blob: bfe4fb0c98428bdb4cfc9fb4b2a6cbb1084a0da1 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/init/main.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * GK 2/5/95 - Changed to support mounting root fs via NFS
7 * Added initrd & change_root: Werner Almesberger & Hans Lermen, Feb '96
8 * Moan early if gcc is old, avoiding bogus kernels - Paul Gortmaker, May '96
9 * Simplified starting of init: Michael A. Griffith <grif@acm.org>
10 */
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/types.h>
13#include <linux/module.h>
14#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/kernel.h>
16#include <linux/syscalls.h>
Ingo Molnar9b5609f2008-02-14 09:41:09 +010017#include <linux/stackprotector.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include <linux/string.h>
19#include <linux/ctype.h>
20#include <linux/delay.h>
21#include <linux/utsname.h>
22#include <linux/ioport.h>
23#include <linux/init.h>
24#include <linux/smp_lock.h>
25#include <linux/initrd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/bootmem.h>
27#include <linux/tty.h>
28#include <linux/gfp.h>
29#include <linux/percpu.h>
30#include <linux/kmod.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070031#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/kernel_stat.h>
Rusty Russelld7cd5612006-12-07 02:14:08 +010033#include <linux/start_kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/security.h>
Jens Axboe3d442232008-06-26 11:21:34 +020035#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/workqueue.h>
37#include <linux/profile.h>
38#include <linux/rcupdate.h>
39#include <linux/moduleparam.h>
40#include <linux/kallsyms.h>
41#include <linux/writeback.h>
42#include <linux/cpu.h>
43#include <linux/cpuset.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070044#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/efi.h>
Thomas Gleixner906568c2007-02-16 01:28:01 -080046#include <linux/tick.h>
Andrew Morton6168a702007-02-17 21:22:39 -080047#include <linux/interrupt.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070048#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070049#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/unistd.h>
51#include <linux/rmap.h>
52#include <linux/mempolicy.h>
53#include <linux/key.h>
Adrian Bunkb6cd0b72006-06-27 02:53:54 -070054#include <linux/buffer_head.h>
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -070055#include <linux/page_cgroup.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070056#include <linux/debug_locks.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070057#include <linux/debugobjects.h>
Ingo Molnarfbb9ce952006-07-03 00:24:50 -070058#include <linux/lockdep.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080059#include <linux/pid_namespace.h>
Adrian Bunk1f217822006-12-19 13:01:28 -080060#include <linux/device.h>
Eric W. Biederman73c27992007-05-09 02:34:32 -070061#include <linux/kthread.h>
Adrian Bunke6fe6642007-11-09 22:39:39 +010062#include <linux/sched.h>
Adrian Bunka1c9eea2008-02-06 01:36:44 -080063#include <linux/signal.h>
Akinobu Mita199f0ca2008-04-29 01:03:13 -070064#include <linux/idr.h>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040065#include <linux/ftrace.h>
Arjan van de Ven22a9d642009-01-07 08:45:46 -080066#include <linux/async.h>
Frederic Weisbecker3f5ec132008-11-11 23:21:31 +010067#include <trace/boot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69#include <asm/io.h>
70#include <asm/bugs.h>
71#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070072#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080073#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
Linus Torvalds1da177e2005-04-16 15:20:36 -070075#ifdef CONFIG_X86_LOCAL_APIC
76#include <asm/smp.h>
77#endif
78
Sam Ravnborgaae5f662007-02-26 16:45:41 +010079static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082extern void fork_init(unsigned long);
83extern void mca_init(void);
84extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085extern void prio_tree_init(void);
86extern void radix_tree_init(void);
87extern void free_initmem(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088#ifdef CONFIG_ACPI
89extern void acpi_early_init(void);
90#else
91static inline void acpi_early_init(void) { }
92#endif
Arjan van de Ven37b73c82006-01-06 00:12:01 -080093#ifndef CONFIG_DEBUG_RODATA
94static inline void mark_rodata_ro(void) { }
95#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070096
97#ifdef CONFIG_TC
98extern void tc_init(void);
99#endif
100
101enum system_states system_state;
102EXPORT_SYMBOL(system_state);
103
104/*
105 * Boot command-line arguments
106 */
107#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
108#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
109
110extern void time_init(void);
111/* Default late time init is NULL. archs can override this later. */
Jan Beulichd2e31922009-01-06 14:41:10 -0800112void (*__initdata late_time_init)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113extern void softirq_init(void);
114
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800115/* Untouched command line saved by arch-specific code. */
116char __initdata boot_command_line[COMMAND_LINE_SIZE];
117/* Untouched saved command line (eg. for /proc) */
118char *saved_command_line;
119/* Command line for parameter parsing */
120static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
122static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700123static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
Jan Beulich8b3b2952007-07-15 23:41:07 -0700125#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100127unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700128
129/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 * Setup routine for controlling SMP activation
131 *
132 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
133 * activation entirely (the MPS table probe still happens, though).
