blob: f1b9f0fdb1b48ff07456b1420d54ee1c72b0e882 [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>
Len Brown4a7a16d2009-06-12 20:42:08 -040027#include <linux/acpi.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/tty.h>
29#include <linux/gfp.h>
30#include <linux/percpu.h>
31#include <linux/kmod.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070032#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/kernel_stat.h>
Rusty Russelld7cd5612006-12-07 02:14:08 +010034#include <linux/start_kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/security.h>
Jens Axboe3d442232008-06-26 11:21:34 +020036#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/workqueue.h>
38#include <linux/profile.h>
39#include <linux/rcupdate.h>
40#include <linux/moduleparam.h>
41#include <linux/kallsyms.h>
42#include <linux/writeback.h>
43#include <linux/cpu.h>
44#include <linux/cpuset.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070045#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046#include <linux/efi.h>
Thomas Gleixner906568c2007-02-16 01:28:01 -080047#include <linux/tick.h>
Andrew Morton6168a702007-02-17 21:22:39 -080048#include <linux/interrupt.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070049#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070050#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051#include <linux/unistd.h>
52#include <linux/rmap.h>
53#include <linux/mempolicy.h>
54#include <linux/key.h>
Adrian Bunkb6cd0b72006-06-27 02:53:54 -070055#include <linux/buffer_head.h>
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -070056#include <linux/page_cgroup.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070057#include <linux/debug_locks.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070058#include <linux/debugobjects.h>
Ingo Molnarfbb9ce952006-07-03 00:24:50 -070059#include <linux/lockdep.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080060#include <linux/pid_namespace.h>
Adrian Bunk1f217822006-12-19 13:01:28 -080061#include <linux/device.h>
Eric W. Biederman73c27992007-05-09 02:34:32 -070062#include <linux/kthread.h>
Adrian Bunke6fe6642007-11-09 22:39:39 +010063#include <linux/sched.h>
Adrian Bunka1c9eea2008-02-06 01:36:44 -080064#include <linux/signal.h>
Akinobu Mita199f0ca2008-04-29 01:03:13 -070065#include <linux/idr.h>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040066#include <linux/ftrace.h>
Arjan van de Ven22a9d642009-01-07 08:45:46 -080067#include <linux/async.h>
Frederic Weisbecker3f5ec132008-11-11 23:21:31 +010068#include <trace/boot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70#include <asm/io.h>
71#include <asm/bugs.h>
72#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070073#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080074#include <asm/cacheflush.h>
Frederic Weisbecker36994e52008-12-29 13:42:23 -080075#include <trace/kmemtrace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
Linus Torvalds1da177e2005-04-16 15:20:36 -070077#ifdef CONFIG_X86_LOCAL_APIC
78#include <asm/smp.h>
79#endif
80
Sam Ravnborgaae5f662007-02-26 16:45:41 +010081static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084extern void fork_init(unsigned long);
85extern void mca_init(void);
86extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087extern void prio_tree_init(void);
88extern void radix_tree_init(void);
89extern void free_initmem(void);
Arjan van de Ven37b73c82006-01-06 00:12:01 -080090#ifndef CONFIG_DEBUG_RODATA
91static inline void mark_rodata_ro(void) { }
92#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
94#ifdef CONFIG_TC
95extern void tc_init(void);
96#endif
97
Paul E. McKenneya6826042009-02-25 18:03:42 -080098enum system_states system_state __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -070099EXPORT_SYMBOL(system_state);
100
101/*
102 * Boot command-line arguments
103 */
104#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
105#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
106
107extern void time_init(void);
108/* Default late time init is NULL. archs can override this later. */
Jan Beulichd2e31922009-01-06 14:41:10 -0800109void (*__initdata late_time_init)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110extern void softirq_init(void);
111
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800112/* Untouched command line saved by arch-specific code. */
113char __initdata boot_command_line[COMMAND_LINE_SIZE];
114/* Untouched saved command line (eg. for /proc) */
115char *saved_command_line;
116/* Command line for parameter parsing */
117static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118
119static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700120static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121
Jan Beulich8b3b2952007-07-15 23:41:07 -0700122#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100124unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700125
126/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127 * Setup routine for controlling SMP activation
128 *
129 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
130 * activation entirely (the MPS table probe still happens, though).
