blob: cd168ebc592439b650042916ab26efd913f7ef42 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/init/main.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * GK 2/5/95 - Changed to support mounting root fs via NFS
7 * Added initrd & change_root: Werner Almesberger & Hans Lermen, Feb '96
8 * Moan early if gcc is old, avoiding bogus kernels - Paul Gortmaker, May '96
9 * Simplified starting of init: Michael A. Griffith <grif@acm.org>
10 */
11
Linus Torvalds1da177e2005-04-16 15:20:36 -070012#include <linux/types.h>
13#include <linux/module.h>
14#include <linux/proc_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/kernel.h>
16#include <linux/syscalls.h>
17#include <linux/string.h>
18#include <linux/ctype.h>
19#include <linux/delay.h>
20#include <linux/utsname.h>
21#include <linux/ioport.h>
22#include <linux/init.h>
23#include <linux/smp_lock.h>
24#include <linux/initrd.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/bootmem.h>
26#include <linux/tty.h>
27#include <linux/gfp.h>
28#include <linux/percpu.h>
29#include <linux/kmod.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070030#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/kernel_stat.h>
Rusty Russelld7cd5612006-12-07 02:14:08 +010032#include <linux/start_kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/security.h>
Jens Axboe3d442232008-06-26 11:21:34 +020034#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/workqueue.h>
36#include <linux/profile.h>
37#include <linux/rcupdate.h>
38#include <linux/moduleparam.h>
39#include <linux/kallsyms.h>
40#include <linux/writeback.h>
41#include <linux/cpu.h>
42#include <linux/cpuset.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070043#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044#include <linux/efi.h>
Thomas Gleixner906568c2007-02-16 01:28:01 -080045#include <linux/tick.h>
Andrew Morton6168a702007-02-17 21:22:39 -080046#include <linux/interrupt.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070047#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070048#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/unistd.h>
50#include <linux/rmap.h>
51#include <linux/mempolicy.h>
52#include <linux/key.h>
Jan Beulich4552d5d2006-06-26 13:57:28 +020053#include <linux/unwind.h>
Adrian Bunkb6cd0b72006-06-27 02:53:54 -070054#include <linux/buffer_head.h>
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -070055#include <linux/page_cgroup.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070056#include <linux/debug_locks.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070057#include <linux/debugobjects.h>
Ingo Molnarfbb9ce92006-07-03 00:24:50 -070058#include <linux/lockdep.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080059#include <linux/pid_namespace.h>
Adrian Bunk1f217822006-12-19 13:01:28 -080060#include <linux/device.h>
Eric W. Biederman73c27992007-05-09 02:34:32 -070061#include <linux/kthread.h>
Adrian Bunke6fe6642007-11-09 22:39:39 +010062#include <linux/sched.h>
Adrian Bunka1c9eea2008-02-06 01:36:44 -080063#include <linux/signal.h>
Akinobu Mita199f0ca2008-04-29 01:03:13 -070064#include <linux/idr.h>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040065#include <linux/ftrace.h>
Frederic Weisbecker3f5ec132008-11-11 23:21:31 +010066#include <trace/boot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070067
68#include <asm/io.h>
69#include <asm/bugs.h>
70#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070071#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080072#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#ifdef CONFIG_X86_LOCAL_APIC
75#include <asm/smp.h>
76#endif
77
Sam Ravnborgaae5f662007-02-26 16:45:41 +010078static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081extern void fork_init(unsigned long);
82extern void mca_init(void);
83extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084extern void prio_tree_init(void);
85extern void radix_tree_init(void);
86extern void free_initmem(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087#ifdef CONFIG_ACPI
88extern void acpi_early_init(void);
89#else
90static inline void acpi_early_init(void) { }
91#endif
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
100enum system_states system_state;
101EXPORT_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. */
111void (*late_time_init)(void);
112extern 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 */
Len Brown61ec7562007-08-16 03:34:22 -0400138#ifndef CONFIG_X86_IO_APIC
139static inline void disable_ioapic_setup(void) {};
140#endif
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142static int __init nosmp(char *str)
143{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100144 setup_max_cpus = 0;
Len Brown61ec7562007-08-16 03:34:22 -0400145 disable_ioapic_setup();
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)
Len Brown61ec7562007-08-16 03:34:22 -0400155 disable_ioapic_setup();
156
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
Andi Kleenca74a6f2008-01-30 13:33:17 +0100162#define 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
356static inline void setup_per_cpu_areas(void) { }
Mike Travise0982e92008-03-26 14:23:48 -0700357static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358static inline void smp_prepare_cpus(unsigned int maxcpus) { }
359
360#else
361
Mike Travis321a8e92008-04-04 18:11:02 -0700362#if NR_CPUS > BITS_PER_LONG
363cpumask_t cpu_mask_all __read_mostly = CPU_MASK_ALL;
