xref: /openbmc/linux/init/main.c (revision 9fb29c73)
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/extable.h>
16 #include <linux/module.h>
17 #include <linux/proc_fs.h>
18 #include <linux/binfmts.h>
19 #include <linux/kernel.h>
20 #include <linux/syscalls.h>
21 #include <linux/stackprotector.h>
22 #include <linux/string.h>
23 #include <linux/ctype.h>
24 #include <linux/delay.h>
25 #include <linux/ioport.h>
26 #include <linux/init.h>
27 #include <linux/initrd.h>
28 #include <linux/memblock.h>
29 #include <linux/acpi.h>
30 #include <linux/console.h>
31 #include <linux/nmi.h>
32 #include <linux/percpu.h>
33 #include <linux/kmod.h>
34 #include <linux/vmalloc.h>
35 #include <linux/kernel_stat.h>
36 #include <linux/start_kernel.h>
37 #include <linux/security.h>
38 #include <linux/smp.h>
39 #include <linux/profile.h>
40 #include <linux/rcupdate.h>
41 #include <linux/moduleparam.h>
42 #include <linux/kallsyms.h>
43 #include <linux/writeback.h>
44 #include <linux/cpu.h>
45 #include <linux/cpuset.h>
46 #include <linux/cgroup.h>
47 #include <linux/efi.h>
48 #include <linux/tick.h>
49 #include <linux/sched/isolation.h>
50 #include <linux/interrupt.h>
51 #include <linux/taskstats_kern.h>
52 #include <linux/delayacct.h>
53 #include <linux/unistd.h>
54 #include <linux/utsname.h>
55 #include <linux/rmap.h>
56 #include <linux/mempolicy.h>
57 #include <linux/key.h>
58 #include <linux/buffer_head.h>
59 #include <linux/page_ext.h>
60 #include <linux/debug_locks.h>
61 #include <linux/debugobjects.h>
62 #include <linux/lockdep.h>
63 #include <linux/kmemleak.h>
64 #include <linux/pid_namespace.h>
65 #include <linux/device.h>
66 #include <linux/kthread.h>
67 #include <linux/sched.h>
68 #include <linux/sched/init.h>
69 #include <linux/signal.h>
70 #include <linux/idr.h>
71 #include <linux/kgdb.h>
72 #include <linux/ftrace.h>
73 #include <linux/async.h>
74 #include <linux/sfi.h>
75 #include <linux/shmem_fs.h>
76 #include <linux/slab.h>
77 #include <linux/perf_event.h>
78 #include <linux/ptrace.h>
79 #include <linux/pti.h>
80 #include <linux/blkdev.h>
81 #include <linux/elevator.h>
82 #include <linux/sched/clock.h>
83 #include <linux/sched/task.h>
84 #include <linux/sched/task_stack.h>
85 #include <linux/context_tracking.h>
86 #include <linux/random.h>
87 #include <linux/list.h>
88 #include <linux/integrity.h>
89 #include <linux/proc_ns.h>
90 #include <linux/io.h>
91 #include <linux/cache.h>
92 #include <linux/rodata_test.h>
93 #include <linux/jump_label.h>
94 #include <linux/mem_encrypt.h>
95 
96 #include <asm/io.h>
97 #include <asm/bugs.h>
98 #include <asm/setup.h>
99 #include <asm/sections.h>
100 #include <asm/cacheflush.h>
101 
102 #define CREATE_TRACE_POINTS
103 #include <trace/events/initcall.h>
104 
105 static int kernel_init(void *);
106 
107 extern void init_IRQ(void);
108 extern void radix_tree_init(void);
109 
110 /*
111  * Debug helper: via this flag we know that we are in 'early bootup code'
112  * where only the boot processor is running with IRQ disabled.  This means
113  * two things - IRQ must not be enabled before the flag is cleared and some
114  * operations which are not allowed with IRQ disabled are allowed while the
115  * flag is set.
116  */
117 bool early_boot_irqs_disabled __read_mostly;
118 
119 enum system_states system_state __read_mostly;
120 EXPORT_SYMBOL(system_state);
121 
122 /*
123  * Boot command-line arguments
124  */
125 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
126 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
127 
128 extern void time_init(void);
129 /* Default late time init is NULL. archs can override this later. */
130 void (*__initdata late_time_init)(void);
131 
132 /* Untouched command line saved by arch-specific code. */
133 char __initdata boot_command_line[COMMAND_LINE_SIZE];
134 /* Untouched saved command line (eg. for /proc) */
135 char *saved_command_line;
136 /* Command line for parameter parsing */
137 static char *static_command_line;
138 /* Command line for per-initcall parameter parsing */
139 static char *initcall_command_line;
140 
141 static char *execute_command;
142 static char *ramdisk_execute_command;
143 
144 /*
145  * Used to generate warnings if static_key manipulation functions are used
146  * before jump_label_init is called.
