xref: /linux/init/main.c (revision 98366c20a275e957416e9516db5dcb7195b4e101)
1 /*
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 
12 #include <linux/types.h>
13 #include <linux/module.h>
14 #include <linux/proc_fs.h>
15 #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>
25 #include <linux/hdreg.h>
26 #include <linux/bootmem.h>
27 #include <linux/tty.h>
28 #include <linux/gfp.h>
29 #include <linux/percpu.h>
30 #include <linux/kmod.h>
31 #include <linux/kernel_stat.h>
32 #include <linux/start_kernel.h>
33 #include <linux/security.h>
34 #include <linux/workqueue.h>
35 #include <linux/profile.h>
36 #include <linux/rcupdate.h>
37 #include <linux/moduleparam.h>
38 #include <linux/kallsyms.h>
39 #include <linux/writeback.h>
40 #include <linux/cpu.h>
41 #include <linux/cpuset.h>
42 #include <linux/cgroup.h>
43 #include <linux/efi.h>
44 #include <linux/tick.h>
45 #include <linux/interrupt.h>
46 #include <linux/taskstats_kern.h>
47 #include <linux/delayacct.h>
48 #include <linux/unistd.h>
49 #include <linux/rmap.h>
50 #include <linux/mempolicy.h>
51 #include <linux/key.h>
52 #include <linux/unwind.h>
53 #include <linux/buffer_head.h>
54 #include <linux/debug_locks.h>
55 #include <linux/lockdep.h>
56 #include <linux/pid_namespace.h>
57 #include <linux/device.h>
58 #include <linux/kthread.h>
59 
60 #include <asm/io.h>
61 #include <asm/bugs.h>
62 #include <asm/setup.h>
63 #include <asm/sections.h>
64 #include <asm/cacheflush.h>
65 
66 #ifdef CONFIG_X86_LOCAL_APIC
67 #include <asm/smp.h>
68 #endif
69 
70 /*
71  * This is one of the first .c files built. Error out early if we have compiler
72  * trouble.
73  */
74 
75 #if __GNUC__ == 4 && __GNUC_MINOR__ == 1 && __GNUC_PATCHLEVEL__ == 0
76 #warning gcc-4.1.0 is known to miscompile the kernel.  A different compiler version is recommended.
77 #endif
78 
79 static int kernel_init(void *);
80 
81 extern void init_IRQ(void);
82 extern void fork_init(unsigned long);
83 extern void mca_init(void);
84 extern void sbus_init(void);
85 extern void signals_init(void);
86 extern void pidhash_init(void);
87 extern void pidmap_init(void);
88 extern void prio_tree_init(void);
89 extern void radix_tree_init(void);
90 extern void free_initmem(void);
91 #ifdef	CONFIG_ACPI
92 extern void acpi_early_init(void);
93 #else
94 static inline void acpi_early_init(void) { }
95 #endif
96 #ifndef CONFIG_DEBUG_RODATA
97 static inline void mark_rodata_ro(void) { }
98 #endif
99 
100 #ifdef CONFIG_TC
101 extern void tc_init(void);
102 #endif
103 
104 enum system_states system_state;
105 EXPORT_SYMBOL(system_state);
106 
107 /*
108  * Boot command-line arguments
109  */
110 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
111 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
112 
113 extern void time_init(void);
114 /* Default late time init is NULL. archs can override this later. */
115 void (*late_time_init)(void);
116 extern void softirq_init(void);
117 
118 /* Untouched command line saved by arch-specific code. */
119 char __initdata boot_command_line[COMMAND_LINE_SIZE];
120 /* Untouched saved command line (eg. for /proc) */
121 char *saved_command_line;
122 /* Command line for parameter parsing */
123 static char *static_command_line;
124 
125 static char *execute_command;
126 static char *ramdisk_execute_command;
127 
128 #ifdef CONFIG_SMP
129 /* Setup configured maximum number of CPUs to activate */
130 static unsigned int __initdata max_cpus = NR_CPUS;
131 
132 /*
133  * Setup routine for controlling SMP activation
134  *
135  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
136  * activation entirely (the MPS table probe still happens, though).
137  *
138  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
139  * greater than 0, limits the maximum number of CPUs activated in
140  * SMP mode to <NUM>.
