blob: b5a892c68375287188f62125267b5cf51dbf2770 [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>
Adrian Bunkb6cd0b72006-06-27 02:53:54 -070053#include <linux/buffer_head.h>
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -070054#include <linux/page_cgroup.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070055#include <linux/debug_locks.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070056#include <linux/debugobjects.h>
Ingo Molnarfbb9ce952006-07-03 00:24:50 -070057#include <linux/lockdep.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080058#include <linux/pid_namespace.h>
Adrian Bunk1f217822006-12-19 13:01:28 -080059#include <linux/device.h>
Eric W. Biederman73c27992007-05-09 02:34:32 -070060#include <linux/kthread.h>
Adrian Bunke6fe6642007-11-09 22:39:39 +010061#include <linux/sched.h>
Adrian Bunka1c9eea2008-02-06 01:36:44 -080062#include <linux/signal.h>
Akinobu Mita199f0ca2008-04-29 01:03:13 -070063#include <linux/idr.h>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040064#include <linux/ftrace.h>
Frederic Weisbecker3f5ec132008-11-11 23:21:31 +010065#include <trace/boot.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070066
67#include <asm/io.h>
68#include <asm/bugs.h>
69#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -070070#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -080071#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#ifdef CONFIG_X86_LOCAL_APIC
74#include <asm/smp.h>
75#endif
76
Sam Ravnborgaae5f662007-02-26 16:45:41 +010077static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080extern void fork_init(unsigned long);
81extern void mca_init(void);
82extern void sbus_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083extern void prio_tree_init(void);
84extern void radix_tree_init(void);
85extern void free_initmem(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070086#ifdef CONFIG_ACPI
87extern void acpi_early_init(void);
88#else
89static inline void acpi_early_init(void) { }
90#endif
Arjan van de Ven37b73c82006-01-06 00:12:01 -080091#ifndef CONFIG_DEBUG_RODATA
92static inline void mark_rodata_ro(void) { }
93#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
95#ifdef CONFIG_TC
96extern void tc_init(void);
97#endif
98
99enum system_states system_state;
100EXPORT_SYMBOL(system_state);
101
102/*
103 * Boot command-line arguments
104 */
105#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
106#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
107
108extern void time_init(void);
109/* Default late time init is NULL. archs can override this later. */
Jan Beulichd2e31922009-01-06 14:41:10 -0800110void (*__initdata late_time_init)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111extern void softirq_init(void);
112
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800113/* Untouched command line saved by arch-specific code. */
114char __initdata boot_command_line[COMMAND_LINE_SIZE];
115/* Untouched saved command line (eg. for /proc) */
116char *saved_command_line;
117/* Command line for parameter parsing */
118static char *static_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119
120static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700121static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122
Jan Beulich8b3b2952007-07-15 23:41:07 -0700123#ifdef CONFIG_SMP
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124/* Setup configured maximum number of CPUs to activate */
Andi Kleenca74a6f2008-01-30 13:33:17 +0100125unsigned int __initdata setup_max_cpus = NR_CPUS;
Vivek Goyal7e962872006-09-27 01:50:44 -0700126
127/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128 * Setup routine for controlling SMP activation
129 *
130 * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
131 * activation entirely (the MPS table probe still happens, though).
132 *
133 * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
134 * greater than 0, limits the maximum number of CPUs activated in
135 * SMP mode to <NUM>.
136 */
Len Brown61ec7562007-08-16 03:34:22 -0400137#ifndef CONFIG_X86_IO_APIC
138static inline void disable_ioapic_setup(void) {};
139#endif
140
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;
Len Brown61ec7562007-08-16 03:34:22 -0400144 disable_ioapic_setup();
Jan Beulich8b3b2952007-07-15 23:41:07 -0700145 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
Jan Beulich8b3b2952007-07-15 23:41:07 -0700148early_param("nosmp", nosmp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
150static int __init maxcpus(char *str)
151{
Andi Kleenca74a6f2008-01-30 13:33:17 +0100152 get_option(&str, &setup_max_cpus);
153 if (setup_max_cpus == 0)
Len Brown61ec7562007-08-16 03:34:22 -0400154 disable_ioapic_setup();
155
156 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157}
158
Hugh Dickins81340972007-08-27 16:02:12 +0100159early_param("maxcpus", maxcpus);
Jan Beulich8b3b2952007-07-15 23:41:07 -0700160#else
Andi Kleenca74a6f2008-01-30 13:33:17 +0100161#define setup_max_cpus NR_CPUS
Jan Beulich8b3b2952007-07-15 23:41:07 -0700162#endif
163
164/*
165 * If set, this is an indication to the drivers that reset the underlying
166 * device before going ahead with the initialization otherwise driver might
167 * rely on the BIOS and skip the reset operation.
