Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/init/main.c
4 *
5 * Copyright (C) 1991, 1992 Linus Torvalds
6 *
7 * GK 2/5/95 - Changed to support mounting root fs via NFS
8 * Added initrd & change_root: Werner Almesberger & Hans Lermen, Feb '96
9 * Moan early if gcc is old, avoiding bogus kernels - Paul Gortmaker, May '96
10 * Simplified starting of init: Michael A. Griffith <grif@acm.org>
11 */
12
13#define DEBUG /* Enable initcall_debug */
14
15#include <linux/types.h>
16#include <linux/extable.h>
17#include <linux/module.h>
18#include <linux/proc_fs.h>
19#include <linux/binfmts.h>
20#include <linux/kernel.h>
21#include <linux/syscalls.h>
22#include <linux/stackprotector.h>
23#include <linux/string.h>
24#include <linux/ctype.h>
25#include <linux/delay.h>
26#include <linux/ioport.h>
27#include <linux/init.h>
28#include <linux/initrd.h>
29#include <linux/memblock.h>
30#include <linux/acpi.h>
31#include <linux/console.h>
32#include <linux/nmi.h>
33#include <linux/percpu.h>
34#include <linux/kmod.h>
35#include <linux/vmalloc.h>
36#include <linux/kernel_stat.h>
37#include <linux/start_kernel.h>
38#include <linux/security.h>
39#include <linux/smp.h>
40#include <linux/profile.h>
41#include <linux/rcupdate.h>
42#include <linux/moduleparam.h>
43#include <linux/kallsyms.h>
44#include <linux/writeback.h>
45#include <linux/cpu.h>
46#include <linux/cpuset.h>
47#include <linux/cgroup.h>
48#include <linux/efi.h>
49#include <linux/tick.h>
50#include <linux/sched/isolation.h>
51#include <linux/interrupt.h>
52#include <linux/taskstats_kern.h>
53#include <linux/delayacct.h>
54#include <linux/unistd.h>
55#include <linux/utsname.h>
56#include <linux/rmap.h>
57#include <linux/mempolicy.h>
58#include <linux/key.h>
59#include <linux/buffer_head.h>
60#include <linux/page_ext.h>
61#include <linux/debug_locks.h>
62#include <linux/debugobjects.h>
63#include <linux/lockdep.h>
64#include <linux/kmemleak.h>
65#include <linux/pid_namespace.h>
66#include <linux/device.h>
67#include <linux/kthread.h>
68#include <linux/sched.h>
69#include <linux/sched/init.h>
70#include <linux/signal.h>
71#include <linux/idr.h>
72#include <linux/kgdb.h>
73#include <linux/ftrace.h>
74#include <linux/async.h>
75#include <linux/sfi.h>
76#include <linux/shmem_fs.h>
77#include <linux/slab.h>
78#include <linux/perf_event.h>
79#include <linux/ptrace.h>
80#include <linux/pti.h>
81#include <linux/blkdev.h>
82#include <linux/elevator.h>
83#include <linux/sched/clock.h>
84#include <linux/sched/task.h>
85#include <linux/sched/task_stack.h>
86#include <linux/context_tracking.h>
87#include <linux/random.h>
88#include <linux/list.h>
89#include <linux/integrity.h>
90#include <linux/proc_ns.h>
91#include <linux/io.h>
92#include <linux/cache.h>
93#include <linux/rodata_test.h>
94#include <linux/jump_label.h>
95#include <linux/mem_encrypt.h>
96
97#include <asm/io.h>
98#include <asm/bugs.h>
99#include <asm/setup.h>
100#include <asm/sections.h>
101#include <asm/cacheflush.h>
102
103#define CREATE_TRACE_POINTS
104#include <trace/events/initcall.h>
105
106static int kernel_init(void *);
107
108extern void init_IRQ(void);
109extern void radix_tree_init(void);
110
111/*
112 * Debug helper: via this flag we know that we are in 'early bootup code'
113 * where only the boot processor is running with IRQ disabled. This means
114 * two things - IRQ must not be enabled before the flag is cleared and some
115 * operations which are not allowed with IRQ disabled are allowed while the
116 * flag is set.
