xref: /linux/init/main.c (revision 9c23c516b54d2f383f0868ac1e2b351e7b3975ec)
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/stackprotector.h>
18 #include <linux/string.h>
19 #include <linux/ctype.h>
20 #include <linux/delay.h>
21 #include <linux/ioport.h>
22 #include <linux/init.h>
23 #include <linux/initrd.h>
24 #include <linux/bootmem.h>
25 #include <linux/acpi.h>
26 #include <linux/tty.h>
27 #include <linux/percpu.h>
28 #include <linux/kmod.h>
29 #include <linux/vmalloc.h>
30 #include <linux/kernel_stat.h>
31 #include <linux/start_kernel.h>
32 #include <linux/security.h>
33 #include <linux/smp.h>
34 #include <linux/profile.h>
35 #include <linux/rcupdate.h>
36 #include <linux/moduleparam.h>
37 #include <linux/kallsyms.h>
38 #include <linux/writeback.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/cgroup.h>
42 #include <linux/efi.h>
43 #include <linux/tick.h>
44 #include <linux/interrupt.h>
45 #include <linux/taskstats_kern.h>
46 #include <linux/delayacct.h>
47 #include <linux/unistd.h>
48 #include <linux/rmap.h>
49 #include <linux/mempolicy.h>
50 #include <linux/key.h>
51 #include <linux/buffer_head.h>
52 #include <linux/page_cgroup.h>
53 #include <linux/debug_locks.h>
54 #include <linux/debugobjects.h>
55 #include <linux/lockdep.h>
56 #include <linux/kmemleak.h>
57 #include <linux/pid_namespace.h>
58 #include <linux/device.h>
59 #include <linux/kthread.h>
60 #include <linux/sched.h>
61 #include <linux/signal.h>
62 #include <linux/idr.h>
63 #include <linux/kgdb.h>
64 #include <linux/ftrace.h>
65 #include <linux/async.h>
66 #include <linux/kmemcheck.h>
67 #include <linux/sfi.h>
68 #include <linux/shmem_fs.h>
69 #include <linux/slab.h>
70 #include <linux/perf_event.h>
71 
72 #include <asm/io.h>
73 #include <asm/bugs.h>
74 #include <asm/setup.h>
75 #include <asm/sections.h>
76 #include <asm/cacheflush.h>
77 
78 #ifdef CONFIG_X86_LOCAL_APIC
79 #include <asm/smp.h>
80 #endif
81 
82 static int kernel_init(void *);
83 
84 extern void init_IRQ(void);
85 extern void fork_init(unsigned long);
86 extern void mca_init(void);
87 extern void sbus_init(void);
88 extern void prio_tree_init(void);
89 extern void radix_tree_init(void);
90 #ifndef CONFIG_DEBUG_RODATA
91 static inline void mark_rodata_ro(void) { }
92 #endif
93 
94 #ifdef CONFIG_TC
95 extern void tc_init(void);
96 #endif
97 
98 /*
99  * Debug helper: via this flag we know that we are in 'early bootup code'
100  * where only the boot processor is running with IRQ disabled.  This means
101  * two things - IRQ must not be enabled before the flag is cleared and some
102  * operations which are not allowed with IRQ disabled are allowed while the
103  * flag is set.
104  */
105 bool early_boot_irqs_disabled __read_mostly;
106 
107 enum system_states system_state __read_mostly;
108 EXPORT_SYMBOL(system_state);
109 
110 /*
111  * Boot command-line arguments
112  */
113 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
114 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
115 
116 extern void time_init(void);
117 /* Default late time init is NULL. archs can override this later. */
118 void (*__initdata late_time_init)(void);
119 extern void softirq_init(void);
120 
121 /* Untouched command line saved by arch-specific code. */
122 char __initdata boot_command_line[COMMAND_LINE_SIZE];
123 /* Untouched saved command line (eg. for /proc) */
124 char *saved_command_line;
125 /* Command line for parameter parsing */
126 static char *static_command_line;
127 
128 static char *execute_command;
129 static char *ramdisk_execute_command;
130 
131 /*
132  * If set, this is an indication to the drivers that reset the underlying
133  * device before going ahead with the initialization otherwise driver might
134  * rely on the BIOS and skip the reset operation.