131 *
132 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
133 * greater than 0, limits the maximum number of CPUs activated in
134 * SMP mode to <NUM>.
135 */
Ingo Molnar65a4e572009-01-31 03:36:17 +0100136
137void __weak arch_disable_smp_support(void) { }
Len Brown61ec7562007-08-16 03:34:22 -0400138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139static int __init nosmp(char *str)
140{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100141 setup_max_cpus = 0;
Ingo Molnar65a4e572009-01-31 03:36:17 +0100142 arch_disable_smp_support();
143
Jan Beulich8b3b2952007-07-15 23:41:07 -0700144 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145}
146
Jan Beulich8b3b2952007-07-15 23:41:07 -0700147early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148
149static int __init maxcpus(char *str)
150{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100151 get_option(&str, &setup_max_cpus);
152 if (setup_max_cpus == 0)
Ingo Molnar65a4e572009-01-31 03:36:17 +0100153 arch_disable_smp_support();
Len Brown61ec7562007-08-16 03:34:22 -0400154
155 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156}
157
Hugh Dickins81340972007-08-27 16:02:12 +0100158early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700159#else
Ingo Molnar65a4e572009-01-31 03:36:17 +0100160const unsigned int setup_max_cpus = NR_CPUS;
Jan Beulich8b3b2952007-07-15 23:41:07 -0700161#endif
162
163/*
164 * If set, this is an indication to the drivers that reset the underlying
165 * device before going ahead with the initialization otherwise driver might
166 * rely on the BIOS and skip the reset operation.
167 *
168 * This is useful if kernel is booting in an unreliable environment.
169 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
170 * skipped and devices will be in unknown state.
171 */
172unsigned int reset_devices;
173EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
Vivek Goyal7e962872006-09-27 01:50:44 -0700175static int __init set_reset_devices(char *str)
176{
177 reset_devices = 1;
178 return 1;
179}
180
181__setup("reset_devices", set_reset_devices);
182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
184char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
185static const char *panic_later, *panic_param;
186
187extern struct obs_kernel_param __setup_start[], __setup_end[];
188
189static int __init obsolete_checksetup(char *line)
190{
191 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200192 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193
194 p = __setup_start;
195 do {
196 int n = strlen(p->str);
197 if (!strncmp(line, p->str, n)) {
198 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200199 /* Already done in parse_early_param?
200 * (Needs exact match on param part).
201 * Keep iterating, as we can have early
202 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200204 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 } else if (!p->setup_func) {
206 printk(KERN_WARNING "Parameter %s is obsolete,"
207 " ignored\n", p->str);
208 return 1;
209 } else if (p->setup_func(line + n))
210 return 1;
211 }
212 p++;
213 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200214
215 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216}
217
218/*
219 * This should be approx 2 Bo*oMips to start (note initial shift), and will
220 * still work even if initially too large, it will just take slightly longer
221 */
222unsigned long loops_per_jiffy = (1<<12);
223
224EXPORT_SYMBOL(loops_per_jiffy);
225
226static int __init debug_kernel(char *str)
227{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800229 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230}
231
232static int __init quiet_kernel(char *str)
233{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800235 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236}
237
Yinghai Luf6f21c812008-02-08 04:21:58 -0800238early_param("debug", debug_kernel);
239early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240
241static int __init loglevel(char *str)
242{
243 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800244 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245}
246
Yinghai Luf6f21c812008-02-08 04:21:58 -0800247early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
249/*
250 * Unknown boot options get handed to init, unless they look like
251 * failed parameters
252 */
253static int __init unknown_bootoption(char *param, char *val)
254{
255 /* Change NUL term back to "=", to make "param" the whole string. */
256 if (val) {
257 /* param=val or param="val"? */
258 if (val == param+strlen(param)+1)
259 val[-1] = '=';
260 else if (val == param+strlen(param)+2) {
261 val[-2] = '=';
262 memmove(val-1, val, strlen(val)+1);
263 val--;
264 } else
265 BUG();
266 }
267
268 /* Handle obsolete-style parameters */
269 if (obsolete_checksetup(param))
270 return 0;
271
272 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200273 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 * line work?"
