xref: /openbmc/linux/init/main.c (revision 5bd8e16d)
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 
80 #include <asm/io.h>
81 #include <asm/bugs.h>
82 #include <asm/setup.h>
83 #include <asm/sections.h>
84 #include <asm/cacheflush.h>
85 
86 #ifdef CONFIG_X86_LOCAL_APIC
87 #include <asm/smp.h>
88 #endif
89 
90 static int kernel_init(void *);
91 
92 extern void init_IRQ(void);
93 extern void fork_init(unsigned long);
94 extern void mca_init(void);
95 extern void sbus_init(void);
96 extern void radix_tree_init(void);
97 #ifndef CONFIG_DEBUG_RODATA
98 static inline void mark_rodata_ro(void) { }
99 #endif
100 
101 #ifdef CONFIG_TC
102 extern void tc_init(void);
103 #endif
104 
105 /*
106  * Debug helper: via this flag we know that we are in 'early bootup code'
107  * where only the boot processor is running with IRQ disabled.  This means
108  * two things - IRQ must not be enabled before the flag is cleared and some
109  * operations which are not allowed with IRQ disabled are allowed while the
110  * flag is set.
111  */
112 bool early_boot_irqs_disabled __read_mostly;
113 
114 enum system_states system_state __read_mostly;
115 EXPORT_SYMBOL(system_state);
116 
117 /*
118  * Boot command-line arguments
119  */
120 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
121 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
122 
123 extern void time_init(void);
124 /* Default late time init is NULL. archs can override this later. */
125 void (*__initdata late_time_init)(void);
126 extern void softirq_init(void);
127 
128 /* Untouched command line saved by arch-specific code. */
129 char __initdata boot_command_line[COMMAND_LINE_SIZE];
130 /* Untouched saved command line (eg. for /proc) */
131 char *saved_command_line;
132 /* Command line for parameter parsing */
133 static char *static_command_line;
134 
135 static char *execute_command;
136 static char *ramdisk_execute_command;
137 
138 /*
139  * If set, this is an indication to the drivers that reset the underlying
140  * device before going ahead with the initialization otherwise driver might
141  * rely on the BIOS and skip the reset operation.
142  *
143  * This is useful if kernel is booting in an unreliable environment.
144  * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
145  * skipped and devices will be in unknown state.
146  */
147 unsigned int reset_devices;
148 EXPORT_SYMBOL(reset_devices);
149 
150 static int __init set_reset_devices(char *str)
151 {
152 	reset_devices = 1;
153 	return 1;
154 }
155 
156 __setup("reset_devices", set_reset_devices);
157 
158 static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
159 const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
160 static const char *panic_later, *panic_param;
161 
162 extern const struct obs_kernel_param __setup_start[], __setup_end[];
163 
164 static int __init obsolete_checksetup(char *line)
165 {
166 	const struct obs_kernel_param *p;
167 	int had_early_param = 0;
168 
169 	p = __setup_start;
170 	do {
171 		int n = strlen(p->str);
172 		if (parameqn(line, p->str, n)) {
173 			if (p->early) {
174 				/* Already done in parse_early_param?
175 				 * (Needs exact match on param part).