168 *
169 * This is useful if kernel is booting in an unreliable environment.
170 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
171 * skipped and devices will be in unknown state.
172 */
173unsigned int reset_devices;
174EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
Vivek Goyal7e962872006-09-27 01:50:44 -0700176static int __init set_reset_devices(char *str)
177{
178 reset_devices = 1;
179 return 1;
180}
181
182__setup("reset_devices", set_reset_devices);
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
185char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
186static const char *panic_later, *panic_param;
187
188extern struct obs_kernel_param __setup_start[], __setup_end[];
189
190static int __init obsolete_checksetup(char *line)
191{
192 struct obs_kernel_param *p;
Rusty Russell33df0d12006-09-26 10:52:32 +0200193 int had_early_param = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195 p = __setup_start;
196 do {
197 int n = strlen(p->str);
198 if (!strncmp(line, p->str, n)) {
199 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200200 /* Already done in parse_early_param?
201 * (Needs exact match on param part).
202 * Keep iterating, as we can have early
203 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 if (line[n] == '\0' || line[n] == '=')
Rusty Russell33df0d12006-09-26 10:52:32 +0200205 had_early_param = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 } else if (!p->setup_func) {
207 printk(KERN_WARNING "Parameter %s is obsolete,"
208 " ignored\n", p->str);
209 return 1;
210 } else if (p->setup_func(line + n))
211 return 1;
212 }
213 p++;
214 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200215
216 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217}
218
219/*
220 * This should be approx 2 Bo*oMips to start (note initial shift), and will
221 * still work even if initially too large, it will just take slightly longer
222 */
223unsigned long loops_per_jiffy = (1<<12);
224
225EXPORT_SYMBOL(loops_per_jiffy);
226
227static int __init debug_kernel(char *str)
228{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 console_loglevel = 10;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800230 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231}
232
233static int __init quiet_kernel(char *str)
234{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 console_loglevel = 4;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800236 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237}
238
Yinghai Luf6f21c812008-02-08 04:21:58 -0800239early_param("debug", debug_kernel);
240early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242static int __init loglevel(char *str)
243{
244 get_option(&str, &console_loglevel);
Alex Riesend9d4fcf2008-03-04 14:28:31 -0800245 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
247
Yinghai Luf6f21c812008-02-08 04:21:58 -0800248early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
250/*
251 * Unknown boot options get handed to init, unless they look like
252 * failed parameters
253 */
254static int __init unknown_bootoption(char *param, char *val)
255{
256 /* Change NUL term back to "=", to make "param" the whole string. */
257 if (val) {
258 /* param=val or param="val"? */
259 if (val == param+strlen(param)+1)
260 val[-1] = '=';
261 else if (val == param+strlen(param)+2) {
262 val[-2] = '=';
263 memmove(val-1, val, strlen(val)+1);
264 val--;
265 } else
266 BUG();
267 }
268
269 /* Handle obsolete-style parameters */
270 if (obsolete_checksetup(param))
271 return 0;
272
273 /*
Simon Arlott211fee82007-10-20 01:28:29 +0200274 * Preemptive maintenance for "why didn't my misspelled command
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 * line work?"