147  */
148 bool static_key_initialized __read_mostly;
149 EXPORT_SYMBOL_GPL(static_key_initialized);
150 
151 /*
152  * If set, this is an indication to the drivers that reset the underlying
153  * device before going ahead with the initialization otherwise driver might
154  * rely on the BIOS and skip the reset operation.
155  *
156  * This is useful if kernel is booting in an unreliable environment.
157  * For ex. kdump situation where previous kernel has crashed, BIOS has been
158  * skipped and devices will be in unknown state.
159  */
160 unsigned int reset_devices;
161 EXPORT_SYMBOL(reset_devices);
162 
163 static int __init set_reset_devices(char *str)
164 {
165 	reset_devices = 1;
166 	return 1;
167 }
168 
169 __setup("reset_devices", set_reset_devices);
170 
171 static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
172 const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
173 static const char *panic_later, *panic_param;
174 
175 extern const struct obs_kernel_param __setup_start[], __setup_end[];
176 
177 static bool __init obsolete_checksetup(char *line)
178 {
179 	const struct obs_kernel_param *p;
180 	bool had_early_param = false;
181 
182 	p = __setup_start;
183 	do {
184 		int n = strlen(p->str);
185 		if (parameqn(line, p->str, n)) {
186 			if (p->early) {
187 				/* Already done in parse_early_param?
188 				 * (Needs exact match on param part).
189 				 * Keep iterating, as we can have early
190 				 * params and __setups of same names 8( */
191 				if (line[n] == '\0' || line[n] == '=')
192 					had_early_param = true;
193 			} else if (!p->setup_func) {
194 				pr_warn("Parameter %s is obsolete, ignored\n",
195 					p->str);
196 				return true;
197 			} else if (p->setup_func(line + n))
198 				return true;
199 		}
200 		p++;
201 	} while (p < __setup_end);
202 
203 	return had_early_param;
204 }
205 
206 /*
207  * This should be approx 2 Bo*oMips to start (note initial shift), and will
208  * still work even if initially too large, it will just take slightly longer
209  */
210 unsigned long loops_per_jiffy = (1<<12);
211 EXPORT_SYMBOL(loops_per_jiffy);
212 
213 static int __init debug_kernel(char *str)
214 {
215 	console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
216 	return 0;
217 }
218 
219 static int __init quiet_kernel(char *str)
220 {
221 	console_loglevel = CONSOLE_LOGLEVEL_QUIET;
222 	return 0;
223 }
224 
225 early_param("debug", debug_kernel);
226 early_param("quiet", quiet_kernel);
227 
228 static int __init loglevel(char *str)
229 {
230 	int newlevel;
231 
232 	/*
233 	 * Only update loglevel value when a correct setting was passed,
234 	 * to prevent blind crashes (when loglevel being set to 0) that
235 	 * are quite hard to debug
236 	 */
237 	if (get_option(&str, &newlevel)) {
238 		console_loglevel = newlevel;
239 		return 0;
240 	}
241 
242 	return -EINVAL;
243 }
244 
245 early_param("loglevel", loglevel);
246 
247 /* Change NUL term back to "=", to make "param" the whole string. */
248 static int __init repair_env_string(char *param, char *val,
249 				    const char *unused, void *arg)
250 {
251 	if (val) {
252 		/* param=val or param="val"? */
253 		if (val == param+strlen(param)+1)
254 			val[-1] = '=';
255 		else if (val == param+strlen(param)+2) {
256 			val[-2] = '=';
257 			memmove(val-1, val, strlen(val)+1);
258 			val--;
259 		} else
260 			BUG();
261 	}
262 	return 0;
263 }
264 
265 /* Anything after -- gets handed straight to init. */
266 static int __init set_init_arg(char *param, char *val,
267 			       const char *unused, void *arg)
268 {
269 	unsigned int i;
270 
271 	if (panic_later)
272 		return 0;
273 
274 	repair_env_string(param, val, unused, NULL);
275 
276 	for (i = 0; argv_init[i]; i++) {
277 		if (i == MAX_INIT_ARGS) {
278 			panic_later = "init";
279 			panic_param = param;
280 			return 0;
281 		}
282 	}
283 	argv_init[i] = param;
284 	return 0;
285 }
286 
287 /*
288  * Unknown boot options get handed to init, unless they look like
289  * unused parameters (modprobe will find them in /proc/cmdline).
