blob: e810196bf2f2ae2522fd7ce1ec0dbae3d7f6ebd9 [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{
383 int cpu, highest_cpu = 0;
384
385 for_each_possible_cpu(cpu)
386 highest_cpu = cpu;
387
388 nr_cpu_ids = highest_cpu + 1;
389}
390
Mike Travisdd5af902008-01-30 13:33:32 +0100391#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800392unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393
394EXPORT_SYMBOL(__per_cpu_offset);
395
396static void __init setup_per_cpu_areas(void)
397{
398 unsigned long size, i;
399 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800400 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
402 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200403 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
404 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800406 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 __per_cpu_offset[i] = ptr - __per_cpu_start;
408 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800409 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 }
411}
Mike Travisdd5af902008-01-30 13:33:32 +0100412#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
414/* Called by boot processor to activate the rest. */
415static void __init smp_init(void)
416{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800417 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
Max Krasnyanskye761b772008-07-15 04:43:49 -0700419 /*
420 * Set up the current CPU as possible to migrate to.
421 * The other ones will be done by cpu_up/cpu_down()
422 */
423 cpu = smp_processor_id();
424 cpu_set(cpu, cpu_active_map);
425
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800427 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100428 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800430 if (!cpu_online(cpu))
431 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 }
433
434 /* Any cleanup work */
435 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100436 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437}
438
439#endif
440
441/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800442 * We need to store the untouched command line for future reference.
443 * We also need to store the touched command line since the parameter
444 * parsing is performed in place, and we should allow a component to
445 * store reference of name/value for future reference.
446 */
447static void __init setup_command_line(char *command_line)
448{
449 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
450 static_command_line = alloc_bootmem(strlen (command_line)+1);
451 strcpy (saved_command_line, boot_command_line);
452 strcpy (static_command_line, command_line);
453}
454
455/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 * We need to finalize in a non-__init function or else race conditions
457 * between the root thread and the init thread may cause start_kernel to
458 * be reaped by free_initmem before the root thread has proceeded to
459 * cpu_idle.
460 *
461 * gcc-3.4 accidentally inlines this function, so use noinline.
462 */
463
Sam Ravnborg92080302007-05-17 20:43:54 +0200464static void noinline __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 __releases(kernel_lock)
466{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700467 int pid;
468
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100469 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700471 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700472 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200474
475 /*
476 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200477 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200478 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200479 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800480 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200481 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800482 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200483
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800484 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200486}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
488/* Check for early params. */
489static int __init do_early_param(char *param, char *val)
490{
491 struct obs_kernel_param *p;
492
493 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700494 if ((p->early && strcmp(param, p->str) == 0) ||
495 (strcmp(param, "console") == 0 &&
496 strcmp(p->str, "earlycon") == 0)
497 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 if (p->setup_func(val) != 0)
499 printk(KERN_WARNING
500 "Malformed early option '%s'\n", param);
501 }
502 }
503 /* We accept everything at this stage. */
504 return 0;
505}
506
507/* Arch code calls this early on, or if not, just before other parsing. */
508void __init parse_early_param(void)
509{
510 static __initdata int done = 0;
511 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
512
513 if (done)
514 return;
515
516 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800517 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
519 done = 1;
520}
521
522/*
523 * Activate the first processor.
524 */
525
Stas Sergeev44fd2292006-03-23 02:59:44 -0800526static void __init boot_cpu_init(void)
527{
528 int cpu = smp_processor_id();
529 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
530 cpu_set(cpu, cpu_online_map);
531 cpu_set(cpu, cpu_present_map);
532 cpu_set(cpu, cpu_possible_map);
533}
534
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000535void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700536{
537}
538
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000539void __init __weak thread_info_cache_init(void)
540{
541}
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543asmlinkage void __init start_kernel(void)
544{
545 char * command_line;
546 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700547
548 smp_setup_processor_id();
549
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700550 /*
551 * Need to run as early as possible, to initialize the
552 * lockdep hash:
553 */
Andi Kleenc9538ed2006-09-26 10:52:34 +0200554 unwind_init();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700555 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700556 debug_objects_early_init();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700557 cgroup_init_early();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700558
559 local_irq_disable();
560 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700561 early_init_irq_lock_class();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/*
564 * Interrupts are still disabled. Do necessary setups, then
565 * enable them
566 */
567 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800568 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800569 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 page_address_init();
571 printk(KERN_NOTICE);
Linus Torvalds89937802006-12-11 09:28:46 -0800572 printk(linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700574 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800575 setup_command_line(command_line);
Jan Beulich690a973f2006-10-21 18:37:01 +0200576 unwind_setup();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700578 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800579 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 /*
582 * Set up the scheduler prior starting any interrupts (such as the
583 * timer interrupt). Full topology setup happens at smp_init()
584 * time - but meanwhile we still have a functioning scheduler.
585 */
586 sched_init();
587 /*
588 * Disable preemption - early bootup scheduling is extremely
589 * fragile until we cpu_idle() for the first time.