176 				 * Keep iterating, as we can have early
177 				 * params and __setups of same names 8( */
178 				if (line[n] == '\0' || line[n] == '=')
179 					had_early_param = 1;
180 			} else if (!p->setup_func) {
181 				pr_warn("Parameter %s is obsolete, ignored\n",
182 					p->str);
183 				return 1;
184 			} else if (p->setup_func(line + n))
185 				return 1;
186 		}
187 		p++;
188 	} while (p < __setup_end);
189 
190 	return had_early_param;
191 }
192 
193 /*
194  * This should be approx 2 Bo*oMips to start (note initial shift), and will
195  * still work even if initially too large, it will just take slightly longer
196  */
197 unsigned long loops_per_jiffy = (1<<12);
198 
199 EXPORT_SYMBOL(loops_per_jiffy);
200 
201 static int __init debug_kernel(char *str)
202 {
203 	console_loglevel = 10;
204 	return 0;
205 }
206 
207 static int __init quiet_kernel(char *str)
208 {
209 	console_loglevel = 4;
210 	return 0;
211 }
212 
213 early_param("debug", debug_kernel);
214 early_param("quiet", quiet_kernel);
215 
216 static int __init loglevel(char *str)
217 {
218 	int newlevel;
219 
220 	/*
221 	 * Only update loglevel value when a correct setting was passed,
222 	 * to prevent blind crashes (when loglevel being set to 0) that
223 	 * are quite hard to debug
224 	 */
225 	if (get_option(&str, &newlevel)) {
226 		console_loglevel = newlevel;
227 		return 0;
228 	}
229 
230 	return -EINVAL;
231 }
232 
233 early_param("loglevel", loglevel);
234 
235 /* Change NUL term back to "=", to make "param" the whole string. */
236 static int __init repair_env_string(char *param, char *val, const char *unused)
237 {
238 	if (val) {
239 		/* param=val or param="val"? */
240 		if (val == param+strlen(param)+1)
241 			val[-1] = '=';
242 		else if (val == param+strlen(param)+2) {
243 			val[-2] = '=';
244 			memmove(val-1, val, strlen(val)+1);
245 			val--;
246 		} else
247 			BUG();
248 	}
249 	return 0;
250 }
251 
252 /*
253  * Unknown boot options get handed to init, unless they look like
254  * unused parameters (modprobe will find them in /proc/cmdline).
255  */
256 static int __init unknown_bootoption(char *param, char *val, const char *unused)
257 {
258 	repair_env_string(param, val, unused);
259 
260 	/* Handle obsolete-style parameters */
261 	if (obsolete_checksetup(param))
262 		return 0;
263 
264 	/* Unused module parameter. */
265 	if (strchr(param, '.') && (!val || strchr(param, '.') < val))
266 		return 0;
267 
268 	if (panic_later)
269 		return 0;
270 
271 	if (val) {
272 		/* Environment option */
273 		unsigned int i;
274 		for (i = 0; envp_init[i]; i++) {
275 			if (i == MAX_INIT_ENVS) {
276 				panic_later = "Too many boot env vars at `%s'";
277 				panic_param = param;
278 			}
279 			if (!strncmp(param, envp_init[i], val - param))
280 				break;
281 		}
282 		envp_init[i] = param;
283 	} else {
284 		/* Command line option */
285 		unsigned int i;
286 		for (i = 0; argv_init[i]; i++) {
287 			if (i == MAX_INIT_ARGS) {
288 				panic_later = "Too many boot init vars at `%s'";
289 				panic_param = param;
290 			}
291 		}
292 		argv_init[i] = param;
293 	}
294 	return 0;
295 }
296 
297 static int __init init_setup(char *str)
298 {
299 	unsigned int i;
300 
301 	execute_command = str;
302 	/*
303 	 * In case LILO is going to boot us with default command line,
304 	 * it prepends "auto" before the whole cmdline which makes
305 	 * the shell think it should execute a script with such name.
306 	 * So we ignore all arguments entered _before_ init=... [MJ]
307 	 */
308 	for (i = 1; i < MAX_INIT_ARGS; i++)
309 		argv_init[i] = NULL;
310 	return 1;
311 }
312 __setup("init=", init_setup);
313 
314 static int __init rdinit_setup(char *str)
315 {
316 	unsigned int i;
317 
318 	ramdisk_execute_command = str;
319 	/* See "auto" comment in init_setup */
320 	for (i = 1; i < MAX_INIT_ARGS; i++)
321 		argv_init[i] = NULL;
322 	return 1;
323 }
324 __setup("rdinit=", rdinit_setup);
325 
326 #ifndef CONFIG_SMP
327 static const unsigned int setup_max_cpus = NR_CPUS;
328 #ifdef CONFIG_X86_LOCAL_APIC
329 static void __init smp_init(void)
330 {
331 	APIC_init_uniprocessor();
332 }
333 #else
334 #define smp_init()	do { } while (0)
335 #endif
336 
337 static inline void setup_nr_cpu_ids(void) { }
338 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
339 #endif
340 
341 /*
342  * We need to store the untouched command line for future reference.
343  * We also need to store the touched command line since the parameter
344  * parsing is performed in place, and we should allow a component to
345  * store reference of name/value for future reference.