135  *
136  * This is useful if kernel is booting in an unreliable environment.
137  * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
138  * skipped and devices will be in unknown state.
139  */
140 unsigned int reset_devices;
141 EXPORT_SYMBOL(reset_devices);
142 
143 static int __init set_reset_devices(char *str)
144 {
145 	reset_devices = 1;
146 	return 1;
147 }
148 
149 __setup("reset_devices", set_reset_devices);
150 
151 static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
152 const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
153 static const char *panic_later, *panic_param;
154 
155 extern const struct obs_kernel_param __setup_start[], __setup_end[];
156 
157 static int __init obsolete_checksetup(char *line)
158 {
159 	const struct obs_kernel_param *p;
160 	int had_early_param = 0;
161 
162 	p = __setup_start;
163 	do {
164 		int n = strlen(p->str);
165 		if (parameqn(line, p->str, n)) {
166 			if (p->early) {
167 				/* Already done in parse_early_param?
168 				 * (Needs exact match on param part).
169 				 * Keep iterating, as we can have early
170 				 * params and __setups of same names 8( */
171 				if (line[n] == '\0' || line[n] == '=')
172 					had_early_param = 1;
173 			} else if (!p->setup_func) {
174 				printk(KERN_WARNING "Parameter %s is obsolete,"
175 				       " ignored\n", p->str);
176 				return 1;
177 			} else if (p->setup_func(line + n))
178 				return 1;
179 		}
180 		p++;
181 	} while (p < __setup_end);
182 
183 	return had_early_param;
184 }
185 
186 /*
187  * This should be approx 2 Bo*oMips to start (note initial shift), and will
188  * still work even if initially too large, it will just take slightly longer
189  */
190 unsigned long loops_per_jiffy = (1<<12);
191 
192 EXPORT_SYMBOL(loops_per_jiffy);
193 
194 static int __init debug_kernel(char *str)
195 {
196 	console_loglevel = 10;
197 	return 0;
198 }
199 
200 static int __init quiet_kernel(char *str)
201 {
202 	console_loglevel = 4;
203 	return 0;
204 }
205 
206 early_param("debug", debug_kernel);
207 early_param("quiet", quiet_kernel);
208 
209 static int __init loglevel(char *str)
210 {
211 	int newlevel;
212 
213 	/*
214 	 * Only update loglevel value when a correct setting was passed,
215 	 * to prevent blind crashes (when loglevel being set to 0) that
216 	 * are quite hard to debug
217 	 */
218 	if (get_option(&str, &newlevel)) {
219 		console_loglevel = newlevel;
220 		return 0;
221 	}
222 
223 	return -EINVAL;
224 }
225 
226 early_param("loglevel", loglevel);
227 
228 /* Change NUL term back to "=", to make "param" the whole string. */
229 static int __init repair_env_string(char *param, char *val)
230 {
231 	if (val) {
232 		/* param=val or param="val"? */
233 		if (val == param+strlen(param)+1)
234 			val[-1] = '=';
235 		else if (val == param+strlen(param)+2) {
236 			val[-2] = '=';
237 			memmove(val-1, val, strlen(val)+1);
238 			val--;
239 		} else
240 			BUG();
241 	}
242 	return 0;
243 }
244 
245 /*
246  * Unknown boot options get handed to init, unless they look like
247  * unused parameters (modprobe will find them in /proc/cmdline).
