blob: 602d724afa5c985296a6bd1bf68ee9335cfeb3bf [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 Molnar9b5609fd2008-02-14 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 Molnarfbb9ce92006-07-03 00:24:50 -070058#include <linux/lockdep.h>
Catalin Marinas3c7b4e62009-06-11 13:22:39 +010059#include <linux/kmemleak.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>
Vegard Nossumdfec0722008-04-04 00:51:41 +020068#include <linux/kmemcheck.h>
Zhaolei02af61b2009-04-10 14:26:18 +080069#include <linux/kmemtrace.h>
Frederic Weisbecker3f5ec132008-11-11 23:21:31 +010070#include <trace/boot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
72#include <asm/io.h>
73#include <asm/bugs.h>
74#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070075#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080076#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070077
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#ifdef CONFIG_X86_LOCAL_APIC
79#include <asm/smp.h>
80#endif
81
Sam Ravnborgaae5f662007-02-26 16:45:41 +010082static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
84extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085extern void fork_init(unsigned long);
86extern void mca_init(void);
87extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070088extern void prio_tree_init(void);
89extern void radix_tree_init(void);
90extern void free_initmem(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091#ifdef CONFIG_ACPI
92extern void acpi_early_init(void);
93#else
94static inline void acpi_early_init(void) { }
95#endif
Arjan van de Ven37b73c82006-01-06 00:12:01 -080096#ifndef CONFIG_DEBUG_RODATA
97static inline void mark_rodata_ro(void) { }
98#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
100#ifdef CONFIG_TC
101extern void tc_init(void);
102#endif
103
Paul E. McKenneya6826042009-02-25 18:03:42 -0800104enum system_states system_state __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105EXPORT_SYMBOL(system_state);
106
107/*
108 * Boot command-line arguments
109 */
110#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
111#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
112
113extern void time_init(void);
114/* Default late time init is NULL. archs can override this later. */
Jan Beulichd2e31922009-01-06 14:41:10 -0800115void (*__initdata late_time_init)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116extern void softirq_init(void);
117
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800118/* Untouched command line saved by arch-specific code. */
119char __initdata boot_command_line[COMMAND_LINE_SIZE];
120/* Untouched saved command line (eg. for /proc) */
121char *saved_command_line;
122/* Command line for parameter parsing */
123static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124
125static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700126static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Jan Beulich8b3b2952007-07-15 23:41:07 -0700128#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100130unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700131
132/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 * Setup routine for controlling SMP activation
134 *
135 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
136 * activation entirely (the MPS table probe still happens, though).
137 *
138 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
139 * greater than 0, limits the maximum number of CPUs activated in
140 * SMP mode to <NUM>.
141 */
Ingo Molnar65a4e572009-01-31 03:36:17 +0100142
143void __weak arch_disable_smp_support(void) { }
Len Brown61ec7562007-08-16 03:34:22 -0400144
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145static int __init nosmp(char *str)
146{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100147 setup_max_cpus = 0;
Ingo Molnar65a4e572009-01-31 03:36:17 +0100148 arch_disable_smp_support();
149
Jan Beulich8b3b2952007-07-15 23:41:07 -0700150 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151}
152
Jan Beulich8b3b2952007-07-15 23:41:07 -0700153early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
155static int __init maxcpus(char *str)
156{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100157 get_option(&str, &setup_max_cpus);
158 if (setup_max_cpus == 0)
Ingo Molnar65a4e572009-01-31 03:36:17 +0100159 arch_disable_smp_support();
Len Brown61ec7562007-08-16 03:34:22 -0400160
161 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162}
163
Hugh Dickins81340972007-08-27 16:02:12 +0100164early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700165#else
Ingo Molnar65a4e572009-01-31 03:36:17 +0100166const unsigned int setup_max_cpus = NR_CPUS;
Jan Beulich8b3b2952007-07-15 23:41:07 -0700167#endif
168
169/*
170 * If set, this is an indication to the drivers that reset the underlying
171 * device before going ahead with the initialization otherwise driver might
172 * rely on the BIOS and skip the reset operation.