141  */
142 #ifndef CONFIG_X86_IO_APIC
143 static inline void disable_ioapic_setup(void) {};
144 #endif
145 
146 static int __init nosmp(char *str)
147 {
148 	max_cpus = 0;
149 	disable_ioapic_setup();
150 	return 0;
151 }
152 
153 early_param("nosmp", nosmp);
154 
155 static int __init maxcpus(char *str)
156 {
157 	get_option(&str, &max_cpus);
158 	if (max_cpus == 0)
159 		disable_ioapic_setup();
160 
161 	return 0;
162 }
163 
164 early_param("maxcpus", maxcpus);
165 #else
166 #define max_cpus NR_CPUS
167 #endif
168 
169 /*
170  * If set, this is an indication to the drivers that reset the underlying
171  * device before going ahead with the initialization otherwise driver might
172  * rely on the BIOS and skip the reset operation.
173  *
174  * This is useful if kernel is booting in an unreliable environment.
175  * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
176  * skipped and devices will be in unknown state.
177  */
178 unsigned int reset_devices;
179 EXPORT_SYMBOL(reset_devices);
180 
181 static int __init set_reset_devices(char *str)
182 {
183 	reset_devices = 1;
184 	return 1;
185 }
186 
187 __setup("reset_devices", set_reset_devices);
188 
189 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
190 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
191 static const char *panic_later, *panic_param;
192 
193 extern struct obs_kernel_param __setup_start[], __setup_end[];
194 
195 static int __init obsolete_checksetup(char *line)
196 {
197 	struct obs_kernel_param *p;
198 	int had_early_param = 0;
199 
200 	p = __setup_start;
201 	do {
202 		int n = strlen(p->str);
203 		if (!strncmp(line, p->str, n)) {
204 			if (p->early) {
205 				/* Already done in parse_early_param?
206 				 * (Needs exact match on param part).
207 				 * Keep iterating, as we can have early
208 				 * params and __setups of same names 8( */
209 				if (line[n] == '\0' || line[n] == '=')
210 					had_early_param = 1;
211 			} else if (!p->setup_func) {
212 				printk(KERN_WARNING "Parameter %s is obsolete,"
213 				       " ignored\n", p->str);
214 				return 1;
215 			} else if (p->setup_func(line + n))
216 				return 1;
217 		}
218 		p++;
219 	} while (p < __setup_end);
220 
221 	return had_early_param;
222 }
223 
224 /*
225  * This should be approx 2 Bo*oMips to start (note initial shift), and will
226  * still work even if initially too large, it will just take slightly longer
227  */
228 unsigned long loops_per_jiffy = (1<<12);
229 
230 EXPORT_SYMBOL(loops_per_jiffy);
231 
232 static int __init debug_kernel(char *str)
233 {
234 	if (*str)
235 		return 0;
236 	console_loglevel = 10;
237 	return 1;
238 }
239 
240 static int __init quiet_kernel(char *str)
241 {
242 	if (*str)
243 		return 0;
244 	console_loglevel = 4;
245 	return 1;
246 }
247 
248 __setup("debug", debug_kernel);
249 __setup("quiet", quiet_kernel);
250 
251 static int __init loglevel(char *str)
252 {
253 	get_option(&str, &console_loglevel);
254 	return 1;
255 }
256 
257 __setup("loglevel=", loglevel);
258 
259 /*
260  * Unknown boot options get handed to init, unless they look like
261  * failed parameters
262  */
263 static int __init unknown_bootoption(char *param, char *val)
264 {
265 	/* Change NUL term back to "=", to make "param" the whole string. */
266 	if (val) {
267 		/* param=val or param="val"? */
268 		if (val == param+strlen(param)+1)
269 			val[-1] = '=';
270 		else if (val == param+strlen(param)+2) {
271 			val[-2] = '=';
272 			memmove(val-1, val, strlen(val)+1);
273 			val--;
274 		} else
275 			BUG();
276 	}
277 
278 	/* Handle obsolete-style parameters */
279 	if (obsolete_checksetup(param))
280 		return 0;
281 
282 	/*
283 	 * Preemptive maintenance for "why didn't my misspelled command
284 	 * line work?"