248  */
249 static int __init unknown_bootoption(char *param, char *val)
250 {
251 	repair_env_string(param, val);
252 
253 	/* Handle obsolete-style parameters */
254 	if (obsolete_checksetup(param))
255 		return 0;
256 
257 	/* Unused module parameter. */
258 	if (strchr(param, '.') && (!val || strchr(param, '.') < val))
259 		return 0;
260 
261 	if (panic_later)
262 		return 0;
263 
264 	if (val) {
265 		/* Environment option */
266 		unsigned int i;
267 		for (i = 0; envp_init[i]; i++) {
268 			if (i == MAX_INIT_ENVS) {
269 				panic_later = "Too many boot env vars at `%s'";
270 				panic_param = param;
271 			}
272 			if (!strncmp(param, envp_init[i], val - param))
273 				break;
274 		}
275 		envp_init[i] = param;
276 	} else {
277 		/* Command line option */
278 		unsigned int i;
279 		for (i = 0; argv_init[i]; i++) {
280 			if (i == MAX_INIT_ARGS) {
281 				panic_later = "Too many boot init vars at `%s'";
282 				panic_param = param;
283 			}
284 		}
285 		argv_init[i] = param;
286 	}
287 	return 0;
288 }
289 
290 static int __init init_setup(char *str)
291 {
292 	unsigned int i;
293 
294 	execute_command = str;
295 	/*
296 	 * In case LILO is going to boot us with default command line,
297 	 * it prepends "auto" before the whole cmdline which makes
298 	 * the shell think it should execute a script with such name.
299 	 * So we ignore all arguments entered _before_ init=... [MJ]
300 	 */
301 	for (i = 1; i < MAX_INIT_ARGS; i++)
302 		argv_init[i] = NULL;
303 	return 1;
304 }
305 __setup("init=", init_setup);
306 
307 static int __init rdinit_setup(char *str)
308 {
309 	unsigned int i;
310 
311 	ramdisk_execute_command = str;
312 	/* See "auto" comment in init_setup */
313 	for (i = 1; i < MAX_INIT_ARGS; i++)
314 		argv_init[i] = NULL;
315 	return 1;
316 }
317 __setup("rdinit=", rdinit_setup);
318 
319 #ifndef CONFIG_SMP
320 static const unsigned int setup_max_cpus = NR_CPUS;
321 #ifdef CONFIG_X86_LOCAL_APIC
322 static void __init smp_init(void)
323 {
324 	APIC_init_uniprocessor();
325 }
326 #else
327 #define smp_init()	do { } while (0)
328 #endif
329 
330 static inline void setup_nr_cpu_ids(void) { }
331 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
332 #endif
333 
334 /*
335  * We need to store the untouched command line for future reference.
336  * We also need to store the touched command line since the parameter
337  * parsing is performed in place, and we should allow a component to
338  * store reference of name/value for future reference.
339  */
340 static void __init setup_command_line(char *command_line)
341 {
342 	saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
343 	static_command_line = alloc_bootmem(strlen (command_line)+1);
344 	strcpy (saved_command_line, boot_command_line);
345 	strcpy (static_command_line, command_line);
346 }
347 
348 /*
349  * We need to finalize in a non-__init function or else race conditions
350  * between the root thread and the init thread may cause start_kernel to
351  * be reaped by free_initmem before the root thread has proceeded to
352  * cpu_idle.
353  *
354  * gcc-3.4 accidentally inlines this function, so use noinline.
355  */
356 
357 static __initdata DECLARE_COMPLETION(kthreadd_done);
358 
359 static noinline void __init_refok rest_init(void)
360 {
361 	int pid;
362 
363 	rcu_scheduler_starting();
364 	/*
365 	 * We need to spawn init first so that it obtains pid 1, however
366 	 * the init task will end up wanting to create kthreads, which, if
367 	 * we schedule it before we create kthreadd, will OOPS.