173 *
174 * This is useful if kernel is booting in an unreliable environment.
175 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
176 * skipped and devices will be in unknown state.
177 */
178unsigned int reset_devices;
179EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180
Vivek Goyal7e962872006-09-27 01:50:44 -0700181static int __init set_reset_devices(char *str)
182{
183 reset_devices = 1;
184 return 1;
185}
186
187__setup("reset_devices", set_reset_devices);
188
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
190char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
191static const char *panic_later, *panic_param;
192
193extern struct obs_kernel_param __setup_start[], __setup_end[];
194
195static int __init obsolete_checksetup(char *line)
196{
197 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200198 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
200 p = __setup_start;
201 do {
202 int n = strlen(p->str);
203 if (!strncmp(line, p->str, n)) {
204 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200205 /* Already done in parse_early_param?
206 * (Needs exact match on param part).
207 * Keep iterating, as we can have early
208 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200210 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 } else if (!p->setup_func) {
212 printk(KERN_WARNING "Parameter %s is obsolete,"
213 " ignored\n", p->str);
214 return 1;
215 } else if (p->setup_func(line + n))
216 return 1;
217 }
218 p++;
219 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200220
221 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222}
223
224/*
225 * This should be approx 2 Bo*oMips to start (note initial shift), and will
226 * still work even if initially too large, it will just take slightly longer
227 */
228unsigned long loops_per_jiffy = (1<<12);
229
230EXPORT_SYMBOL(loops_per_jiffy);
231
232static int __init debug_kernel(char *str)
233{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800235 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236}
237
238static int __init quiet_kernel(char *str)
239{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800241 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242}
243
Yinghai Luf6f21c812008-02-08 04:21:58 -0800244early_param("debug", debug_kernel);
245early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246
247static int __init loglevel(char *str)
248{
249 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800250 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
Yinghai Luf6f21c812008-02-08 04:21:58 -0800253early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
255/*
256 * Unknown boot options get handed to init, unless they look like
257 * failed parameters
258 */
259static int __init unknown_bootoption(char *param, char *val)
260{
261 /* Change NUL term back to "=", to make "param" the whole string. */
262 if (val) {
263 /* param=val or param="val"? */
264 if (val == param+strlen(param)+1)
265 val[-1] = '=';
266 else if (val == param+strlen(param)+2) {
267 val[-2] = '=';
268 memmove(val-1, val, strlen(val)+1);
269 val--;
270 } else
271 BUG();
272 }
273
274 /* Handle obsolete-style parameters */
275 if (obsolete_checksetup(param))
276 return 0;
277
278 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200279 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 * line work?"
281 */
282 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
283 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
284 return 0;
285 }
286
287 if (panic_later)
288 return 0;
289
290 if (val) {
291 /* Environment option */
292 unsigned int i;
293 for (i = 0; envp_init[i]; i++) {
294 if (i == MAX_INIT_ENVS) {
295 panic_later = "Too many boot env vars at `%s'";
296 panic_param = param;
297 }
298 if (!strncmp(param, envp_init[i], val - param))
299 break;
300 }
301 envp_init[i] = param;
302 } else {
303 /* Command line option */
304 unsigned int i;
305 for (i = 0; argv_init[i]; i++) {
306 if (i == MAX_INIT_ARGS) {
307 panic_later = "Too many boot init vars at `%s'";
308 panic_param = param;
309 }
310 }
311 argv_init[i] = param;
312 }
313 return 0;
314}
315
Ingo Molnar12d6f212008-01-30 13:33:58 +0100316#ifdef CONFIG_DEBUG_PAGEALLOC
317int __read_mostly debug_pagealloc_enabled = 0;
318#endif
319
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320static int __init init_setup(char *str)
321{
322 unsigned int i;
323
324 execute_command = str;
325 /*
326 * In case LILO is going to boot us with default command line,
327 * it prepends "auto" before the whole cmdline which makes
328 * the shell think it should execute a script with such name.