134 *
135 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
136 * greater than 0, limits the maximum number of CPUs activated in
137 * SMP mode to <NUM>.
138 */
Len Brown61ec7562007-08-16 03:34:22 -0400139#ifndef CONFIG_X86_IO_APIC
140static inline void disable_ioapic_setup(void) {};
141#endif
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143static int __init nosmp(char *str)
144{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100145 setup_max_cpus = 0;
Len Brown61ec7562007-08-16 03:34:22 -0400146 disable_ioapic_setup();
Jan Beulich8b3b2952007-07-15 23:41:07 -0700147 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
Jan Beulich8b3b2952007-07-15 23:41:07 -0700150early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
152static int __init maxcpus(char *str)
153{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100154 get_option(&str, &setup_max_cpus);
155 if (setup_max_cpus == 0)
Len Brown61ec7562007-08-16 03:34:22 -0400156 disable_ioapic_setup();
157
158 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159}
160
Hugh Dickins81340972007-08-27 16:02:12 +0100161early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700162#else
Andi Kleenca74a6f2008-01-30 13:33:17 +0100163#define setup_max_cpus NR_CPUS
Jan Beulich8b3b2952007-07-15 23:41:07 -0700164#endif
165
166/*
167 * If set, this is an indication to the drivers that reset the underlying
168 * device before going ahead with the initialization otherwise driver might
169 * rely on the BIOS and skip the reset operation.
170 *
171 * This is useful if kernel is booting in an unreliable environment.
172 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
173 * skipped and devices will be in unknown state.
174 */
175unsigned int reset_devices;
176EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
Vivek Goyal7e962872006-09-27 01:50:44 -0700178static int __init set_reset_devices(char *str)
179{
180 reset_devices = 1;
181 return 1;
182}
183
184__setup("reset_devices", set_reset_devices);
185
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
187char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
188static const char *panic_later, *panic_param;
189
190extern struct obs_kernel_param __setup_start[], __setup_end[];
191
192static int __init obsolete_checksetup(char *line)
193{
194 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200195 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
197 p = __setup_start;
198 do {
199 int n = strlen(p->str);
200 if (!strncmp(line, p->str, n)) {
201 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200202 /* Already done in parse_early_param?
203 * (Needs exact match on param part).
204 * Keep iterating, as we can have early
205 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200207 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 } else if (!p->setup_func) {
209 printk(KERN_WARNING "Parameter %s is obsolete,"
210 " ignored\n", p->str);
211 return 1;
212 } else if (p->setup_func(line + n))
213 return 1;
214 }
215 p++;
216 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200217
218 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
220
221/*
222 * This should be approx 2 Bo*oMips to start (note initial shift), and will
223 * still work even if initially too large, it will just take slightly longer
224 */
225unsigned long loops_per_jiffy = (1<<12);
226
227EXPORT_SYMBOL(loops_per_jiffy);
228
229static int __init debug_kernel(char *str)
230{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800232 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
235static int __init quiet_kernel(char *str)
236{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800238 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
Yinghai Luf6f21c812008-02-08 04:21:58 -0800241early_param("debug", debug_kernel);
242early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243
244static int __init loglevel(char *str)
245{
246 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800247 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248}
249
Yinghai Luf6f21c812008-02-08 04:21:58 -0800250early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251
252/*
253 * Unknown boot options get handed to init, unless they look like
254 * failed parameters
255 */
256static int __init unknown_bootoption(char *param, char *val)
257{
258 /* Change NUL term back to "=", to make "param" the whole string. */
259 if (val) {
260 /* param=val or param="val"? */
261 if (val == param+strlen(param)+1)
262 val[-1] = '=';
263 else if (val == param+strlen(param)+2) {
264 val[-2] = '=';
265 memmove(val-1, val, strlen(val)+1);
266 val--;
267 } else
268 BUG();
269 }
270
271 /* Handle obsolete-style parameters */
272 if (obsolete_checksetup(param))
273 return 0;
274
275 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200276 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 * line work?"