285 	 */
286 	if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
287 		printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
288 		return 0;
289 	}
290 
291 	if (panic_later)
292 		return 0;
293 
294 	if (val) {
295 		/* Environment option */
296 		unsigned int i;
297 		for (i = 0; envp_init[i]; i++) {
298 			if (i == MAX_INIT_ENVS) {
299 				panic_later = "Too many boot env vars at `%s'";
300 				panic_param = param;
301 			}
302 			if (!strncmp(param, envp_init[i], val - param))
303 				break;
304 		}
305 		envp_init[i] = param;
306 	} else {
307 		/* Command line option */
308 		unsigned int i;
309 		for (i = 0; argv_init[i]; i++) {
310 			if (i == MAX_INIT_ARGS) {
311 				panic_later = "Too many boot init vars at `%s'";
312 				panic_param = param;
313 			}
314 		}
315 		argv_init[i] = param;
316 	}
317 	return 0;
318 }
319 
320 static int __init init_setup(char *str)
321 {
322 	unsigned int i;
323 
324 	execute_command = str;
325 	/*
326 	 * In case LILO is going to boot us with default command line,
327 	 * it prepends "auto" before the whole cmdline which makes
328 	 * the shell think it should execute a script with such name.
329 	 * So we ignore all arguments entered _before_ init=... [MJ]
330 	 */
331 	for (i = 1; i < MAX_INIT_ARGS; i++)
332 		argv_init[i] = NULL;
333 	return 1;
334 }
335 __setup("init=", init_setup);
336 
337 static int __init rdinit_setup(char *str)
338 {
339 	unsigned int i;
340 
341 	ramdisk_execute_command = str;
342 	/* See "auto" comment in init_setup */
343 	for (i = 1; i < MAX_INIT_ARGS; i++)
344 		argv_init[i] = NULL;
345 	return 1;
346 }
347 __setup("rdinit=", rdinit_setup);
348 
349 #ifndef CONFIG_SMP
350 
351 #ifdef CONFIG_X86_LOCAL_APIC
352 static void __init smp_init(void)
353 {
354 	APIC_init_uniprocessor();
355 }
356 #else
357 #define smp_init()	do { } while (0)
358 #endif
359 
360 static inline void setup_per_cpu_areas(void) { }
361 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
362 
363 #else
364 
365 #ifdef __GENERIC_PER_CPU
366 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
367 
368 EXPORT_SYMBOL(__per_cpu_offset);
369 
370 static void __init setup_per_cpu_areas(void)
371 {
372 	unsigned long size, i;
373 	char *ptr;
374 	unsigned long nr_possible_cpus = num_possible_cpus();
375 
376 	/* Copy section for each CPU (we discard the original) */
377 	size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
378 	ptr = alloc_bootmem_pages(size * nr_possible_cpus);
379 
380 	for_each_possible_cpu(i) {
381 		__per_cpu_offset[i] = ptr - __per_cpu_start;
382 		memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
383 		ptr += size;
384 	}
385 }
386 #endif /* !__GENERIC_PER_CPU */
387 
388 /* Called by boot processor to activate the rest. */
389 static void __init smp_init(void)
390 {
391 	unsigned int cpu;
392 
393 	/* FIXME: This should be done in userspace --RR */
394 	for_each_present_cpu(cpu) {
395 		if (num_online_cpus() >= max_cpus)
396 			break;
397 		if (!cpu_online(cpu))
398 			cpu_up(cpu);
399 	}
400 
401 	/* Any cleanup work */
402 	printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
403 	smp_cpus_done(max_cpus);
404 }
405 
406 #endif
407 
408 /*
409  * We need to store the untouched command line for future reference.
410  * We also need to store the touched command line since the parameter
411  * parsing is performed in place, and we should allow a component to
412  * store reference of name/value for future reference.
413  */
414 static void __init setup_command_line(char *command_line)
415 {
416 	saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
417 	static_command_line = alloc_bootmem(strlen (command_line)+1);
418 	strcpy (saved_command_line, boot_command_line);
419 	strcpy (static_command_line, command_line);
420 }
421 
422 /*
423  * We need to finalize in a non-__init function or else race conditions
424  * between the root thread and the init thread may cause start_kernel to
425  * be reaped by free_initmem before the root thread has proceeded to
426  * cpu_idle.
427  *
428  * gcc-3.4 accidentally inlines this function, so use noinline.