364EXPORT_SYMBOL(cpu_mask_all);
365#endif
366
Mike Travise0982e92008-03-26 14:23:48 -0700367/* Setup number of possible processor ids */
368int nr_cpu_ids __read_mostly = NR_CPUS;
369EXPORT_SYMBOL(nr_cpu_ids);
370
371/* An arch may set nr_cpu_ids earlier if needed, so this would be redundant */
372static void __init setup_nr_cpu_ids(void)
373{
Rusty Russelle0c0ba72009-01-01 10:12:19 +1030374 nr_cpu_ids = find_last_bit(cpumask_bits(cpu_possible_mask),NR_CPUS) + 1;
Mike Travise0982e92008-03-26 14:23:48 -0700375}
376
Mike Travisdd5af902008-01-30 13:33:32 +0100377#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800378unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
380EXPORT_SYMBOL(__per_cpu_offset);
381
382static void __init setup_per_cpu_areas(void)
383{
384 unsigned long size, i;
385 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800386 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
388 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200389 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
390 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800392 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 __per_cpu_offset[i] = ptr - __per_cpu_start;
394 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800395 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 }
397}
Mike Travisdd5af902008-01-30 13:33:32 +0100398#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399
400/* Called by boot processor to activate the rest. */
401static void __init smp_init(void)
402{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800403 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404
Max Krasnyanskye761b772008-07-15 04:43:49 -0700405 /*
406 * Set up the current CPU as possible to migrate to.
407 * The other ones will be done by cpu_up/cpu_down()
408 */
409 cpu = smp_processor_id();
410 cpu_set(cpu, cpu_active_map);
411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800413 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100414 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800416 if (!cpu_online(cpu))
417 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 }
419
420 /* Any cleanup work */
421 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100422 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
425#endif
426
427/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800428 * We need to store the untouched command line for future reference.
429 * We also need to store the touched command line since the parameter
430 * parsing is performed in place, and we should allow a component to
431 * store reference of name/value for future reference.
432 */
433static void __init setup_command_line(char *command_line)
434{
435 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
436 static_command_line = alloc_bootmem(strlen (command_line)+1);
437 strcpy (saved_command_line, boot_command_line);
438 strcpy (static_command_line, command_line);
439}
440
441/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * We need to finalize in a non-__init function or else race conditions
443 * between the root thread and the init thread may cause start_kernel to
444 * be reaped by free_initmem before the root thread has proceeded to
445 * cpu_idle.
446 *
447 * gcc-3.4 accidentally inlines this function, so use noinline.
448 */
449
Sam Ravnborg92080302007-05-17 20:43:54 +0200450static void noinline __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 __releases(kernel_lock)
452{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700453 int pid;
454
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100455 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700457 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700458 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200460
461 /*
462 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200463 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200464 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200465 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800466 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200467 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800468 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200469
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800470 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200472}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473
474/* Check for early params. */
475static int __init do_early_param(char *param, char *val)
476{
477 struct obs_kernel_param *p;
478
479 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700480 if ((p->early && strcmp(param, p->str) == 0) ||
481 (strcmp(param, "console") == 0 &&
482 strcmp(p->str, "earlycon") == 0)
483 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (p->setup_func(val) != 0)
485 printk(KERN_WARNING
486 "Malformed early option '%s'\n", param);
487 }
488 }
489 /* We accept everything at this stage. */
490 return 0;
491}
492
493/* Arch code calls this early on, or if not, just before other parsing. */
494void __init parse_early_param(void)
495{
496 static __initdata int done = 0;
497 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
498
499 if (done)
500 return;
501
502 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800503 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
505 done = 1;
506}
507
508/*
509 * Activate the first processor.