117 */
118bool early_boot_irqs_disabled __read_mostly;
119
120enum system_states system_state __read_mostly;
121EXPORT_SYMBOL(system_state);
122
123/*
124 * Boot command-line arguments
125 */
126#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
127#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
128
129extern void time_init(void);
130/* Default late time init is NULL. archs can override this later. */
131void (*__initdata late_time_init)(void);
132
133/* Untouched command line saved by arch-specific code. */
134char __initdata boot_command_line[COMMAND_LINE_SIZE];
135/* Untouched saved command line (eg. for /proc) */
136char *saved_command_line;
137/* Command line for parameter parsing */
138static char *static_command_line;
139/* Command line for per-initcall parameter parsing */
140static char *initcall_command_line;
141
142static char *execute_command;
143static char *ramdisk_execute_command;
144
145/*
146 * Used to generate warnings if static_key manipulation functions are used
147 * before jump_label_init is called.
148 */
149bool static_key_initialized __read_mostly;
150EXPORT_SYMBOL_GPL(static_key_initialized);
151
152/*
153 * If set, this is an indication to the drivers that reset the underlying
154 * device before going ahead with the initialization otherwise driver might
155 * rely on the BIOS and skip the reset operation.
156 *
157 * This is useful if kernel is booting in an unreliable environment.
158 * For ex. kdump situation where previous kernel has crashed, BIOS has been
159 * skipped and devices will be in unknown state.
160 */
161unsigned int reset_devices;
162EXPORT_SYMBOL(reset_devices);
163
164static int __init set_reset_devices(char *str)
165{
166 reset_devices = 1;
167 return 1;
168}
169
170__setup("reset_devices", set_reset_devices);
171
172static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
173const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
174static const char *panic_later, *panic_param;
175
176extern const struct obs_kernel_param __setup_start[], __setup_end[];
177
178static bool __init obsolete_checksetup(char *line)
179{
180 const struct obs_kernel_param *p;
181 bool had_early_param = false;
182
183 p = __setup_start;
184 do {
185 int n = strlen(p->str);
186 if (parameqn(line, p->str, n)) {
187 if (p->early) {
188 /* Already done in parse_early_param?
189 * (Needs exact match on param part).
190 * Keep iterating, as we can have early
191 * params and __setups of same names 8( */
192 if (line[n] == '\0' || line[n] == '=')
193 had_early_param = true;
194 } else if (!p->setup_func) {
195 pr_warn("Parameter %s is obsolete, ignored\n",
196 p->str);
197 return true;
198 } else if (p->setup_func(line + n))
199 return true;
200 }
201 p++;
202 } while (p < __setup_end);
203
204 return had_early_param;
205}
206
207/*
208 * This should be approx 2 Bo*oMips to start (note initial shift), and will
209 * still work even if initially too large, it will just take slightly longer
210 */
211unsigned long loops_per_jiffy = (1<<12);
212EXPORT_SYMBOL(loops_per_jiffy);
213
214static int __init debug_kernel(char *str)
215{
216 console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
217 return 0;
218}
219
220static int __init quiet_kernel(char *str)
221{
222 console_loglevel = CONSOLE_LOGLEVEL_QUIET;
223 return 0;
224}
225
226early_param("debug", debug_kernel);
227early_param("quiet", quiet_kernel);
228
229static int __init loglevel(char *str)
230{
231 int newlevel;
232
233 /*
234 * Only update loglevel value when a correct setting was passed,
235 * to prevent blind crashes (when loglevel being set to 0) that
236 * are quite hard to debug
237 */
238 if (get_option(&str, &newlevel)) {
239 console_loglevel = newlevel;
240 return 0;
241 }
242
243 return -EINVAL;
244}
245
246early_param("loglevel", loglevel);
247
248/* Change NUL term back to "=", to make "param" the whole string. */
249static int __init repair_env_string(char *param, char *val,
250 const char *unused, void *arg)
251{
252 if (val) {
253 /* param=val or param="val"? */
254 if (val == param+strlen(param)+1)
255 val[-1] = '=';
256 else if (val == param+strlen(param)+2) {
257 val[-2] = '=';
258 memmove(val-1, val, strlen(val)+1);
259 val--;
260 } else
261 BUG();
262 }
263 return 0;
264}
265
266/* Anything after -- gets handed straight to init. */
267static int __init set_init_arg(char *param, char *val,
268 const char *unused, void *arg)
269{
270 unsigned int i;
271
272 if (panic_later)
273 return 0;
274
275 repair_env_string(param, val, unused, NULL);
276
277 for (i = 0; argv_init[i]; i++) {
278 if (i == MAX_INIT_ARGS) {
279 panic_later = "init";
280 panic_param = param;
281 return 0;
282 }
283 }
284 argv_init[i] = param;
285 return 0;
286}
287
288/*
289 * Unknown boot options get handed to init, unless they look like
290 * unused parameters (modprobe will find them in /proc/cmdline).