429  */
430 
431 static void noinline __init_refok rest_init(void)
432 	__releases(kernel_lock)
433 {
434 	int pid;
435 
436 	kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
437 	numa_default_policy();
438 	pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
439 	kthreadd_task = find_task_by_pid(pid);
440 	unlock_kernel();
441 
442 	/*
443 	 * The boot idle thread must execute schedule()
444 	 * at least once to get things moving:
445 	 */
446 	init_idle_bootup_task(current);
447 	preempt_enable_no_resched();
448 	schedule();
449 	preempt_disable();
450 
451 	/* Call into cpu_idle with preempt disabled */
452 	cpu_idle();
453 }
454 
455 /* Check for early params. */
456 static int __init do_early_param(char *param, char *val)
457 {
458 	struct obs_kernel_param *p;
459 
460 	for (p = __setup_start; p < __setup_end; p++) {
461 		if ((p->early && strcmp(param, p->str) == 0) ||
462 		    (strcmp(param, "console") == 0 &&
463 		     strcmp(p->str, "earlycon") == 0)
464 		) {
465 			if (p->setup_func(val) != 0)
466 				printk(KERN_WARNING
467 				       "Malformed early option '%s'\n", param);
468 		}
469 	}
470 	/* We accept everything at this stage. */
471 	return 0;
472 }
473 
474 /* Arch code calls this early on, or if not, just before other parsing. */
475 void __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. */
484 	strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
485 	parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
486 	done = 1;
487 }
488 
489 /*
490  *	Activate the first processor.
491  */
492 
493 static 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 */
497 	cpu_set(cpu, cpu_online_map);
498 	cpu_set(cpu, cpu_present_map);
499 	cpu_set(cpu, cpu_possible_map);
500 }
501 
502 void __init __attribute__((weak)) smp_setup_processor_id(void)
503 {
504 }
505 
506 asmlinkage void __init start_kernel(void)
507 {
508 	char * command_line;
509 	extern struct kernel_param __start___param[], __stop___param[];
510 
511 	smp_setup_processor_id();
512 
513 	/*
514 	 * Need to run as early as possible, to initialize the
515 	 * lockdep hash:
516 	 */
517 	unwind_init();
518 	lockdep_init();
519 	cgroup_init_early();
520 
521 	local_irq_disable();
522 	early_boot_irqs_off();
523 	early_init_irq_lock_class();
524 
525 /*
526  * Interrupts are still disabled. Do necessary setups, then
527  * enable them
528  */
529 	lock_kernel();
530 	tick_init();
531 	boot_cpu_init();
532 	page_address_init();
533 	printk(KERN_NOTICE);
534 	printk(linux_banner);
535 	setup_arch(&command_line);
536 	setup_command_line(command_line);
537 	unwind_setup();
538 	setup_per_cpu_areas();
539 	smp_prepare_boot_cpu();	/* arch-specific boot-cpu hooks */
540 
541 	/*
542 	 * Set up the scheduler prior starting any interrupts (such as the
543 	 * timer interrupt). Full topology setup happens at smp_init()
544 	 * time - but meanwhile we still have a functioning scheduler.
545 	 */
546 	sched_init();
547 	/*
548 	 * Disable preemption - early bootup scheduling is extremely
549 	 * fragile until we cpu_idle() for the first time.
550 	 */
551 	preempt_disable();
552 	build_all_zonelists();
553 	page_alloc_init();
554 	printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
555 	parse_early_param();
556 	parse_args("Booting kernel", static_command_line, __start___param,
557 		   __stop___param - __start___param,
558 		   &unknown_bootoption);
559 	if (!irqs_disabled()) {
560 		printk(KERN_WARNING "start_kernel(): bug: interrupts were "
561 				"enabled *very* early, fixing it\n");
562 		local_irq_disable();
563 	}
564 	sort_main_extable();
565 	trap_init();
566 	rcu_init();
567 	init_IRQ();
568 	pidhash_init();
569 	init_timers();
570 	hrtimers_init();
571 	softirq_init();
572 	timekeeping_init();
573 	time_init();
574 	profile_init();
575 	if (!irqs_disabled())
576 		printk("start_kernel(): bug: interrupts were enabled early\n");
577 	early_boot_irqs_on();
578 	local_irq_enable();
579 
580 	/*
581 	 * HACK ALERT! This is early. We're enabling the console before
582 	 * we've done PCI setups etc, and console_init() must be aware of
583 	 * this. But we do want output early, in case something goes wrong.