346  */
347 static void __init setup_command_line(char *command_line)
348 {
349 	saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
350 	static_command_line = alloc_bootmem(strlen (command_line)+1);
351 	strcpy (saved_command_line, boot_command_line);
352 	strcpy (static_command_line, command_line);
353 }
354 
355 /*
356  * We need to finalize in a non-__init function or else race conditions
357  * between the root thread and the init thread may cause start_kernel to
358  * be reaped by free_initmem before the root thread has proceeded to
359  * cpu_idle.
360  *
361  * gcc-3.4 accidentally inlines this function, so use noinline.
362  */
363 
364 static __initdata DECLARE_COMPLETION(kthreadd_done);
365 
366 static noinline void __init_refok rest_init(void)
367 {
368 	int pid;
369 
370 	rcu_scheduler_starting();
371 	/*
372 	 * We need to spawn init first so that it obtains pid 1, however
373 	 * the init task will end up wanting to create kthreads, which, if
374 	 * we schedule it before we create kthreadd, will OOPS.
375 	 */
376 	kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
377 	numa_default_policy();
378 	pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
379 	rcu_read_lock();
380 	kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
381 	rcu_read_unlock();
382 	complete(&kthreadd_done);
383 
384 	/*
385 	 * The boot idle thread must execute schedule()
386 	 * at least once to get things moving:
387 	 */
388 	init_idle_bootup_task(current);
389 	schedule_preempt_disabled();
390 	/* Call into cpu_idle with preempt disabled */
391 	cpu_startup_entry(CPUHP_ONLINE);
392 }
393 
394 /* Check for early params. */
395 static int __init do_early_param(char *param, char *val, const char *unused)
396 {
397 	const struct obs_kernel_param *p;
398 
399 	for (p = __setup_start; p < __setup_end; p++) {
400 		if ((p->early && parameq(param, p->str)) ||
401 		    (strcmp(param, "console") == 0 &&
402 		     strcmp(p->str, "earlycon") == 0)
403 		) {
404 			if (p->setup_func(val) != 0)
405 				pr_warn("Malformed early option '%s'\n", param);
406 		}
407 	}
408 	/* We accept everything at this stage. */
409 	return 0;
410 }
411 
412 void __init parse_early_options(char *cmdline)
413 {
414 	parse_args("early options", cmdline, NULL, 0, 0, 0, do_early_param);
415 }
416 
417 /* Arch code calls this early on, or if not, just before other parsing. */
418 void __init parse_early_param(void)
419 {
420 	static __initdata int done = 0;
421 	static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
422 
423 	if (done)
424 		return;
425 
426 	/* All fall through to do_early_param. */
427 	strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
428 	parse_early_options(tmp_cmdline);
429 	done = 1;
430 }
431 
432 /*
433  *	Activate the first processor.
434  */
435 
436 static void __init boot_cpu_init(void)
437 {
438 	int cpu = smp_processor_id();
439 	/* Mark the boot cpu "present", "online" etc for SMP and UP case */
440 	set_cpu_online(cpu, true);
441 	set_cpu_active(cpu, true);
442 	set_cpu_present(cpu, true);
443 	set_cpu_possible(cpu, true);
444 }
445 
446 void __init __weak smp_setup_processor_id(void)
447 {
448 }
449 
450 # if THREAD_SIZE >= PAGE_SIZE
451 void __init __weak thread_info_cache_init(void)
452 {
453 }
454 #endif
455 
456 /*
457  * Set up kernel memory allocators
458  */
459 static void __init mm_init(void)
460 {
461 	/*
462 	 * page_cgroup requires contiguous pages,
463 	 * bigger than MAX_ORDER unless SPARSEMEM.