368 	 */
369 	kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
370 	numa_default_policy();
371 	pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
372 	rcu_read_lock();
373 	kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
374 	rcu_read_unlock();
375 	complete(&kthreadd_done);
376 
377 	/*
378 	 * The boot idle thread must execute schedule()
379 	 * at least once to get things moving:
380 	 */
381 	init_idle_bootup_task(current);
382 	schedule_preempt_disabled();
383 	/* Call into cpu_idle with preempt disabled */
384 	cpu_idle();
385 }
386 
387 /* Check for early params. */
388 static int __init do_early_param(char *param, char *val)
389 {
390 	const struct obs_kernel_param *p;
391 
392 	for (p = __setup_start; p < __setup_end; p++) {
393 		if ((p->early && parameq(param, p->str)) ||
394 		    (strcmp(param, "console") == 0 &&
395 		     strcmp(p->str, "earlycon") == 0)
396 		) {
397 			if (p->setup_func(val) != 0)
398 				printk(KERN_WARNING
399 				       "Malformed early option '%s'\n", param);
400 		}
401 	}
402 	/* We accept everything at this stage. */
403 	return 0;
404 }
405 
406 void __init parse_early_options(char *cmdline)
407 {
408 	parse_args("early options", cmdline, NULL, 0, 0, 0, do_early_param);
409 }
410 
411 /* Arch code calls this early on, or if not, just before other parsing. */
412 void __init parse_early_param(void)
413 {
414 	static __initdata int done = 0;
415 	static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
416 
417 	if (done)
418 		return;
419 
420 	/* All fall through to do_early_param. */
421 	strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
422 	parse_early_options(tmp_cmdline);
423 	done = 1;
424 }
425 
426 /*
427  *	Activate the first processor.
428  */
429 
430 static void __init boot_cpu_init(void)
431 {
432 	int cpu = smp_processor_id();
433 	/* Mark the boot cpu "present", "online" etc for SMP and UP case */
434 	set_cpu_online(cpu, true);
435 	set_cpu_active(cpu, true);
436 	set_cpu_present(cpu, true);
437 	set_cpu_possible(cpu, true);
438 }
439 
440 void __init __weak smp_setup_processor_id(void)
441 {
442 }
443 
444 void __init __weak thread_info_cache_init(void)
445 {
446 }
447 
448 /*
449  * Set up kernel memory allocators
450  */
451 static void __init mm_init(void)
452 {
453 	/*
454 	 * page_cgroup requires contiguous pages,
455 	 * bigger than MAX_ORDER unless SPARSEMEM.
456 	 */
457 	page_cgroup_init_flatmem();
458 	mem_init();
459 	kmem_cache_init();
460 	percpu_init_late();
461 	pgtable_cache_init();
462 	vmalloc_init();
463 }
464 
465 asmlinkage void __init start_kernel(void)
466 {
467 	char * command_line;
468 	extern const struct kernel_param __start___param[], __stop___param[];
469 
470 	/*
471 	 * Need to run as early as possible, to initialize the
472 	 * lockdep hash:
473 	 */
474 	lockdep_init();
475 	smp_setup_processor_id();
476 	debug_objects_early_init();
477 
478 	/*
479 	 * Set up the the initial canary ASAP:
480 	 */
481 	boot_init_stack_canary();
482 
483 	cgroup_init_early();
484 
485 	local_irq_disable();
486 	early_boot_irqs_disabled = true;
487 
488 /*
489  * Interrupts are still disabled. Do necessary setups, then
490  * enable them
491  */
492 	tick_init();
493 	boot_cpu_init();
494 	page_address_init();
495 	printk(KERN_NOTICE "%s", linux_banner);
496 	setup_arch(&command_line);
497 	mm_init_owner(&init_mm, &init_task);
498 	mm_init_cpumask(&init_mm);
499 	setup_command_line(command_line);
500 	setup_nr_cpu_ids();
501 	setup_per_cpu_areas();
502 	smp_prepare_boot_cpu();	/* arch-specific boot-cpu hooks */
503 
504 	build_all_zonelists(NULL);
505 	page_alloc_init();
506 
507 	printk(KERN_NOTICE "Kernel command line: %s\n", boot_command_line);
508 	parse_early_param();
509 	parse_args("Booting kernel", static_command_line, __start___param,
510 		   __stop___param - __start___param,
511 		   0, 0, &unknown_bootoption);
512 
513 	jump_label_init();
514 
515 	/*
516 	 * These use large bootmem allocations and must precede
517 	 * kmem_cache_init()
518 	 */
519 	setup_log_buf(0);
520 	pidhash_init();
521 	vfs_caches_init_early();
522 	sort_main_extable();
523 	trap_init();
524 	mm_init();
525 
526 	/*
527 	 * Set up the scheduler prior starting any interrupts (such as the
528 	 * timer interrupt). Full topology setup happens at smp_init()
529 	 * time - but meanwhile we still have a functioning scheduler.