290  */
291 static int __init unknown_bootoption(char *param, char *val,
292 				     const char *unused, void *arg)
293 {
294 	repair_env_string(param, val, unused, NULL);
295 
296 	/* Handle obsolete-style parameters */
297 	if (obsolete_checksetup(param))
298 		return 0;
299 
300 	/* Unused module parameter. */
301 	if (strchr(param, '.') && (!val || strchr(param, '.') < val))
302 		return 0;
303 
304 	if (panic_later)
305 		return 0;
306 
307 	if (val) {
308 		/* Environment option */
309 		unsigned int i;
310 		for (i = 0; envp_init[i]; i++) {
311 			if (i == MAX_INIT_ENVS) {
312 				panic_later = "env";
313 				panic_param = param;
314 			}
315 			if (!strncmp(param, envp_init[i], val - param))
316 				break;
317 		}
318 		envp_init[i] = param;
319 	} else {
320 		/* Command line option */
321 		unsigned int i;
322 		for (i = 0; argv_init[i]; i++) {
323 			if (i == MAX_INIT_ARGS) {
324 				panic_later = "init";
325 				panic_param = param;
326 			}
327 		}
328 		argv_init[i] = param;
329 	}
330 	return 0;
331 }
332 
333 static int __init init_setup(char *str)
334 {
335 	unsigned int i;
336 
337 	execute_command = str;
338 	/*
339 	 * In case LILO is going to boot us with default command line,
340 	 * it prepends "auto" before the whole cmdline which makes
341 	 * the shell think it should execute a script with such name.
342 	 * So we ignore all arguments entered _before_ init=... [MJ]
343 	 */
344 	for (i = 1; i < MAX_INIT_ARGS; i++)
345 		argv_init[i] = NULL;
346 	return 1;
347 }
348 __setup("init=", init_setup);
349 
350 static int __init rdinit_setup(char *str)
351 {
352 	unsigned int i;
353 
354 	ramdisk_execute_command = str;
355 	/* See "auto" comment in init_setup */
356 	for (i = 1; i < MAX_INIT_ARGS; i++)
357 		argv_init[i] = NULL;
358 	return 1;
359 }
360 __setup("rdinit=", rdinit_setup);
361 
362 #ifndef CONFIG_SMP
363 static const unsigned int setup_max_cpus = NR_CPUS;
364 static inline void setup_nr_cpu_ids(void) { }
365 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
366 #endif
367 
368 /*
369  * We need to store the untouched command line for future reference.
370  * We also need to store the touched command line since the parameter
371  * parsing is performed in place, and we should allow a component to
372  * store reference of name/value for future reference.
373  */
374 static void __init setup_command_line(char *command_line)
375 {
376 	saved_command_line =
377 		memblock_alloc(strlen(boot_command_line) + 1, SMP_CACHE_BYTES);
378 	initcall_command_line =
379 		memblock_alloc(strlen(boot_command_line) + 1, SMP_CACHE_BYTES);
380 	static_command_line = memblock_alloc(strlen(command_line) + 1,
381 					     SMP_CACHE_BYTES);
382 	strcpy(saved_command_line, boot_command_line);
383 	strcpy(static_command_line, command_line);
384 }
385 
386 /*
387  * We need to finalize in a non-__init function or else race conditions
388  * between the root thread and the init thread may cause start_kernel to
389  * be reaped by free_initmem before the root thread has proceeded to
390  * cpu_idle.
391  *
392  * gcc-3.4 accidentally inlines this function, so use noinline.
393  */
394 
395 static __initdata DECLARE_COMPLETION(kthreadd_done);
396 
397 noinline void __ref rest_init(void)
398 {
399 	struct task_struct *tsk;
400 	int pid;
401 
402 	rcu_scheduler_starting();
403 	/*
404 	 * We need to spawn init first so that it obtains pid 1, however
405 	 * the init task will end up wanting to create kthreads, which, if
406 	 * we schedule it before we create kthreadd, will OOPS.
407 	 */
408 	pid = kernel_thread(kernel_init, NULL, CLONE_FS);
409 	/*
410 	 * Pin init on the boot CPU. Task migration is not properly working
411 	 * until sched_init_smp() has been run. It will set the allowed
412 	 * CPUs for init to the non isolated CPUs.
413 	 */
414 	rcu_read_lock();
415 	tsk = find_task_by_pid_ns(pid, &init_pid_ns);
416 	set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id()));
417 	rcu_read_unlock();
418 
419 	numa_default_policy();
420 	pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
421 	rcu_read_lock();
422 	kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
423 	rcu_read_unlock();
424 
425 	/*
426 	 * Enable might_sleep() and smp_processor_id() checks.
427 	 * They cannot be enabled earlier because with CONFIG_PREEMPT=y
428 	 * kernel_thread() would trigger might_sleep() splats. With
429 	 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled
430 	 * already, but it's stuck on the kthreadd_done completion.