275 */
276 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
277 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
278 return 0;
279 }
280
281 if (panic_later)
282 return 0;
283
284 if (val) {
285 /* Environment option */
286 unsigned int i;
287 for (i = 0; envp_init[i]; i++) {
288 if (i == MAX_INIT_ENVS) {
289 panic_later = "Too many boot env vars at `%s'";
290 panic_param = param;
291 }
292 if (!strncmp(param, envp_init[i], val - param))
293 break;
294 }
295 envp_init[i] = param;
296 } else {
297 /* Command line option */
298 unsigned int i;
299 for (i = 0; argv_init[i]; i++) {
300 if (i == MAX_INIT_ARGS) {
301 panic_later = "Too many boot init vars at `%s'";
302 panic_param = param;
303 }
304 }
305 argv_init[i] = param;
306 }
307 return 0;
308}
309
Ingo Molnar12d6f212008-01-30 13:33:58 +0100310#ifdef CONFIG_DEBUG_PAGEALLOC
311int __read_mostly debug_pagealloc_enabled = 0;
312#endif
313
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314static int __init init_setup(char *str)
315{
316 unsigned int i;
317
318 execute_command = str;
319 /*
320 * In case LILO is going to boot us with default command line,
321 * it prepends "auto" before the whole cmdline which makes
322 * the shell think it should execute a script with such name.
323 * So we ignore all arguments entered _before_ init=... [MJ]
324 */
325 for (i = 1; i < MAX_INIT_ARGS; i++)
326 argv_init[i] = NULL;
327 return 1;
328}
329__setup("init=", init_setup);
330
Olof Johanssonffdfc402005-09-06 15:17:19 -0700331static int __init rdinit_setup(char *str)
332{
333 unsigned int i;
334
335 ramdisk_execute_command = str;
336 /* See "auto" comment in init_setup */
337 for (i = 1; i < MAX_INIT_ARGS; i++)
338 argv_init[i] = NULL;
339 return 1;
340}
341__setup("rdinit=", rdinit_setup);
342
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343#ifndef CONFIG_SMP
344
345#ifdef CONFIG_X86_LOCAL_APIC
346static void __init smp_init(void)
347{
348 APIC_init_uniprocessor();
349}
350#else
351#define smp_init() do { } while (0)
352#endif
353
354static inline void setup_per_cpu_areas(void) { }
Mike Travise0982e92008-03-26 14:23:48 -0700355static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356static inline void smp_prepare_cpus(unsigned int maxcpus) { }
357
358#else
359
Mike Travis321a8e92008-04-04 18:11:02 -0700360#if NR_CPUS > BITS_PER_LONG
361cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
362EXPORT_SYMBOL(cpu_mask_all);
363#endif
364
Mike Travise0982e92008-03-26 14:23:48 -0700365/* Setup number of possible processor ids */
366int nr_cpu_ids __read_mostly = NR_CPUS;
367EXPORT_SYMBOL(nr_cpu_ids);
368
369/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
370static void __init setup_nr_cpu_ids(void)
371{
Rusty Russelle0c0ba72009-01-01 10:12:19 +1030372 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
Mike Travise0982e92008-03-26 14:23:48 -0700373}
374
Mike Travisdd5af902008-01-30 13:33:32 +0100375#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800376unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
378EXPORT_SYMBOL(__per_cpu_offset);
379
380static void __init setup_per_cpu_areas(void)
381{
382 unsigned long size, i;
383 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800384 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
386 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200387 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
388 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800390 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 __per_cpu_offset[i] = ptr - __per_cpu_start;
392 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800393 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 }
395}
Mike Travisdd5af902008-01-30 13:33:32 +0100396#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
398/* Called by boot processor to activate the rest. */
399static void __init smp_init(void)
400{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800401 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Max Krasnyanskye761b772008-07-15 04:43:49 -0700403 /*
404 * Set up the current CPU as possible to migrate to.
