xref: /openbmc/linux/init/main.c (revision f7777dcc)
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 #define DEBUG		/* Enable initcall_debug */
13 
14 #include <linux/types.h>
15 #include <linux/module.h>
16 #include <linux/proc_fs.h>
17 #include <linux/kernel.h>
18 #include <linux/syscalls.h>
19 #include <linux/stackprotector.h>
20 #include <linux/string.h>
21 #include <linux/ctype.h>
22 #include <linux/delay.h>
23 #include <linux/ioport.h>
24 #include <linux/init.h>
25 #include <linux/initrd.h>
26 #include <linux/bootmem.h>
27 #include <linux/acpi.h>
28 #include <linux/tty.h>
29 #include <linux/percpu.h>
30 #include <linux/kmod.h>
31 #include <linux/vmalloc.h>
32 #include <linux/kernel_stat.h>
33 #include <linux/start_kernel.h>
34 #include <linux/security.h>
35 #include <linux/smp.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>
43 #include <linux/cgroup.h>
44 #include <linux/efi.h>
45 #include <linux/tick.h>
46 #include <linux/interrupt.h>
47 #include <linux/taskstats_kern.h>
48 #include <linux/delayacct.h>
49 #include <linux/unistd.h>
50 #include <linux/rmap.h>
51 #include <linux/mempolicy.h>
52 #include <linux/key.h>
53 #include <linux/buffer_head.h>
54 #include <linux/page_cgroup.h>
55 #include <linux/debug_locks.h>
56 #include <linux/debugobjects.h>
57 #include <linux/lockdep.h>
58 #include <linux/kmemleak.h>
59 #include <linux/pid_namespace.h>
60 #include <linux/device.h>
61 #include <linux/kthread.h>
62 #include <linux/sched.h>
63 #include <linux/signal.h>
64 #include <linux/idr.h>
65 #include <linux/kgdb.h>
66 #include <linux/ftrace.h>
67 #include <linux/async.h>
68 #include <linux/kmemcheck.h>
69 #include <linux/sfi.h>
70 #include <linux/shmem_fs.h>
71 #include <linux/slab.h>
72 #include <linux/perf_event.h>
73 #include <linux/file.h>
74 #include <linux/ptrace.h>
75 #include <linux/blkdev.h>
76 #include <linux/elevator.h>
77 #include <linux/sched_clock.h>
78 #include <linux/context_tracking.h>
79 #include <linux/random.h>
80 
81 #include <asm/io.h>
82 #include <asm/bugs.h>
83 #include <asm/setup.h>
84 #include <asm/sections.h>
85 #include <asm/cacheflush.h>
86 
87 #ifdef CONFIG_X86_LOCAL_APIC
88 #include <asm/smp.h>
89 #endif
90 
91 static int kernel_init(void *);
92 
93 extern void init_IRQ(void);
94 extern void fork_init(unsigned long);
95 extern void mca_init(void);
96 extern void sbus_init(void);
97 extern void radix_tree_init(void);
98 #ifndef CONFIG_DEBUG_RODATA
99 static inline void mark_rodata_ro(void) { }
100 #endif
101 
102 #ifdef CONFIG_TC
103 extern void tc_init(void);
104 #endif
105 
106 /*
107  * Debug helper: via this flag we know that we are in 'early bootup code'
108  * where only the boot processor is running with IRQ disabled.  This means
109  * two things - IRQ must not be enabled before the flag is cleared and some
110  * operations which are not allowed with IRQ disabled are allowed while the
111  * flag is set.
112  */
113 bool early_boot_irqs_disabled __read_mostly;
114 
115 enum system_states system_state __read_mostly;
116 EXPORT_SYMBOL(system_state);
117 
118 /*
119  * Boot command-line arguments
120  */
121 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
122 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
123 
124 extern void time_init(void);
125 /* Default late time init is NULL. archs can override this later. */
126 void (*__initdata late_time_init)(void);
127 extern void softirq_init(void);
128 
129 /* Untouched command line saved by arch-specific code. */
130 char __initdata boot_command_line[COMMAND_LINE_SIZE];
131 /* Untouched saved command line (eg. for /proc) */
132 char *saved_command_line;
133 /* Command line for parameter parsing */
134 static char *static_command_line;
135 
136 static char *execute_command;
137 static char *ramdisk_execute_command;
138 
139 /*
140  * If set, this is an indication to the drivers that reset the underlying
141  * device before going ahead with the initialization otherwise driver might
142  * rely on the BIOS and skip the reset operation.
