xref: /openbmc/linux/init/main.c (revision d5cb9783536a41df9f9cba5b0a1d78047ed787f7)
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 __KERNEL_SYSCALLS__
13 
14 #include <linux/config.h>
15 #include <linux/types.h>
16 #include <linux/module.h>
17 #include <linux/proc_fs.h>
18 #include <linux/devfs_fs_kernel.h>
19 #include <linux/kernel.h>
20 #include <linux/syscalls.h>
21 #include <linux/string.h>
22 #include <linux/ctype.h>
23 #include <linux/delay.h>
24 #include <linux/utsname.h>
25 #include <linux/ioport.h>
26 #include <linux/init.h>
27 #include <linux/smp_lock.h>
28 #include <linux/initrd.h>
29 #include <linux/hdreg.h>
30 #include <linux/bootmem.h>
31 #include <linux/tty.h>
32 #include <linux/gfp.h>
33 #include <linux/percpu.h>
34 #include <linux/kmod.h>
35 #include <linux/kernel_stat.h>
36 #include <linux/security.h>
37 #include <linux/workqueue.h>
38 #include <linux/profile.h>
39 #include <linux/rcupdate.h>
40 #include <linux/moduleparam.h>
41 #include <linux/kallsyms.h>
42 #include <linux/writeback.h>
43 #include <linux/cpu.h>
44 #include <linux/cpuset.h>
45 #include <linux/efi.h>
46 #include <linux/unistd.h>
47 #include <linux/rmap.h>
48 #include <linux/mempolicy.h>
49 #include <linux/key.h>
50 #include <net/sock.h>
51 
52 #include <asm/io.h>
53 #include <asm/bugs.h>
54 #include <asm/setup.h>
55 #include <asm/sections.h>
56 
57 /*
58  * This is one of the first .c files built. Error out early
59  * if we have compiler trouble..
60  */
61 #if __GNUC__ == 2 && __GNUC_MINOR__ == 96
62 #ifdef CONFIG_FRAME_POINTER
63 #error This compiler cannot compile correctly with frame pointers enabled
64 #endif
65 #endif
66 
67 #ifdef CONFIG_X86_LOCAL_APIC
68 #include <asm/smp.h>
69 #endif
70 
71 /*
72  * Versions of gcc older than that listed below may actually compile
73  * and link okay, but the end product can have subtle run time bugs.
74  * To avoid associated bogus bug reports, we flatly refuse to compile
75  * with a gcc that is known to be too old from the very beginning.
76  */
77 #if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 95)
78 #error Sorry, your GCC is too old. It builds incorrect kernels.
79 #endif
80 
81 static int init(void *);
82 
83 extern void init_IRQ(void);
84 extern void fork_init(unsigned long);
85 extern void mca_init(void);
86 extern void sbus_init(void);
87 extern void sysctl_init(void);
88 extern void signals_init(void);
89 extern void buffer_init(void);
90 extern void pidhash_init(void);
91 extern void pidmap_init(void);
92 extern void prio_tree_init(void);
93 extern void radix_tree_init(void);
94 extern void free_initmem(void);
95 extern void populate_rootfs(void);
96 extern void driver_init(void);
97 extern void prepare_namespace(void);
98 #ifdef	CONFIG_ACPI
99 extern void acpi_early_init(void);
100 #else
101 static inline void acpi_early_init(void) { }
102 #endif
103 
104 #ifdef CONFIG_TC
105 extern void tc_init(void);
106 #endif
107 
108 enum system_states system_state;
109 EXPORT_SYMBOL(system_state);
110 
111 /*
112  * Boot command-line arguments
113  */
114 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
115 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
116 
117 extern void time_init(void);
118 /* Default late time init is NULL. archs can override this later. */
119 void (*late_time_init)(void);
120 extern void softirq_init(void);
121 
122 /* Untouched command line (eg. for /proc) saved by arch-specific code. */
123 char saved_command_line[COMMAND_LINE_SIZE];
124 
125 static char *execute_command;
126 static char *ramdisk_execute_command;
127 
128 /* Setup configured maximum number of CPUs to activate */
129 static unsigned int max_cpus = NR_CPUS;
130 
131 /*
132  * Setup routine for controlling SMP activation
133  *
134  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
135  * activation entirely (the MPS table probe still happens, though).
136  *
137  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
138  * greater than 0, limits the maximum number of CPUs activated in
139  * SMP mode to <NUM>.