329 * So we ignore all arguments entered _before_ init=... [MJ]
330 */
331 for (i = 1; i < MAX_INIT_ARGS; i++)
332 argv_init[i] = NULL;
333 return 1;
334}
335__setup("init=", init_setup);
336
Olof Johanssonffdfc402005-09-06 15:17:19 -0700337static int __init rdinit_setup(char *str)
338{
339 unsigned int i;
340
341 ramdisk_execute_command = str;
342 /* See "auto" comment in init_setup */
343 for (i = 1; i < MAX_INIT_ARGS; i++)
344 argv_init[i] = NULL;
345 return 1;
346}
347__setup("rdinit=", rdinit_setup);
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349#ifndef CONFIG_SMP
350
351#ifdef CONFIG_X86_LOCAL_APIC
352static void __init smp_init(void)
353{
354 APIC_init_uniprocessor();
355}
356#else
357#define smp_init() do { } while (0)
358#endif
359
Mike Travise0982e92008-03-26 14:23:48 -0700360static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361static inline void smp_prepare_cpus(unsigned int maxcpus) { }
362
363#else
364
Mike Travis321a8e92008-04-04 18:11:02 -0700365#if NR_CPUS > BITS_PER_LONG
366cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
367EXPORT_SYMBOL(cpu_mask_all);
368#endif
369
Mike Travise0982e92008-03-26 14:23:48 -0700370/* Setup number of possible processor ids */
371int nr_cpu_ids __read_mostly = NR_CPUS;
372EXPORT_SYMBOL(nr_cpu_ids);
373
374/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
375static void __init setup_nr_cpu_ids(void)
376{
Rusty Russelle0c0ba72009-01-01 10:12:19 +1030377 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
Mike Travise0982e92008-03-26 14:23:48 -0700378}
379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380/* Called by boot processor to activate the rest. */
381static void __init smp_init(void)
382{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800383 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
Max Krasnyanskye761b772008-07-15 04:43:49 -0700385 /*
386 * Set up the current CPU as possible to migrate to.
387 * The other ones will be done by cpu_up/cpu_down()
388 */
Rusty Russell2b17fa52009-03-30 22:05:12 -0600389 set_cpu_active(smp_processor_id(), true);
Max Krasnyanskye761b772008-07-15 04:43:49 -0700390
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800392 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100393 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800395 if (!cpu_online(cpu))
396 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 }
398
399 /* Any cleanup work */
400 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100401 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402}
403
404#endif
405
406/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800407 * We need to store the untouched command line for future reference.
408 * We also need to store the touched command line since the parameter
409 * parsing is performed in place, and we should allow a component to
410 * store reference of name/value for future reference.
411 */
412static void __init setup_command_line(char *command_line)
413{
414 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
415 static_command_line = alloc_bootmem(strlen (command_line)+1);
416 strcpy (saved_command_line, boot_command_line);
417 strcpy (static_command_line, command_line);
418}
419
420/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 * We need to finalize in a non-__init function or else race conditions
422 * between the root thread and the init thread may cause start_kernel to
423 * be reaped by free_initmem before the root thread has proceeded to
424 * cpu_idle.
425 *
426 * gcc-3.4 accidentally inlines this function, so use noinline.
