blob: 546ebd2f44ba3acf15daedfb4f7b4a935a8e474b [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 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>
Frederic Weisbecker3f5ec132008-11-11 23:21:31 +010066#include <trace/boot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68#include <asm/io.h>
69#include <asm/bugs.h>
70#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070071#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080072#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#ifdef CONFIG_X86_LOCAL_APIC
75#include <asm/smp.h>
76#endif
77
78/*
Coywolf Qi Hunt69c99ac2006-01-14 13:21:11 -080079 * This is one of the first .c files built. Error out early if we have compiler
80 * trouble.
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
Andrew Mortonf1a60db2006-12-06 20:39:59 -080083#if __GNUC__ == 4 && __GNUC_MINOR__ == 1 && __GNUC_PATCHLEVEL__ == 0
84#warning gcc-4.1.0 is known to miscompile the kernel. A different compiler version is recommended.
85#endif
86
Sam Ravnborgaae5f662007-02-26 16:45:41 +010087static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088
89extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090extern void fork_init(unsigned long);
91extern void mca_init(void);
92extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093extern void prio_tree_init(void);
94extern void radix_tree_init(void);
95extern void free_initmem(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070096#ifdef CONFIG_ACPI
97extern void acpi_early_init(void);
98#else
99static inline void acpi_early_init(void) { }
100#endif
Arjan van de Ven37b73c82006-01-06 00:12:01 -0800101#ifndef CONFIG_DEBUG_RODATA
102static inline void mark_rodata_ro(void) { }
103#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104
105#ifdef CONFIG_TC
106extern void tc_init(void);
107#endif
108
109enum system_states system_state;
110EXPORT_SYMBOL(system_state);
111
112/*
113 * Boot command-line arguments
114 */
115#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
116#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
117
118extern void time_init(void);
119/* Default late time init is NULL. archs can override this later. */
120void (*late_time_init)(void);
121extern void softirq_init(void);
122
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800123/* Untouched command line saved by arch-specific code. */
124char __initdata boot_command_line[COMMAND_LINE_SIZE];
125/* Untouched saved command line (eg. for /proc) */
126char *saved_command_line;
127/* Command line for parameter parsing */
128static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129
130static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700131static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Jan Beulich8b3b2952007-07-15 23:41:07 -0700133#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100135unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700136
137/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 * Setup routine for controlling SMP activation
139 *
140 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
141 * activation entirely (the MPS table probe still happens, though).
142 *
143 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
144 * greater than 0, limits the maximum number of CPUs activated in
145 * SMP mode to <NUM>.
146 */
Len Brown61ec7562007-08-16 03:34:22 -0400147#ifndef CONFIG_X86_IO_APIC
148static inline void disable_ioapic_setup(void) {};
149#endif
150
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151static int __init nosmp(char *str)
152{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100153 setup_max_cpus = 0;
Len Brown61ec7562007-08-16 03:34:22 -0400154 disable_ioapic_setup();
Jan Beulich8b3b2952007-07-15 23:41:07 -0700155 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
Jan Beulich8b3b2952007-07-15 23:41:07 -0700158early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
160static int __init maxcpus(char *str)
161{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100162 get_option(&str, &setup_max_cpus);
163 if (setup_max_cpus == 0)
Len Brown61ec7562007-08-16 03:34:22 -0400164 disable_ioapic_setup();
165
166 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167}
168
Hugh Dickins81340972007-08-27 16:02:12 +0100169early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700170#else
Andi Kleenca74a6f2008-01-30 13:33:17 +0100171#define setup_max_cpus NR_CPUS
Jan Beulich8b3b2952007-07-15 23:41:07 -0700172#endif
173
174/*
175 * If set, this is an indication to the drivers that reset the underlying
176 * device before going ahead with the initialization otherwise driver might
177 * rely on the BIOS and skip the reset operation.
178 *
179 * This is useful if kernel is booting in an unreliable environment.
180 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
181 * skipped and devices will be in unknown state.
182 */
183unsigned int reset_devices;
184EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
Vivek Goyal7e962872006-09-27 01:50:44 -0700186static int __init set_reset_devices(char *str)
187{
188 reset_devices = 1;
189 return 1;
190}
191
192__setup("reset_devices", set_reset_devices);
193
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
195char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
196static const char *panic_later, *panic_param;
197
198extern struct obs_kernel_param __setup_start[], __setup_end[];
199
200static int __init obsolete_checksetup(char *line)
201{
202 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200203 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204
205 p = __setup_start;
206 do {
207 int n = strlen(p->str);
208 if (!strncmp(line, p->str, n)) {
209 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200210 /* Already done in parse_early_param?