291 */
292static int __init unknown_bootoption(char *param, char *val,
293 const char *unused, void *arg)
294{
295 repair_env_string(param, val, unused, NULL);
296
297 /* Handle obsolete-style parameters */
298 if (obsolete_checksetup(param))
299 return 0;
300
301 /* Unused module parameter. */
302 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
303 return 0;
304
305 if (panic_later)
306 return 0;
307
308 if (val) {
309 /* Environment option */
310 unsigned int i;
311 for (i = 0; envp_init[i]; i++) {
312 if (i == MAX_INIT_ENVS) {
313 panic_later = "env";
314 panic_param = param;
315 }
316 if (!strncmp(param, envp_init[i], val - param))
317 break;
318 }
319 envp_init[i] = param;
320 } else {
321 /* Command line option */
322 unsigned int i;
323 for (i = 0; argv_init[i]; i++) {
324 if (i == MAX_INIT_ARGS) {
325 panic_later = "init";
326 panic_param = param;
327 }
328 }
329 argv_init[i] = param;
330 }
331 return 0;
332}
333
334static int __init init_setup(char *str)
335{
336 unsigned int i;
337
338 execute_command = str;
339 /*
340 * In case LILO is going to boot us with default command line,
341 * it prepends "auto" before the whole cmdline which makes
342 * the shell think it should execute a script with such name.
343 * So we ignore all arguments entered _before_ init=... [MJ]
344 */
345 for (i = 1; i < MAX_INIT_ARGS; i++)
346 argv_init[i] = NULL;
347 return 1;
348}
349__setup("init=", init_setup);
350
351static int __init rdinit_setup(char *str)
352{
353 unsigned int i;
354
355 ramdisk_execute_command = str;
356 /* See "auto" comment in init_setup */
357 for (i = 1; i < MAX_INIT_ARGS; i++)
358 argv_init[i] = NULL;
359 return 1;
360}
361__setup("rdinit=", rdinit_setup);
362
363#ifndef CONFIG_SMP
364static const unsigned int setup_max_cpus = NR_CPUS;
365static inline void setup_nr_cpu_ids(void) { }
366static inline void smp_prepare_cpus(unsigned int maxcpus) { }
367#endif
368
369/*
370 * We need to store the untouched command line for future reference.
371 * We also need to store the touched command line since the parameter
372 * parsing is performed in place, and we should allow a component to
373 * store reference of name/value for future reference.
374 */
375static void __init setup_command_line(char *command_line)
376{
377 size_t len = strlen(boot_command_line) + 1;
378
379 saved_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
380 if (!saved_command_line)
381 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
382
383 initcall_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
384 if (!initcall_command_line)
385 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
386
387 static_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
388 if (!static_command_line)
389 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
390
391 strcpy(saved_command_line, boot_command_line);
392 strcpy(static_command_line, command_line);
393}
394
395/*
396 * We need to finalize in a non-__init function or else race conditions
397 * between the root thread and the init thread may cause start_kernel to
398 * be reaped by free_initmem before the root thread has proceeded to
399 * cpu_idle.
400 *
401 * gcc-3.4 accidentally inlines this function, so use noinline.
402 */
403
404static __initdata DECLARE_COMPLETION(kthreadd_done);
405
406noinline void __ref rest_init(void)
407{
408 struct task_struct *tsk;
409 int pid;
410
411 rcu_scheduler_starting();
412 /*
413 * We need to spawn init first so that it obtains pid 1, however
414 * the init task will end up wanting to create kthreads, which, if
415 * we schedule it before we create kthreadd, will OOPS.
416 */
417 pid = kernel_thread(kernel_init, NULL, CLONE_FS);
418 /*
419 * Pin init on the boot CPU. Task migration is not properly working
420 * until sched_init_smp() has been run. It will set the allowed
421 * CPUs for init to the non isolated CPUs.
422 */
423 rcu_read_lock();
424 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
425 set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id()));
426 rcu_read_unlock();
427
428 numa_default_policy();
429 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
430 rcu_read_lock();
431 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
432 rcu_read_unlock();
433
434 /*
435 * Enable might_sleep() and smp_processor_id() checks.
436 * They cannot be enabled earlier because with CONFIG_PREEMPTION=y
437 * kernel_thread() would trigger might_sleep() splats. With
438 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled
439 * already, but it's stuck on the kthreadd_done completion.