405 * The other ones will be done by cpu_up/cpu_down()
406 */
Rusty Russell2b17fa52009-03-30 22:05:12 -0600407 set_cpu_active(smp_processor_id(), true);
Max Krasnyanskye761b772008-07-15 04:43:49 -0700408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800410 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100411 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800413 if (!cpu_online(cpu))
414 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 }
416
417 /* Any cleanup work */
418 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100419 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
422#endif
423
424/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800425 * We need to store the untouched command line for future reference.
426 * We also need to store the touched command line since the parameter
427 * parsing is performed in place, and we should allow a component to
428 * store reference of name/value for future reference.
429 */
430static void __init setup_command_line(char *command_line)
431{
432 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
433 static_command_line = alloc_bootmem(strlen (command_line)+1);
434 strcpy (saved_command_line, boot_command_line);
435 strcpy (static_command_line, command_line);
436}
437
438/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 * We need to finalize in a non-__init function or else race conditions
440 * between the root thread and the init thread may cause start_kernel to
441 * be reaped by free_initmem before the root thread has proceeded to
442 * cpu_idle.
443 *
444 * gcc-3.4 accidentally inlines this function, so use noinline.
445 */
446
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800447static noinline void __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 __releases(kernel_lock)
449{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700450 int pid;
451
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100452 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700454 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700455 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200457
458 /*
459 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200460 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200461 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200462 init_idle_bootup_task(current);
Paul E. McKenneya6826042009-02-25 18:03:42 -0800463 rcu_scheduler_starting();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800464 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200465 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800466 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200467
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800468 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200470}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472/* Check for early params. */
473static int __init do_early_param(char *param, char *val)
474{
475 struct obs_kernel_param *p;
476
477 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700478 if ((p->early && strcmp(param, p->str) == 0) ||
479 (strcmp(param, "console") == 0 &&
480 strcmp(p->str, "earlycon") == 0)
481 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 if (p->setup_func(val) != 0)
483 printk(KERN_WARNING
484 "Malformed early option '%s'\n", param);
485 }
486 }
487 /* We accept everything at this stage. */
488 return 0;
489}
490
Magnus Damm13977092009-03-30 14:37:25 -0700491void __init parse_early_options(char *cmdline)
492{
493 parse_args("early options", cmdline, NULL, 0, do_early_param);
494}
495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496/* Arch code calls this early on, or if not, just before other parsing. */
497void __init parse_early_param(void)
498{
499 static __initdata int done = 0;
500 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
501
502 if (done)
503 return;
504
505 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800506 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Magnus Damm13977092009-03-30 14:37:25 -0700507 parse_early_options(tmp_cmdline);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 done = 1;
509}
510
511/*
512 * Activate the first processor.
513 */
514
Stas Sergeev44fd2292006-03-23 02:59:44 -0800515static void __init boot_cpu_init(void)
516{
517 int cpu = smp_processor_id();
518 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030519 set_cpu_online(cpu, true);
520 set_cpu_present(cpu, true);
521 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800522}
523
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000524void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700525{
526}
527
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000528void __init __weak thread_info_cache_init(void)
529{
530}
531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532asmlinkage void __init start_kernel(void)
533{
534 char * command_line;
535 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700536
537 smp_setup_processor_id();
538
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700539 /*
540 * Need to run as early as possible, to initialize the
541 * lockdep hash:
542 */
543 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700544 debug_objects_early_init();
Ingo Molnar42059422008-02-14 09:44:08 +0100545
546 /*
547 * Set up the the initial canary ASAP:
548 */
549 boot_init_stack_canary();
550
Paul Menageddbcc7e2007-10-18 23:39:30 -0700551 cgroup_init_early();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700552
553 local_irq_disable();
554 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700555 early_init_irq_lock_class();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700556
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557/*
558 * Interrupts are still disabled. Do necessary setups, then
559 * enable them
560 */
561 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800562 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800563 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 page_address_init();
Alex Riesen657cafa2009-05-24 15:30:48 +0200565 printk(KERN_NOTICE "%s", linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700567 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800568 setup_command_line(command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700570 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800571 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 /*
574 * Set up the scheduler prior starting any interrupts (such as the
575 * timer interrupt). Full topology setup happens at smp_init()
576 * time - but meanwhile we still have a functioning scheduler.