211 * (Needs exact match on param part).
212 * Keep iterating, as we can have early
213 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200215 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 } else if (!p->setup_func) {
217 printk(KERN_WARNING "Parameter %s is obsolete,"
218 " ignored\n", p->str);
219 return 1;
220 } else if (p->setup_func(line + n))
221 return 1;
222 }
223 p++;
224 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200225
226 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227}
228
229/*
230 * This should be approx 2 Bo*oMips to start (note initial shift), and will
231 * still work even if initially too large, it will just take slightly longer
232 */
233unsigned long loops_per_jiffy = (1<<12);
234
235EXPORT_SYMBOL(loops_per_jiffy);
236
237static int __init debug_kernel(char *str)
238{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800240 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
243static int __init quiet_kernel(char *str)
244{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800246 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
Yinghai Luf6f21c812008-02-08 04:21:58 -0800249early_param("debug", debug_kernel);
250early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252static int __init loglevel(char *str)
253{
254 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800255 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256}
257
Yinghai Luf6f21c812008-02-08 04:21:58 -0800258early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
260/*
261 * Unknown boot options get handed to init, unless they look like
262 * failed parameters
263 */
264static int __init unknown_bootoption(char *param, char *val)
265{
266 /* Change NUL term back to "=", to make "param" the whole string. */
267 if (val) {
268 /* param=val or param="val"? */
269 if (val == param+strlen(param)+1)
270 val[-1] = '=';
271 else if (val == param+strlen(param)+2) {
272 val[-2] = '=';
273 memmove(val-1, val, strlen(val)+1);
274 val--;
275 } else
276 BUG();
277 }
278
279 /* Handle obsolete-style parameters */
280 if (obsolete_checksetup(param))
281 return 0;
282
283 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200284 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 * line work?"
286 */
287 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
288 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
289 return 0;
290 }
291
292 if (panic_later)
293 return 0;
294
295 if (val) {
296 /* Environment option */
297 unsigned int i;
298 for (i = 0; envp_init[i]; i++) {
299 if (i == MAX_INIT_ENVS) {
300 panic_later = "Too many boot env vars at `%s'";
301 panic_param = param;
302 }
303 if (!strncmp(param, envp_init[i], val - param))
304 break;
305 }
306 envp_init[i] = param;
307 } else {
308 /* Command line option */
309 unsigned int i;
310 for (i = 0; argv_init[i]; i++) {
311 if (i == MAX_INIT_ARGS) {
312 panic_later = "Too many boot init vars at `%s'";
313 panic_param = param;
314 }
315 }
316 argv_init[i] = param;
317 }
318 return 0;
319}
320
Ingo Molnar12d6f212008-01-30 13:33:58 +0100321#ifdef CONFIG_DEBUG_PAGEALLOC
322int __read_mostly debug_pagealloc_enabled = 0;
323#endif
324
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325static int __init init_setup(char *str)
326{
327 unsigned int i;
328
329 execute_command = str;
330 /*
331 * In case LILO is going to boot us with default command line,
332 * it prepends "auto" before the whole cmdline which makes
333 * the shell think it should execute a script with such name.