427 */
428
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800429static noinline void __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 __releases(kernel_lock)
431{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700432 int pid;
433
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100434 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700436 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700437 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200439
440 /*
441 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200442 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200443 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200444 init_idle_bootup_task(current);
Paul E. McKenneya6826042009-02-25 18:03:42 -0800445 rcu_scheduler_starting();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800446 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200447 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800448 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200449
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800450 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200452}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453
454/* Check for early params. */
455static int __init do_early_param(char *param, char *val)
456{
457 struct obs_kernel_param *p;
458
459 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700460 if ((p->early && strcmp(param, p->str) == 0) ||
461 (strcmp(param, "console") == 0 &&
462 strcmp(p->str, "earlycon") == 0)
463 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 if (p->setup_func(val) != 0)
465 printk(KERN_WARNING
466 "Malformed early option '%s'\n", param);
467 }
468 }
469 /* We accept everything at this stage. */
470 return 0;
471}
472
Magnus Damm13977092009-03-30 14:37:25 -0700473void __init parse_early_options(char *cmdline)
474{
475 parse_args("early options", cmdline, NULL, 0, do_early_param);
476}
477
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478/* Arch code calls this early on, or if not, just before other parsing. */
479void __init parse_early_param(void)
480{
481 static __initdata int done = 0;
482 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
483
484 if (done)
485 return;
486
487 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800488 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Magnus Damm13977092009-03-30 14:37:25 -0700489 parse_early_options(tmp_cmdline);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 done = 1;
491}
492
493/*
494 * Activate the first processor.
495 */
496
Stas Sergeev44fd2292006-03-23 02:59:44 -0800497static void __init boot_cpu_init(void)
498{
499 int cpu = smp_processor_id();
500 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030501 set_cpu_online(cpu, true);
502 set_cpu_present(cpu, true);
503 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800504}
505
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000506void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700507{
508}
509
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000510void __init __weak thread_info_cache_init(void)
511{
512}
513
Pekka Enberg444f4782009-06-11 18:29:06 +0300514/*
515 * Set up kernel memory allocators
516 */
517static void __init mm_init(void)
518{
KAMEZAWA Hiroyukica371c02009-06-12 10:33:53 +0300519 /*
520 * page_cgroup requires countinous pages as memmap
521 * and it's bigger than MAX_ORDER unless SPARSEMEM.
522 */
523 page_cgroup_init_flatmem();
Pekka Enberg444f4782009-06-11 18:29:06 +0300524 mem_init();
525 kmem_cache_init();
Benjamin Herrenschmidtc868d552009-06-17 13:48:39 +1000526 pgtable_cache_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300527 vmalloc_init();
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 Molnarfbb9ce92006-07-03 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 Molnarfbb9ce92006-07-03 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 Molnarfbb9ce92006-07-03 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();
Alex Riesen657cafa2009-05-24 15:30:48 +0200563 printk(KERN_NOTICE "%s", linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700565 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800566 setup_command_line(command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700568 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800569 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Pekka Enberg83b519e2009-06-10 19:40:04 +0300571 build_all_zonelists();
572 page_alloc_init();
573
574 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
575 parse_early_param();
576 parse_args("Booting kernel", static_command_line, __start___param,
577 __stop___param - __start___param,
578 &unknown_bootoption);
579 /*
580 * These use large bootmem allocations and must precede
581 * kmem_cache_init()
582 */
583 pidhash_init();
Pekka Enberg83b519e2009-06-10 19:40:04 +0300584 vfs_caches_init_early();
585 sort_main_extable();
586 trap_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300587 mm_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 /*
589 * Set up the scheduler prior starting any interrupts (such as the
590 * timer interrupt). Full topology setup happens at smp_init()
591 * time - but meanwhile we still have a functioning scheduler.
592 */
593 sched_init();
594 /*
595 * Disable preemption - early bootup scheduling is extremely
596 * fragile until we cpu_idle() for the first time.
597 */
598 preempt_disable();
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 rcu_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800605 /* init some links before init_ISA_irqs() */
606 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 init_IRQ();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100608 prio_tree_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 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())
Ron Lee24d431d2008-11-27 02:31:57 +1030617 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
618 "enabled early\n");
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700619 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700620 local_irq_enable();
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000621
622 /* Interrupts are enabled now so all GFP allocations are safe. */
623 set_gfp_allowed_mask(__GFP_BITS_MASK);
624
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300625 kmem_cache_init_late();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
627 /*
628 * HACK ALERT! This is early. We're enabling the console before
629 * we've done PCI setups etc, and console_init() must be aware of
630 * this. But we do want output early, in case something goes wrong.