143  *
144  * This is useful if kernel is booting in an unreliable environment.
145  * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
146  * skipped and devices will be in unknown state.
147  */
148 unsigned int reset_devices;
149 EXPORT_SYMBOL(reset_devices);
150 
151 static int __init set_reset_devices(char *str)
152 {
153 	reset_devices = 1;
154 	return 1;
155 }
156 
157 __setup("reset_devices", set_reset_devices);
158 
159 static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
160 const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
161 static const char *panic_later, *panic_param;
162 
163 extern const struct obs_kernel_param __setup_start[], __setup_end[];
164 
165 static int __init obsolete_checksetup(char *line)
166 {
167 	const struct obs_kernel_param *p;
168 	int had_early_param = 0;
169 
170 	p = __setup_start;
171 	do {
172 		int n = strlen(p->str);
173 		if (parameqn(line, p->str, n)) {
174 			if (p->early) {
175 				/* Already done in parse_early_param?
176 				 * (Needs exact match on param part).
177 				 * Keep iterating, as we can have early
178 				 * params and __setups of same names 8( */
179 				if (line[n] == '\0' || line[n] == '=')
180 					had_early_param = 1;
181 			} else if (!p->setup_func) {
182 				pr_warn("Parameter %s is obsolete, ignored\n",
183 					p->str);
184 				return 1;
185 			} else if (p->setup_func(line + n))
186 				return 1;
187 		}
188 		p++;
189 	} while (p < __setup_end);
190 
191 	return had_early_param;
192 }
193 
194 /*
195  * This should be approx 2 Bo*oMips to start (note initial shift), and will
196  * still work even if initially too large, it will just take slightly longer
197  */
198 unsigned long loops_per_jiffy = (1<<12);
199 
200 EXPORT_SYMBOL(loops_per_jiffy);
201 
202 static int __init debug_kernel(char *str)
203 {
204 	console_loglevel = 10;
205 	return 0;
206 }
207 
208 static int __init quiet_kernel(char *str)
209 {
210 	console_loglevel = 4;
211 	return 0;
212 }
213 
214 early_param("debug", debug_kernel);
215 early_param("quiet", quiet_kernel);
216 
217 static int __init loglevel(char *str)
218 {
219 	int newlevel;
220 
221 	/*
222 	 * Only update loglevel value when a correct setting was passed,
223 	 * to prevent blind crashes (when loglevel being set to 0) that
224 	 * are quite hard to debug
225 	 */
226 	if (get_option(&str, &newlevel)) {
227 		console_loglevel = newlevel;
228 		return 0;
229 	}
230 
231 	return -EINVAL;
232 }
233 
234 early_param("loglevel", loglevel);
235 
236 /* Change NUL term back to "=", to make "param" the whole string. */
237 static int __init repair_env_string(char *param, char *val, const char *unused)
238 {
239 	if (val) {
240 		/* param=val or param="val"? */
241 		if (val == param+strlen(param)+1)
242 			val[-1] = '=';
243 		else if (val == param+strlen(param)+2) {
244 			val[-2] = '=';
245 			memmove(val-1, val, strlen(val)+1);
246 			val--;
247 		} else
248 			BUG();
249 	}
250 	return 0;
251 }
252 
253 /*
254  * Unknown boot options get handed to init, unless they look like
255  * unused parameters (modprobe will find them in /proc/cmdline).