577 */
578 sched_init();
579 /*
580 * Disable preemption - early bootup scheduling is extremely
581 * fragile until we cpu_idle() for the first time.
582 */
583 preempt_disable();
584 build_all_zonelists();
585 page_alloc_init();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800586 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 parse_early_param();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800588 parse_args("Booting kernel", static_command_line, __start___param,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 __stop___param - __start___param,
590 &unknown_bootoption);
Ard van Breemenc4a68302007-01-05 16:36:19 -0800591 if (!irqs_disabled()) {
592 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
593 "enabled *very* early, fixing it\n");
594 local_irq_disable();
595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 sort_main_extable();
597 trap_init();
598 rcu_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800599 /* init some links before init_ISA_irqs() */
600 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601 init_IRQ();
602 pidhash_init();
603 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800604 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700606 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700607 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200608 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700609 profile_init();
610 if (!irqs_disabled())
Ron Lee24d431d2008-11-27 02:31:57 +1030611 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
612 "enabled early\n");
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700613 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700614 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
616 /*
617 * HACK ALERT! This is early. We're enabling the console before
618 * we've done PCI setups etc, and console_init() must be aware of
619 * this. But we do want output early, in case something goes wrong.
620 */
621 console_init();
622 if (panic_later)
623 panic(panic_later, panic_param);
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700624
625 lockdep_info();
626
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700627 /*
628 * Need to run this when irqs are enabled, because it wants
629 * to self-test [hard/soft]-irqs on/off lock inversion bugs
630 * too:
631 */
632 locking_selftest();
633
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634#ifdef CONFIG_BLK_DEV_INITRD
635 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700636 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200638 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700639 page_to_pfn(virt_to_page((void *)initrd_start)),
640 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 initrd_start = 0;
642 }
643#endif
Nick Piggindb64fe02008-10-18 20:27:03 -0700644 vmalloc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 vfs_caches_init_early();
Paul Jacksonc417f022006-01-08 01:02:01 -0800646 cpuset_init_early();
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700647 page_cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 mem_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100649 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100650 cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 kmem_cache_init();
Eduard - Gabriel Munteanub9ce08c2008-08-10 20:14:03 +0300652 kmemtrace_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700653 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700654 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700655 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 numa_policy_init();
657 if (late_time_init)
658 late_time_init();
659 calibrate_delay();
660 pidmap_init();
661 pgtable_cache_init();
662 prio_tree_init();
663 anon_vma_init();
664#ifdef CONFIG_X86
665 if (efi_enabled)
666 efi_enter_virtual_mode();
667#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000668 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100669 cred_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 fork_init(num_physpages);
671 proc_caches_init();
672 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 key_init();
674 security_init();
675 vfs_caches_init(num_physpages);
676 radix_tree_init();
677 signals_init();
678 /* rootfs populating might need page-writeback */
679 page_writeback_init();
680#ifdef CONFIG_PROC_FS
681 proc_root_init();
682#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700683 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700685 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700686 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687
688 check_bugs();
689
690 acpi_early_init(); /* before LAPIC and SMP init */
691
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400692 ftrace_init();
693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 /* Do the rest non-__init'ed, we're now alive */
695 rest_init();
696}
697
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800698int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500699core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Arjan van de Ven59f94152008-07-30 12:49:02 -0700701int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700702{
703 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100704 ktime_t calltime, delta, rettime;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700705 char msgbuf[64];
Frederic Weisbecker74239072008-11-11 23:24:42 +0100706 struct boot_trace_call call;
707 struct boot_trace_ret ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700708
709 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100710 call.caller = task_pid_nr(current);
711 printk("calling %pF @ %i\n", fn, call.caller);
712 calltime = ktime_get();
713 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100714 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700715 }
716
Frederic Weisbecker74239072008-11-11 23:24:42 +0100717 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700718
719 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100720 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100721 rettime = ktime_get();
722 delta = ktime_sub(rettime, calltime);
Will Newton1d926f22008-11-21 14:08:59 -0800723 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Frederic Weisbecker74239072008-11-11 23:24:42 +0100724 trace_boot_ret(&ret, fn);
Tim Birdca538f62008-10-09 15:23:05 -0700725 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbecker74239072008-11-11 23:24:42 +0100726 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700727 }
728
729 msgbuf[0] = 0;
730
Frederic Weisbecker74239072008-11-11 23:24:42 +0100731 if (ret.result && ret.result != -ENODEV && initcall_debug)
732 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700733
734 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700735 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700736 preempt_count() = count;
737 }
738 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700739 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700740 local_irq_enable();
741 }
742 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700743 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700744 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700745
Frederic Weisbecker74239072008-11-11 23:24:42 +0100746 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700747}
748
749
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700750extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751
752static void __init do_initcalls(void)
753{
754 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700756 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700757 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758
759 /* Make sure there is no pending stuff from the initcall sequence */
760 flush_scheduled_work();
761}
762
763/*
764 * Ok, the machine is now initialized. None of the devices
765 * have been touched yet, but the CPU subsystem is up and
766 * running, and memory and process management works.