631 */
632 console_init();
633 if (panic_later)
634 panic(panic_later, panic_param);
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700635
636 lockdep_info();
637
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700638 /*
639 * Need to run this when irqs are enabled, because it wants
640 * to self-test [hard/soft]-irqs on/off lock inversion bugs
641 * too:
642 */
643 locking_selftest();
644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645#ifdef CONFIG_BLK_DEV_INITRD
646 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700647 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200649 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700650 page_to_pfn(virt_to_page((void *)initrd_start)),
651 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 initrd_start = 0;
653 }
654#endif
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700655 page_cgroup_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100656 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100657 cpu_hotplug_init();
Eduard - Gabriel Munteanub9ce08c2008-08-10 20:14:03 +0300658 kmemtrace_init();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100659 kmemleak_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700660 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700661 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700662 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 numa_policy_init();
664 if (late_time_init)
665 late_time_init();
666 calibrate_delay();
667 pidmap_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 anon_vma_init();
669#ifdef CONFIG_X86
670 if (efi_enabled)
671 efi_enter_virtual_mode();
672#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000673 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100674 cred_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 fork_init(num_physpages);
676 proc_caches_init();
677 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 key_init();
679 security_init();
680 vfs_caches_init(num_physpages);
681 radix_tree_init();
682 signals_init();
683 /* rootfs populating might need page-writeback */
684 page_writeback_init();
685#ifdef CONFIG_PROC_FS
686 proc_root_init();
687#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700688 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700690 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700691 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
693 check_bugs();
694
695 acpi_early_init(); /* before LAPIC and SMP init */
696
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400697 ftrace_init();
698
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 /* Do the rest non-__init'ed, we're now alive */
700 rest_init();
701}
702
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700703/* Call all constructor functions linked into the kernel. */
704static void __init do_ctors(void)
705{
706#ifdef CONFIG_CONSTRUCTORS
707 ctor_fn_t *call = (ctor_fn_t *) __ctors_start;
708
709 for (; call < (ctor_fn_t *) __ctors_end; call++)
710 (*call)();
711#endif
712}
713
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800714int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500715core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716
Arjan van de Ven59f94152008-07-30 12:49:02 -0700717int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700718{
719 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100720 ktime_t calltime, delta, rettime;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700721 char msgbuf[64];
Frederic Weisbecker74239072008-11-11 23:24:42 +0100722 struct boot_trace_call call;
723 struct boot_trace_ret ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700724
725 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100726 call.caller = task_pid_nr(current);
727 printk("calling %pF @ %i\n", fn, call.caller);
728 calltime = ktime_get();
729 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100730 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700731 }
732
Frederic Weisbecker74239072008-11-11 23:24:42 +0100733 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700734
735 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100736 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100737 rettime = ktime_get();
738 delta = ktime_sub(rettime, calltime);
Will Newton1d926f22008-11-21 14:08:59 -0800739 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Frederic Weisbecker74239072008-11-11 23:24:42 +0100740 trace_boot_ret(&ret, fn);
Tim Birdca538f62008-10-09 15:23:05 -0700741 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbecker74239072008-11-11 23:24:42 +0100742 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700743 }
744
745 msgbuf[0] = 0;
746
Frederic Weisbecker74239072008-11-11 23:24:42 +0100747 if (ret.result && ret.result != -ENODEV && initcall_debug)
748 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700749
750 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700751 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700752 preempt_count() = count;
753 }
754 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700755 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700756 local_irq_enable();
757 }
758 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700759 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700760 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700761
Frederic Weisbecker74239072008-11-11 23:24:42 +0100762 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700763}
764
765
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700766extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
768static void __init do_initcalls(void)
769{
770 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700772 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700773 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
775 /* Make sure there is no pending stuff from the initcall sequence */
776 flush_scheduled_work();
777}
778
779/*
780 * Ok, the machine is now initialized. None of the devices
781 * have been touched yet, but the CPU subsystem is up and
782 * running, and memory and process management works.