464 	 */
465 	page_cgroup_init_flatmem();
466 	mem_init();
467 	kmem_cache_init();
468 	percpu_init_late();
469 	pgtable_cache_init();
470 	vmalloc_init();
471 }
472 
473 asmlinkage void __init start_kernel(void)
474 {
475 	char * command_line;
476 	extern const struct kernel_param __start___param[], __stop___param[];
477 
478 	/*
479 	 * Need to run as early as possible, to initialize the
480 	 * lockdep hash:
481 	 */
482 	lockdep_init();
483 	smp_setup_processor_id();
484 	debug_objects_early_init();
485 
486 	/*
487 	 * Set up the the initial canary ASAP:
488 	 */
489 	boot_init_stack_canary();
490 
491 	cgroup_init_early();
492 
493 	local_irq_disable();
494 	early_boot_irqs_disabled = true;
495 
496 /*
497  * Interrupts are still disabled. Do necessary setups, then
498  * enable them
499  */
500 	boot_cpu_init();
501 	page_address_init();
502 	pr_notice("%s", linux_banner);
503 	setup_arch(&command_line);
504 	mm_init_owner(&init_mm, &init_task);
505 	mm_init_cpumask(&init_mm);
506 	setup_command_line(command_line);
507 	setup_nr_cpu_ids();
508 	setup_per_cpu_areas();
509 	smp_prepare_boot_cpu();	/* arch-specific boot-cpu hooks */
510 
511 	build_all_zonelists(NULL, NULL);
512 	page_alloc_init();
513 
514 	pr_notice("Kernel command line: %s\n", boot_command_line);
515 	parse_early_param();
516 	parse_args("Booting kernel", static_command_line, __start___param,
517 		   __stop___param - __start___param,
518 		   -1, -1, &unknown_bootoption);
519 
520 	jump_label_init();
521 
522 	/*
523 	 * These use large bootmem allocations and must precede
524 	 * kmem_cache_init()
525 	 */
526 	setup_log_buf(0);
527 	pidhash_init();
528 	vfs_caches_init_early();
529 	sort_main_extable();
530 	trap_init();
531 	mm_init();
532 
533 	/*
534 	 * Set up the scheduler prior starting any interrupts (such as the
535 	 * timer interrupt). Full topology setup happens at smp_init()
536 	 * time - but meanwhile we still have a functioning scheduler.
537 	 */
538 	sched_init();
539 	/*
540 	 * Disable preemption - early bootup scheduling is extremely
541 	 * fragile until we cpu_idle() for the first time.
542 	 */
543 	preempt_disable();
544 	if (WARN(!irqs_disabled(), "Interrupts were enabled *very* early, fixing it\n"))
545 		local_irq_disable();
546 	idr_init_cache();
547 	rcu_init();
548 	tick_nohz_init();
549 	context_tracking_init();
550 	radix_tree_init();
551 	/* init some links before init_ISA_irqs() */
552 	early_irq_init();
553 	init_IRQ();
554 	tick_init();
555 	init_timers();
556 	hrtimers_init();
557 	softirq_init();
558 	timekeeping_init();
559 	time_init();
560 	sched_clock_postinit();
561 	perf_event_init();
562 	profile_init();
563 	call_function_init();
564 	WARN(!irqs_disabled(), "Interrupts were enabled early\n");
565 	early_boot_irqs_disabled = false;
566 	local_irq_enable();
567 
568 	kmem_cache_init_late();
569 
570 	/*
571 	 * HACK ALERT! This is early. We're enabling the console before
572 	 * we've done PCI setups etc, and console_init() must be aware of
573 	 * this. But we do want output early, in case something goes wrong.