278 */
279 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
280 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
281 return 0;
282 }
283
284 if (panic_later)
285 return 0;
286
287 if (val) {
288 /* Environment option */
289 unsigned int i;
290 for (i = 0; envp_init[i]; i++) {
291 if (i == MAX_INIT_ENVS) {
292 panic_later = "Too many boot env vars at `%s'";
293 panic_param = param;
294 }
295 if (!strncmp(param, envp_init[i], val - param))
296 break;
297 }
298 envp_init[i] = param;
299 } else {
300 /* Command line option */
301 unsigned int i;
302 for (i = 0; argv_init[i]; i++) {
303 if (i == MAX_INIT_ARGS) {
304 panic_later = "Too many boot init vars at `%s'";
305 panic_param = param;
306 }
307 }
308 argv_init[i] = param;
309 }
310 return 0;
311}
312
Ingo Molnar12d6f212008-01-30 13:33:58 +0100313#ifdef CONFIG_DEBUG_PAGEALLOC
314int __read_mostly debug_pagealloc_enabled = 0;
315#endif
316
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317static int __init init_setup(char *str)
318{
319 unsigned int i;
320
321 execute_command = str;
322 /*
323 * In case LILO is going to boot us with default command line,
324 * it prepends "auto" before the whole cmdline which makes
325 * the shell think it should execute a script with such name.
326 * So we ignore all arguments entered _before_ init=... [MJ]
327 */
328 for (i = 1; i < MAX_INIT_ARGS; i++)
329 argv_init[i] = NULL;
330 return 1;
331}
332__setup("init=", init_setup);
333
Olof Johanssonffdfc402005-09-06 15:17:19 -0700334static int __init rdinit_setup(char *str)
335{
336 unsigned int i;
337
338 ramdisk_execute_command = str;
339 /* See "auto" comment in init_setup */
340 for (i = 1; i < MAX_INIT_ARGS; i++)
341 argv_init[i] = NULL;
342 return 1;
343}
344__setup("rdinit=", rdinit_setup);
345
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346#ifndef CONFIG_SMP
347
348#ifdef CONFIG_X86_LOCAL_APIC
349static void __init smp_init(void)
350{
351 APIC_init_uniprocessor();
352}
353#else
354#define smp_init() do { } while (0)
355#endif
356
357static inline void setup_per_cpu_areas(void) { }
Mike Travise0982e92008-03-26 14:23:48 -0700358static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359static inline void smp_prepare_cpus(unsigned int maxcpus) { }
360
361#else
362
Mike Travis321a8e92008-04-04 18:11:02 -0700363#if NR_CPUS > BITS_PER_LONG
364cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
365EXPORT_SYMBOL(cpu_mask_all);
366#endif
367
Mike Travise0982e92008-03-26 14:23:48 -0700368/* Setup number of possible processor ids */
369int nr_cpu_ids __read_mostly = NR_CPUS;
370EXPORT_SYMBOL(nr_cpu_ids);
371
372/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
373static void __init setup_nr_cpu_ids(void)
374{
Rusty Russelle0c0ba72009-01-01 10:12:19 +1030375 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
Mike Travise0982e92008-03-26 14:23:48 -0700376}
377
Mike Travisdd5af902008-01-30 13:33:32 +0100378#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800379unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
381EXPORT_SYMBOL(__per_cpu_offset);
382
383static void __init setup_per_cpu_areas(void)
384{
385 unsigned long size, i;
386 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800387 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
389 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200390 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
391 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800393 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 __per_cpu_offset[i] = ptr - __per_cpu_start;
395 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800396 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 }
398}
Mike Travisdd5af902008-01-30 13:33:32 +0100399#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
401/* Called by boot processor to activate the rest. */
402static void __init smp_init(void)
403{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800404 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405
Max Krasnyanskye761b772008-07-15 04:43:49 -0700406 /*
407 * Set up the current CPU as possible to migrate to.