140  */
141 static int __init nosmp(char *str)
142 {
143 	max_cpus = 0;
144 	return 1;
145 }
146 
147 __setup("nosmp", nosmp);
148 
149 static int __init maxcpus(char *str)
150 {
151 	get_option(&str, &max_cpus);
152 	return 1;
153 }
154 
155 __setup("maxcpus=", maxcpus);
156 
157 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
158 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
159 static const char *panic_later, *panic_param;
160 
161 extern struct obs_kernel_param __setup_start[], __setup_end[];
162 
163 static int __init obsolete_checksetup(char *line)
164 {
165 	struct obs_kernel_param *p;
166 
167 	p = __setup_start;
168 	do {
169 		int n = strlen(p->str);
170 		if (!strncmp(line, p->str, n)) {
171 			if (p->early) {
172 				/* Already done in parse_early_param?  (Needs
173 				 * exact match on param part) */
174 				if (line[n] == '\0' || line[n] == '=')
175 					return 1;
176 			} else if (!p->setup_func) {
177 				printk(KERN_WARNING "Parameter %s is obsolete,"
178 				       " ignored\n", p->str);
179 				return 1;
180 			} else if (p->setup_func(line + n))
181 				return 1;
182 		}
183 		p++;
184 	} while (p < __setup_end);
185 	return 0;
186 }
187 
188 /*
189  * This should be approx 2 Bo*oMips to start (note initial shift), and will
190  * still work even if initially too large, it will just take slightly longer
191  */
192 unsigned long loops_per_jiffy = (1<<12);
193 
194 EXPORT_SYMBOL(loops_per_jiffy);
195 
196 static int __init debug_kernel(char *str)
197 {
198 	if (*str)
199 		return 0;
200 	console_loglevel = 10;
201 	return 1;
202 }
203 
204 static int __init quiet_kernel(char *str)
205 {
206 	if (*str)
207 		return 0;
208 	console_loglevel = 4;
209 	return 1;
210 }
211 
212 __setup("debug", debug_kernel);
213 __setup("quiet", quiet_kernel);
214 
215 static int __init loglevel(char *str)
216 {
217 	get_option(&str, &console_loglevel);
218 	return 1;
219 }
220 
221 __setup("loglevel=", loglevel);
222 
223 /*
224  * Unknown boot options get handed to init, unless they look like
225  * failed parameters
226  */
227 static int __init unknown_bootoption(char *param, char *val)
228 {
229 	/* Change NUL term back to "=", to make "param" the whole string. */
230 	if (val) {
231 		/* param=val or param="val"? */
232 		if (val == param+strlen(param)+1)
233 			val[-1] = '=';
234 		else if (val == param+strlen(param)+2) {
235 			val[-2] = '=';
236 			memmove(val-1, val, strlen(val)+1);
237 			val--;
238 		} else
239 			BUG();
240 	}
241 
242 	/* Handle obsolete-style parameters */
243 	if (obsolete_checksetup(param))
244 		return 0;
245 
246 	/*
247 	 * Preemptive maintenance for "why didn't my mispelled command
248 	 * line work?"
249 	 */
250 	if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
251 		printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
252 		return 0;
253 	}
254 
255 	if (panic_later)
256 		return 0;
257 
258 	if (val) {
259 		/* Environment option */
260 		unsigned int i;
261 		for (i = 0; envp_init[i]; i++) {
262 			if (i == MAX_INIT_ENVS) {
263 				panic_later = "Too many boot env vars at `%s'";
264 				panic_param = param;
265 			}
266 			if (!strncmp(param, envp_init[i], val - param))
267 				break;
268 		}
269 		envp_init[i] = param;
270 	} else {
271 		/* Command line option */
272 		unsigned int i;
273 		for (i = 0; argv_init[i]; i++) {
274 			if (i == MAX_INIT_ARGS) {
275 				panic_later = "Too many boot init vars at `%s'";
276 				panic_param = param;
277 			}
278 		}
279 		argv_init[i] = param;
280 	}
281 	return 0;
282 }
283 
284 static int __init init_setup(char *str)
285 {
286 	unsigned int i;
287 
288 	execute_command = str;
289 	/*
290 	 * In case LILO is going to boot us with default command line,
291 	 * it prepends "auto" before the whole cmdline which makes
292 	 * the shell think it should execute a script with such name.
