blob: 63904bb6ae37bf30b1e0b59fc266af7a080491cf [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>
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 Molnarfbb9ce92006-07-03 00:24:50 -070059#include <linux/lockdep.h>
Catalin Marinas3c7b4e62009-06-11 13:22:39 +010060#include <linux/kmemleak.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080061#include <linux/pid_namespace.h>
Adrian Bunk1f217822006-12-19 13:01:28 -080062#include <linux/device.h>
Eric W. Biederman73c27992007-05-09 02:34:32 -070063#include <linux/kthread.h>
Adrian Bunke6fe6642007-11-09 22:39:39 +010064#include <linux/sched.h>
Adrian Bunka1c9eea2008-02-06 01:36:44 -080065#include <linux/signal.h>
Akinobu Mita199f0ca2008-04-29 01:03:13 -070066#include <linux/idr.h>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040067#include <linux/ftrace.h>
Arjan van de Ven22a9d642009-01-07 08:45:46 -080068#include <linux/async.h>
Vegard Nossumdfec0722008-04-04 00:51:41 +020069#include <linux/kmemcheck.h>
Zhaolei02af61b2009-04-10 14:26:18 +080070#include <linux/kmemtrace.h>
Frederic Weisbecker3f5ec132008-11-11 23:21:31 +010071#include <trace/boot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73#include <asm/io.h>
74#include <asm/bugs.h>
75#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070076#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080077#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
Linus Torvalds1da177e2005-04-16 15:20:36 -070079#ifdef CONFIG_X86_LOCAL_APIC
80#include <asm/smp.h>
81#endif
82
Sam Ravnborgaae5f662007-02-26 16:45:41 +010083static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086extern void fork_init(unsigned long);
87extern void mca_init(void);
88extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070089extern void prio_tree_init(void);
90extern void radix_tree_init(void);
91extern void free_initmem(void);
Arjan van de Ven37b73c82006-01-06 00:12:01 -080092#ifndef CONFIG_DEBUG_RODATA
93static inline void mark_rodata_ro(void) { }
94#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070095
96#ifdef CONFIG_TC
97extern void tc_init(void);
98#endif
99
Paul E. McKenneya6826042009-02-25 18:03:42 -0800100enum system_states system_state __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101EXPORT_SYMBOL(system_state);
102
103/*
104 * Boot command-line arguments
105 */
106#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
107#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
108
109extern void time_init(void);
110/* Default late time init is NULL. archs can override this later. */
Jan Beulichd2e31922009-01-06 14:41:10 -0800111void (*__initdata late_time_init)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112extern void softirq_init(void);
113
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800114/* Untouched command line saved by arch-specific code. */
115char __initdata boot_command_line[COMMAND_LINE_SIZE];
116/* Untouched saved command line (eg. for /proc) */
117char *saved_command_line;
118/* Command line for parameter parsing */
119static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700122static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
Jan Beulich8b3b2952007-07-15 23:41:07 -0700124#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100126unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700127
128/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 * Setup routine for controlling SMP activation
130 *
131 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
132 * activation entirely (the MPS table probe still happens, though).
133 *
134 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
135 * greater than 0, limits the maximum number of CPUs activated in
136 * SMP mode to <NUM>.
137 */
Ingo Molnar65a4e572009-01-31 03:36:17 +0100138
139void __weak arch_disable_smp_support(void) { }
Len Brown61ec7562007-08-16 03:34:22 -0400140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141static int __init nosmp(char *str)
142{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100143 setup_max_cpus = 0;
Ingo Molnar65a4e572009-01-31 03:36:17 +0100144 arch_disable_smp_support();
145
Jan Beulich8b3b2952007-07-15 23:41:07 -0700146 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147}
148
Jan Beulich8b3b2952007-07-15 23:41:07 -0700149early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150
151static int __init maxcpus(char *str)
152{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100153 get_option(&str, &setup_max_cpus);
154 if (setup_max_cpus == 0)
Ingo Molnar65a4e572009-01-31 03:36:17 +0100155 arch_disable_smp_support();
Len Brown61ec7562007-08-16 03:34:22 -0400156
157 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158}
159
Hugh Dickins81340972007-08-27 16:02:12 +0100160early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700161#else
Ingo Molnar65a4e572009-01-31 03:36:17 +0100162const unsigned int setup_max_cpus = NR_CPUS;
Jan Beulich8b3b2952007-07-15 23:41:07 -0700163#endif
164
165/*
166 * If set, this is an indication to the drivers that reset the underlying
167 * device before going ahead with the initialization otherwise driver might
168 * rely on the BIOS and skip the reset operation.