334 * So we ignore all arguments entered _before_ init=... [MJ]
335 */
336 for (i = 1; i < MAX_INIT_ARGS; i++)
337 argv_init[i] = NULL;
338 return 1;
339}
340__setup("init=", init_setup);
341
Olof Johanssonffdfc402005-09-06 15:17:19 -0700342static int __init rdinit_setup(char *str)
343{
344 unsigned int i;
345
346 ramdisk_execute_command = str;
347 /* See "auto" comment in init_setup */
348 for (i = 1; i < MAX_INIT_ARGS; i++)
349 argv_init[i] = NULL;
350 return 1;
351}
352__setup("rdinit=", rdinit_setup);
353
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354#ifndef CONFIG_SMP
355
356#ifdef CONFIG_X86_LOCAL_APIC
357static void __init smp_init(void)
358{
359 APIC_init_uniprocessor();
360}
361#else
362#define smp_init() do { } while (0)
363#endif
364
365static inline void setup_per_cpu_areas(void) { }
Mike Travise0982e92008-03-26 14:23:48 -0700366static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367static inline void smp_prepare_cpus(unsigned int maxcpus) { }
368
369#else
370
Mike Travis321a8e92008-04-04 18:11:02 -0700371#if NR_CPUS > BITS_PER_LONG
372cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
373EXPORT_SYMBOL(cpu_mask_all);
374#endif
375
Mike Travise0982e92008-03-26 14:23:48 -0700376/* Setup number of possible processor ids */
377int nr_cpu_ids __read_mostly = NR_CPUS;
378EXPORT_SYMBOL(nr_cpu_ids);
379
380/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
381static void __init setup_nr_cpu_ids(void)
382{
Rusty Russelle0c0ba72009-01-01 10:12:19 +1030383 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
Mike Travise0982e92008-03-26 14:23:48 -0700384}
385
Mike Travisdd5af902008-01-30 13:33:32 +0100386#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800387unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389EXPORT_SYMBOL(__per_cpu_offset);
390
391static void __init setup_per_cpu_areas(void)
392{
393 unsigned long size, i;
394 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800395 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396
397 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200398 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
399 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800401 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 __per_cpu_offset[i] = ptr - __per_cpu_start;
403 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800404 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 }
406}
Mike Travisdd5af902008-01-30 13:33:32 +0100407#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
409/* Called by boot processor to activate the rest. */
410static void __init smp_init(void)
411{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800412 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
Max Krasnyanskye761b772008-07-15 04:43:49 -0700414 /*
415 * Set up the current CPU as possible to migrate to.
416 * The other ones will be done by cpu_up/cpu_down()
417 */
418 cpu = smp_processor_id();
419 cpu_set(cpu, cpu_active_map);
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800422 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100423 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800425 if (!cpu_online(cpu))
426 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 }
428
429 /* Any cleanup work */
430 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100431 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432}
433
434#endif
435
436/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800437 * We need to store the untouched command line for future reference.
438 * We also need to store the touched command line since the parameter
439 * parsing is performed in place, and we should allow a component to
440 * store reference of name/value for future reference.
441 */
442static void __init setup_command_line(char *command_line)
443{
444 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
445 static_command_line = alloc_bootmem(strlen (command_line)+1);
446 strcpy (saved_command_line, boot_command_line);
447 strcpy (static_command_line, command_line);
448}
449
450/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 * We need to finalize in a non-__init function or else race conditions
452 * between the root thread and the init thread may cause start_kernel to
453 * be reaped by free_initmem before the root thread has proceeded to
454 * cpu_idle.
455 *
456 * gcc-3.4 accidentally inlines this function, so use noinline.
457 */
458
Sam Ravnborg92080302007-05-17 20:43:54 +0200459static void noinline __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 __releases(kernel_lock)
461{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700462 int pid;
463
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100464 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700466 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700467 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200469
470 /*
471 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200472 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200473 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200474 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800475 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200476 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800477 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200478
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800479 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200481}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483/* Check for early params. */
484static int __init do_early_param(char *param, char *val)
485{
486 struct obs_kernel_param *p;
487
488 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700489 if ((p->early && strcmp(param, p->str) == 0) ||
490 (strcmp(param, "console") == 0 &&
491 strcmp(p->str, "earlycon") == 0)
492 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 if (p->setup_func(val) != 0)
494 printk(KERN_WARNING
495 "Malformed early option '%s'\n", param);
496 }
497 }
498 /* We accept everything at this stage. */
499 return 0;
500}
501
502/* Arch code calls this early on, or if not, just before other parsing. */
503void __init parse_early_param(void)
504{
505 static __initdata int done = 0;
506 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
507
508 if (done)
509 return;
510
511 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800512 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
514 done = 1;
515}
516
517/*
518 * Activate the first processor.