293 	 * So we ignore all arguments entered _before_ init=... [MJ]
294 	 */
295 	for (i = 1; i < MAX_INIT_ARGS; i++)
296 		argv_init[i] = NULL;
297 	return 1;
298 }
299 __setup("init=", init_setup);
300 
301 static int __init rdinit_setup(char *str)
302 {
303 	unsigned int i;
304 
305 	ramdisk_execute_command = str;
306 	/* See "auto" comment in init_setup */
307 	for (i = 1; i < MAX_INIT_ARGS; i++)
308 		argv_init[i] = NULL;
309 	return 1;
310 }
311 __setup("rdinit=", rdinit_setup);
312 
313 extern void setup_arch(char **);
314 
315 #ifndef CONFIG_SMP
316 
317 #ifdef CONFIG_X86_LOCAL_APIC
318 static void __init smp_init(void)
319 {
320 	APIC_init_uniprocessor();
321 }
322 #else
323 #define smp_init()	do { } while (0)
324 #endif
325 
326 static inline void setup_per_cpu_areas(void) { }
327 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
328 
329 #else
330 
331 #ifdef __GENERIC_PER_CPU
332 unsigned long __per_cpu_offset[NR_CPUS];
333 
334 EXPORT_SYMBOL(__per_cpu_offset);
335 
336 static void __init setup_per_cpu_areas(void)
337 {
338 	unsigned long size, i;
339 	char *ptr;
340 
341 	/* Copy section for each CPU (we discard the original) */
342 	size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
343 #ifdef CONFIG_MODULES
344 	if (size < PERCPU_ENOUGH_ROOM)
345 		size = PERCPU_ENOUGH_ROOM;
346 #endif
347 
348 	ptr = alloc_bootmem(size * NR_CPUS);
349 
350 	for (i = 0; i < NR_CPUS; i++, ptr += size) {
351 		__per_cpu_offset[i] = ptr - __per_cpu_start;
352 		memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
353 	}
354 }
355 #endif /* !__GENERIC_PER_CPU */
356 
357 /* Called by boot processor to activate the rest. */
358 static void __init smp_init(void)
359 {
360 	unsigned int i;
361 
362 	/* FIXME: This should be done in userspace --RR */
363 	for_each_present_cpu(i) {
364 		if (num_online_cpus() >= max_cpus)
365 			break;
366 		if (!cpu_online(i))
367 			cpu_up(i);
368 	}
369 
370 	/* Any cleanup work */
371 	printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
372 	smp_cpus_done(max_cpus);
373 #if 0
374 	/* Get other processors into their bootup holding patterns. */
375 
376 	smp_commence();
377 #endif
378 }
379 
380 #endif
381 
382 /*
383  * We need to finalize in a non-__init function or else race conditions
384  * between the root thread and the init thread may cause start_kernel to
385  * be reaped by free_initmem before the root thread has proceeded to
386  * cpu_idle.
387  *
388  * gcc-3.4 accidentally inlines this function, so use noinline.
389  */
390 
391 static void noinline rest_init(void)
392 	__releases(kernel_lock)
393 {
394 	kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
395 	numa_default_policy();
396 	unlock_kernel();
397 	preempt_enable_no_resched();
398 
399 	/*
400 	 * The boot idle thread must execute schedule()
401 	 * at least one to get things moving:
402 	 */
403 	schedule();
404 
405 	cpu_idle();
406 }
407 
408 /* Check for early params. */
409 static int __init do_early_param(char *param, char *val)
410 {
411 	struct obs_kernel_param *p;
412 
413 	for (p = __setup_start; p < __setup_end; p++) {
414 		if (p->early && strcmp(param, p->str) == 0) {
415 			if (p->setup_func(val) != 0)
416 				printk(KERN_WARNING
417 				       "Malformed early option '%s'\n", param);
418 		}
419 	}
420 	/* We accept everything at this stage. */
421 	return 0;
422 }
423 
424 /* Arch code calls this early on, or if not, just before other parsing. */
425 void __init parse_early_param(void)
426 {
427 	static __initdata int done = 0;
428 	static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
429 
430 	if (done)
431 		return;
432 
433 	/* All fall through to do_early_param. */
434 	strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
435 	parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
436 	done = 1;
437 }
438 
439 /*
440  *	Activate the first processor.
441  */
442 
443 asmlinkage void __init start_kernel(void)
444 {
445 	char * command_line;
446 	extern struct kernel_param __start___param[], __stop___param[];
447 /*
448  * Interrupts are still disabled. Do necessary setups, then
449  * enable them
450  */
451 	lock_kernel();
452 	page_address_init();
453 	printk(KERN_NOTICE);
454 	printk(linux_banner);
455 	setup_arch(&command_line);
456 	setup_per_cpu_areas();
457 
458 	/*
459 	 * Mark the boot cpu "online" so that it can call console drivers in
460 	 * printk() and can access its per-cpu storage.