276 */
277 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
278 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
279 return 0;
280 }
281
282 if (panic_later)
283 return 0;
284
285 if (val) {
286 /* Environment option */
287 unsigned int i;
288 for (i = 0; envp_init[i]; i++) {
289 if (i == MAX_INIT_ENVS) {
290 panic_later = "Too many boot env vars at `%s'";
291 panic_param = param;
292 }
293 if (!strncmp(param, envp_init[i], val - param))
294 break;
295 }
296 envp_init[i] = param;
297 } else {
298 /* Command line option */
299 unsigned int i;
300 for (i = 0; argv_init[i]; i++) {
301 if (i == MAX_INIT_ARGS) {
302 panic_later = "Too many boot init vars at `%s'";
303 panic_param = param;
304 }
305 }
306 argv_init[i] = param;
307 }
308 return 0;
309}
310
Ingo Molnar12d6f212008-01-30 13:33:58 +0100311#ifdef CONFIG_DEBUG_PAGEALLOC
312int __read_mostly debug_pagealloc_enabled = 0;
313#endif
314
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315static int __init init_setup(char *str)
316{
317 unsigned int i;
318
319 execute_command = str;
320 /*
321 * In case LILO is going to boot us with default command line,
322 * it prepends "auto" before the whole cmdline which makes
323 * the shell think it should execute a script with such name.
324 * So we ignore all arguments entered _before_ init=... [MJ]
325 */
326 for (i = 1; i < MAX_INIT_ARGS; i++)
327 argv_init[i] = NULL;
328 return 1;
329}
330__setup("init=", init_setup);
331
Olof Johanssonffdfc402005-09-06 15:17:19 -0700332static int __init rdinit_setup(char *str)
333{
334 unsigned int i;
335
336 ramdisk_execute_command = str;
337 /* See "auto" comment in init_setup */
338 for (i = 1; i < MAX_INIT_ARGS; i++)
339 argv_init[i] = NULL;
340 return 1;
341}
342__setup("rdinit=", rdinit_setup);
343
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344#ifndef CONFIG_SMP
345
346#ifdef CONFIG_X86_LOCAL_APIC
347static void __init smp_init(void)
348{
349 APIC_init_uniprocessor();
350}
351#else
352#define smp_init() do { } while (0)
353#endif
354
355static inline void setup_per_cpu_areas(void) { }
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
Mike Travisdd5af902008-01-30 13:33:32 +0100376#ifndef CONFIG_HAVE_SETUP_PER_CPU_AREA
Eric Dumazetb73b4592006-03-23 03:01:07 -0800377unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
379EXPORT_SYMBOL(__per_cpu_offset);
380
381static void __init setup_per_cpu_areas(void)
382{
383 unsigned long size, i;
384 char *ptr;
Eric Dumazet63872f82006-03-23 03:01:04 -0800385 unsigned long nr_possible_cpus = num_possible_cpus();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
387 /* Copy section for each CPU (we discard the original) */
Jeremy Fitzhardingeb6e35902007-05-02 19:27:12 +0200388 size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
389 ptr = alloc_bootmem_pages(size * nr_possible_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
KAMEZAWA Hiroyuki0a945022006-03-28 01:56:37 -0800391 for_each_possible_cpu(i) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 __per_cpu_offset[i] = ptr - __per_cpu_start;
393 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
Eric Dumazet63872f82006-03-23 03:01:04 -0800394 ptr += size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 }
396}
Mike Travisdd5af902008-01-30 13:33:32 +0100397#endif /* CONFIG_HAVE_SETUP_PER_CPU_AREA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399/* Called by boot processor to activate the rest. */
400static void __init smp_init(void)
401{
Christoph Lameter53b8a312007-02-20 13:57:51 -0800402 unsigned int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700403
Max Krasnyanskye761b772008-07-15 04:43:49 -0700404 /*
405 * Set up the current CPU as possible to migrate to.
406 * The other ones will be done by cpu_up/cpu_down()
407 */
408 cpu = smp_processor_id();
409 cpu_set(cpu, cpu_active_map);
410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 /* FIXME: This should be done in userspace --RR */
Christoph Lameter53b8a312007-02-20 13:57:51 -0800412 for_each_present_cpu(cpu) {
Andi Kleenca74a6f2008-01-30 13:33:17 +0100413 if (num_online_cpus() >= setup_max_cpus)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 break;
Christoph Lameter53b8a312007-02-20 13:57:51 -0800415 if (!cpu_online(cpu))
416 cpu_up(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 }
418
419 /* Any cleanup work */
420 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
Andi Kleenca74a6f2008-01-30 13:33:17 +0100421 smp_cpus_done(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422}
423
424#endif
425
426/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800427 * We need to store the untouched command line for future reference.