256  */
257 static int __init unknown_bootoption(char *param, char *val, const char *unused)
258 {
259 	repair_env_string(param, val, unused);
260 
261 	/* Handle obsolete-style parameters */
262 	if (obsolete_checksetup(param))
263 		return 0;
264 
265 	/* Unused module parameter. */
266 	if (strchr(param, '.') && (!val || strchr(param, '.') < val))
267 		return 0;
268 
269 	if (panic_later)
270 		return 0;
271 
272 	if (val) {
273 		/* Environment option */
274 		unsigned int i;
275 		for (i = 0; envp_init[i]; i++) {
276 			if (i == MAX_INIT_ENVS) {
277 				panic_later = "Too many boot env vars at `%s'";
278 				panic_param = param;
279 			}
280 			if (!strncmp(param, envp_init[i], val - param))
281 				break;
282 		}
283 		envp_init[i] = param;
284 	} else {
285 		/* Command line option */
286 		unsigned int i;
287 		for (i = 0; argv_init[i]; i++) {
288 			if (i == MAX_INIT_ARGS) {
289 				panic_later = "Too many boot init vars at `%s'";
290 				panic_param = param;
291 			}
292 		}
293 		argv_init[i] = param;
294 	}
295 	return 0;
296 }
297 
298 static int __init init_setup(char *str)
299 {
300 	unsigned int i;
301 
302 	execute_command = str;
303 	/*
304 	 * In case LILO is going to boot us with default command line,
305 	 * it prepends "auto" before the whole cmdline which makes
306 	 * the shell think it should execute a script with such name.
307 	 * So we ignore all arguments entered _before_ init=... [MJ]
308 	 */
309 	for (i = 1; i < MAX_INIT_ARGS; i++)
310 		argv_init[i] = NULL;
311 	return 1;
312 }
313 __setup("init=", init_setup);
314 
315 static int __init rdinit_setup(char *str)
316 {
317 	unsigned int i;
318 
319 	ramdisk_execute_command = str;
320 	/* See "auto" comment in init_setup */
321 	for (i = 1; i < MAX_INIT_ARGS; i++)
322 		argv_init[i] = NULL;
323 	return 1;
324 }
325 __setup("rdinit=", rdinit_setup);
326 
327 #ifndef CONFIG_SMP
328 static const unsigned int setup_max_cpus = NR_CPUS;
329 #ifdef CONFIG_X86_LOCAL_APIC
330 static void __init smp_init(void)
331 {
332 	APIC_init_uniprocessor();
333 }
334 #else
335 #define smp_init()	do { } while (0)
336 #endif
337 
338 static inline void setup_nr_cpu_ids(void) { }
339 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
340 #endif
341 
342 /*
343  * We need to store the untouched command line for future reference.
344  * We also need to store the touched command line since the parameter
345  * parsing is performed in place, and we should allow a component to
346  * store reference of name/value for future reference.
347  */
348 static void __init setup_command_line(char *command_line)
349 {
350 	saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
351 	static_command_line = alloc_bootmem(strlen (command_line)+1);
352 	strcpy (saved_command_line, boot_command_line);
353 	strcpy (static_command_line, command_line);
354 }
355 
356 /*
357  * We need to finalize in a non-__init function or else race conditions
358  * between the root thread and the init thread may cause start_kernel to
359  * be reaped by free_initmem before the root thread has proceeded to
360  * cpu_idle.
361  *
362  * gcc-3.4 accidentally inlines this function, so use noinline.
363  */
364 
365 static __initdata DECLARE_COMPLETION(kthreadd_done);
366 
367 static noinline void __init_refok rest_init(void)
368 {
369 	int pid;
370 
371 	rcu_scheduler_starting();
372 	/*
373 	 * We need to spawn init first so that it obtains pid 1, however
374 	 * the init task will end up wanting to create kthreads, which, if
375 	 * we schedule it before we create kthreadd, will OOPS.