440 */
441 system_state = SYSTEM_SCHEDULING;
442
443 complete(&kthreadd_done);
444
445 /*
446 * The boot idle thread must execute schedule()
447 * at least once to get things moving:
448 */
449 schedule_preempt_disabled();
450 /* Call into cpu_idle with preempt disabled */
451 cpu_startup_entry(CPUHP_ONLINE);
452}
453
454/* Check for early params. */
455static int __init do_early_param(char *param, char *val,
456 const char *unused, void *arg)
457{
458 const struct obs_kernel_param *p;
459
460 for (p = __setup_start; p < __setup_end; p++) {
461 if ((p->early && parameq(param, p->str)) ||
462 (strcmp(param, "console") == 0 &&
463 strcmp(p->str, "earlycon") == 0)
464 ) {
465 if (p->setup_func(val) != 0)
466 pr_warn("Malformed early option '%s'\n", param);
467 }
468 }
469 /* We accept everything at this stage. */
470 return 0;
471}
472
473void __init parse_early_options(char *cmdline)
474{
475 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
476 do_early_param);
477}
478
479/* Arch code calls this early on, or if not, just before other parsing. */
480void __init parse_early_param(void)
481{
482 static int done __initdata;
483 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
484
485 if (done)
486 return;
487
488 /* All fall through to do_early_param. */
489 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
490 parse_early_options(tmp_cmdline);
491 done = 1;
492}
493
494void __init __weak arch_post_acpi_subsys_init(void) { }
495
496void __init __weak smp_setup_processor_id(void)
497{
498}
499
500# if THREAD_SIZE >= PAGE_SIZE
501void __init __weak thread_stack_cache_init(void)
502{
503}
504#endif
505
506void __init __weak mem_encrypt_init(void) { }
507
508void __init __weak poking_init(void) { }
509
510void __init __weak pgtable_cache_init(void) { }
511
512bool initcall_debug;
513core_param(initcall_debug, initcall_debug, bool, 0644);
514
515#ifdef TRACEPOINTS_ENABLED
516static void __init initcall_debug_enable(void);
517#else
518static inline void initcall_debug_enable(void)
519{
520}
521#endif
522
523/* Report memory auto-initialization states for this boot. */
524static void __init report_meminit(void)
525{
526 const char *stack;
527
528 if (IS_ENABLED(CONFIG_INIT_STACK_ALL))
529 stack = "all";
530 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL))
531 stack = "byref_all";
532 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF))
533 stack = "byref";
534 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_USER))
535 stack = "__user";
536 else
537 stack = "off";
538
539 pr_info("mem auto-init: stack:%s, heap alloc:%s, heap free:%s\n",
540 stack, want_init_on_alloc(GFP_KERNEL) ? "on" : "off",
541 want_init_on_free() ? "on" : "off");
542 if (want_init_on_free())
543 pr_info("mem auto-init: clearing system memory may take some time...\n");
544}
545
546/*
547 * Set up kernel memory allocators
548 */
549static void __init mm_init(void)
550{
551 /*
552 * page_ext requires contiguous pages,
553 * bigger than MAX_ORDER unless SPARSEMEM.
554 */
555 page_ext_init_flatmem();
556 report_meminit();
557 mem_init();
558 kmem_cache_init();
559 kmemleak_init();
560 pgtable_init();
561 debug_objects_mem_init();
562 vmalloc_init();
563 ioremap_huge_init();
564 /* Should be run before the first non-init thread is created */
565 init_espfix_bsp();
566 /* Should be run after espfix64 is set up. */
567 pti_init();
568}
569
570void __init __weak arch_call_rest_init(void)
571{
572 rest_init();
573}
574
575asmlinkage __visible void __init start_kernel(void)
576{
577 char *command_line;
578 char *after_dashes;
579
580 set_task_stack_end_magic(&init_task);
581 smp_setup_processor_id();
582 debug_objects_early_init();
583
584 cgroup_init_early();
585
586 local_irq_disable();
587 early_boot_irqs_disabled = true;
588
589 /*
590 * Interrupts are still disabled. Do necessary setups, then
591 * enable them.