767 *
768 * Now we can finally start doing some real work..
769 */
770static void __init do_basic_setup(void)
771{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200772 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700773 init_workqueues();
Lai Jiangshan759ee092009-03-25 17:06:30 +0800774 cpuset_init_smp();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 usermodehelper_init();
776 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800777 init_irq_proc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 do_initcalls();
779}
780
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700781static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700782{
783 initcall_t *call;
784
785 for (call = __initcall_start; call < __early_initcall_end; call++)
786 do_one_initcall(*call);
787}
788
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789static void run_init_process(char *init_filename)
790{
791 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700792 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793}
794
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100795/* This is a non __init function. Force it to be noinline otherwise gcc
796 * makes it inline to init() and it becomes part of init.text section
797 */
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800798static noinline int init_post(void)
Hannes Ederacdd0522009-03-31 15:23:50 -0700799 __releases(kernel_lock)
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100800{
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800801 /* need to finish all async __init code before freeing the memory */
802 async_synchronize_full();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100803 free_initmem();
804 unlock_kernel();
805 mark_rodata_ro();
806 system_state = SYSTEM_RUNNING;
807 numa_default_policy();
808
809 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
810 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
811
812 (void) sys_dup(0);
813 (void) sys_dup(0);
814
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700815 current->signal->flags |= SIGNAL_UNKILLABLE;
816
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100817 if (ramdisk_execute_command) {
818 run_init_process(ramdisk_execute_command);
819 printk(KERN_WARNING "Failed to execute %s\n",
820 ramdisk_execute_command);
821 }
822
823 /*
824 * We try each of these until one succeeds.
825 *
826 * The Bourne shell can be used instead of init if we are
827 * trying to recover a really broken machine.
828 */
829 if (execute_command) {
830 run_init_process(execute_command);
831 printk(KERN_WARNING "Failed to execute %s. Attempting "
832 "defaults...\n", execute_command);
833 }
834 run_init_process("/sbin/init");
835 run_init_process("/etc/init");
836 run_init_process("/bin/init");
837 run_init_process("/bin/sh");
838
839 panic("No init found. Try passing init= option to kernel.");
840}
841
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100842static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843{
844 lock_kernel();
845 /*
846 * init can run on any cpu.
847 */
Rusty Russell1a2142a2009-03-30 22:05:10 -0600848 set_cpus_allowed_ptr(current, cpu_all_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 /*
850 * Tell the world that we're going to be the grim
851 * reaper of innocent orphaned children.
852 *
853 * We don't want people to have to make incorrect
854 * assumptions about where in the task array this
855 * can be found.
856 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800857 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700859 cad_pid = task_pid(current);
860
Andi Kleenca74a6f2008-01-30 13:33:17 +0100861 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862
863 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100864 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866 smp_init();
867 sched_init_smp();
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 do_basic_setup();
870
871 /*
872 * check if there is an early userspace init. If yes, let it do all
873 * the work
874 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700875
876 if (!ramdisk_execute_command)
877 ramdisk_execute_command = "/init";
878
879 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
880 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700882 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
884 /*
885 * Ok, we have completed the initial bootup, and
886 * we're essentially up and running. Get rid of the
887 * initmem segments and start the user-mode stuff..
888 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100889
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100890 init_post();
891 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}