530 	 */
531 	sched_init();
532 	/*
533 	 * Disable preemption - early bootup scheduling is extremely
534 	 * fragile until we cpu_idle() for the first time.
535 	 */
536 	preempt_disable();
537 	if (!irqs_disabled()) {
538 		printk(KERN_WARNING "start_kernel(): bug: interrupts were "
539 				"enabled *very* early, fixing it\n");
540 		local_irq_disable();
541 	}
542 	idr_init_cache();
543 	perf_event_init();
544 	rcu_init();
545 	radix_tree_init();
546 	/* init some links before init_ISA_irqs() */
547 	early_irq_init();
548 	init_IRQ();
549 	prio_tree_init();
550 	init_timers();
551 	hrtimers_init();
552 	softirq_init();
553 	timekeeping_init();
554 	time_init();
555 	profile_init();
556 	call_function_init();
557 	if (!irqs_disabled())
558 		printk(KERN_CRIT "start_kernel(): bug: interrupts were "
559 				 "enabled early\n");
560 	early_boot_irqs_disabled = false;
561 	local_irq_enable();
562 
563 	/* Interrupts are enabled now so all GFP allocations are safe. */
564 	gfp_allowed_mask = __GFP_BITS_MASK;
565 
566 	kmem_cache_init_late();
567 
568 	/*
569 	 * HACK ALERT! This is early. We're enabling the console before
570 	 * we've done PCI setups etc, and console_init() must be aware of
571 	 * this. But we do want output early, in case something goes wrong.
572 	 */
573 	console_init();
574 	if (panic_later)
575 		panic(panic_later, panic_param);
576 
577 	lockdep_info();
578 
579 	/*
580 	 * Need to run this when irqs are enabled, because it wants
581 	 * to self-test [hard/soft]-irqs on/off lock inversion bugs
582 	 * too:
583 	 */
584 	locking_selftest();
585 
586 #ifdef CONFIG_BLK_DEV_INITRD
587 	if (initrd_start && !initrd_below_start_ok &&
588 	    page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
589 		printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
590 		    "disabling it.\n",
591 		    page_to_pfn(virt_to_page((void *)initrd_start)),
592 		    min_low_pfn);
593 		initrd_start = 0;
594 	}
595 #endif
596 	page_cgroup_init();
597 	debug_objects_mem_init();
598 	kmemleak_init();
599 	setup_per_cpu_pageset();
600 	numa_policy_init();
601 	if (late_time_init)
602 		late_time_init();
603 	sched_clock_init();
604 	calibrate_delay();
605 	pidmap_init();
606 	anon_vma_init();
607 #ifdef CONFIG_X86
608 	if (efi_enabled)
609 		efi_enter_virtual_mode();
610 #endif
611 	thread_info_cache_init();
612 	cred_init();
613 	fork_init(totalram_pages);
614 	proc_caches_init();
615 	buffer_init();
616 	key_init();
617 	security_init();
618 	dbg_late_init();
619 	vfs_caches_init(totalram_pages);
620 	signals_init();
621 	/* rootfs populating might need page-writeback */
622 	page_writeback_init();
623 #ifdef CONFIG_PROC_FS
624 	proc_root_init();
625 #endif
626 	cgroup_init();
627 	cpuset_init();
628 	taskstats_init_early();
629 	delayacct_init();
630 
631 	check_bugs();
632 
633 	acpi_early_init(); /* before LAPIC and SMP init */
634 	sfi_init_late();
635 
636 	ftrace_init();
637 
638 	/* Do the rest non-__init'ed, we're now alive */
639 	rest_init();
640 }
641 
642 /* Call all constructor functions linked into the kernel. */
643 static void __init do_ctors(void)
644 {
645 #ifdef CONFIG_CONSTRUCTORS
646 	ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
647 
648 	for (; fn < (ctor_fn_t *) __ctors_end; fn++)