376 	 */
377 	kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
378 	numa_default_policy();
379 	pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
380 	rcu_read_lock();
381 	kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
382 	rcu_read_unlock();
383 	complete(&kthreadd_done);
384 
385 	/*
386 	 * The boot idle thread must execute schedule()
387 	 * at least once to get things moving:
388 	 */
389 	init_idle_bootup_task(current);
390 	schedule_preempt_disabled();
391 	/* Call into cpu_idle with preempt disabled */
392 	cpu_startup_entry(CPUHP_ONLINE);
393 }
394 
395 /* Check for early params. */
396 static int __init do_early_param(char *param, char *val, const char *unused)
397 {
398 	const struct obs_kernel_param *p;
399 
400 	for (p = __setup_start; p < __setup_end; p++) {
401 		if ((p->early && parameq(param, p->str)) ||
402 		    (strcmp(param, "console") == 0 &&
403 		     strcmp(p->str, "earlycon") == 0)
404 		) {
405 			if (p->setup_func(val) != 0)
406 				pr_warn("Malformed early option '%s'\n", param);
407 		}
408 	}
409 	/* We accept everything at this stage. */
410 	return 0;
411 }
412 
413 void __init parse_early_options(char *cmdline)
414 {
415 	parse_args("early options", cmdline, NULL, 0, 0, 0, do_early_param);
416 }
417 
418 /* Arch code calls this early on, or if not, just before other parsing. */
419 void __init parse_early_param(void)
420 {
421 	static __initdata int done = 0;
422 	static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
423 
424 	if (done)
425 		return;
426 
427 	/* All fall through to do_early_param. */
428 	strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
429 	parse_early_options(tmp_cmdline);
430 	done = 1;
431 }
432 
433 /*
434  *	Activate the first processor.
435  */
436 
437 static void __init boot_cpu_init(void)
438 {
439 	int cpu = smp_processor_id();
440 	/* Mark the boot cpu "present", "online" etc for SMP and UP case */
441 	set_cpu_online(cpu, true);
442 	set_cpu_active(cpu, true);
443 	set_cpu_present(cpu, true);
444 	set_cpu_possible(cpu, true);
445 }
446 
447 void __init __weak smp_setup_processor_id(void)
448 {
449 }
450 
451 # if THREAD_SIZE >= PAGE_SIZE
452 void __init __weak thread_info_cache_init(void)
453 {
454 }
455 #endif
456 
457 /*
458  * Set up kernel memory allocators
459  */
460 static void __init mm_init(void)
461 {
462 	/*
463 	 * page_cgroup requires contiguous pages,
464 	 * bigger than MAX_ORDER unless SPARSEMEM.
465 	 */
466 	page_cgroup_init_flatmem();
467 	mem_init();
468 	kmem_cache_init();
469 	percpu_init_late();
470 	pgtable_cache_init();
471 	vmalloc_init();
472 }
473 
474 asmlinkage void __init start_kernel(void)
475 {
476 	char * command_line;
477 	extern const struct kernel_param __start___param[], __stop___param[];
478 
479 	/*
480 	 * Need to run as early as possible, to initialize the
481 	 * lockdep hash:
482 	 */
483 	lockdep_init();
484 	smp_setup_processor_id();
485 	debug_objects_early_init();
486 
487 	/*
488 	 * Set up the the initial canary ASAP:
489 	 */
490 	boot_init_stack_canary();
491 
492 	cgroup_init_early();
493 
494 	local_irq_disable();
495 	early_boot_irqs_disabled = true;
496 
497 /*
498  * Interrupts are still disabled. Do necessary setups, then
499  * enable them
500  */
501 	boot_cpu_init();
502 	page_address_init();
503 	pr_notice("%s", linux_banner);
504 	setup_arch(&command_line);
505 	mm_init_owner(&init_mm, &init_task);
506 	mm_init_cpumask(&init_mm);
507 	setup_command_line(command_line);
508 	setup_nr_cpu_ids();
509 	setup_per_cpu_areas();
510 	smp_prepare_boot_cpu();	/* arch-specific boot-cpu hooks */
511 
512 	build_all_zonelists(NULL, NULL);
513 	page_alloc_init();
514 
515 	pr_notice("Kernel command line: %s\n", boot_command_line);
516 	parse_early_param();
517 	parse_args("Booting kernel", static_command_line, __start___param,
518 		   __stop___param - __start___param,
519 		   -1, -1, &unknown_bootoption);
520 
521 	jump_label_init();
522 
523 	/*
524 	 * These use large bootmem allocations and must precede
525 	 * kmem_cache_init()
526 	 */
527 	setup_log_buf(0);
528 	pidhash_init();
529 	vfs_caches_init_early();
530 	sort_main_extable();
531 	trap_init();
532 	mm_init();
533 
534 	/*
535 	 * Set up the scheduler prior starting any interrupts (such as the
536 	 * timer interrupt). Full topology setup happens at smp_init()
537 	 * time - but meanwhile we still have a functioning scheduler.