431 	 */
432 	system_state = SYSTEM_SCHEDULING;
433 
434 	complete(&kthreadd_done);
435 
436 	/*
437 	 * The boot idle thread must execute schedule()
438 	 * at least once to get things moving:
439 	 */
440 	schedule_preempt_disabled();
441 	/* Call into cpu_idle with preempt disabled */
442 	cpu_startup_entry(CPUHP_ONLINE);
443 }
444 
445 /* Check for early params. */
446 static int __init do_early_param(char *param, char *val,
447 				 const char *unused, void *arg)
448 {
449 	const struct obs_kernel_param *p;
450 
451 	for (p = __setup_start; p < __setup_end; p++) {
452 		if ((p->early && parameq(param, p->str)) ||
453 		    (strcmp(param, "console") == 0 &&
454 		     strcmp(p->str, "earlycon") == 0)
455 		) {
456 			if (p->setup_func(val) != 0)
457 				pr_warn("Malformed early option '%s'\n", param);
458 		}
459 	}
460 	/* We accept everything at this stage. */
461 	return 0;
462 }
463 
464 void __init parse_early_options(char *cmdline)
465 {
466 	parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
467 		   do_early_param);
468 }
469 
470 /* Arch code calls this early on, or if not, just before other parsing. */
471 void __init parse_early_param(void)
472 {
473 	static int done __initdata;
474 	static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
475 
476 	if (done)
477 		return;
478 
479 	/* All fall through to do_early_param. */
480 	strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
481 	parse_early_options(tmp_cmdline);
482 	done = 1;
483 }
484 
485 void __init __weak arch_post_acpi_subsys_init(void) { }
486 
487 void __init __weak smp_setup_processor_id(void)
488 {
489 }
490 
491 # if THREAD_SIZE >= PAGE_SIZE
492 void __init __weak thread_stack_cache_init(void)
493 {
494 }
495 #endif
496 
497 void __init __weak mem_encrypt_init(void) { }
498 
499 bool initcall_debug;
500 core_param(initcall_debug, initcall_debug, bool, 0644);
501 
502 #ifdef TRACEPOINTS_ENABLED
503 static void __init initcall_debug_enable(void);
504 #else
505 static inline void initcall_debug_enable(void)
506 {
507 }
508 #endif
509 
510 /*
511  * Set up kernel memory allocators
512  */
513 static void __init mm_init(void)
514 {
515 	/*
516 	 * page_ext requires contiguous pages,
517 	 * bigger than MAX_ORDER unless SPARSEMEM.
518 	 */
519 	page_ext_init_flatmem();
520 	mem_init();
521 	kmem_cache_init();
522 	pgtable_init();
523 	debug_objects_mem_init();
524 	vmalloc_init();
525 	ioremap_huge_init();
526 	/* Should be run before the first non-init thread is created */
527 	init_espfix_bsp();
528 	/* Should be run after espfix64 is set up. */
529 	pti_init();
530 }
531 
532 void __init __weak arch_call_rest_init(void)
533 {
534 	rest_init();
535 }
536 
537 asmlinkage __visible void __init start_kernel(void)
538 {
539 	char *command_line;
540 	char *after_dashes;
541 
542 	set_task_stack_end_magic(&init_task);
543 	smp_setup_processor_id();
544 	debug_objects_early_init();
545 
546 	cgroup_init_early();
547 
548 	local_irq_disable();
549 	early_boot_irqs_disabled = true;
550 
551 	/*
552 	 * Interrupts are still disabled. Do necessary setups, then
553 	 * enable them.
554 	 */
555 	boot_cpu_init();
556 	page_address_init();
557 	pr_notice("%s", linux_banner);
558 	setup_arch(&command_line);
559 	/*
560 	 * Set up the the initial canary and entropy after arch
561 	 * and after adding latent and command line entropy.
562 	 */
563 	add_latent_entropy();
564 	add_device_randomness(command_line, strlen(command_line));
565 	boot_init_stack_canary();
566 	mm_init_cpumask(&init_mm);
567 	setup_command_line(command_line);
568 	setup_nr_cpu_ids();
569 	setup_per_cpu_areas();
570 	smp_prepare_boot_cpu();	/* arch-specific boot-cpu hooks */
571 	boot_cpu_hotplug_init();
572 
573 	build_all_zonelists(NULL);
574 	page_alloc_init();
575 
576 	pr_notice("Kernel command line: %s\n", boot_command_line);
577 	parse_early_param();
578 	after_dashes = parse_args("Booting kernel",
579 				  static_command_line, __start___param,
580 				  __stop___param - __start___param,
581 				  -1, -1, NULL, &unknown_bootoption);
582 	if (!IS_ERR_OR_NULL(after_dashes))
583 		parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
584 			   NULL, set_init_arg);
585 
586 	jump_label_init();
587 
588 	/*
589 	 * These use large bootmem allocations and must precede
590 	 * kmem_cache_init()
591 	 */
592 	setup_log_buf(0);
593 	vfs_caches_init_early();
594 	sort_main_extable();
595 	trap_init();
596 	mm_init();
597 
598 	ftrace_init();
599 
600 	/* trace_printk can be enabled here */
601 	early_trace_init();
602 
603 	/*
604 	 * Set up the scheduler prior starting any interrupts (such as the
605 	 * timer interrupt). Full topology setup happens at smp_init()
606 	 * time - but meanwhile we still have a functioning scheduler.