169 *
170 * This is useful if kernel is booting in an unreliable environment.
171 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
172 * skipped and devices will be in unknown state.
173 */
174unsigned int reset_devices;
175EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Vivek Goyal7e962872006-09-27 01:50:44 -0700177static int __init set_reset_devices(char *str)
178{
179 reset_devices = 1;
180 return 1;
181}
182
183__setup("reset_devices", set_reset_devices);
184
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
186char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
187static const char *panic_later, *panic_param;
188
189extern struct obs_kernel_param __setup_start[], __setup_end[];
190
191static int __init obsolete_checksetup(char *line)
192{
193 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200194 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196 p = __setup_start;
197 do {
198 int n = strlen(p->str);
199 if (!strncmp(line, p->str, n)) {
200 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200201 /* Already done in parse_early_param?
202 * (Needs exact match on param part).
203 * Keep iterating, as we can have early
204 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200206 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 } else if (!p->setup_func) {
208 printk(KERN_WARNING "Parameter %s is obsolete,"
209 " ignored\n", p->str);
210 return 1;
211 } else if (p->setup_func(line + n))
212 return 1;
213 }
214 p++;
215 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200216
217 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218}
219
220/*
221 * This should be approx 2 Bo*oMips to start (note initial shift), and will
222 * still work even if initially too large, it will just take slightly longer
223 */
224unsigned long loops_per_jiffy = (1<<12);
225
226EXPORT_SYMBOL(loops_per_jiffy);
227
228static int __init debug_kernel(char *str)
229{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800231 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232}
233
234static int __init quiet_kernel(char *str)
235{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800237 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
Yinghai Luf6f21c812008-02-08 04:21:58 -0800240early_param("debug", debug_kernel);
241early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243static int __init loglevel(char *str)
244{
245 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800246 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247}
248
Yinghai Luf6f21c812008-02-08 04:21:58 -0800249early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
251/*
252 * Unknown boot options get handed to init, unless they look like
253 * failed parameters
254 */
255static int __init unknown_bootoption(char *param, char *val)
256{
257 /* Change NUL term back to "=", to make "param" the whole string. */
258 if (val) {
259 /* param=val or param="val"? */
260 if (val == param+strlen(param)+1)
261 val[-1] = '=';
262 else if (val == param+strlen(param)+2) {
263 val[-2] = '=';
264 memmove(val-1, val, strlen(val)+1);
265 val--;
266 } else
267 BUG();
268 }
269
270 /* Handle obsolete-style parameters */
271 if (obsolete_checksetup(param))
272 return 0;
273
274 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200275 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 * line work?"
277 */
278 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
279 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
280 return 0;
281 }
282
283 if (panic_later)
284 return 0;
285
286 if (val) {
287 /* Environment option */
288 unsigned int i;
289 for (i = 0; envp_init[i]; i++) {
290 if (i == MAX_INIT_ENVS) {
291 panic_later = "Too many boot env vars at `%s'";
292 panic_param = param;
293 }
294 if (!strncmp(param, envp_init[i], val - param))
295 break;
296 }
297 envp_init[i] = param;
298 } else {
299 /* Command line option */
300 unsigned int i;
301 for (i = 0; argv_init[i]; i++) {
302 if (i == MAX_INIT_ARGS) {
303 panic_later = "Too many boot init vars at `%s'";
304 panic_param = param;
305 }
306 }
307 argv_init[i] = param;
308 }
309 return 0;
310}
311
Ingo Molnar12d6f212008-01-30 13:33:58 +0100312#ifdef CONFIG_DEBUG_PAGEALLOC
313int __read_mostly debug_pagealloc_enabled = 0;
314#endif
315
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316static int __init init_setup(char *str)
317{
318 unsigned int i;
319
320 execute_command = str;
321 /*
322 * In case LILO is going to boot us with default command line,
323 * it prepends "auto" before the whole cmdline which makes
324 * the shell think it should execute a script with such name.