408 * The other ones will be done by cpu_up/cpu_down()
409 */
410 cpu = smp_processor_id();
411 cpu_set(cpu, cpu_active_map);
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800414 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100415 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800417 if (!cpu_online(cpu))
418 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 }
420
421 /* Any cleanup work */
422 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100423 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
426#endif
427
428/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800429 * We need to store the untouched command line for future reference.
430 * We also need to store the touched command line since the parameter
431 * parsing is performed in place, and we should allow a component to
432 * store reference of name/value for future reference.
433 */
434static void __init setup_command_line(char *command_line)
435{
436 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
437 static_command_line = alloc_bootmem(strlen (command_line)+1);
438 strcpy (saved_command_line, boot_command_line);
439 strcpy (static_command_line, command_line);
440}
441
442/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443 * We need to finalize in a non-__init function or else race conditions
444 * between the root thread and the init thread may cause start_kernel to
445 * be reaped by free_initmem before the root thread has proceeded to
446 * cpu_idle.
447 *
448 * gcc-3.4 accidentally inlines this function, so use noinline.
449 */
450
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800451static noinline void __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 __releases(kernel_lock)
453{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700454 int pid;
455
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100456 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700458 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700459 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200461
462 /*
463 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200464 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200465 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200466 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800467 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200468 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800469 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200470
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800471 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200473}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
475/* Check for early params. */
476static int __init do_early_param(char *param, char *val)
477{
478 struct obs_kernel_param *p;
479
480 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700481 if ((p->early && strcmp(param, p->str) == 0) ||
482 (strcmp(param, "console") == 0 &&
483 strcmp(p->str, "earlycon") == 0)
484 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 if (p->setup_func(val) != 0)
486 printk(KERN_WARNING
487 "Malformed early option '%s'\n", param);
488 }
489 }
490 /* We accept everything at this stage. */
491 return 0;
492}
493
494/* Arch code calls this early on, or if not, just before other parsing. */
495void __init parse_early_param(void)
496{
497 static __initdata int done = 0;
498 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
499
500 if (done)
501 return;
502
503 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800504 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
506 done = 1;
507}
508
509/*
510 * Activate the first processor.
511 */
512
Stas Sergeev44fd2292006-03-23 02:59:44 -0800513static void __init boot_cpu_init(void)
514{
515 int cpu = smp_processor_id();
516 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030517 set_cpu_online(cpu, true);
518 set_cpu_present(cpu, true);
519 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800520}
521
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000522void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700523{
524}
525
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000526void __init __weak thread_info_cache_init(void)
527{
528}
529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530asmlinkage void __init start_kernel(void)
531{
532 char * command_line;
533 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700534
535 smp_setup_processor_id();
536
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700537 /*
538 * Need to run as early as possible, to initialize the
539 * lockdep hash:
540 */
541 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700542 debug_objects_early_init();
Ingo Molnar42059422008-02-14 09:44:08 +0100543
544 /*
545 * Set up the the initial canary ASAP:
546 */
547 boot_init_stack_canary();
548
Paul Menageddbcc7e2007-10-18 23:39:30 -0700549 cgroup_init_early();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700550
551 local_irq_disable();
552 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700553 early_init_irq_lock_class();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700554
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555/*
556 * Interrupts are still disabled. Do necessary setups, then
557 * enable them
558 */
559 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800560 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800561 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 page_address_init();
563 printk(KERN_NOTICE);
Linus Torvalds89937802006-12-11 09:28:46 -0800564 printk(linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700566 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800567 setup_command_line(command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700569 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800570 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
572 /*
573 * Set up the scheduler prior starting any interrupts (such as the
574 * timer interrupt). Full topology setup happens at smp_init()
575 * time - but meanwhile we still have a functioning scheduler.