607 	 */
608 	sched_init();
609 	/*
610 	 * Disable preemption - early bootup scheduling is extremely
611 	 * fragile until we cpu_idle() for the first time.
612 	 */
613 	preempt_disable();
614 	if (WARN(!irqs_disabled(),
615 		 "Interrupts were enabled *very* early, fixing it\n"))
616 		local_irq_disable();
617 	radix_tree_init();
618 
619 	/*
620 	 * Set up housekeeping before setting up workqueues to allow the unbound
621 	 * workqueue to take non-housekeeping into account.
622 	 */
623 	housekeeping_init();
624 
625 	/*
626 	 * Allow workqueue creation and work item queueing/cancelling
627 	 * early.  Work item execution depends on kthreads and starts after
628 	 * workqueue_init().
629 	 */
630 	workqueue_init_early();
631 
632 	rcu_init();
633 
634 	/* Trace events are available after this */
635 	trace_init();
636 
637 	if (initcall_debug)
638 		initcall_debug_enable();
639 
640 	context_tracking_init();
641 	/* init some links before init_ISA_irqs() */
642 	early_irq_init();
643 	init_IRQ();
644 	tick_init();
645 	rcu_init_nohz();
646 	init_timers();
647 	hrtimers_init();
648 	softirq_init();
649 	timekeeping_init();
650 	time_init();
651 	printk_safe_init();
652 	perf_event_init();
653 	profile_init();
654 	call_function_init();
655 	WARN(!irqs_disabled(), "Interrupts were enabled early\n");
656 
657 	early_boot_irqs_disabled = false;
658 	local_irq_enable();
659 
660 	kmem_cache_init_late();
661 
662 	/*
663 	 * HACK ALERT! This is early. We're enabling the console before
664 	 * we've done PCI setups etc, and console_init() must be aware of
665 	 * this. But we do want output early, in case something goes wrong.
666 	 */
667 	console_init();
668 	if (panic_later)
669 		panic("Too many boot %s vars at `%s'", panic_later,
670 		      panic_param);
671 
672 	lockdep_init();
673 
674 	/*
675 	 * Need to run this when irqs are enabled, because it wants
676 	 * to self-test [hard/soft]-irqs on/off lock inversion bugs
677 	 * too:
678 	 */
679 	locking_selftest();
680 
681 	/*
682 	 * This needs to be called before any devices perform DMA
683 	 * operations that might use the SWIOTLB bounce buffers. It will
684 	 * mark the bounce buffers as decrypted so that their usage will
685 	 * not cause "plain-text" data to be decrypted when accessed.