592 */
593 boot_cpu_init();
594 page_address_init();
595 pr_notice("%s", linux_banner);
596 early_security_init();
597 setup_arch(&command_line);
598 setup_command_line(command_line);
599 setup_nr_cpu_ids();
600 setup_per_cpu_areas();
601 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
602 boot_cpu_hotplug_init();
603
604 build_all_zonelists(NULL);
605 page_alloc_init();
606
607 pr_notice("Kernel command line: %s\n", boot_command_line);
608 /* parameters may set static keys */
609 jump_label_init();
610 parse_early_param();
611 after_dashes = parse_args("Booting kernel",
612 static_command_line, __start___param,
613 __stop___param - __start___param,
614 -1, -1, NULL, &unknown_bootoption);
615 if (!IS_ERR_OR_NULL(after_dashes))
616 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
617 NULL, set_init_arg);
618
619 /*
620 * These use large bootmem allocations and must precede
621 * kmem_cache_init()
622 */
623 setup_log_buf(0);
624 vfs_caches_init_early();
625 sort_main_extable();
626 trap_init();
627 mm_init();
628
629 ftrace_init();
630
631 /* trace_printk can be enabled here */
632 early_trace_init();
633
634 /*
635 * Set up the scheduler prior starting any interrupts (such as the
636 * timer interrupt). Full topology setup happens at smp_init()
637 * time - but meanwhile we still have a functioning scheduler.
638 */
639 sched_init();
640 /*
641 * Disable preemption - early bootup scheduling is extremely
642 * fragile until we cpu_idle() for the first time.
643 */
644 preempt_disable();
645 if (WARN(!irqs_disabled(),
646 "Interrupts were enabled *very* early, fixing it\n"))
647 local_irq_disable();
648 radix_tree_init();
649
650 /*
651 * Set up housekeeping before setting up workqueues to allow the unbound
652 * workqueue to take non-housekeeping into account.
653 */
654 housekeeping_init();
655
656 /*
657 * Allow workqueue creation and work item queueing/cancelling
658 * early. Work item execution depends on kthreads and starts after
659 * workqueue_init().
660 */
661 workqueue_init_early();
662
663 rcu_init();
664
665 /* Trace events are available after this */
666 trace_init();
667
668 if (initcall_debug)
669 initcall_debug_enable();
670
671 context_tracking_init();
672 /* init some links before init_ISA_irqs() */
673 early_irq_init();
674 init_IRQ();
675 tick_init();
676 rcu_init_nohz();
677 init_timers();
678 hrtimers_init();
679 softirq_init();
680 timekeeping_init();
681
682 /*
683 * For best initial stack canary entropy, prepare it after:
684 * - setup_arch() for any UEFI RNG entropy and boot cmdline access
685 * - timekeeping_init() for ktime entropy used in rand_initialize()
686 * - rand_initialize() to get any arch-specific entropy like RDRAND
687 * - add_latent_entropy() to get any latent entropy
688 * - adding command line entropy
689 */
690 rand_initialize();
691 add_latent_entropy();
692 add_device_randomness(command_line, strlen(command_line));
693 boot_init_stack_canary();
694
695 time_init();
696 printk_safe_init();
697 perf_event_init();
698 profile_init();
699 call_function_init();
700 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
701
702 early_boot_irqs_disabled = false;
703 local_irq_enable();
704
705 kmem_cache_init_late();
706
707 /*
708 * HACK ALERT! This is early. We're enabling the console before
709 * we've done PCI setups etc, and console_init() must be aware of
710 * this. But we do want output early, in case something goes wrong.
711 */
712 console_init();
713 if (panic_later)
714 panic("Too many boot %s vars at `%s'", panic_later,
715 panic_param);
716
717 lockdep_init();
718
719 /*
720 * Need to run this when irqs are enabled, because it wants
721 * to self-test [hard/soft]-irqs on/off lock inversion bugs
722 * too:
723 */
724 locking_selftest();
725
726 /*
727 * This needs to be called before any devices perform DMA
728 * operations that might use the SWIOTLB bounce buffers. It will
729 * mark the bounce buffers as decrypted so that their usage will
730 * not cause "plain-text" data to be decrypted when accessed.