519 */
520
Stas Sergeev44fd2292006-03-23 02:59:44 -0800521static void __init boot_cpu_init(void)
522{
523 int cpu = smp_processor_id();
524 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030525 set_cpu_online(cpu, true);
526 set_cpu_present(cpu, true);
527 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800528}
529
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000530void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700531{
532}
533
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000534void __init __weak thread_info_cache_init(void)
535{
536}
537
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800538void __init __weak arch_early_irq_init(void)
539{
540}
541
542void __init __weak early_irq_init(void)
543{
544 arch_early_irq_init();
545}
546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547asmlinkage void __init start_kernel(void)
548{
549 char * command_line;
550 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700551
552 smp_setup_processor_id();
553
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700554 /*
555 * Need to run as early as possible, to initialize the
556 * lockdep hash:
557 */
Andi Kleenc9538ed2006-09-26 10:52:34 +0200558 unwind_init();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700559 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700560 debug_objects_early_init();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700561 cgroup_init_early();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700562
563 local_irq_disable();
564 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700565 early_init_irq_lock_class();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567/*
568 * Interrupts are still disabled. Do necessary setups, then
569 * enable them
570 */
571 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800572 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800573 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 page_address_init();
575 printk(KERN_NOTICE);
Linus Torvalds89937802006-12-11 09:28:46 -0800576 printk(linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700578 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800579 setup_command_line(command_line);
Jan Beulich690a9732006-10-21 18:37:01 +0200580 unwind_setup();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700582 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800583 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
585 /*
586 * Set up the scheduler prior starting any interrupts (such as the
587 * timer interrupt). Full topology setup happens at smp_init()
588 * time - but meanwhile we still have a functioning scheduler.
589 */
590 sched_init();
591 /*
592 * Disable preemption - early bootup scheduling is extremely
593 * fragile until we cpu_idle() for the first time.
594 */
595 preempt_disable();
596 build_all_zonelists();
597 page_alloc_init();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800598 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 parse_early_param();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800600 parse_args("Booting kernel", static_command_line, __start___param,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 __stop___param - __start___param,
602 &unknown_bootoption);
Ard van Breemenc4a68302007-01-05 16:36:19 -0800603 if (!irqs_disabled()) {
604 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
605 "enabled *very* early, fixing it\n");
606 local_irq_disable();
607 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 sort_main_extable();
609 trap_init();
610 rcu_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800611 /* init some links before init_ISA_irqs() */
612 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 init_IRQ();
614 pidhash_init();
615 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800616 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700618 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700619 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200620 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700621 profile_init();
622 if (!irqs_disabled())
623 printk("start_kernel(): bug: interrupts were enabled early\n");
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700624 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700625 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
627 /*
628 * HACK ALERT! This is early. We're enabling the console before
629 * we've done PCI setups etc, and console_init() must be aware of
630 * this. But we do want output early, in case something goes wrong.
631 */
632 console_init();
633 if (panic_later)
634 panic(panic_later, panic_param);
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700635
636 lockdep_info();
637
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700638 /*
639 * Need to run this when irqs are enabled, because it wants
640 * to self-test [hard/soft]-irqs on/off lock inversion bugs
641 * too:
642 */
643 locking_selftest();
644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645#ifdef CONFIG_BLK_DEV_INITRD
646 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700647 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200649 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700650 page_to_pfn(virt_to_page((void *)initrd_start)),
651 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 initrd_start = 0;
653 }
654#endif
Nick Piggindb64fe02008-10-18 20:27:03 -0700655 vmalloc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 vfs_caches_init_early();
Paul Jacksonc417f022006-01-08 01:02:01 -0800657 cpuset_init_early();
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700658 page_cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 mem_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100660 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100661 cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 kmem_cache_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700663 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700664 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700665 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 numa_policy_init();
667 if (late_time_init)
668 late_time_init();
669 calibrate_delay();
670 pidmap_init();
671 pgtable_cache_init();
672 prio_tree_init();
673 anon_vma_init();
674#ifdef CONFIG_X86
675 if (efi_enabled)
676 efi_enter_virtual_mode();
677#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000678 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100679 cred_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 fork_init(num_physpages);
681 proc_caches_init();
682 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 key_init();
684 security_init();
685 vfs_caches_init(num_physpages);
686 radix_tree_init();
687 signals_init();
688 /* rootfs populating might need page-writeback */
689 page_writeback_init();
690#ifdef CONFIG_PROC_FS
691 proc_root_init();
692#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700693 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700695 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700696 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697
698 check_bugs();
699
700 acpi_early_init(); /* before LAPIC and SMP init */
701
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400702 ftrace_init();
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 /* Do the rest non-__init'ed, we're now alive */