686 	 */
687 	mem_encrypt_init();
688 
689 #ifdef CONFIG_BLK_DEV_INITRD
690 	if (initrd_start && !initrd_below_start_ok &&
691 	    page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
692 		pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
693 		    page_to_pfn(virt_to_page((void *)initrd_start)),
694 		    min_low_pfn);
695 		initrd_start = 0;
696 	}
697 #endif
698 	page_ext_init();
699 	kmemleak_init();
700 	setup_per_cpu_pageset();
701 	numa_policy_init();
702 	acpi_early_init();
703 	if (late_time_init)
704 		late_time_init();
705 	sched_clock_init();
706 	calibrate_delay();
707 	pid_idr_init();
708 	anon_vma_init();
709 #ifdef CONFIG_X86
710 	if (efi_enabled(EFI_RUNTIME_SERVICES))
711 		efi_enter_virtual_mode();
712 #endif
713 	thread_stack_cache_init();
714 	cred_init();
715 	fork_init();
716 	proc_caches_init();
717 	uts_ns_init();
718 	buffer_init();
719 	key_init();
720 	security_init();
721 	dbg_late_init();
722 	vfs_caches_init();
723 	pagecache_init();
724 	signals_init();
725 	seq_file_init();
726 	proc_root_init();
727 	nsfs_init();
728 	cpuset_init();
729 	cgroup_init();
730 	taskstats_init_early();
731 	delayacct_init();
732 
733 	check_bugs();
734 
735 	acpi_subsystem_init();
736 	arch_post_acpi_subsys_init();
737 	sfi_init_late();
738 
739 	/* Do the rest non-__init'ed, we're now alive */
740 	arch_call_rest_init();
741 }
742 
743 /* Call all constructor functions linked into the kernel. */
744 static void __init do_ctors(void)
745 {
746 #ifdef CONFIG_CONSTRUCTORS
747 	ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
748 
749 	for (; fn < (ctor_fn_t *) __ctors_end; fn++)
750 		(*fn)();
751 #endif
752 }
753 
754 #ifdef CONFIG_KALLSYMS
755 struct blacklist_entry {
756 	struct list_head next;
757 	char *buf;
758 };
759 
760 static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
761 
762 static int __init initcall_blacklist(char *str)
763 {
764 	char *str_entry;
765 	struct blacklist_entry *entry;
766 
767 	/* str argument is a comma-separated list of functions */
768 	do {
769 		str_entry = strsep(&str, ",");
770 		if (str_entry) {
771 			pr_debug("blacklisting initcall %s\n", str_entry);
772 			entry = memblock_alloc(sizeof(*entry),
773 					       SMP_CACHE_BYTES);
774 			entry->buf = memblock_alloc(strlen(str_entry) + 1,
775 						    SMP_CACHE_BYTES);
776 			strcpy(entry->buf, str_entry);
777 			list_add(&entry->next, &blacklisted_initcalls);
778 		}
779 	} while (str_entry);
780 
781 	return 0;
782 }
783 
784 static bool __init_or_module initcall_blacklisted(initcall_t fn)
785 {
786 	struct blacklist_entry *entry;
787 	char fn_name[KSYM_SYMBOL_LEN];
788 	unsigned long addr;
789 
790 	if (list_empty(&blacklisted_initcalls))
791 		return false;
792 
793 	addr = (unsigned long) dereference_function_descriptor(fn);
794 	sprint_symbol_no_offset(fn_name, addr);
795 
796 	/*
797 	 * fn will be "function_name [module_name]" where [module_name] is not
798 	 * displayed for built-in init functions.  Strip off the [module_name].
799 	 */
800 	strreplace(fn_name, ' ', '\0');
801 
802 	list_for_each_entry(entry, &blacklisted_initcalls, next) {
803 		if (!strcmp(fn_name, entry->buf)) {
804 			pr_debug("initcall %s blacklisted\n", fn_name);
805 			return true;
806 		}
807 	}
808 
809 	return false;
810 }
811 #else
812 static int __init initcall_blacklist(char *str)
813 {
814 	pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
815 	return 0;
816 }
817 
818 static bool __init_or_module initcall_blacklisted(initcall_t fn)
819 {
820 	return false;
821 }
822 #endif
823 __setup("initcall_blacklist=", initcall_blacklist);
824 
825 static __init_or_module void
826 trace_initcall_start_cb(void *data, initcall_t fn)
827 {
828 	ktime_t *calltime = (ktime_t *)data;
829 
830 	printk(KERN_DEBUG "calling  %pF @ %i\n", fn, task_pid_nr(current));
831 	*calltime = ktime_get();
832 }
833 
834 static __init_or_module void
835 trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
836 {
837 	ktime_t *calltime = (ktime_t *)data;
838 	ktime_t delta, rettime;
839 	unsigned long long duration;
840 
841 	rettime = ktime_get();
842 	delta = ktime_sub(rettime, *calltime);
843 	duration = (unsigned long long) ktime_to_ns(delta) >> 10;
844 	printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n",
845 		 fn, ret, duration);
846 }
847 
848 static ktime_t initcall_calltime;
849 
850 #ifdef TRACEPOINTS_ENABLED
851 static void __init initcall_debug_enable(void)
852 {
853 	int ret;
854 
855 	ret = register_trace_initcall_start(trace_initcall_start_cb,
856 					    &initcall_calltime);
857 	ret |= register_trace_initcall_finish(trace_initcall_finish_cb,
858 					      &initcall_calltime);
859 	WARN(ret, "Failed to register initcall tracepoints\n");
860 }
861 # define do_trace_initcall_start	trace_initcall_start
862 # define do_trace_initcall_finish	trace_initcall_finish
863 #else