538 	 */
539 	sched_init();
540 	/*
541 	 * Disable preemption - early bootup scheduling is extremely
542 	 * fragile until we cpu_idle() for the first time.
543 	 */
544 	preempt_disable();
545 	if (WARN(!irqs_disabled(), "Interrupts were enabled *very* early, fixing it\n"))
546 		local_irq_disable();
547 	idr_init_cache();
548 	rcu_init();
549 	tick_nohz_init();
550 	context_tracking_init();
551 	radix_tree_init();
552 	/* init some links before init_ISA_irqs() */
553 	early_irq_init();
554 	init_IRQ();
555 	tick_init();
556 	init_timers();
557 	hrtimers_init();
558 	softirq_init();
559 	timekeeping_init();
560 	time_init();
561 	sched_clock_postinit();
562 	perf_event_init();
563 	profile_init();
564 	call_function_init();
565 	WARN(!irqs_disabled(), "Interrupts were enabled early\n");
566 	early_boot_irqs_disabled = false;
567 	local_irq_enable();
568 
569 	kmem_cache_init_late();
570 
571 	/*
572 	 * HACK ALERT! This is early. We're enabling the console before
573 	 * we've done PCI setups etc, and console_init() must be aware of
574 	 * this. But we do want output early, in case something goes wrong.
575 	 */
576 	console_init();
577 	if (panic_later)
578 		panic(panic_later, panic_param);
579 
580 	lockdep_info();
581 
582 	/*
583 	 * Need to run this when irqs are enabled, because it wants
584 	 * to self-test [hard/soft]-irqs on/off lock inversion bugs
585 	 * too:
586 	 */
587 	locking_selftest();
588 
589 #ifdef CONFIG_BLK_DEV_INITRD
590 	if (initrd_start && !initrd_below_start_ok &&
591 	    page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
592 		pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
593 		    page_to_pfn(virt_to_page((void *)initrd_start)),
594 		    min_low_pfn);
595 		initrd_start = 0;
596 	}
597 #endif
598 	page_cgroup_init();
599 	debug_objects_mem_init();
600 	kmemleak_init();
601 	setup_per_cpu_pageset();
602 	numa_policy_init();
603 	if (late_time_init)
604 		late_time_init();
605 	sched_clock_init();
606 	calibrate_delay();
607 	pidmap_init();
608 	anon_vma_init();
609 #ifdef CONFIG_X86
610 	if (efi_enabled(EFI_RUNTIME_SERVICES))
611 		efi_enter_virtual_mode();
612 #endif
613 	thread_info_cache_init();
614 	cred_init();
615 	fork_init(totalram_pages);
616 	proc_caches_init();
617 	buffer_init();
618 	key_init();
619 	security_init();
620 	dbg_late_init();
621 	vfs_caches_init(totalram_pages);
622 	signals_init();
623 	/* rootfs populating might need page-writeback */
624 	page_writeback_init();
625 #ifdef CONFIG_PROC_FS
626 	proc_root_init();
627 #endif
628 	cgroup_init();
629 	cpuset_init();
630 	taskstats_init_early();
631 	delayacct_init();
632 
633 	check_bugs();
634 
635 	acpi_early_init(); /* before LAPIC and SMP init */
636 	sfi_init_late();
637 
638 	if (efi_enabled(EFI_RUNTIME_SERVICES)) {
639 		efi_late_init();
640 		efi_free_boot_services();
641 	}
642 
643 	ftrace_init();
644 
645 	/* Do the rest non-__init'ed, we're now alive */
646 	rest_init();
647 }
648 
649 /* Call all constructor functions linked into the kernel. */
650 static void __init do_ctors(void)
651 {
652 #ifdef CONFIG_CONSTRUCTORS
653 	ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
654 