576 */
577 sched_init();
578 /*
579 * Disable preemption - early bootup scheduling is extremely
580 * fragile until we cpu_idle() for the first time.
581 */
582 preempt_disable();
583 build_all_zonelists();
584 page_alloc_init();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800585 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 parse_early_param();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800587 parse_args("Booting kernel", static_command_line, __start___param,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 __stop___param - __start___param,
589 &unknown_bootoption);
Ard van Breemenc4a68302007-01-05 16:36:19 -0800590 if (!irqs_disabled()) {
591 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
592 "enabled *very* early, fixing it\n");
593 local_irq_disable();
594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 sort_main_extable();
596 trap_init();
597 rcu_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800598 /* init some links before init_ISA_irqs() */
599 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 init_IRQ();
601 pidhash_init();
602 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800603 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700605 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700606 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200607 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700608 profile_init();
609 if (!irqs_disabled())
Ron Lee24d431d2008-11-27 02:31:57 +1030610 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
611 "enabled early\n");
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700612 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700613 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
615 /*
616 * HACK ALERT! This is early. We're enabling the console before
617 * we've done PCI setups etc, and console_init() must be aware of
618 * this. But we do want output early, in case something goes wrong.
619 */
620 console_init();
621 if (panic_later)
622 panic(panic_later, panic_param);
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700623
624 lockdep_info();
625
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700626 /*
627 * Need to run this when irqs are enabled, because it wants
628 * to self-test [hard/soft]-irqs on/off lock inversion bugs
629 * too:
630 */
631 locking_selftest();
632
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633#ifdef CONFIG_BLK_DEV_INITRD
634 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700635 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200637 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700638 page_to_pfn(virt_to_page((void *)initrd_start)),
639 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 initrd_start = 0;
641 }
642#endif
Nick Piggindb64fe02008-10-18 20:27:03 -0700643 vmalloc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 vfs_caches_init_early();
Paul Jacksonc417f022006-01-08 01:02:01 -0800645 cpuset_init_early();
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700646 page_cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 mem_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100648 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100649 cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 kmem_cache_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700651 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700652 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700653 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 numa_policy_init();
655 if (late_time_init)
656 late_time_init();
657 calibrate_delay();
658 pidmap_init();
659 pgtable_cache_init();
660 prio_tree_init();
661 anon_vma_init();
662#ifdef CONFIG_X86
663 if (efi_enabled)
664 efi_enter_virtual_mode();
665#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000666 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100667 cred_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 fork_init(num_physpages);
669 proc_caches_init();
670 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 key_init();
672 security_init();
673 vfs_caches_init(num_physpages);
674 radix_tree_init();
675 signals_init();
676 /* rootfs populating might need page-writeback */
677 page_writeback_init();
678#ifdef CONFIG_PROC_FS
679 proc_root_init();
680#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700681 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700683 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700684 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
686 check_bugs();
687
688 acpi_early_init(); /* before LAPIC and SMP init */
689
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400690 ftrace_init();
691
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 /* Do the rest non-__init'ed, we're now alive */
693 rest_init();
694}
695
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800696int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500697core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698
Arjan van de Ven59f94152008-07-30 12:49:02 -0700699int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700700{
701 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100702 ktime_t calltime, delta, rettime;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700703 char msgbuf[64];
Frederic Weisbecker74239072008-11-11 23:24:42 +0100704 struct boot_trace_call call;
705 struct boot_trace_ret ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700706
707 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100708 call.caller = task_pid_nr(current);
709 printk("calling %pF @ %i\n", fn, call.caller);
710 calltime = ktime_get();
711 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100712 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700713 }
714
Frederic Weisbecker74239072008-11-11 23:24:42 +0100715 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700716
717 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100718 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100719 rettime = ktime_get();
720 delta = ktime_sub(rettime, calltime);
Will Newton1d926f22008-11-21 14:08:59 -0800721 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Frederic Weisbecker74239072008-11-11 23:24:42 +0100722 trace_boot_ret(&ret, fn);
Tim Birdca538f62008-10-09 15:23:05 -0700723 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbecker74239072008-11-11 23:24:42 +0100724 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700725 }
726
727 msgbuf[0] = 0;
728
Frederic Weisbecker74239072008-11-11 23:24:42 +0100729 if (ret.result && ret.result != -ENODEV && initcall_debug)
730 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700731
732 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700733 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700734 preempt_count() = count;
735 }
736 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700737 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700738 local_irq_enable();
739 }
740 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700741 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700742 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700743
Frederic Weisbecker74239072008-11-11 23:24:42 +0100744 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700745}
746
747
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700748extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749
750static void __init do_initcalls(void)
751{
752 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700754 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700755 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756
757 /* Make sure there is no pending stuff from the initcall sequence */
758 flush_scheduled_work();
759}
760
761/*
762 * Ok, the machine is now initialized. None of the devices
763 * have been touched yet, but the CPU subsystem is up and
764 * running, and memory and process management works.