574 	 */
575 	console_init();
576 	if (panic_later)
577 		panic(panic_later, panic_param);
578 
579 	lockdep_info();
580 
581 	/*
582 	 * Need to run this when irqs are enabled, because it wants
583 	 * to self-test [hard/soft]-irqs on/off lock inversion bugs
584 	 * too:
585 	 */
586 	locking_selftest();
587 
588 #ifdef CONFIG_BLK_DEV_INITRD
589 	if (initrd_start && !initrd_below_start_ok &&
590 	    page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
591 		pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
592 		    page_to_pfn(virt_to_page((void *)initrd_start)),
593 		    min_low_pfn);
594 		initrd_start = 0;
595 	}
596 #endif
597 	page_cgroup_init();
598 	debug_objects_mem_init();
599 	kmemleak_init();
600 	setup_per_cpu_pageset();
601 	numa_policy_init();
602 	if (late_time_init)
603 		late_time_init();
604 	sched_clock_init();
605 	calibrate_delay();
606 	pidmap_init();
607 	anon_vma_init();
608 #ifdef CONFIG_X86
609 	if (efi_enabled(EFI_RUNTIME_SERVICES))
610 		efi_enter_virtual_mode();
611 #endif
612 	thread_info_cache_init();
613 	cred_init();
614 	fork_init(totalram_pages);
615 	proc_caches_init();
616 	buffer_init();
617 	key_init();
618 	security_init();
619 	dbg_late_init();
620 	vfs_caches_init(totalram_pages);
621 	signals_init();
622 	/* rootfs populating might need page-writeback */
623 	page_writeback_init();
624 #ifdef CONFIG_PROC_FS
625 	proc_root_init();
626 #endif
627 	cgroup_init();
628 	cpuset_init();
629 	taskstats_init_early();
630 	delayacct_init();
631 
632 	check_bugs();
633 
634 	acpi_early_init(); /* before LAPIC and SMP init */
635 	sfi_init_late();
636 
637 	if (efi_enabled(EFI_RUNTIME_SERVICES)) {
638 		efi_late_init();
639 		efi_free_boot_services();
640 	}
641 
642 	ftrace_init();
643 
644 	/* Do the rest non-__init'ed, we're now alive */
645 	rest_init();
646 }
647 
648 /* Call all constructor functions linked into the kernel. */
649 static void __init do_ctors(void)
650 {
651 #ifdef CONFIG_CONSTRUCTORS
652 	ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
653 
654 	for (; fn < (ctor_fn_t *) __ctors_end; fn++)
655 		(*fn)();
656 #endif
657 }
658 
659 bool initcall_debug;
660 core_param(initcall_debug, initcall_debug, bool, 0644);
661 
662 static int __init_or_module do_one_initcall_debug(initcall_t fn)
663 {
664 	ktime_t calltime, delta, rettime;
665 	unsigned long long duration;
666 	int ret;
667 
668 	pr_debug("calling  %pF @ %i\n", fn, task_pid_nr(current));
669 	calltime = ktime_get();
670 	ret = fn();
671 	rettime = ktime_get();
672 	delta = ktime_sub(rettime, calltime);
673 	duration = (unsigned long long) ktime_to_ns(delta) >> 10;
674 	pr_debug("initcall %pF returned %d after %lld usecs\n",
675 		 fn, ret, duration);
676 
677 	return ret;
678 }
679 
680 int __init_or_module do_one_initcall(initcall_t fn)
681 {
682 	int count = preempt_count();
683 	int ret;
684 	char msgbuf[64];
685 
686 	if (initcall_debug)
687 		ret = do_one_initcall_debug(fn);
688 	else
689 		ret = fn();
690 
691 	msgbuf[0] = 0;
692 
693 	if (preempt_count() != count) {
694 		sprintf(msgbuf, "preemption imbalance ");
695 		preempt_count() = count;
696 	}
697 	if (irqs_disabled()) {
698 		strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
699 		local_irq_enable();
700 	}
701 	WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
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 /* Keep these in sync with initcalls in include/linux/init.h */
731 static char *initcall_level_names[] __initdata = {
732 	"early",
733 	"core",
734 	"postcore",
735 	"arch",
736 	"subsys",
737 	"fs",
738 	"device",
739 	"late",
740 };
741 
742 static void __init do_initcall_level(int level)
743 {
744 	extern const struct kernel_param __start___param[], __stop___param[];
745 	initcall_t *fn;
746 
747 	strcpy(static_command_line, saved_command_line);
748 	parse_args(initcall_level_names[level],
749 		   static_command_line, __start___param,
750 		   __stop___param - __start___param,
751 		   level, level,
752 		   &repair_env_string);
753 
754 	for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
755 		do_one_initcall(*fn);
756 }
757 
758 static void __init do_initcalls(void)
759 {
760 	int level;
761 
762 	for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
763 		do_initcall_level(level);
764 }
765 
766 /*
767  * Ok, the machine is now initialized. None of the devices
768  * have been touched yet, but the CPU subsystem is up and
769  * running, and memory and process management works.
770  *
771  * Now we can finally start doing some real work..