325 * So we ignore all arguments entered _before_ init=... [MJ]
326 */
327 for (i = 1; i < MAX_INIT_ARGS; i++)
328 argv_init[i] = NULL;
329 return 1;
330}
331__setup("init=", init_setup);
332
Olof Johanssonffdfc402005-09-06 15:17:19 -0700333static int __init rdinit_setup(char *str)
334{
335 unsigned int i;
336
337 ramdisk_execute_command = str;
338 /* See "auto" comment in init_setup */
339 for (i = 1; i < MAX_INIT_ARGS; i++)
340 argv_init[i] = NULL;
341 return 1;
342}
343__setup("rdinit=", rdinit_setup);
344
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345#ifndef CONFIG_SMP
346
347#ifdef CONFIG_X86_LOCAL_APIC
348static void __init smp_init(void)
349{
350 APIC_init_uniprocessor();
351}
352#else
353#define smp_init() do { } while (0)
354#endif
355
Mike Travise0982e92008-03-26 14:23:48 -0700356static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357static inline void smp_prepare_cpus(unsigned int maxcpus) { }
358
359#else
360
Mike Travis321a8e92008-04-04 18:11:02 -0700361#if NR_CPUS > BITS_PER_LONG
362cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
363EXPORT_SYMBOL(cpu_mask_all);
364#endif
365
Mike Travise0982e92008-03-26 14:23:48 -0700366/* Setup number of possible processor ids */
367int nr_cpu_ids __read_mostly = NR_CPUS;
368EXPORT_SYMBOL(nr_cpu_ids);
369
370/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
371static void __init setup_nr_cpu_ids(void)
372{
Rusty Russelle0c0ba72009-01-01 10:12:19 +1030373 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
Mike Travise0982e92008-03-26 14:23:48 -0700374}
375
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376/* Called by boot processor to activate the rest. */
377static void __init smp_init(void)
378{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800379 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Max Krasnyanskye761b772008-07-15 04:43:49 -0700381 /*
382 * Set up the current CPU as possible to migrate to.
383 * The other ones will be done by cpu_up/cpu_down()
384 */
Rusty Russell2b17fa52009-03-30 22:05:12 -0600385 set_cpu_active(smp_processor_id(), true);
Max Krasnyanskye761b772008-07-15 04:43:49 -0700386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800388 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100389 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800391 if (!cpu_online(cpu))
392 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 }
394
395 /* Any cleanup work */
396 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100397 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398}
399
400#endif
401
402/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800403 * We need to store the untouched command line for future reference.
404 * We also need to store the touched command line since the parameter
405 * parsing is performed in place, and we should allow a component to
406 * store reference of name/value for future reference.
407 */
408static void __init setup_command_line(char *command_line)
409{
410 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
411 static_command_line = alloc_bootmem(strlen (command_line)+1);
412 strcpy (saved_command_line, boot_command_line);
413 strcpy (static_command_line, command_line);
414}
415
416/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 * We need to finalize in a non-__init function or else race conditions
418 * between the root thread and the init thread may cause start_kernel to
419 * be reaped by free_initmem before the root thread has proceeded to
420 * cpu_idle.
421 *
422 * gcc-3.4 accidentally inlines this function, so use noinline.