705 rest_init();
706}
707
Arjan van de Ven59f94152008-07-30 12:49:02 -0700708static int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500709core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710
Arjan van de Ven59f94152008-07-30 12:49:02 -0700711int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700712{
713 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100714 ktime_t calltime, delta, rettime;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700715 char msgbuf[64];
Frederic Weisbecker74239072008-11-11 23:24:42 +0100716 struct boot_trace_call call;
717 struct boot_trace_ret ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700718
719 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100720 call.caller = task_pid_nr(current);
721 printk("calling %pF @ %i\n", fn, call.caller);
722 calltime = ktime_get();
723 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100724 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700725 }
726
Frederic Weisbecker74239072008-11-11 23:24:42 +0100727 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700728
729 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100730 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100731 rettime = ktime_get();
732 delta = ktime_sub(rettime, calltime);
Will Newton1d926f22008-11-21 14:08:59 -0800733 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Frederic Weisbecker74239072008-11-11 23:24:42 +0100734 trace_boot_ret(&ret, fn);
Tim Birdca538f62008-10-09 15:23:05 -0700735 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbecker74239072008-11-11 23:24:42 +0100736 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700737 }
738
739 msgbuf[0] = 0;
740
Frederic Weisbecker74239072008-11-11 23:24:42 +0100741 if (ret.result && ret.result != -ENODEV && initcall_debug)
742 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700743
744 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700745 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700746 preempt_count() = count;
747 }
748 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700749 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700750 local_irq_enable();
751 }
752 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700753 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700754 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700755
Frederic Weisbecker74239072008-11-11 23:24:42 +0100756 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700757}
758
759
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700760extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
762static void __init do_initcalls(void)
763{
764 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700766 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700767 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
769 /* Make sure there is no pending stuff from the initcall sequence */
770 flush_scheduled_work();
771}
772
773/*
774 * Ok, the machine is now initialized. None of the devices
775 * have been touched yet, but the CPU subsystem is up and
776 * running, and memory and process management works.
777 *
778 * Now we can finally start doing some real work..
779 */
780static void __init do_basic_setup(void)
781{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200782 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700783 init_workqueues();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 usermodehelper_init();
785 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800786 init_irq_proc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 do_initcalls();
788}
789
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700790static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700791{
792 initcall_t *call;
793
794 for (call = __initcall_start; call < __early_initcall_end; call++)
795 do_one_initcall(*call);
796}
797
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798static void run_init_process(char *init_filename)
799{
800 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700801 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802}
803
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100804/* This is a non __init function. Force it to be noinline otherwise gcc
805 * makes it inline to init() and it becomes part of init.text section
806 */
807static int noinline init_post(void)
808{
809 free_initmem();
810 unlock_kernel();
811 mark_rodata_ro();
812 system_state = SYSTEM_RUNNING;
813 numa_default_policy();
814
815 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
816 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
817
818 (void) sys_dup(0);
819 (void) sys_dup(0);
820
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700821 current->signal->flags |= SIGNAL_UNKILLABLE;
822
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100823 if (ramdisk_execute_command) {
824 run_init_process(ramdisk_execute_command);
825 printk(KERN_WARNING "Failed to execute %s\n",
826 ramdisk_execute_command);
827 }
828
829 /*
830 * We try each of these until one succeeds.
831 *
832 * The Bourne shell can be used instead of init if we are
833 * trying to recover a really broken machine.
834 */
835 if (execute_command) {
836 run_init_process(execute_command);
837 printk(KERN_WARNING "Failed to execute %s. Attempting "
838 "defaults...\n", execute_command);
839 }
840 run_init_process("/sbin/init");
841 run_init_process("/etc/init");
842 run_init_process("/bin/init");
843 run_init_process("/bin/sh");
844
845 panic("No init found. Try passing init= option to kernel.");
846}
847
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100848static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849{
850 lock_kernel();
851 /*
852 * init can run on any cpu.
853 */
Mike Travis321a8e92008-04-04 18:11:02 -0700854 set_cpus_allowed_ptr(current, CPU_MASK_ALL_PTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 /*
856 * Tell the world that we're going to be the grim
857 * reaper of innocent orphaned children.
858 *
859 * We don't want people to have to make incorrect
860 * assumptions about where in the task array this
861 * can be found.
862 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800863 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700865 cad_pid = task_pid(current);
866
Andi Kleenca74a6f2008-01-30 13:33:17 +0100867 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868
869 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100870 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872 smp_init();
873 sched_init_smp();
874
875 cpuset_init_smp();
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 do_basic_setup();
878
879 /*
880 * check if there is an early userspace init. If yes, let it do all
881 * the work
882 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700883
884 if (!ramdisk_execute_command)
885 ramdisk_execute_command = "/init";
886
887 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
888 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891
892 /*
893 * Ok, we have completed the initial bootup, and
894 * we're essentially up and running. Get rid of the
895 * initmem segments and start the user-mode stuff..
896 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100897
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100898 init_post();
899 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900}