864 static inline void do_trace_initcall_start(initcall_t fn)
865 {
866 	if (!initcall_debug)
867 		return;
868 	trace_initcall_start_cb(&initcall_calltime, fn);
869 }
870 static inline void do_trace_initcall_finish(initcall_t fn, int ret)
871 {
872 	if (!initcall_debug)
873 		return;
874 	trace_initcall_finish_cb(&initcall_calltime, fn, ret);
875 }
876 #endif /* !TRACEPOINTS_ENABLED */
877 
878 int __init_or_module do_one_initcall(initcall_t fn)
879 {
880 	int count = preempt_count();
881 	char msgbuf[64];
882 	int ret;
883 
884 	if (initcall_blacklisted(fn))
885 		return -EPERM;
886 
887 	do_trace_initcall_start(fn);
888 	ret = fn();
889 	do_trace_initcall_finish(fn, ret);
890 
891 	msgbuf[0] = 0;
892 
893 	if (preempt_count() != count) {
894 		sprintf(msgbuf, "preemption imbalance ");
895 		preempt_count_set(count);
896 	}
897 	if (irqs_disabled()) {
898 		strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
899 		local_irq_enable();
900 	}
901 	WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
902 
903 	add_latent_entropy();
904 	return ret;
905 }
906 
907 
908 extern initcall_entry_t __initcall_start[];
909 extern initcall_entry_t __initcall0_start[];
910 extern initcall_entry_t __initcall1_start[];
911 extern initcall_entry_t __initcall2_start[];
912 extern initcall_entry_t __initcall3_start[];
913 extern initcall_entry_t __initcall4_start[];
914 extern initcall_entry_t __initcall5_start[];
915 extern initcall_entry_t __initcall6_start[];
916 extern initcall_entry_t __initcall7_start[];
917 extern initcall_entry_t __initcall_end[];
918 
919 static initcall_entry_t *initcall_levels[] __initdata = {
920 	__initcall0_start,
921 	__initcall1_start,
922 	__initcall2_start,
923 	__initcall3_start,
924 	__initcall4_start,
925 	__initcall5_start,
926 	__initcall6_start,
927 	__initcall7_start,
928 	__initcall_end,
929 };
930 
931 /* Keep these in sync with initcalls in include/linux/init.h */
932 static const char *initcall_level_names[] __initdata = {
933 	"pure",
934 	"core",
935 	"postcore",
936 	"arch",
937 	"subsys",
938 	"fs",
939 	"device",
940 	"late",
941 };
942 
943 static void __init do_initcall_level(int level)
944 {
945 	initcall_entry_t *fn;
946 
947 	strcpy(initcall_command_line, saved_command_line);
948 	parse_args(initcall_level_names[level],
949 		   initcall_command_line, __start___param,
950 		   __stop___param - __start___param,
951 		   level, level,
952 		   NULL, &repair_env_string);
953 
954 	trace_initcall_level(initcall_level_names[level]);
955 	for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
956 		do_one_initcall(initcall_from_entry(fn));
957 }
958 
959 static void __init do_initcalls(void)
960 {
961 	int level;
962 
963 	for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
964 		do_initcall_level(level);
965 }
966 
967 /*
968  * Ok, the machine is now initialized. None of the devices
969  * have been touched yet, but the CPU subsystem is up and
970  * running, and memory and process management works.
971  *
972  * Now we can finally start doing some real work..
973  */
974 static void __init do_basic_setup(void)
975 {
976 	cpuset_init_smp();
977 	shmem_init();
978 	driver_init();
979 	init_irq_proc();
980 	do_ctors();
981 	usermodehelper_enable();
982 	do_initcalls();
983 }
984 
985 static void __init do_pre_smp_initcalls(void)
986 {
987 	initcall_entry_t *fn;
988 
989 	trace_initcall_level("early");
990 	for (fn = __initcall_start; fn < __initcall0_start; fn++)
991 		do_one_initcall(initcall_from_entry(fn));
992 }
993 
994 static int run_init_process(const char *init_filename)
995 {
996 	argv_init[0] = init_filename;
997 	pr_info("Run %s as init process\n", init_filename);
998 	return do_execve(getname_kernel(init_filename),
999 		(const char __user *const __user *)argv_init,
1000 		(const char __user *const __user *)envp_init);
1001 }
1002 
1003 static int try_to_run_init_process(const char *init_filename)
1004 {
1005 	int ret;
1006 
1007 	ret = run_init_process(init_filename);
1008 
1009 	if (ret && ret != -ENOENT) {
1010 		pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
1011 		       init_filename, ret);
1012 	}
1013 
1014 	return ret;
1015 }
1016 
1017 static noinline void __init kernel_init_freeable(void);
1018 
1019 #if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
1020 bool rodata_enabled __ro_after_init = true;
1021 static int __init set_debug_rodata(char *str)
1022 {
1023 	return strtobool(str, &rodata_enabled);
1024 }
1025 __setup("rodata=", set_debug_rodata);
1026 #endif
1027 
1028 #ifdef CONFIG_STRICT_KERNEL_RWX
1029 static void mark_readonly(void)
1030 {
1031 	if (rodata_enabled) {
1032 		/*
1033 		 * load_module() results in W+X mappings, which are cleaned
1034 		 * up with call_rcu().  Let's make sure that queued work is
1035 		 * flushed so that we don't hit false positives looking for
1036 		 * insecure pages which are W+X.