423 */
424
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800425static noinline void __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 __releases(kernel_lock)
427{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700428 int pid;
429
Paul E. McKenney7db905e2009-09-02 14:01:24 -0700430 rcu_scheduler_starting();
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100431 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700433 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700434 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200436
437 /*
438 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200439 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200440 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200441 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800442 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200443 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800444 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200445
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800446 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200448}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449
450/* Check for early params. */
451static int __init do_early_param(char *param, char *val)
452{
453 struct obs_kernel_param *p;
454
455 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700456 if ((p->early && strcmp(param, p->str) == 0) ||
457 (strcmp(param, "console") == 0 &&
458 strcmp(p->str, "earlycon") == 0)
459 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 if (p->setup_func(val) != 0)
461 printk(KERN_WARNING
462 "Malformed early option '%s'\n", param);
463 }
464 }
465 /* We accept everything at this stage. */
466 return 0;
467}
468
Magnus Damm13977092009-03-30 14:37:25 -0700469void __init parse_early_options(char *cmdline)
470{
471 parse_args("early options", cmdline, NULL, 0, do_early_param);
472}
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474/* Arch code calls this early on, or if not, just before other parsing. */
475void __init parse_early_param(void)
476{
477 static __initdata int done = 0;
478 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
479
480 if (done)
481 return;
482
483 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800484 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Magnus Damm13977092009-03-30 14:37:25 -0700485 parse_early_options(tmp_cmdline);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 done = 1;
487}
488
489/*
490 * Activate the first processor.
491 */
492
Stas Sergeev44fd2292006-03-23 02:59:44 -0800493static void __init boot_cpu_init(void)
494{
495 int cpu = smp_processor_id();
496 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030497 set_cpu_online(cpu, true);
498 set_cpu_present(cpu, true);
499 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800500}
501
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000502void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700503{
504}
505
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000506void __init __weak thread_info_cache_init(void)
507{
508}
509
Pekka Enberg444f4782009-06-11 18:29:06 +0300510/*
511 * Set up kernel memory allocators
512 */
513static void __init mm_init(void)
514{
KAMEZAWA Hiroyukica371c02009-06-12 10:33:53 +0300515 /*
516 * page_cgroup requires countinous pages as memmap
517 * and it's bigger than MAX_ORDER unless SPARSEMEM.
518 */
519 page_cgroup_init_flatmem();
Pekka Enberg444f4782009-06-11 18:29:06 +0300520 mem_init();
521 kmem_cache_init();
Benjamin Herrenschmidtc868d552009-06-17 13:48:39 +1000522 pgtable_cache_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300523 vmalloc_init();
524}
525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526asmlinkage void __init start_kernel(void)
527{
528 char * command_line;
529 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700530
531 smp_setup_processor_id();
532
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700533 /*
534 * Need to run as early as possible, to initialize the
535 * lockdep hash:
536 */
537 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700538 debug_objects_early_init();
Ingo Molnar42059422008-02-14 09:44:08 +0100539
540 /*
541 * Set up the the initial canary ASAP:
542 */
543 boot_init_stack_canary();
544
Paul Menageddbcc7e2007-10-18 23:39:30 -0700545 cgroup_init_early();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700546
547 local_irq_disable();
548 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700549 early_init_irq_lock_class();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700550
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551/*
552 * Interrupts are still disabled. Do necessary setups, then
553 * enable them
554 */
555 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800556 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800557 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 page_address_init();
Alex Riesen657cafa2009-05-24 15:30:48 +0200559 printk(KERN_NOTICE "%s", linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700561 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800562 setup_command_line(command_line);
Mike Travise0982e92008-03-26 14:23:48 -0700563 setup_nr_cpu_ids();
Tejun Heod6647bd2009-07-21 17:11:50 +0900564 setup_per_cpu_areas();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800565 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Pekka Enberg83b519e2009-06-10 19:40:04 +0300567 build_all_zonelists();
568 page_alloc_init();
569
570 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
571 parse_early_param();
572 parse_args("Booting kernel", static_command_line, __start___param,
573 __stop___param - __start___param,
574 &unknown_bootoption);
575 /*
576 * These use large bootmem allocations and must precede
577 * kmem_cache_init()
578 */
579 pidhash_init();
Pekka Enberg83b519e2009-06-10 19:40:04 +0300580 vfs_caches_init_early();
581 sort_main_extable();
582 trap_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300583 mm_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 /*
585 * Set up the scheduler prior starting any interrupts (such as the
586 * timer interrupt). Full topology setup happens at smp_init()
587 * time - but meanwhile we still have a functioning scheduler.