655 	for (; fn < (ctor_fn_t *) __ctors_end; fn++)
656 		(*fn)();
657 #endif
658 }
659 
660 bool initcall_debug;
661 core_param(initcall_debug, initcall_debug, bool, 0644);
662 
663 static int __init_or_module do_one_initcall_debug(initcall_t fn)
664 {
665 	ktime_t calltime, delta, rettime;
666 	unsigned long long duration;
667 	int ret;
668 
669 	pr_debug("calling  %pF @ %i\n", fn, task_pid_nr(current));
670 	calltime = ktime_get();
671 	ret = fn();
672 	rettime = ktime_get();
673 	delta = ktime_sub(rettime, calltime);
674 	duration = (unsigned long long) ktime_to_ns(delta) >> 10;
675 	pr_debug("initcall %pF returned %d after %lld usecs\n",
676 		 fn, ret, duration);
677 
678 	return ret;
679 }
680 
681 int __init_or_module do_one_initcall(initcall_t fn)
682 {
683 	int count = preempt_count();
684 	int ret;
685 	char msgbuf[64];
686 
687 	if (initcall_debug)
688 		ret = do_one_initcall_debug(fn);
689 	else
690 		ret = fn();
691 
692 	msgbuf[0] = 0;
693 
694 	if (preempt_count() != count) {
695 		sprintf(msgbuf, "preemption imbalance ");
696 		preempt_count() = count;
697 	}
698 	if (irqs_disabled()) {
699 		strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
700 		local_irq_enable();
701 	}
702 	WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
703 
704 	return ret;
705 }
706 
707 
708 extern initcall_t __initcall_start[];
709 extern initcall_t __initcall0_start[];
710 extern initcall_t __initcall1_start[];
711 extern initcall_t __initcall2_start[];
712 extern initcall_t __initcall3_start[];
713 extern initcall_t __initcall4_start[];
714 extern initcall_t __initcall5_start[];
715 extern initcall_t __initcall6_start[];
716 extern initcall_t __initcall7_start[];
717 extern initcall_t __initcall_end[];
718 
719 static initcall_t *initcall_levels[] __initdata = {
720 	__initcall0_start,
721 	__initcall1_start,
722 	__initcall2_start,
723 	__initcall3_start,
724 	__initcall4_start,
725 	__initcall5_start,
726 	__initcall6_start,
727 	__initcall7_start,
728 	__initcall_end,
729 };
730 
731 /* Keep these in sync with initcalls in include/linux/init.h */
732 static char *initcall_level_names[] __initdata = {
733 	"early",
734 	"core",
735 	"postcore",
736 	"arch",
737 	"subsys",
738 	"fs",
739 	"device",
740 	"late",
741 };
742 
743 static void __init do_initcall_level(int level)
744 {
745 	extern const struct kernel_param __start___param[], __stop___param[];
746 	initcall_t *fn;
747 
748 	strcpy(static_command_line, saved_command_line);
749 	parse_args(initcall_level_names[level],
750 		   static_command_line, __start___param,
751 		   __stop___param - __start___param,
752 		   level, level,
753 		   &repair_env_string);
754 
755 	for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
756 		do_one_initcall(*fn);
757 }
758 
759 static void __init do_initcalls(void)
760 {
761 	int level;
762 
763 	for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
764 		do_initcall_level(level);
765 }
766 
767 /*
768  * Ok, the machine is now initialized. None of the devices
769  * have been touched yet, but the CPU subsystem is up and
770  * running, and memory and process management works.
771  *
772  * Now we can finally start doing some real work..