461 	 */
462 	smp_prepare_boot_cpu();
463 
464 	/*
465 	 * Set up the scheduler prior starting any interrupts (such as the
466 	 * timer interrupt). Full topology setup happens at smp_init()
467 	 * time - but meanwhile we still have a functioning scheduler.
468 	 */
469 	sched_init();
470 	/*
471 	 * Disable preemption - early bootup scheduling is extremely
472 	 * fragile until we cpu_idle() for the first time.
473 	 */
474 	preempt_disable();
475 	build_all_zonelists();
476 	page_alloc_init();
477 	printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
478 	parse_early_param();
479 	parse_args("Booting kernel", command_line, __start___param,
480 		   __stop___param - __start___param,
481 		   &unknown_bootoption);
482 	sort_main_extable();
483 	trap_init();
484 	rcu_init();
485 	init_IRQ();
486 	pidhash_init();
487 	init_timers();
488 	softirq_init();
489 	time_init();
490 
491 	/*
492 	 * HACK ALERT! This is early. We're enabling the console before
493 	 * we've done PCI setups etc, and console_init() must be aware of
494 	 * this. But we do want output early, in case something goes wrong.
495 	 */
496 	console_init();
497 	if (panic_later)
498 		panic(panic_later, panic_param);
499 	profile_init();
500 	local_irq_enable();
501 #ifdef CONFIG_BLK_DEV_INITRD
502 	if (initrd_start && !initrd_below_start_ok &&
503 			initrd_start < min_low_pfn << PAGE_SHIFT) {
504 		printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
505 		    "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
506 		initrd_start = 0;
507 	}
508 #endif
509 	vfs_caches_init_early();
510 	mem_init();
511 	kmem_cache_init();
512 	setup_per_cpu_pageset();
513 	numa_policy_init();
514 	if (late_time_init)
515 		late_time_init();
516 	calibrate_delay();
517 	pidmap_init();
518 	pgtable_cache_init();
519 	prio_tree_init();
520 	anon_vma_init();
521 #ifdef CONFIG_X86
522 	if (efi_enabled)
523 		efi_enter_virtual_mode();
524 #endif
525 	fork_init(num_physpages);
526 	proc_caches_init();
527 	buffer_init();
528 	unnamed_dev_init();
529 	key_init();
530 	security_init();
531 	vfs_caches_init(num_physpages);
532 	radix_tree_init();
533 	signals_init();
534 	/* rootfs populating might need page-writeback */
535 	page_writeback_init();
536 #ifdef CONFIG_PROC_FS
537 	proc_root_init();
538 #endif
539 	cpuset_init();
540 
541 	check_bugs();
542 
543 	acpi_early_init(); /* before LAPIC and SMP init */
544 
545 	/* Do the rest non-__init'ed, we're now alive */
546 	rest_init();
547 }
548 
549 static int __initdata initcall_debug;
550 
551 static int __init initcall_debug_setup(char *str)
552 {
553 	initcall_debug = 1;
554 	return 1;
555 }
556 __setup("initcall_debug", initcall_debug_setup);
557 
558 struct task_struct *child_reaper = &init_task;
559 
560 extern initcall_t __initcall_start[], __initcall_end[];
561 
562 static void __init do_initcalls(void)
563 {
564 	initcall_t *call;
565 	int count = preempt_count();
566 
567 	for (call = __initcall_start; call < __initcall_end; call++) {
568 		char *msg;
569 
570 		if (initcall_debug) {
571 			printk(KERN_DEBUG "Calling initcall 0x%p", *call);
572 			print_fn_descriptor_symbol(": %s()", (unsigned long) *call);
573 			printk("\n");
574 		}
575 
576 		(*call)();
577 
578 		msg = NULL;
579 		if (preempt_count() != count) {
580 			msg = "preemption imbalance";
581 			preempt_count() = count;
582 		}
583 		if (irqs_disabled()) {
584 			msg = "disabled interrupts";
585 			local_irq_enable();
586 		}
587 		if (msg) {
588 			printk(KERN_WARNING "error in initcall at 0x%p: "
589 				"returned with %s\n", *call, msg);
590 		}
591 	}
592 
593 	/* Make sure there is no pending stuff from the initcall sequence */
594 	flush_scheduled_work();
595 }
596 
597 /*
598  * Ok, the machine is now initialized. None of the devices
599  * have been touched yet, but the CPU subsystem is up and
600  * running, and memory and process management works.