428 * We also need to store the touched command line since the parameter
429 * parsing is performed in place, and we should allow a component to
430 * store reference of name/value for future reference.
431 */
432static void __init setup_command_line(char *command_line)
433{
434 saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
435 static_command_line = alloc_bootmem(strlen (command_line)+1);
436 strcpy (saved_command_line, boot_command_line);
437 strcpy (static_command_line, command_line);
438}
439
440/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 * We need to finalize in a non-__init function or else race conditions
442 * between the root thread and the init thread may cause start_kernel to
443 * be reaped by free_initmem before the root thread has proceeded to
444 * cpu_idle.
445 *
446 * gcc-3.4 accidentally inlines this function, so use noinline.
447 */
448
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800449static noinline void __init_refok rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 __releases(kernel_lock)
451{
Eric W. Biederman73c27992007-05-09 02:34:32 -0700452 int pid;
453
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100454 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700456 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700457 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 unlock_kernel();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200459
460 /*
461 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200462 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200463 */
Ingo Molnar1df21052007-07-09 18:51:58 +0200464 init_idle_bootup_task(current);
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800465 preempt_enable_no_resched();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200466 schedule();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800467 preempt_disable();
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200468
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800469 /* Call into cpu_idle with preempt disabled */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 cpu_idle();
Ingo Molnar1df21052007-07-09 18:51:58 +0200471}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473/* Check for early params. */
474static int __init do_early_param(char *param, char *val)
475{
476 struct obs_kernel_param *p;
477
478 for (p = __setup_start; p < __setup_end; p++) {
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700479 if ((p->early && strcmp(param, p->str) == 0) ||
480 (strcmp(param, "console") == 0 &&
481 strcmp(p->str, "earlycon") == 0)
482 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 if (p->setup_func(val) != 0)
484 printk(KERN_WARNING
485 "Malformed early option '%s'\n", param);
486 }
487 }
488 /* We accept everything at this stage. */
489 return 0;
490}
491
492/* Arch code calls this early on, or if not, just before other parsing. */
493void __init parse_early_param(void)
494{
495 static __initdata int done = 0;
496 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
497
498 if (done)
499 return;
500
501 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800502 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
504 done = 1;
505}
506
507/*
508 * Activate the first processor.
509 */
510
Stas Sergeev44fd2292006-03-23 02:59:44 -0800511static void __init boot_cpu_init(void)
512{
513 int cpu = smp_processor_id();
514 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
Rusty Russell915441b2009-01-01 10:12:15 +1030515 set_cpu_online(cpu, true);
516 set_cpu_present(cpu, true);
517 set_cpu_possible(cpu, true);
Stas Sergeev44fd2292006-03-23 02:59:44 -0800518}
519
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000520void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700521{
522}
523
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000524void __init __weak thread_info_cache_init(void)
525{
526}
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528asmlinkage void __init start_kernel(void)
529{
530 char * command_line;
531 extern struct kernel_param __start___param[], __stop___param[];
Andrew Morton033ab7f2006-06-30 01:55:50 -0700532
533 smp_setup_processor_id();
534
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700535 /*
536 * Need to run as early as possible, to initialize the
537 * lockdep hash:
538 */
539 lockdep_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700540 debug_objects_early_init();
Paul Menageddbcc7e2007-10-18 23:39:30 -0700541 cgroup_init_early();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700542
543 local_irq_disable();
544 early_boot_irqs_off();
Ingo Molnar243c7622006-07-03 00:25:06 -0700545 early_init_irq_lock_class();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700546
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547/*
548 * Interrupts are still disabled. Do necessary setups, then
549 * enable them
550 */
551 lock_kernel();
Thomas Gleixner906568c2007-02-16 01:28:01 -0800552 tick_init();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800553 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 page_address_init();
555 printk(KERN_NOTICE);
Linus Torvalds89937802006-12-11 09:28:46 -0800556 printk(linux_banner);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 setup_arch(&command_line);
Balbir Singhcf475ad2008-04-29 01:00:16 -0700558 mm_init_owner(&init_mm, &init_task);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800559 setup_command_line(command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 setup_per_cpu_areas();
Mike Travise0982e92008-03-26 14:23:48 -0700561 setup_nr_cpu_ids();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800562 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
564 /*
565 * Set up the scheduler prior starting any interrupts (such as the
566 * timer interrupt). Full topology setup happens at smp_init()
567 * time - but meanwhile we still have a functioning scheduler.