765 *
766 * Now we can finally start doing some real work..
767 */
768static void __init do_basic_setup(void)
769{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200770 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 init_workqueues();
772 usermodehelper_init();
773 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800774 init_irq_proc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 do_initcalls();
776}
777
Vivek Goyal88d20322007-01-11 01:52:44 +0100778static void __init do_pre_smp_initcalls(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779{
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700780 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700782 for (call = __initcall_start; call < __early_initcall_end; call++)
783 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
785
786static void run_init_process(char *init_filename)
787{
788 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700789 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790}
791
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100792/* This is a non __init function. Force it to be noinline otherwise gcc
793 * makes it inline to init() and it becomes part of init.text section
794 */
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800795static noinline int init_post(void)
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100796{
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800797 /* need to finish all async __init code before freeing the memory */
798 async_synchronize_full();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100799 free_initmem();
800 unlock_kernel();
801 mark_rodata_ro();
802 system_state = SYSTEM_RUNNING;
803 numa_default_policy();
804
805 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
806 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
807
808 (void) sys_dup(0);
809 (void) sys_dup(0);
810
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700811 current->signal->flags |= SIGNAL_UNKILLABLE;
812
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100813 if (ramdisk_execute_command) {
814 run_init_process(ramdisk_execute_command);
815 printk(KERN_WARNING "Failed to execute %s\n",
816 ramdisk_execute_command);
817 }
818
819 /*
820 * We try each of these until one succeeds.
821 *
822 * The Bourne shell can be used instead of init if we are
823 * trying to recover a really broken machine.
824 */
825 if (execute_command) {
826 run_init_process(execute_command);
827 printk(KERN_WARNING "Failed to execute %s. Attempting "
828 "defaults...\n", execute_command);
829 }
830 run_init_process("/sbin/init");
831 run_init_process("/etc/init");
832 run_init_process("/bin/init");
833 run_init_process("/bin/sh");
834
835 panic("No init found. Try passing init= option to kernel.");
836}
837
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100838static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
840 lock_kernel();
841 /*
842 * init can run on any cpu.
843 */
Mike Travis321a8e92008-04-04 18:11:02 -0700844 set_cpus_allowed_ptr(current, CPU_MASK_ALL_PTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 /*
846 * Tell the world that we're going to be the grim
847 * reaper of innocent orphaned children.
848 *
849 * We don't want people to have to make incorrect
850 * assumptions about where in the task array this
851 * can be found.
852 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800853 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700855 cad_pid = task_pid(current);
856
Andi Kleenca74a6f2008-01-30 13:33:17 +0100857 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
859 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100860 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 smp_init();
863 sched_init_smp();
864
865 cpuset_init_smp();
866
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 do_basic_setup();
868
869 /*
870 * check if there is an early userspace init. If yes, let it do all
871 * the work
872 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700873
874 if (!ramdisk_execute_command)
875 ramdisk_execute_command = "/init";
876
877 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
878 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700880 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
882 /*
883 * Ok, we have completed the initial bootup, and
884 * we're essentially up and running. Get rid of the
885 * initmem segments and start the user-mode stuff..
886 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100887
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100888 init_post();
889 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890}