584 	 */
585 	console_init();
586 	if (panic_later)
587 		panic(panic_later, panic_param);
588 
589 	lockdep_info();
590 
591 	/*
592 	 * Need to run this when irqs are enabled, because it wants
593 	 * to self-test [hard/soft]-irqs on/off lock inversion bugs
594 	 * too:
595 	 */
596 	locking_selftest();
597 
598 #ifdef CONFIG_BLK_DEV_INITRD
599 	if (initrd_start && !initrd_below_start_ok &&
600 			initrd_start < min_low_pfn << PAGE_SHIFT) {
601 		printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
602 		    "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
603 		initrd_start = 0;
604 	}
605 #endif
606 	vfs_caches_init_early();
607 	cpuset_init_early();
608 	mem_init();
609 	kmem_cache_init();
610 	setup_per_cpu_pageset();
611 	numa_policy_init();
612 	if (late_time_init)
613 		late_time_init();
614 	calibrate_delay();
615 	pidmap_init();
616 	pgtable_cache_init();
617 	prio_tree_init();
618 	anon_vma_init();
619 #ifdef CONFIG_X86
620 	if (efi_enabled)
621 		efi_enter_virtual_mode();
622 #endif
623 	fork_init(num_physpages);
624 	proc_caches_init();
625 	buffer_init();
626 	unnamed_dev_init();
627 	key_init();
628 	security_init();
629 	vfs_caches_init(num_physpages);
630 	radix_tree_init();
631 	signals_init();
632 	/* rootfs populating might need page-writeback */
633 	page_writeback_init();
634 #ifdef CONFIG_PROC_FS
635 	proc_root_init();
636 #endif
637 	cgroup_init();
638 	cpuset_init();
639 	taskstats_init_early();
640 	delayacct_init();
641 
642 	check_bugs();
643 
644 	acpi_early_init(); /* before LAPIC and SMP init */
645 
646 	/* Do the rest non-__init'ed, we're now alive */
647 	rest_init();
648 }
649 
650 static int __initdata initcall_debug;
651 
652 static int __init initcall_debug_setup(char *str)
653 {
654 	initcall_debug = 1;
655 	return 1;
656 }
657 __setup("initcall_debug", initcall_debug_setup);
658 
659 extern initcall_t __initcall_start[], __initcall_end[];
660 
661 static void __init do_initcalls(void)
662 {
663 	initcall_t *call;
664 	int count = preempt_count();
665 
666 	for (call = __initcall_start; call < __initcall_end; call++) {
667 		ktime_t t0, t1, delta;
668 		char *msg = NULL;
669 		char msgbuf[40];
670 		int result;
671 
672 		if (initcall_debug) {
673 			printk("Calling initcall 0x%p", *call);
674 			print_fn_descriptor_symbol(": %s()",
675 					(unsigned long) *call);
676 			printk("\n");
677 			t0 = ktime_get();
678 		}
679 
680 		result = (*call)();
681 
682 		if (initcall_debug) {
683 			t1 = ktime_get();
684 			delta = ktime_sub(t1, t0);
685 
686 			printk("initcall 0x%p", *call);
687 			print_fn_descriptor_symbol(": %s()",
688 					(unsigned long) *call);
689 			printk(" returned %d.\n", result);
690 
691 			printk("initcall 0x%p ran for %Ld msecs: ",
692 				*call, (unsigned long long)delta.tv64 >> 20);
693 			print_fn_descriptor_symbol("%s()\n",
694 				(unsigned long) *call);
695 		}
696 
697 		if (result && result != -ENODEV && initcall_debug) {
698 			sprintf(msgbuf, "error code %d", result);
699 			msg = msgbuf;
700 		}
701 		if (preempt_count() != count) {
702 			msg = "preemption imbalance";
703 			preempt_count() = count;
704 		}
705 		if (irqs_disabled()) {
706 			msg = "disabled interrupts";
707 			local_irq_enable();
708 		}
709 		if (msg) {
710 			printk(KERN_WARNING "initcall at 0x%p", *call);
711 			print_fn_descriptor_symbol(": %s()",
712 					(unsigned long) *call);
713 			printk(": returned with %s\n", msg);
714 		}
715 	}
716 
717 	/* Make sure there is no pending stuff from the initcall sequence */
718 	flush_scheduled_work();
719 }
720 
721 /*
722  * Ok, the machine is now initialized. None of the devices
723  * have been touched yet, but the CPU subsystem is up and
724  * running, and memory and process management works.