590 */
591 preempt_disable();
592 build_all_zonelists();
593 page_alloc_init();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800594 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 parse_early_param();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800596 parse_args("Booting kernel", static_command_line, __start___param,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 __stop___param - __start___param,
598 &unknown_bootoption);
Ard van Breemenc4a68302007-01-05 16:36:19 -0800599 if (!irqs_disabled()) {
600 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
601 "enabled *very* early, fixing it\n");
602 local_irq_disable();
603 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 sort_main_extable();
605 trap_init();
606 rcu_init();
607 init_IRQ();
608 pidhash_init();
609 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800610 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700612 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700613 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200614 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700615 profile_init();
616 if (!irqs_disabled())
617 printk("start_kernel(): bug: interrupts were enabled early\n");
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700618 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700619 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 /*
622 * HACK ALERT! This is early. We're enabling the console before
623 * we've done PCI setups etc, and console_init() must be aware of
624 * this. But we do want output early, in case something goes wrong.
625 */
626 console_init();
627 if (panic_later)
628 panic(panic_later, panic_param);
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700629
630 lockdep_info();
631
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700632 /*
633 * Need to run this when irqs are enabled, because it wants
634 * to self-test [hard/soft]-irqs on/off lock inversion bugs
635 * too:
636 */
637 locking_selftest();
638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639#ifdef CONFIG_BLK_DEV_INITRD
640 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700641 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200643 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700644 page_to_pfn(virt_to_page((void *)initrd_start)),
645 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 initrd_start = 0;
647 }
648#endif
Nick Piggindb64fe02008-10-18 20:27:03 -0700649 vmalloc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 vfs_caches_init_early();
Paul Jacksonc417f022006-01-08 01:02:01 -0800651 cpuset_init_early();
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700652 page_cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 mem_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100654 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100655 cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 kmem_cache_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700657 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700658 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700659 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 numa_policy_init();
661 if (late_time_init)
662 late_time_init();
663 calibrate_delay();
664 pidmap_init();
665 pgtable_cache_init();
666 prio_tree_init();
667 anon_vma_init();
668#ifdef CONFIG_X86
669 if (efi_enabled)
670 efi_enter_virtual_mode();
671#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000672 thread_info_cache_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 fork_init(num_physpages);
674 proc_caches_init();
675 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 key_init();
677 security_init();
678 vfs_caches_init(num_physpages);
679 radix_tree_init();
680 signals_init();
681 /* rootfs populating might need page-writeback */
682 page_writeback_init();
683#ifdef CONFIG_PROC_FS
684 proc_root_init();
685#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700686 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700688 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700689 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690
691 check_bugs();
692
693 acpi_early_init(); /* before LAPIC and SMP init */
694
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400695 ftrace_init();
696
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 /* Do the rest non-__init'ed, we're now alive */
698 rest_init();
699}
700
Arjan van de Ven59f94152008-07-30 12:49:02 -0700701static int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500702core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Arjan van de Ven59f94152008-07-30 12:49:02 -0700704int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700705{
706 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100707 ktime_t calltime, delta, rettime;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700708 char msgbuf[64];
Frederic Weisbecker74239072008-11-11 23:24:42 +0100709 struct boot_trace_call call;
710 struct boot_trace_ret ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700711
712 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100713 call.caller = task_pid_nr(current);
714 printk("calling %pF @ %i\n", fn, call.caller);
715 calltime = ktime_get();
716 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100717 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700718 }
719
Frederic Weisbecker74239072008-11-11 23:24:42 +0100720 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700721
722 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100723 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100724 rettime = ktime_get();
725 delta = ktime_sub(rettime, calltime);
726 ret.duration = (unsigned long long) delta.tv64 >> 10;
727 trace_boot_ret(&ret, fn);
728 printk("initcall %pF returned %d after %Ld usecs\n", fn,
729 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700730 }
731
732 msgbuf[0] = 0;
733
Frederic Weisbecker74239072008-11-11 23:24:42 +0100734 if (ret.result && ret.result != -ENODEV && initcall_debug)
735 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700736
737 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700738 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700739 preempt_count() = count;
740 }
741 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700742 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700743 local_irq_enable();
744 }
745 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700746 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700747 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700748
Frederic Weisbecker74239072008-11-11 23:24:42 +0100749 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700750}
751
752
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700753extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
755static void __init do_initcalls(void)
756{
757 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700759 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700760 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
762 /* Make sure there is no pending stuff from the initcall sequence */
763 flush_scheduled_work();
764}
765
766/*
767 * Ok, the machine is now initialized. None of the devices
768 * have been touched yet, but the CPU subsystem is up and
769 * running, and memory and process management works.
770 *
771 * Now we can finally start doing some real work..
772 */
773static void __init do_basic_setup(void)
774{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200775 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700776 init_workqueues();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 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 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100890
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100891 init_post();
892 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893}