510 */
511
Stas Sergeev44fd2292006-03-23 02:59:44 -0800512static void __init boot_cpu_init(void)
513{
514 int cpu = smp_processor_id();
515 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030516 set_cpu_online(cpu, true);
517 set_cpu_present(cpu, true);
518 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800519}
520
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000521void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700522{
523}
524
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000525void __init __weak thread_info_cache_init(void)
526{
527}
528
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529asmlinkage void __init start_kernel(void)
530{
531 char * command_line;
532 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700533
534 smp_setup_processor_id();
535
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700536 /*
537 * Need to run as early as possible, to initialize the
538 * lockdep hash:
539 */
Andi Kleenc9538ed2006-09-26 10:52:34 +0200540 unwind_init();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700541 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700542 debug_objects_early_init();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700543 cgroup_init_early();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700544
545 local_irq_disable();
546 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700547 early_init_irq_lock_class();
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700548
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549/*
550 * Interrupts are still disabled. Do necessary setups, then
551 * enable them
552 */
553 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800554 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800555 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 page_address_init();
557 printk(KERN_NOTICE);
Linus Torvalds89937802006-12-11 09:28:46 -0800558 printk(linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700560 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800561 setup_command_line(command_line);
Jan Beulich690a9732006-10-21 18:37:01 +0200562 unwind_setup();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700564 setup_nr_cpu_ids();
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
567 /*
568 * Set up the scheduler prior starting any interrupts (such as the
569 * timer interrupt). Full topology setup happens at smp_init()
570 * time - but meanwhile we still have a functioning scheduler.
571 */
572 sched_init();
573 /*
574 * Disable preemption - early bootup scheduling is extremely
575 * fragile until we cpu_idle() for the first time.
576 */
577 preempt_disable();
578 build_all_zonelists();
579 page_alloc_init();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800580 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 parse_early_param();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800582 parse_args("Booting kernel", static_command_line, __start___param,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 __stop___param - __start___param,
584 &unknown_bootoption);
Ard van Breemenc4a68302007-01-05 16:36:19 -0800585 if (!irqs_disabled()) {
586 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
587 "enabled *very* early, fixing it\n");
588 local_irq_disable();
589 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 sort_main_extable();
591 trap_init();
592 rcu_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800593 /* init some links before init_ISA_irqs() */
594 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 init_IRQ();
596 pidhash_init();
597 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800598 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700600 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700601 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200602 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700603 profile_init();
604 if (!irqs_disabled())
605 printk("start_kernel(): bug: interrupts were enabled early\n");
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700606 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700607 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608
609 /*
610 * HACK ALERT! This is early. We're enabling the console before
611 * we've done PCI setups etc, and console_init() must be aware of
612 * this. But we do want output early, in case something goes wrong.
613 */
614 console_init();
615 if (panic_later)
616 panic(panic_later, panic_param);
Ingo Molnarfbb9ce92006-07-03 00:24:50 -0700617
618 lockdep_info();
619
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700620 /*
621 * Need to run this when irqs are enabled, because it wants
622 * to self-test [hard/soft]-irqs on/off lock inversion bugs
623 * too:
624 */
625 locking_selftest();
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627#ifdef CONFIG_BLK_DEV_INITRD
628 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700629 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200631 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700632 page_to_pfn(virt_to_page((void *)initrd_start)),
633 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 initrd_start = 0;
635 }
636#endif
Nick Piggindb64fe02008-10-18 20:27:03 -0700637 vmalloc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 vfs_caches_init_early();
Paul Jacksonc417f022006-01-08 01:02:01 -0800639 cpuset_init_early();
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700640 page_cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 mem_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100642 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100643 cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 kmem_cache_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700645 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700646 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700647 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 numa_policy_init();
649 if (late_time_init)
650 late_time_init();
651 calibrate_delay();
652 pidmap_init();
653 pgtable_cache_init();
654 prio_tree_init();
655 anon_vma_init();
656#ifdef CONFIG_X86
657 if (efi_enabled)
658 efi_enter_virtual_mode();
659#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000660 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100661 cred_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 fork_init(num_physpages);
663 proc_caches_init();
664 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 key_init();
666 security_init();
667 vfs_caches_init(num_physpages);
668 radix_tree_init();
669 signals_init();
670 /* rootfs populating might need page-writeback */
671 page_writeback_init();
672#ifdef CONFIG_PROC_FS
673 proc_root_init();
674#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700675 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700677 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700678 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
680 check_bugs();
681
682 acpi_early_init(); /* before LAPIC and SMP init */
683
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400684 ftrace_init();
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 /* Do the rest non-__init'ed, we're now alive */
687 rest_init();
688}
689