1037 		 */
1038 		rcu_barrier();
1039 		mark_rodata_ro();
1040 		rodata_test();
1041 	} else
1042 		pr_info("Kernel memory protection disabled.\n");
1043 }
1044 #else
1045 static inline void mark_readonly(void)
1046 {
1047 	pr_warn("This architecture does not have kernel memory protection.\n");
1048 }
1049 #endif
1050 
1051 static int __ref kernel_init(void *unused)
1052 {
1053 	int ret;
1054 
1055 	kernel_init_freeable();
1056 	/* need to finish all async __init code before freeing the memory */
1057 	async_synchronize_full();
1058 	ftrace_free_init_mem();
1059 	free_initmem();
1060 	mark_readonly();
1061 
1062 	/*
1063 	 * Kernel mappings are now finalized - update the userspace page-table
1064 	 * to finalize PTI.
1065 	 */
1066 	pti_finalize();
1067 
1068 	system_state = SYSTEM_RUNNING;
1069 	numa_default_policy();
1070 
1071 	rcu_end_inkernel_boot();
1072 
1073 	if (ramdisk_execute_command) {
1074 		ret = run_init_process(ramdisk_execute_command);
1075 		if (!ret)
1076 			return 0;
1077 		pr_err("Failed to execute %s (error %d)\n",
1078 		       ramdisk_execute_command, ret);
1079 	}
1080 
1081 	/*
1082 	 * We try each of these until one succeeds.
1083 	 *
1084 	 * The Bourne shell can be used instead of init if we are
1085 	 * trying to recover a really broken machine.
1086 	 */
1087 	if (execute_command) {
1088 		ret = run_init_process(execute_command);
1089 		if (!ret)
1090 			return 0;
1091 		panic("Requested init %s failed (error %d).",
1092 		      execute_command, ret);
1093 	}
1094 	if (!try_to_run_init_process("/sbin/init") ||
1095 	    !try_to_run_init_process("/etc/init") ||
1096 	    !try_to_run_init_process("/bin/init") ||
1097 	    !try_to_run_init_process("/bin/sh"))
1098 		return 0;
1099 
1100 	panic("No working init found.  Try passing init= option to kernel. "
1101 	      "See Linux Documentation/admin-guide/init.rst for guidance.");
1102 }
1103 
1104 static noinline void __init kernel_init_freeable(void)
1105 {
1106 	/*
1107 	 * Wait until kthreadd is all set-up.
1108 	 */
1109 	wait_for_completion(&kthreadd_done);
1110 
1111 	/* Now the scheduler is fully set up and can do blocking allocations */
1112 	gfp_allowed_mask = __GFP_BITS_MASK;
1113 
1114 	/*
1115 	 * init can allocate pages on any node
1116 	 */
1117 	set_mems_allowed(node_states[N_MEMORY]);
1118 
1119 	cad_pid = task_pid(current);
1120 
1121 	smp_prepare_cpus(setup_max_cpus);
1122 
1123 	workqueue_init();
1124 
1125 	init_mm_internals();
1126 
1127 	do_pre_smp_initcalls();
1128 	lockup_detector_init();
1129 
1130 	smp_init();
1131 	sched_init_smp();
1132 
1133 	page_alloc_init_late();
1134 
1135 	do_basic_setup();
1136 
1137 	/* Open the /dev/console on the rootfs, this should never fail */
1138 	if (ksys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
1139 		pr_err("Warning: unable to open an initial console.\n");
1140 
1141 	(void) ksys_dup(0);
1142 	(void) ksys_dup(0);
1143 	/*
1144 	 * check if there is an early userspace init.  If yes, let it do all
1145 	 * the work
1146 	 */
1147 
1148 	if (!ramdisk_execute_command)
1149 		ramdisk_execute_command = "/init";
1150 
1151 	if (ksys_access((const char __user *)
1152 			ramdisk_execute_command, 0) != 0) {
1153 		ramdisk_execute_command = NULL;
1154 		prepare_namespace();
1155 	}
1156 
1157 	/*
1158 	 * Ok, we have completed the initial bootup, and
1159 	 * we're essentially up and running. Get rid of the
1160 	 * initmem segments and start the user-mode stuff..
1161 	 *
1162 	 * rootfs is available now, try loading the public keys
1163 	 * and default modules
1164 	 */
1165 
1166 	integrity_load_keys();
1167 }
1168