731 */
732 mem_encrypt_init();
733
734#ifdef CONFIG_BLK_DEV_INITRD
735 if (initrd_start && !initrd_below_start_ok &&
736 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
737 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
738 page_to_pfn(virt_to_page((void *)initrd_start)),
739 min_low_pfn);
740 initrd_start = 0;
741 }
742#endif
743 setup_per_cpu_pageset();
744 numa_policy_init();
745 acpi_early_init();
746 if (late_time_init)
747 late_time_init();
748 sched_clock_init();
749 calibrate_delay();
750 pid_idr_init();
751 anon_vma_init();
752#ifdef CONFIG_X86
753 if (efi_enabled(EFI_RUNTIME_SERVICES))
754 efi_enter_virtual_mode();
755#endif
756 thread_stack_cache_init();
757 cred_init();
758 fork_init();
759 proc_caches_init();
760 uts_ns_init();
761 buffer_init();
762 key_init();
763 security_init();
764 dbg_late_init();
765 vfs_caches_init();
766 pagecache_init();
767 signals_init();
768 seq_file_init();
769 proc_root_init();
770 nsfs_init();
771 cpuset_init();
772 cgroup_init();
773 taskstats_init_early();
774 delayacct_init();
775
776 poking_init();
777 check_bugs();
778
779 acpi_subsystem_init();
780 arch_post_acpi_subsys_init();
781 sfi_init_late();
782
783 /* Do the rest non-__init'ed, we're now alive */
784 arch_call_rest_init();
785}
786
787/* Call all constructor functions linked into the kernel. */
788static void __init do_ctors(void)
789{
790#ifdef CONFIG_CONSTRUCTORS
791 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
792
793 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
794 (*fn)();
795#endif
796}
797
798#ifdef CONFIG_KALLSYMS
799struct blacklist_entry {
800 struct list_head next;
801 char *buf;
802};
803
804static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
805
806static int __init initcall_blacklist(char *str)
807{
808 char *str_entry;
809 struct blacklist_entry *entry;
810
811 /* str argument is a comma-separated list of functions */
812 do {
813 str_entry = strsep(&str, ",");
814 if (str_entry) {
815 pr_debug("blacklisting initcall %s\n", str_entry);
816 entry = memblock_alloc(sizeof(*entry),
817 SMP_CACHE_BYTES);
818 if (!entry)
819 panic("%s: Failed to allocate %zu bytes\n",
820 __func__, sizeof(*entry));
821 entry->buf = memblock_alloc(strlen(str_entry) + 1,
822 SMP_CACHE_BYTES);
823 if (!entry->buf)
824 panic("%s: Failed to allocate %zu bytes\n",
825 __func__, strlen(str_entry) + 1);
826 strcpy(entry->buf, str_entry);
827 list_add(&entry->next, &blacklisted_initcalls);
828 }
829 } while (str_entry);
830
831 return 0;
832}
833
834static bool __init_or_module initcall_blacklisted(initcall_t fn)
835{
836 struct blacklist_entry *entry;
837 char fn_name[KSYM_SYMBOL_LEN];
838 unsigned long addr;
839
840 if (list_empty(&blacklisted_initcalls))
841 return false;
842
843 addr = (unsigned long) dereference_function_descriptor(fn);
844 sprint_symbol_no_offset(fn_name, addr);
845
846 /*
847 * fn will be "function_name [module_name]" where [module_name] is not
848 * displayed for built-in init functions. Strip off the [module_name].
849 */
850 strreplace(fn_name, ' ', '\0');
851
852 list_for_each_entry(entry, &blacklisted_initcalls, next) {
853 if (!strcmp(fn_name, entry->buf)) {
854 pr_debug("initcall %s blacklisted\n", fn_name);
855 return true;
856 }
857 }
858
859 return false;
860}
861#else
862static int __init initcall_blacklist(char *str)
863{
864 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
865 return 0;
866}
867
868static bool __init_or_module initcall_blacklisted(initcall_t fn)
869{
870 return false;
871}
872#endif
873__setup("initcall_blacklist=", initcall_blacklist);
874
875static __init_or_module void
876trace_initcall_start_cb(void *data, initcall_t fn)
877{
878 ktime_t *calltime = (ktime_t *)data;
879
880 printk(KERN_DEBUG "calling %pS @ %i\n", fn, task_pid_nr(current));
881 *calltime = ktime_get();
882}
883
884static __init_or_module void
885trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
886{
887 ktime_t *calltime = (ktime_t *)data;
888 ktime_t delta, rettime;
889 unsigned long long duration;
890
891 rettime = ktime_get();
892 delta = ktime_sub(rettime, *calltime);
893 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
894 printk(KERN_DEBUG "initcall %pS returned %d after %lld usecs\n",
895 fn, ret, duration);
896}
897
898static ktime_t initcall_calltime;
899
900#ifdef TRACEPOINTS_ENABLED
901static void __init initcall_debug_enable(void)
902{
903 int ret;
904
905 ret = register_trace_initcall_start(trace_initcall_start_cb,
906 &initcall_calltime);
907 ret |= register_trace_initcall_finish(trace_initcall_finish_cb,
908 &initcall_calltime);
909 WARN(ret, "Failed to register initcall tracepoints\n");
910}
911# define do_trace_initcall_start trace_initcall_start