649 		(*fn)();
650 #endif
651 }
652 
653 bool initcall_debug;
654 core_param(initcall_debug, initcall_debug, bool, 0644);
655 
656 static char msgbuf[64];
657 
658 static int __init_or_module do_one_initcall_debug(initcall_t fn)
659 {
660 	ktime_t calltime, delta, rettime;
661 	unsigned long long duration;
662 	int ret;
663 
664 	printk(KERN_DEBUG "calling  %pF @ %i\n", fn, task_pid_nr(current));
665 	calltime = ktime_get();
666 	ret = fn();
667 	rettime = ktime_get();
668 	delta = ktime_sub(rettime, calltime);
669 	duration = (unsigned long long) ktime_to_ns(delta) >> 10;
670 	printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n", fn,
671 		ret, duration);
672 
673 	return ret;
674 }
675 
676 int __init_or_module do_one_initcall(initcall_t fn)
677 {
678 	int count = preempt_count();
679 	int ret;
680 
681 	if (initcall_debug)
682 		ret = do_one_initcall_debug(fn);
683 	else
684 		ret = fn();
685 
686 	msgbuf[0] = 0;
687 
688 	if (ret && ret != -ENODEV && initcall_debug)
689 		sprintf(msgbuf, "error code %d ", ret);
690 
691 	if (preempt_count() != count) {
692 		strlcat(msgbuf, "preemption imbalance ", sizeof(msgbuf));
693 		preempt_count() = count;
694 	}
695 	if (irqs_disabled()) {
696 		strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
697 		local_irq_enable();
698 	}
699 	if (msgbuf[0]) {
700 		printk("initcall %pF returned with %s\n", fn, msgbuf);
701 	}
702 
703 	return ret;
704 }
705 
706 
707 extern initcall_t __initcall_start[];
708 extern initcall_t __initcall0_start[];
709 extern initcall_t __initcall1_start[];
710 extern initcall_t __initcall2_start[];
711 extern initcall_t __initcall3_start[];
712 extern initcall_t __initcall4_start[];
713 extern initcall_t __initcall5_start[];
714 extern initcall_t __initcall6_start[];
715 extern initcall_t __initcall7_start[];
716 extern initcall_t __initcall_end[];
717 
718 static initcall_t *initcall_levels[] __initdata = {
719 	__initcall0_start,
720 	__initcall1_start,
721 	__initcall2_start,
722 	__initcall3_start,
723 	__initcall4_start,
724 	__initcall5_start,
725 	__initcall6_start,
726 	__initcall7_start,
727 	__initcall_end,
728 };
729 
730 static char *initcall_level_names[] __initdata = {
731 	"early parameters",
732 	"core parameters",
733 	"postcore parameters",
734 	"arch parameters",
735 	"subsys parameters",
736 	"fs parameters",
737 	"device parameters",
738 	"late parameters",
739 };
740 
741 static void __init do_initcall_level(int level)
742 {
743 	extern const struct kernel_param __start___param[], __stop___param[];
744 	initcall_t *fn;
745 
746 	strcpy(static_command_line, saved_command_line);
747 	parse_args(initcall_level_names[level],
748 		   static_command_line, __start___param,
749 		   __stop___param - __start___param,
750 		   level, level,
751 		   repair_env_string);
752 
753 	for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
754 		do_one_initcall(*fn);
755 }
756 
757 static void __init do_initcalls(void)
758 {
759 	int level;
760 
761 	for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
762 		do_initcall_level(level);
763 }
764 
765 /*
766  * Ok, the machine is now initialized. None of the devices
767  * have been touched yet, but the CPU subsystem is up and
768  * running, and memory and process management works.
