initcall(*call);
789}
790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791static void run_init_process(char *init_filename)
792{
793 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700794 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795}
796
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100797/* This is a non __init function. Force it to be noinline otherwise gcc
798 * makes it inline to init() and it becomes part of init.text section
799 */
800static int noinline init_post(void)
801{
802 free_initmem();
803 unlock_kernel();
804 mark_rodata_ro();
805 system_state = SYSTEM_RUNNING;
806 numa_default_policy();
807
808 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
809 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
810
811 (void) sys_dup(0);
812 (void) sys_dup(0);
813
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700814 current->signal->flags |= SIGNAL_UNKILLABLE;
815
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100816 if (ramdisk_execute_command) {
817 run_init_process(ramdisk_execute_command);
818 printk(KERN_WARNING "Failed to execute %s\n",
819 ramdisk_execute_command);
820 }
821
822 /*
823 * We try each of these until one succeeds.
824 *
825 * The Bourne shell can be used instead of init if we are
826 * trying to recover a really broken machine.
827 */
828 if (execute_command) {
829 run_init_process(execute_command);
830 printk(KERN_WARNING "Failed to execute %s. Attempting "
831 "defaults...\n", execute_command);
832 }
833 run_init_process("/sbin/init");
834 run_init_process("/etc/init");
835 run_init_process("/bin/init");
836 run_init_process("/bin/sh");
837
838 panic("No init found. Try passing init= option to kernel.");
839}
840
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100841static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842{
843 lock_kernel();
844 /*
845 * init can run on any cpu.
846 */
Mike Travis321a8e92008-04-04 18:11:02 -0700847 set_cpus_allowed_ptr(current, CPU_MASK_ALL_PTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 /*
849 * Tell the world that we're going to be the grim
850 * reaper of innocent orphaned children.
851 *
852 * We don't want people to have to make incorrect
853 * assumptions about where in the task array this
854 * can be found.
855 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800856 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700858 cad_pid = task_pid(current);
859
Andi Kleenca74a6f2008-01-30 13:33:17 +0100860 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
862 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100863 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 smp_init();
866 sched_init_smp();
867
868 cpuset_init_smp();
869
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 do_basic_setup();
871
872 /*
873 * check if there is an early userspace init. If yes, let it do all
874 * the work
875 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700876
877 if (!ramdisk_execute_command)
878 ramdisk_execute_command = "/init";
879
880 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
881 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
885 /*
886 * Ok, we have completed the initial bootup, and
887 * we're essentially up and running. Get rid of the
888 * initmem segments and start the user-mode stuff..
889 */
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100890 init_post();
891 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}