568 */
569 sched_init();
570 /*
571 * Disable preemption - early bootup scheduling is extremely
572 * fragile until we cpu_idle() for the first time.
573 */
574 preempt_disable();
575 build_all_zonelists();
576 page_alloc_init();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800577 printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 parse_early_param();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800579 parse_args("Booting kernel", static_command_line, __start___param,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 __stop___param - __start___param,
581 &unknown_bootoption);
Ard van Breemenc4a68302007-01-05 16:36:19 -0800582 if (!irqs_disabled()) {
583 printk(KERN_WARNING "start_kernel(): bug: interrupts were "
584 "enabled *very* early, fixing it\n");
585 local_irq_disable();
586 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 sort_main_extable();
588 trap_init();
589 rcu_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800590 /* init some links before init_ISA_irqs() */
591 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 init_IRQ();
593 pidhash_init();
594 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800595 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700597 timekeeping_init();
john stultz88fecaa2006-07-03 00:24:04 -0700598 time_init();
Peter Zijlstra3e51f332008-05-03 18:29:28 +0200599 sched_clock_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700600 profile_init();
601 if (!irqs_disabled())
602 printk("start_kernel(): bug: interrupts were enabled early\n");
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700603 early_boot_irqs_on();
Heiko Carstens93e02812006-07-03 00:24:24 -0700604 local_irq_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605
606 /*
607 * HACK ALERT! This is early. We're enabling the console before
608 * we've done PCI setups etc, and console_init() must be aware of
609 * this. But we do want output early, in case something goes wrong.
610 */
611 console_init();
612 if (panic_later)
613 panic(panic_later, panic_param);
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700614
615 lockdep_info();
616
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700617 /*
618 * Need to run this when irqs are enabled, because it wants
619 * to self-test [hard/soft]-irqs on/off lock inversion bugs
620 * too:
621 */
622 locking_selftest();
623
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624#ifdef CONFIG_BLK_DEV_INITRD
625 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700626 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
Geert Uytterhoevenfb6624e2008-07-17 21:16:36 +0200628 "disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700629 page_to_pfn(virt_to_page((void *)initrd_start)),
630 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631 initrd_start = 0;
632 }
633#endif
Nick Piggindb64fe02008-10-18 20:27:03 -0700634 vmalloc_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 vfs_caches_init_early();
Paul Jacksonc417f022006-01-08 01:02:01 -0800636 cpuset_init_early();
KAMEZAWA Hiroyuki94b6da52008-10-22 14:15:05 -0700637 page_cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 mem_init();
Thomas Gleixnera03c2a42008-02-09 23:24:09 +0100639 enable_debug_pagealloc();
Gautham R Shenoyd2219382008-01-25 21:08:01 +0100640 cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 kmem_cache_init();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700642 debug_objects_mem_init();
Akinobu Mita199f0ca2008-04-29 01:03:13 -0700643 idr_init_cache();
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700644 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 numa_policy_init();
646 if (late_time_init)
647 late_time_init();
648 calibrate_delay();
649 pidmap_init();
650 pgtable_cache_init();
651 prio_tree_init();
652 anon_vma_init();
653#ifdef CONFIG_X86
654 if (efi_enabled)
655 efi_enter_virtual_mode();
656#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000657 thread_info_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100658 cred_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 fork_init(num_physpages);
660 proc_caches_init();
661 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 key_init();
663 security_init();
664 vfs_caches_init(num_physpages);
665 radix_tree_init();
666 signals_init();
667 /* rootfs populating might need page-writeback */