588 */
589 sched_init();
590 /*
591 * Disable preemption - early bootup scheduling is extremely
592 * fragile until we cpu_idle() for the first time.
593 */
594 preempt_disable();
Ard van Breemenc4a68302007-01-05 16:36:19 -0800595 if (!irqs_disabled()) {
596 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
597 "enabled *very* early, fixing it\n");
598 local_irq_disable();
599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 rcu_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800601 /* init some links before init_ISA_irqs() */
602 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 init_IRQ();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100604 prio_tree_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800606 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700608 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700609 time_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700610 profile_init();
611 if (!irqs_disabled())
Ron Lee24d431d2008-11-27 02:31:57 +1030612 printk(KERN_CRIT "start_kernel(): bug: interrupts were "
613 "enabled early\n");
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700614 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700615 local_irq_enable();
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000616
617 /* Interrupts are enabled now so all GFP allocations are safe. */
618 set_gfp_allowed_mask(__GFP_BITS_MASK);
619
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300620 kmem_cache_init_late();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621
622 /*
623 * HACK ALERT! This is early. We're enabling the console before
624 * we've done PCI setups etc, and console_init() must be aware of
625 * this. But we do want output early, in case something goes wrong.
626 */
627 console_init();
628 if (panic_later)
629 panic(panic_later, panic_param);
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700630
631 lockdep_info();
632
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700633 /*
634 * Need to run this when irqs are enabled, because it wants
635 * to self-test [hard/soft]-irqs on/off lock inversion bugs
636 * too:
637 */
638 locking_selftest();
639
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640#ifdef CONFIG_BLK_DEV_INITRD
641 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700642 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200644 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700645 page_to_pfn(virt_to_page((void *)initrd_start)),
646 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 initrd_start = 0;
648 }
649#endif
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700650 page_cgroup_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100651 enable_debug_pagealloc();
Eduard - Gabriel Munteanub9ce08c2008-08-10 20:14:03 +0300652 kmemtrace_init();
Catalin Marinas3c7b4e62009-06-11 13:22:39 +0100653 kmemleak_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700654 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700655 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700656 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 numa_policy_init();
658 if (late_time_init)
659 late_time_init();
Thomas Gleixnerfa84e9e2009-08-21 22:01:12 +0200660 sched_clock_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 calibrate_delay();
662 pidmap_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 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
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700698/* Call all constructor functions linked into the kernel. */
699static void __init do_ctors(void)
700{
701#ifdef CONFIG_CONSTRUCTORS
702 ctor_fn_t *call = (ctor_fn_t *) __ctors_start;
703
704 for (; call < (ctor_fn_t *) __ctors_end; call++)
705 (*call)();
706#endif
707}
708
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800709int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500710core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Ingo Molnar4a683bf2009-08-21 12:53:36 +0200712static char msgbuf[64];
713static struct boot_trace_call call;
714static struct boot_trace_ret ret;
715
Arjan van de Ven59f94152008-07-30 12:49:02 -0700716int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700717{
718 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100719 ktime_t calltime, delta, rettime;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700720
721 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100722 call.caller = task_pid_nr(current);
723 printk("calling %pF @ %i\n", fn, call.caller);
724 calltime = ktime_get();
725 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100726 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700727 }
728
Frederic Weisbecker74239072008-11-11 23:24:42 +0100729 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700730
731 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100732 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100733 rettime = ktime_get();
734 delta = ktime_sub(rettime, calltime);
Will Newton1d926f22008-11-21 14:08:59 -0800735 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Frederic Weisbecker74239072008-11-11 23:24:42 +0100736 trace_boot_ret(&ret, fn);
Tim Birdca538f62008-10-09 15:23:05 -0700737 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbecker74239072008-11-11 23:24:42 +0100738 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700739 }
740
741 msgbuf[0] = 0;
742
Frederic Weisbecker74239072008-11-11 23:24:42 +0100743 if (ret.result && ret.result != -ENODEV && initcall_debug)
744 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700745
746 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700747 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700748 preempt_count() = count;
749 }
750 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700751 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700752 local_irq_enable();
753 }
754 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700755 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700756 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700757
Frederic Weisbecker74239072008-11-11 23:24:42 +0100758 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700759}
760
761
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700762extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763
764static void __init do_initcalls(void)
765{
766 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700768 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700769 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771 /* Make sure there is no pending stuff from the initcall sequence */
772 flush_scheduled_work();
773}
774
775/*
776 * Ok, the machine is now initialized. None of the devices
777 * have been touched yet, but the CPU subsystem is up and
778 * running, and memory and process management works.