783 *
784 * Now we can finally start doing some real work..
785 */
786static void __init do_basic_setup(void)
787{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200788 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700789 init_workqueues();
Lai Jiangshan759ee092009-03-25 17:06:30 +0800790 cpuset_init_smp();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 usermodehelper_init();
792 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800793 init_irq_proc();
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700794 do_ctors();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 do_initcalls();
796}
797
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700798static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700799{
800 initcall_t *call;
801
802 for (call = __initcall_start; call < __early_initcall_end; call++)
803 do_one_initcall(*call);
804}
805
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806static void run_init_process(char *init_filename)
807{
808 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700809 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700810}
811
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100812/* This is a non __init function. Force it to be noinline otherwise gcc
813 * makes it inline to init() and it becomes part of init.text section
814 */
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800815static noinline int init_post(void)
Hannes Ederacdd0522009-03-31 15:23:50 -0700816 __releases(kernel_lock)
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100817{
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800818 /* need to finish all async __init code before freeing the memory */
819 async_synchronize_full();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100820 free_initmem();
821 unlock_kernel();
822 mark_rodata_ro();
823 system_state = SYSTEM_RUNNING;
824 numa_default_policy();
825
826 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
827 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
828
829 (void) sys_dup(0);
830 (void) sys_dup(0);
831
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700832 current->signal->flags |= SIGNAL_UNKILLABLE;
833
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100834 if (ramdisk_execute_command) {
835 run_init_process(ramdisk_execute_command);
836 printk(KERN_WARNING "Failed to execute %s\n",
837 ramdisk_execute_command);
838 }
839
840 /*
841 * We try each of these until one succeeds.
842 *
843 * The Bourne shell can be used instead of init if we are
844 * trying to recover a really broken machine.
845 */
846 if (execute_command) {
847 run_init_process(execute_command);
848 printk(KERN_WARNING "Failed to execute %s. Attempting "
849 "defaults...\n", execute_command);
850 }
851 run_init_process("/sbin/init");
852 run_init_process("/etc/init");
853 run_init_process("/bin/init");
854 run_init_process("/bin/sh");
855
856 panic("No init found. Try passing init= option to kernel.");
857}
858
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100859static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860{
861 lock_kernel();
Miao Xie58568d22009-06-16 15:31:49 -0700862
863 /*
864 * init can allocate pages on any node
865 */
866 set_mems_allowed(node_possible_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 /*
868 * init can run on any cpu.
869 */
Rusty Russell1a2142a2009-03-30 22:05:10 -0600870 set_cpus_allowed_ptr(current, cpu_all_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 /*
872 * Tell the world that we're going to be the grim
873 * reaper of innocent orphaned children.
874 *
875 * We don't want people to have to make incorrect
876 * assumptions about where in the task array this
877 * can be found.
878 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800879 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700881 cad_pid = task_pid(current);
882
Andi Kleenca74a6f2008-01-30 13:33:17 +0100883 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884
885 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100886 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 smp_init();
889 sched_init_smp();
890
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 do_basic_setup();
892
893 /*
894 * check if there is an early userspace init. If yes, let it do all
895 * the work
896 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700897
898 if (!ramdisk_execute_command)
899 ramdisk_execute_command = "/init";
900
901 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
902 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700904 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905
906 /*
907 * Ok, we have completed the initial bootup, and
908 * we're essentially up and running. Get rid of the
909 * initmem segments and start the user-mode stuff..
910 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100911
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100912 init_post();
913 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914}