773  */
774 static void __init do_basic_setup(void)
775 {
776 	cpuset_init_smp();
777 	usermodehelper_init();
778 	shmem_init();
779 	driver_init();
780 	init_irq_proc();
781 	do_ctors();
782 	usermodehelper_enable();
783 	do_initcalls();
784 	random_int_secret_init();
785 }
786 
787 static void __init do_pre_smp_initcalls(void)
788 {
789 	initcall_t *fn;
790 
791 	for (fn = __initcall_start; fn < __initcall0_start; fn++)
792 		do_one_initcall(*fn);
793 }
794 
795 /*
796  * This function requests modules which should be loaded by default and is
797  * called twice right after initrd is mounted and right before init is
798  * exec'd.  If such modules are on either initrd or rootfs, they will be
799  * loaded before control is passed to userland.
800  */
801 void __init load_default_modules(void)
802 {
803 	load_default_elevator_module();
804 }
805 
806 static int run_init_process(const char *init_filename)
807 {
808 	argv_init[0] = init_filename;
809 	return do_execve(init_filename,
810 		(const char __user *const __user *)argv_init,
811 		(const char __user *const __user *)envp_init);
812 }
813 
814 static noinline void __init kernel_init_freeable(void);
815 
816 static int __ref kernel_init(void *unused)
817 {
818 	kernel_init_freeable();
819 	/* need to finish all async __init code before freeing the memory */
820 	async_synchronize_full();
821 	free_initmem();
822 	mark_rodata_ro();
823 	system_state = SYSTEM_RUNNING;
824 	numa_default_policy();
825 
826 	flush_delayed_fput();
827 
828 	if (ramdisk_execute_command) {
829 		if (!run_init_process(ramdisk_execute_command))
830 			return 0;
831 		pr_err("Failed to execute %s\n", ramdisk_execute_command);
832 	}
833 
834 	/*
835 	 * We try each of these until one succeeds.
836 	 *
837 	 * The Bourne shell can be used instead of init if we are
838 	 * trying to recover a really broken machine.
839 	 */
840 	if (execute_command) {
841 		if (!run_init_process(execute_command))
842 			return 0;
843 		pr_err("Failed to execute %s.  Attempting defaults...\n",
844 			execute_command);
845 	}
846 	if (!run_init_process("/sbin/init") ||
847 	    !run_init_process("/etc/init") ||
848 	    !run_init_process("/bin/init") ||
849 	    !run_init_process("/bin/sh"))
850 		return 0;
851 
852 	panic("No init found.  Try passing init= option to kernel. "
853 	      "See Linux Documentation/init.txt for guidance.");
854 }
855 
856 static noinline void __init kernel_init_freeable(void)
857 {
858 	/*
859 	 * Wait until kthreadd is all set-up.
860 	 */
861 	wait_for_completion(&kthreadd_done);
862 
863 	/* Now the scheduler is fully set up and can do blocking allocations */
864 	gfp_allowed_mask = __GFP_BITS_MASK;
865 
866 	/*
867 	 * init can allocate pages on any node
868 	 */
869 	set_mems_allowed(node_states[N_MEMORY]);
870 	/*
871 	 * init can run on any cpu.
872 	 */
873 	set_cpus_allowed_ptr(current, cpu_all_mask);
874 
875 	cad_pid = task_pid(current);
876 
877 	smp_prepare_cpus(setup_max_cpus);
878 
879 	do_pre_smp_initcalls();
880 	lockup_detector_init();
881 
882 	smp_init();
883 	sched_init_smp();
884 
885 	do_basic_setup();
886 
887 	/* Open the /dev/console on the rootfs, this should never fail */
888 	if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
889 		pr_err("Warning: unable to open an initial console.\n");
890 
891 	(void) sys_dup(0);
892 	(void) sys_dup(0);
893 	/*
894 	 * check if there is an early userspace init.  If yes, let it do all
895 	 * the work
896 	 */
897 
898 	if (!ramdisk_execute_command)
899 		ramdisk_execute_command = "/init";
900 
901 	if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
902 		ramdisk_execute_command = NULL;
903 		prepare_namespace();
904 	}
905 
906 	/*
907 	 * Ok, we have completed the initial bootup, and
908 	 * we're essentially up and running. Get rid of the
909 	 * initmem segments and start the user-mode stuff..
910 	 */
911 
912 	/* rootfs is available now, try loading default modules */
913 	load_default_modules();
914 }
915