769  *
770  * Now we can finally start doing some real work..
771  */
772 static void __init do_basic_setup(void)
773 {
774 	cpuset_init_smp();
775 	usermodehelper_init();
776 	shmem_init();
777 	driver_init();
778 	init_irq_proc();
779 	do_ctors();
780 	usermodehelper_enable();
781 	do_initcalls();
782 }
783 
784 static void __init do_pre_smp_initcalls(void)
785 {
786 	initcall_t *fn;
787 
788 	for (fn = __initcall_start; fn < __initcall0_start; fn++)
789 		do_one_initcall(*fn);
790 }
791 
792 static void run_init_process(const char *init_filename)
793 {
794 	argv_init[0] = init_filename;
795 	kernel_execve(init_filename, argv_init, envp_init);
796 }
797 
798 /* This is a non __init function. Force it to be noinline otherwise gcc
799  * makes it inline to init() and it becomes part of init.text section
800  */
801 static noinline int init_post(void)
802 {
803 	/* need to finish all async __init code before freeing the memory */
804 	async_synchronize_full();
805 	free_initmem();
806 	mark_rodata_ro();
807 	system_state = SYSTEM_RUNNING;
808 	numa_default_policy();
809 
810 
811 	current->signal->flags |= SIGNAL_UNKILLABLE;
812 
813 	if (ramdisk_execute_command) {
814 		run_init_process(ramdisk_execute_command);
815 		printk(KERN_WARNING "Failed to execute %s\n",
816 				ramdisk_execute_command);
817 	}
818 
819 	/*
820 	 * We try each of these until one succeeds.
821 	 *
822 	 * The Bourne shell can be used instead of init if we are
823 	 * trying to recover a really broken machine.
824 	 */
825 	if (execute_command) {
826 		run_init_process(execute_command);
827 		printk(KERN_WARNING "Failed to execute %s.  Attempting "
828 					"defaults...\n", execute_command);
829 	}
830 	run_init_process("/sbin/init");
831 	run_init_process("/etc/init");
832 	run_init_process("/bin/init");
833 	run_init_process("/bin/sh");
834 
835 	panic("No init found.  Try passing init= option to kernel. "
836 	      "See Linux Documentation/init.txt for guidance.");
837 }
838 
839 static int __init kernel_init(void * unused)
840 {
841 	/*
842 	 * Wait until kthreadd is all set-up.
843 	 */
844 	wait_for_completion(&kthreadd_done);
845 	/*
846 	 * init can allocate pages on any node
847 	 */
848 	set_mems_allowed(node_states[N_HIGH_MEMORY]);
849 	/*
850 	 * init can run on any cpu.
851 	 */
852 	set_cpus_allowed_ptr(current, cpu_all_mask);
853 
854 	cad_pid = task_pid(current);
855 
856 	smp_prepare_cpus(setup_max_cpus);
857 
858 	do_pre_smp_initcalls();
859 	lockup_detector_init();
860 
861 	smp_init();
862 	sched_init_smp();
863 
864 	do_basic_setup();
865 
866 	/* Open the /dev/console on the rootfs, this should never fail */
867 	if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
868 		printk(KERN_WARNING "Warning: unable to open an initial console.\n");
869 
870 	(void) sys_dup(0);
871 	(void) sys_dup(0);
872 	/*
873 	 * check if there is an early userspace init.  If yes, let it do all
874 	 * the work
875 	 */
876 
877 	if (!ramdisk_execute_command)
878 		ramdisk_execute_command = "/init";
879 
880 	if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
881 		ramdisk_execute_command = NULL;
882 		prepare_namespace();
883 	}
884 
885 	/*
886 	 * Ok, we have completed the initial bootup, and
887 	 * we're essentially up and running. Get rid of the
888 	 * initmem segments and start the user-mode stuff..
889 	 */
890 
891 	init_post();
892 	return 0;
893 }
894