668 page_writeback_init();
669#ifdef CONFIG_PROC_FS
670 proc_root_init();
671#endif
Paul Menageddbcc7e2007-10-18 23:39:30 -0700672 cgroup_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 cpuset_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700674 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700675 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
677 check_bugs();
678
679 acpi_early_init(); /* before LAPIC and SMP init */
680
Steven Rostedt68bf21a2008-08-14 15:45:08 -0400681 ftrace_init();
682
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 /* Do the rest non-__init'ed, we're now alive */
684 rest_init();
685}
686
Arjan van de Ven59f94152008-07-30 12:49:02 -0700687static int initcall_debug;
Rusty Russelld0ea3d72008-10-22 10:00:23 -0500688core_param(initcall_debug, initcall_debug, bool, 0644);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Arjan van de Ven59f94152008-07-30 12:49:02 -0700690int do_one_initcall(initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700691{
692 int count = preempt_count();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100693 ktime_t calltime, delta, rettime;
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700694 char msgbuf[64];
Frederic Weisbecker74239072008-11-11 23:24:42 +0100695 struct boot_trace_call call;
696 struct boot_trace_ret ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700697
698 if (initcall_debug) {
Frederic Weisbecker74239072008-11-11 23:24:42 +0100699 call.caller = task_pid_nr(current);
700 printk("calling %pF @ %i\n", fn, call.caller);
701 calltime = ktime_get();
702 trace_boot_call(&call, fn);
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100703 enable_boot_trace();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700704 }
705
Frederic Weisbecker74239072008-11-11 23:24:42 +0100706 ret.result = fn();
Linus Torvaldse0df1542008-05-15 18:14:01 -0700707
708 if (initcall_debug) {
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100709 disable_boot_trace();
Frederic Weisbecker74239072008-11-11 23:24:42 +0100710 rettime = ktime_get();
711 delta = ktime_sub(rettime, calltime);
Will Newton1d926f22008-11-21 14:08:59 -0800712 ret.duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Frederic Weisbecker74239072008-11-11 23:24:42 +0100713 trace_boot_ret(&ret, fn);
Tim Birdca538f62008-10-09 15:23:05 -0700714 printk("initcall %pF returned %d after %Ld usecs\n", fn,
Frederic Weisbecker74239072008-11-11 23:24:42 +0100715 ret.result, ret.duration);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700716 }
717
718 msgbuf[0] = 0;
719
Frederic Weisbecker74239072008-11-11 23:24:42 +0100720 if (ret.result && ret.result != -ENODEV && initcall_debug)
721 sprintf(msgbuf, "error code %d ", ret.result);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700722
723 if (preempt_count() != count) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700724 strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700725 preempt_count() = count;
726 }
727 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -0700728 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -0700729 local_irq_enable();
730 }
731 if (msgbuf[0]) {
Linus Torvalds96d746c2008-10-03 13:38:07 -0700732 printk("initcall %pF returned with %s\n", fn, msgbuf);
Linus Torvaldse0df1542008-05-15 18:14:01 -0700733 }
Arjan van de Ven59f94152008-07-30 12:49:02 -0700734
Frederic Weisbecker74239072008-11-11 23:24:42 +0100735 return ret.result;
Linus Torvaldse0df1542008-05-15 18:14:01 -0700736}
737
738
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700739extern initcall_t __initcall_start[], __initcall_end[], __early_initcall_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
741static void __init do_initcalls(void)
742{
743 initcall_t *call;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700745 for (call = __early_initcall_end; call < __initcall_end; call++)
Linus Torvaldse0df1542008-05-15 18:14:01 -0700746 do_one_initcall(*call);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
748 /* Make sure there is no pending stuff from the initcall sequence */
749 flush_scheduled_work();
750}
751
752/*
753 * Ok, the machine is now initialized. None of the devices
754 * have been touched yet, but the CPU subsystem is up and
755 * running, and memory and process management works.