912# define do_trace_initcall_finish trace_initcall_finish
913#else
914static inline void do_trace_initcall_start(initcall_t fn)
915{
916 if (!initcall_debug)
917 return;
918 trace_initcall_start_cb(&initcall_calltime, fn);
919}
920static inline void do_trace_initcall_finish(initcall_t fn, int ret)
921{
922 if (!initcall_debug)
923 return;
924 trace_initcall_finish_cb(&initcall_calltime, fn, ret);
925}
926#endif /* !TRACEPOINTS_ENABLED */
927
928int __init_or_module do_one_initcall(initcall_t fn)
929{
930 int count = preempt_count();
931 char msgbuf[64];
932 int ret;
933
934 if (initcall_blacklisted(fn))
935 return -EPERM;
936
937 do_trace_initcall_start(fn);
938 ret = fn();
939 do_trace_initcall_finish(fn, ret);
940
941 msgbuf[0] = 0;
942
943 if (preempt_count() != count) {
944 sprintf(msgbuf, "preemption imbalance ");
945 preempt_count_set(count);
946 }
947 if (irqs_disabled()) {
948 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
949 local_irq_enable();
950 }
951 WARN(msgbuf[0], "initcall %pS returned with %s\n", fn, msgbuf);
952
953 add_latent_entropy();
954 return ret;
955}
956
957
958extern initcall_entry_t __initcall_start[];
959extern initcall_entry_t __initcall0_start[];
960extern initcall_entry_t __initcall1_start[];
961extern initcall_entry_t __initcall2_start[];
962extern initcall_entry_t __initcall3_start[];
963extern initcall_entry_t __initcall4_start[];
964extern initcall_entry_t __initcall5_start[];
965extern initcall_entry_t __initcall6_start[];
966extern initcall_entry_t __initcall7_start[];
967extern initcall_entry_t __initcall_end[];
968
969static initcall_entry_t *initcall_levels[] __initdata = {
970 __initcall0_start,
971 __initcall1_start,
972 __initcall2_start,
973 __initcall3_start,
974 __initcall4_start,
975 __initcall5_start,
976 __initcall6_start,
977 __initcall7_start,
978 __initcall_end,
979};
980
981/* Keep these in sync with initcalls in include/linux/init.h */
982static const char *initcall_level_names[] __initdata = {
983 "pure",
984 "core",
985 "postcore",
986 "arch",
987 "subsys",
988 "fs",
989 "device",
990 "late",
991};
992
993static void __init do_initcall_level(int level)
994{
995 initcall_entry_t *fn;
996
997 strcpy(initcall_command_line, saved_command_line);
998 parse_args(initcall_level_names[level],
999 initcall_command_line, __start___param,
1000 __stop___param - __start___param,
1001 level, level,
1002 NULL, &repair_env_string);
1003
1004 trace_initcall_level(initcall_level_names[level]);
1005 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
1006 do_one_initcall(initcall_from_entry(fn));
1007}
1008
1009static void __init do_initcalls(void)
1010{
1011 int level;
1012
1013 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
1014 do_initcall_level(level);
1015}
1016
1017/*
1018 * Ok, the machine is now initialized. None of the devices
1019 * have been touched yet, but the CPU subsystem is up and
1020 * running, and memory and process management works.
1021 *
1022 * Now we can finally start doing some real work..
1023 */
1024static void __init do_basic_setup(void)
1025{
1026 cpuset_init_smp();
1027 driver_init();
1028 init_irq_proc();
1029 do_ctors();
1030 usermodehelper_enable();
1031 do_initcalls();
1032}
1033
1034static void __init do_pre_smp_initcalls(void)
1035{
1036 initcall_entry_t *fn;
1037
1038 trace_initcall_level("early");
1039 for (fn = __initcall_start; fn < __initcall0_start; fn++)
1040 do_one_initcall(initcall_from_entry(fn));
1041}
1042
1043static int run_init_process(const char *init_filename)
1044{
1045 argv_init[0] = init_filename;
1046 pr_info("Run %s as init process\n", init_filename);
1047 return do_execve(getname_kernel(init_filename),
1048 (const char __user *const __user *)argv_init,
1049 (const char __user *const __user *)envp_init);
1050}
1051
1052static int try_to_run_init_process(const char *init_filename)
1053{
1054 int ret;
1055
1056 ret = run_init_process(init_filename);
1057
1058 if (ret && ret != -ENOENT) {
1059 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
1060 init_filename, ret);
1061 }
1062
1063 return ret;
1064}
1065
1066static noinline void __init kernel_init_freeable(void);
1067
1068#if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
1069bool rodata_enabled __ro_after_init = true;
1070static int __init set_debug_rodata(char *str)
1071{
1072 return strtobool(str, &rodata_enabled);
1073}
1074__setup("rodata=", set_debug_rodata);
1075#endif
1076
1077#ifdef CONFIG_STRICT_KERNEL_RWX
1078static void mark_readonly(void)
1079{
1080 if (rodata_enabled) {
1081 /*
1082 * load_module() results in W+X mappings, which are cleaned
1083 * up with call_rcu(). Let's make sure that queued work is
1084 * flushed so that we don't hit false positives looking for
1085 * insecure pages which are W+X.