772  */
773 static void __init do_basic_setup(void)
774 {
775 	cpuset_init_smp();
776 	usermodehelper_init();
777 	shmem_init();
778 	driver_init();
779 	init_irq_proc();
780 	do_ctors();
781 	usermodehelper_enable();
782 	do_initcalls();
783 }
784 
785 static void __init do_pre_smp_initcalls(void)
786 {
787 	initcall_t *fn;
788 
789 	for (fn = __initcall_start; fn < __initcall0_start; fn++)
790 		do_one_initcall(*fn);
791 }
792 
793 /*
794  * This function requests modules which should be loaded by default and is
795  * called twice right after initrd is mounted and right before init is
796  * exec'd.  If such modules are on either initrd or rootfs, they will be
797  * loaded before control is passed to userland.
798  */
799 void __init load_default_modules(void)
800 {
801 	load_default_elevator_module();
802 }
803 
804 static int run_init_process(const char *init_filename)
805 {
806 	argv_init[0] = init_filename;
807 	return do_execve(init_filename,
808 		(const char __user *const __user *)argv_init,
809 		(const char __user *const __user *)envp_init);
810 }
811 
812 static noinline void __init kernel_init_freeable(void);
813 
814 static int __ref kernel_init(void *unused)
815 {
816 	kernel_init_freeable();
817 	/* need to finish all async __init code before freeing the memory */
818 	async_synchronize_full();
819 	free_initmem();
820 	mark_rodata_ro();
821 	system_state = SYSTEM_RUNNING;
822 	numa_default_policy();
823 
824 	flush_delayed_fput();
825 
826 	if (ramdisk_execute_command) {
827 		if (!run_init_process(ramdisk_execute_command))
828 			return 0;
829 		pr_err("Failed to execute %s\n", ramdisk_execute_command);
830 	}
831 
832 	/*
833 	 * We try each of these until one succeeds.
834 	 *
835 	 * The Bourne shell can be used instead of init if we are
836 	 * trying to recover a really broken machine.
837 	 */
838 	if (execute_command) {
839 		if (!run_init_process(execute_command))
840 			return 0;
841 		pr_err("Failed to execute %s.  Attempting defaults...\n",
842 			execute_command);
843 	}
844 	if (!run_init_process("/sbin/init") ||
845 	    !run_init_process("/etc/init") ||
846 	    !run_init_process("/bin/init") ||
847 	    !run_init_process("/bin/sh"))
848 		return 0;
849 
850 	panic("No init found.  Try passing init= option to kernel. "
851 	      "See Linux Documentation/init.txt for guidance.");
852 }
853 
854 static noinline void __init kernel_init_freeable(void)
855 {
856 	/*
857 	 * Wait until kthreadd is all set-up.
858 	 */
859 	wait_for_completion(&kthreadd_done);
860 
861 	/* Now the scheduler is fully set up and can do blocking allocations */
862 	gfp_allowed_mask = __GFP_BITS_MASK;
863 
864 	/*
865 	 * init can allocate pages on any node
866 	 */
867 	set_mems_allowed(node_states[N_MEMORY]);
868 	/*
869 	 * init can run on any cpu.
870 	 */
871 	set_cpus_allowed_ptr(current, cpu_all_mask);
872 
873 	cad_pid = task_pid(current);
874 
875 	smp_prepare_cpus(setup_max_cpus);
876 
877 	do_pre_smp_initcalls();
878 	lockup_detector_init();
879 
880 	smp_init();
881 	sched_init_smp();
882 
883 	do_basic_setup();
884 
885 	/* Open the /dev/console on the rootfs, this should never fail */
886 	if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
887 		pr_err("Warning: unable to open an initial console.\n");
888 
889 	(void) sys_dup(0);
890 	(void) sys_dup(0);
891 	/*
892 	 * check if there is an early userspace init.  If yes, let it do all
893 	 * the work
894 	 */
895 
896 	if (!ramdisk_execute_command)
897 		ramdisk_execute_command = "/init";
898 
899 	if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
900 		ramdisk_execute_command = NULL;
901 		prepare_namespace();
902 	}
903 
904 	/*
905 	 * Ok, we have completed the initial bootup, and
906 	 * we're essentially up and running. Get rid of the
907 	 * initmem segments and start the user-mode stuff..
908 	 */
909 
910 	/* rootfs is available now, try loading default modules */
911 	load_default_modules();
912 }
913