756 *
757 * Now we can finally start doing some real work..
758 */
759static void __init do_basic_setup(void)
760{
Paul E. McKenney4446a362008-05-12 21:21:05 +0200761 rcu_init_sched(); /* needed by module_init stage. */
Linus Torvalds4403b402008-10-25 19:53:38 -0700762 init_workqueues();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 usermodehelper_init();
764 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -0800765 init_irq_proc();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 do_initcalls();
767}
768
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -0700769static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -0700770{
771 initcall_t *call;
772
773 for (call = __initcall_start; call < __early_initcall_end; call++)
774 do_one_initcall(*call);
775}
776
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777static void run_init_process(char *init_filename)
778{
779 argv_init[0] = init_filename;
Arnd Bergmann67608562006-10-02 02:18:26 -0700780 kernel_execve(init_filename, argv_init, envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781}
782
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100783/* This is a non __init function. Force it to be noinline otherwise gcc
784 * makes it inline to init() and it becomes part of init.text section
785 */
Rakib Mullickf99ebf02009-01-06 14:40:38 -0800786static noinline int init_post(void)
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100787{
788 free_initmem();
789 unlock_kernel();
790 mark_rodata_ro();
791 system_state = SYSTEM_RUNNING;
792 numa_default_policy();
793
794 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
795 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
796
797 (void) sys_dup(0);
798 (void) sys_dup(0);
799
Oleg Nesterovfae5fa42008-04-30 00:53:03 -0700800 current->signal->flags |= SIGNAL_UNKILLABLE;
801
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100802 if (ramdisk_execute_command) {
803 run_init_process(ramdisk_execute_command);
804 printk(KERN_WARNING "Failed to execute %s\n",
805 ramdisk_execute_command);
806 }
807
808 /*
809 * We try each of these until one succeeds.
810 *
811 * The Bourne shell can be used instead of init if we are
812 * trying to recover a really broken machine.
813 */
814 if (execute_command) {
815 run_init_process(execute_command);
816 printk(KERN_WARNING "Failed to execute %s. Attempting "
817 "defaults...\n", execute_command);
818 }
819 run_init_process("/sbin/init");
820 run_init_process("/etc/init");
821 run_init_process("/bin/init");
822 run_init_process("/bin/sh");
823
824 panic("No init found. Try passing init= option to kernel.");
825}
826
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100827static int __init kernel_init(void * unused)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828{
829 lock_kernel();
830 /*
831 * init can run on any cpu.
832 */
Mike Travis321a8e92008-04-04 18:11:02 -0700833 set_cpus_allowed_ptr(current, CPU_MASK_ALL_PTR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 /*
835 * Tell the world that we're going to be the grim
836 * reaper of innocent orphaned children.
837 *
838 * We don't want people to have to make incorrect
839 * assumptions about where in the task array this
840 * can be found.
841 */
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -0800842 init_pid_ns.child_reaper = current;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843
Cedric Le Goater9ec52092006-10-02 02:19:00 -0700844 cad_pid = task_pid(current);
845
Andi Kleenca74a6f2008-01-30 13:33:17 +0100846 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847
848 do_pre_smp_initcalls();
Frédéric Weisbecker3bf77af2008-09-23 11:38:18 +0100849 start_boot_trace();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 smp_init();
852 sched_init_smp();
853
854 cpuset_init_smp();
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 do_basic_setup();
857
858 /*
859 * check if there is an early userspace init. If yes, let it do all
860 * the work
861 */
Olof Johanssonffdfc402005-09-06 15:17:19 -0700862
863 if (!ramdisk_execute_command)
864 ramdisk_execute_command = "/init";
865
866 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
867 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -0700869 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870
871 /*
872 * Ok, we have completed the initial bootup, and
873 * we're essentially up and running. Get rid of the
874 * initmem segments and start the user-mode stuff..
875 */
Frederic Weisbecker71566a02008-10-31 12:57:20 +0100876
Vivek Goyalee5bfa62007-02-13 13:26:22 +0100877 init_post();
878 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879}