601  *
602  * Now we can finally start doing some real work..
603  */
604 static void __init do_basic_setup(void)
605 {
606 	/* drivers will send hotplug events */
607 	init_workqueues();
608 	usermodehelper_init();
609 	driver_init();
610 
611 #ifdef CONFIG_SYSCTL
612 	sysctl_init();
613 #endif
614 
615 	/* Networking initialization needs a process context */
616 	sock_init();
617 
618 	do_initcalls();
619 }
620 
621 static void do_pre_smp_initcalls(void)
622 {
623 	extern int spawn_ksoftirqd(void);
624 #ifdef CONFIG_SMP
625 	extern int migration_init(void);
626 
627 	migration_init();
628 #endif
629 	spawn_ksoftirqd();
630 	spawn_softlockup_task();
631 }
632 
633 static void run_init_process(char *init_filename)
634 {
635 	argv_init[0] = init_filename;
636 	execve(init_filename, argv_init, envp_init);
637 }
638 
639 static inline void fixup_cpu_present_map(void)
640 {
641 #ifdef CONFIG_SMP
642 	int i;
643 
644 	/*
645 	 * If arch is not hotplug ready and did not populate
646 	 * cpu_present_map, just make cpu_present_map same as cpu_possible_map
647 	 * for other cpu bringup code to function as normal. e.g smp_init() etc.
648 	 */
649 	if (cpus_empty(cpu_present_map)) {
650 		for_each_cpu(i) {
651 			cpu_set(i, cpu_present_map);
652 		}
653 	}
654 #endif
655 }
656 
657 static int init(void * unused)
658 {
659 	lock_kernel();
660 	/*
661 	 * init can run on any cpu.
662 	 */
663 	set_cpus_allowed(current, CPU_MASK_ALL);
664 	/*
665 	 * Tell the world that we're going to be the grim
666 	 * reaper of innocent orphaned children.
667 	 *
668 	 * We don't want people to have to make incorrect
669 	 * assumptions about where in the task array this
670 	 * can be found.
671 	 */
672 	child_reaper = current;
673 
674 	/* Sets up cpus_possible() */
675 	smp_prepare_cpus(max_cpus);
676 
677 	do_pre_smp_initcalls();
678 
679 	fixup_cpu_present_map();
680 	smp_init();
681 	sched_init_smp();
682 
683 	cpuset_init_smp();
684 
685 	/*
686 	 * Do this before initcalls, because some drivers want to access
687 	 * firmware files.
688 	 */
689 	populate_rootfs();
690 
691 	do_basic_setup();
692 
693 	/*
694 	 * check if there is an early userspace init.  If yes, let it do all
695 	 * the work
696 	 */
697 
698 	if (!ramdisk_execute_command)
699 		ramdisk_execute_command = "/init";
700 
701 	if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
702 		ramdisk_execute_command = NULL;
703 		prepare_namespace();
704 	}
705 
706 	/*
707 	 * Ok, we have completed the initial bootup, and
708 	 * we're essentially up and running. Get rid of the
709 	 * initmem segments and start the user-mode stuff..
710 	 */
711 	free_initmem();
712 	unlock_kernel();
713 	system_state = SYSTEM_RUNNING;
714 	numa_default_policy();
715 
716 	if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
717 		printk(KERN_WARNING "Warning: unable to open an initial console.\n");
718 
719 	(void) sys_dup(0);
720 	(void) sys_dup(0);
721 
722 	if (ramdisk_execute_command) {
723 		run_init_process(ramdisk_execute_command);
724 		printk(KERN_WARNING "Failed to execute %s\n",
725 				ramdisk_execute_command);
726 	}
727 
728 	/*
729 	 * We try each of these until one succeeds.
730 	 *
731 	 * The Bourne shell can be used instead of init if we are
732 	 * trying to recover a really broken machine.
733 	 */
734 	if (execute_command) {
735 		run_init_process(execute_command);
736 		printk(KERN_WARNING "Failed to execute %s.  Attempting "
737 					"defaults...\n", execute_command);
738 	}
739 	run_init_process("/sbin/init");
740 	run_init_process("/etc/init");
741 	run_init_process("/bin/init");
742 	run_init_process("/bin/sh");
743 
744 	panic("No init found.  Try passing init= option to kernel.");
745 }
746