725  *
726  * Now we can finally start doing some real work..
727  */
728 static void __init do_basic_setup(void)
729 {
730 	/* drivers will send hotplug events */
731 	init_workqueues();
732 	usermodehelper_init();
733 	driver_init();
734 	init_irq_proc();
735 	do_initcalls();
736 }
737 
738 static int __initdata nosoftlockup;
739 
740 static int __init nosoftlockup_setup(char *str)
741 {
742 	nosoftlockup = 1;
743 	return 1;
744 }
745 __setup("nosoftlockup", nosoftlockup_setup);
746 
747 static void __init do_pre_smp_initcalls(void)
748 {
749 	extern int spawn_ksoftirqd(void);
750 #ifdef CONFIG_SMP
751 	extern int migration_init(void);
752 
753 	migration_init();
754 #endif
755 	spawn_ksoftirqd();
756 	if (!nosoftlockup)
757 		spawn_softlockup_task();
758 }
759 
760 static void run_init_process(char *init_filename)
761 {
762 	argv_init[0] = init_filename;
763 	kernel_execve(init_filename, argv_init, envp_init);
764 }
765 
766 /* This is a non __init function. Force it to be noinline otherwise gcc
767  * makes it inline to init() and it becomes part of init.text section
768  */
769 static int noinline init_post(void)
770 {
771 	free_initmem();
772 	unlock_kernel();
773 	mark_rodata_ro();
774 	system_state = SYSTEM_RUNNING;
775 	numa_default_policy();
776 
777 	if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
778 		printk(KERN_WARNING "Warning: unable to open an initial console.\n");
779 
780 	(void) sys_dup(0);
781 	(void) sys_dup(0);
782 
783 	if (ramdisk_execute_command) {
784 		run_init_process(ramdisk_execute_command);
785 		printk(KERN_WARNING "Failed to execute %s\n",
786 				ramdisk_execute_command);
787 	}
788 
789 	/*
790 	 * We try each of these until one succeeds.
791 	 *
792 	 * The Bourne shell can be used instead of init if we are
793 	 * trying to recover a really broken machine.
794 	 */
795 	if (execute_command) {
796 		run_init_process(execute_command);
797 		printk(KERN_WARNING "Failed to execute %s.  Attempting "
798 					"defaults...\n", execute_command);
799 	}
800 	run_init_process("/sbin/init");
801 	run_init_process("/etc/init");
802 	run_init_process("/bin/init");
803 	run_init_process("/bin/sh");
804 
805 	panic("No init found.  Try passing init= option to kernel.");
806 }
807 
808 static int __init kernel_init(void * unused)
809 {
810 	lock_kernel();
811 	/*
812 	 * init can run on any cpu.
813 	 */
814 	set_cpus_allowed(current, CPU_MASK_ALL);
815 	/*
816 	 * Tell the world that we're going to be the grim
817 	 * reaper of innocent orphaned children.
818 	 *
819 	 * We don't want people to have to make incorrect
820 	 * assumptions about where in the task array this
821 	 * can be found.
822 	 */
823 	init_pid_ns.child_reaper = current;
824 
825 	__set_special_pids(1, 1);
826 	cad_pid = task_pid(current);
827 
828 	smp_prepare_cpus(max_cpus);
829 
830 	do_pre_smp_initcalls();
831 
832 	smp_init();
833 	sched_init_smp();
834 
835 	cpuset_init_smp();
836 
837 	do_basic_setup();
838 
839 	/*
840 	 * check if there is an early userspace init.  If yes, let it do all
841 	 * the work
842 	 */
843 
844 	if (!ramdisk_execute_command)
845 		ramdisk_execute_command = "/init";
846 
847 	if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
848 		ramdisk_execute_command = NULL;
849 		prepare_namespace();
850 	}
851 
852 	/*
853 	 * Ok, we have completed the initial bootup, and
854 	 * we're essentially up and running. Get rid of the
855 	 * initmem segments and start the user-mode stuff..
856 	 */
857 	init_post();
858 	return 0;
859 }
860