779 *
780 * Now we can finally start doing some real work..
781 */
782static void __init do_basic_setup(void)
783{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200784 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700785 init_workqueues();
Lai Jiangshan759ee092009-03-25 17:06:30 +0800786 cpuset_init_smp();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 usermodehelper_init();
788 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800789 init_irq_proc();
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700790 do_ctors();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 do_initcalls();
792}
793
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700794static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700795{
796 initcall_t *call;
797
798 for (call = __initcall_start; call < __early_initcall_end; call++)
799 do_one_initcall(*call);
800}
801
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802static void run_init_process(char *init_filename)
803{
804 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700805 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806}
807
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100808/* This is a non __init function. Force it to be noinline otherwise gcc
809 * makes it inline to init() and it becomes part of init.text section
810 */
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800811static noinline int init_post(void)
Hannes Ederacdd0522009-03-31 15:23:50 -0700812 __releases(kernel_lock)
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100813{
Arjan van de Ven22a9d642009-01-07 08:45:46 -0800814 /* need to finish all async __init code before freeing the memory */
815 async_synchronize_full();
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100816 free_initmem();
817 unlock_kernel();
818 mark_rodata_ro();
819 system_state = SYSTEM_RUNNING;
820 numa_default_policy();
821
822 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
823 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
824
825 (void) sys_dup(0);
826 (void) sys_dup(0);
827
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700828 current->signal->flags |= SIGNAL_UNKILLABLE;
829
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100830 if (ramdisk_execute_command) {
831 run_init_process(ramdisk_execute_command);
832 printk(KERN_WARNING "Failed to execute %s\n",
833 ramdisk_execute_command);
834 }
835
836 /*
837 * We try each of these until one succeeds.
838 *
839 * The Bourne shell can be used instead of init if we are
840 * trying to recover a really broken machine.
841 */
842 if (execute_command) {
843 run_init_process(execute_command);
844 printk(KERN_WARNING "Failed to execute %s. Attempting "
845 "defaults...\n", execute_command);
846 }
847 run_init_process("/sbin/init");
848 run_init_process("/etc/init");
849 run_init_process("/bin/init");
850 run_init_process("/bin/sh");
851
852 panic("No init found. Try passing init= option to kernel.");
853}
854
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100855static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856{
857 lock_kernel();
Miao Xie58568d22009-06-16 15:31:49 -0700858
859 /*
860 * init can allocate pages on any node
861 */
862 set_mems_allowed(node_possible_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 /*
864 * init can run on any cpu.
865 */
Rusty Russell1a2142a2009-03-30 22:05:10 -0600866 set_cpus_allowed_ptr(current, cpu_all_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 /*
868 * Tell the world that we're going to be the grim
869 * reaper of innocent orphaned children.
870 *
871 * We don't want people to have to make incorrect
872 * assumptions about where in the task array this
873 * can be found.
874 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800875 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700877 cad_pid = task_pid(current);
878
Andi Kleenca74a6f2008-01-30 13:33:17 +0100879 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880
881 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100882 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 smp_init();
885 sched_init_smp();
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 do_basic_setup();
888
889 /*
890 * check if there is an early userspace init. If yes, let it do all
891 * the work
892 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700893
894 if (!ramdisk_execute_command)
895 ramdisk_execute_command = "/init";
896
897 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
898 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
902 /*
903 * Ok, we have completed the initial bootup, and
904 * we're essentially up and running. Get rid of the
905 * initmem segments and start the user-mode stuff..
906 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100907
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100908 init_post();
909 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910}