Arjan van de Ven59f94152008-07-30 12:49:02 -0700690static int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500691core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692
Arjan van de Ven59f94152008-07-30 12:49:02 -0700693int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700694{
695 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100696 ktime_t calltime, delta, rettime;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700697 char msgbuf[64];
Frederic Weisbecker74239072008-11-11 23:24:42 +0100698 struct boot_trace_call call;
699 struct boot_trace_ret ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700700
701 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100702 call.caller = task_pid_nr(current);
703 printk("calling %pF @ %i\n", fn, call.caller);
704 calltime = ktime_get();
705 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100706 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700707 }
708
Frederic Weisbecker74239072008-11-11 23:24:42 +0100709 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700710
711 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100712 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100713 rettime = ktime_get();
714 delta = ktime_sub(rettime, calltime);
Will Newton1d926f22008-11-21 14:08:59 -0800715 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Frederic Weisbecker74239072008-11-11 23:24:42 +0100716 trace_boot_ret(&ret, fn);
Tim Birdca538f62008-10-09 15:23:05 -0700717 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbecker74239072008-11-11 23:24:42 +0100718 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700719 }
720
721 msgbuf[0] = 0;
722
Frederic Weisbecker74239072008-11-11 23:24:42 +0100723 if (ret.result && ret.result != -ENODEV && initcall_debug)
724 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700725
726 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700727 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700728 preempt_count() = count;
729 }
730 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700731 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700732 local_irq_enable();
733 }
734 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700735 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700736 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700737
Frederic Weisbecker74239072008-11-11 23:24:42 +0100738 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700739}
740
741
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700742extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743
744static void __init do_initcalls(void)
745{
746 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700748 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700749 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
751 /* Make sure there is no pending stuff from the initcall sequence */
752 flush_scheduled_work();
753}
754
755/*
756 * Ok, the machine is now initialized. None of the devices
757 * have been touched yet, but the CPU subsystem is up and
758 * running, and memory and process management works.
759 *
760 * Now we can finally start doing some real work..
761 */
762static void __init do_basic_setup(void)
763{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200764 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700765 init_workqueues();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 usermodehelper_init();
767 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800768 init_irq_proc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769 do_initcalls();
770}
771
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700772static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700773{
774 initcall_t *call;
775
776 for (call = __initcall_start; call < __early_initcall_end; call++)
777 do_one_initcall(*call);
778}
779
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780static void run_init_process(char *init_filename)
781{
782 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700783 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
785
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100786/* This is a non __init function. Force it to be noinline otherwise gcc
787 * makes it inline to init() and it becomes part of init.text section
788 */
789static int noinline init_post(void)
790{
791 free_initmem();
792 unlock_kernel();
793 mark_rodata_ro();
794 system_state = SYSTEM_RUNNING;
795 numa_default_policy();
796
797 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
798 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
799
800 (void) sys_dup(0);
801 (void) sys_dup(0);
802
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700803 current->signal->flags |= SIGNAL_UNKILLABLE;
804
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100805 if (ramdisk_execute_command) {
806 run_init_process(ramdisk_execute_command);
807 printk(KERN_WARNING "Failed to execute %s\n",
808 ramdisk_execute_command);
809 }
810
811 /*
812 * We try each of these until one succeeds.
813 *
814 * The Bourne shell can be used instead of init if we are
815 * trying to recover a really broken machine.
816 */
817 if (execute_command) {
818 run_init_process(execute_command);
819 printk(KERN_WARNING "Failed to execute %s. Attempting "
820 "defaults...\n", execute_command);
821 }
822 run_init_process("/sbin/init");
823 run_init_process("/etc/init");
824 run_init_process("/bin/init");
825 run_init_process("/bin/sh");
826
827 panic("No init found. Try passing init= option to kernel.");
828}
829
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100830static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831{
832 lock_kernel();
833 /*
834 * init can run on any cpu.
835 */
Mike Travis321a8e92008-04-04 18:11:02 -0700836 set_cpus_allowed_ptr(current, CPU_MASK_ALL_PTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 /*
838 * Tell the world that we're going to be the grim
839 * reaper of innocent orphaned children.
840 *
841 * We don't want people to have to make incorrect
842 * assumptions about where in the task array this
843 * can be found.
844 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800845 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700847 cad_pid = task_pid(current);
848
Andi Kleenca74a6f2008-01-30 13:33:17 +0100849 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
851 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100852 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 smp_init();
855 sched_init_smp();
856
857 cpuset_init_smp();
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859 do_basic_setup();
860
861 /*
862 * check if there is an early userspace init. If yes, let it do all
863 * the work
864 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700865
866 if (!ramdisk_execute_command)
867 ramdisk_execute_command = "/init";
868
869 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
870 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873
874 /*
875 * Ok, we have completed the initial bootup, and
876 * we're essentially up and running. Get rid of the
877 * initmem segments and start the user-mode stuff..
878 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100879
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100880 init_post();
881 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}