1086 */
1087 rcu_barrier();
1088 mark_rodata_ro();
1089 rodata_test();
1090 } else
1091 pr_info("Kernel memory protection disabled.\n");
1092}
1093#else
1094static inline void mark_readonly(void)
1095{
1096 pr_warn("This architecture does not have kernel memory protection.\n");
1097}
1098#endif
1099
1100void __weak free_initmem(void)
1101{
1102 free_initmem_default(POISON_FREE_INITMEM);
1103}
1104
1105static int __ref kernel_init(void *unused)
1106{
1107 int ret;
1108
1109 kernel_init_freeable();
1110 /* need to finish all async __init code before freeing the memory */
1111 async_synchronize_full();
1112 ftrace_free_init_mem();
1113 free_initmem();
1114 mark_readonly();
1115
1116 /*
1117 * Kernel mappings are now finalized - update the userspace page-table
1118 * to finalize PTI.
1119 */
1120 pti_finalize();
1121
1122 system_state = SYSTEM_RUNNING;
1123 numa_default_policy();
1124
1125 rcu_end_inkernel_boot();
1126
1127 if (ramdisk_execute_command) {
1128 ret = run_init_process(ramdisk_execute_command);
1129 if (!ret)
1130 return 0;
1131 pr_err("Failed to execute %s (error %d)\n",
1132 ramdisk_execute_command, ret);
1133 }
1134
1135 /*
1136 * We try each of these until one succeeds.
1137 *
1138 * The Bourne shell can be used instead of init if we are
1139 * trying to recover a really broken machine.
1140 */
1141 if (execute_command) {
1142 ret = run_init_process(execute_command);
1143 if (!ret)
1144 return 0;
1145 panic("Requested init %s failed (error %d).",
1146 execute_command, ret);
1147 }
1148 if (!try_to_run_init_process("/sbin/init") ||
1149 !try_to_run_init_process("/etc/init") ||
1150 !try_to_run_init_process("/bin/init") ||
1151 !try_to_run_init_process("/bin/sh"))
1152 return 0;
1153
1154 panic("No working init found. Try passing init= option to kernel. "
1155 "See Linux Documentation/admin-guide/init.rst for guidance.");
1156}
1157
1158static noinline void __init kernel_init_freeable(void)
1159{
1160 /*
1161 * Wait until kthreadd is all set-up.
1162 */
1163 wait_for_completion(&kthreadd_done);
1164
1165 /* Now the scheduler is fully set up and can do blocking allocations */
1166 gfp_allowed_mask = __GFP_BITS_MASK;
1167
1168 /*
1169 * init can allocate pages on any node
1170 */
1171 set_mems_allowed(node_states[N_MEMORY]);
1172
1173 cad_pid = task_pid(current);
1174
1175 smp_prepare_cpus(setup_max_cpus);
1176
1177 workqueue_init();
1178
1179 init_mm_internals();
1180
1181 do_pre_smp_initcalls();
1182 lockup_detector_init();
1183
1184 smp_init();
1185 sched_init_smp();
1186
1187 page_alloc_init_late();
1188 /* Initialize page ext after all struct pages are initialized. */
1189 page_ext_init();
1190
1191 do_basic_setup();
1192
1193 /* Open the /dev/console on the rootfs, this should never fail */
1194 if (ksys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
1195 pr_err("Warning: unable to open an initial console.\n");
1196
1197 (void) ksys_dup(0);
1198 (void) ksys_dup(0);
1199 /*
1200 * check if there is an early userspace init. If yes, let it do all
1201 * the work
1202 */
1203
1204 if (!ramdisk_execute_command)
1205 ramdisk_execute_command = "/init";
1206
1207 if (ksys_access((const char __user *)
1208 ramdisk_execute_command, 0) != 0) {
1209 ramdisk_execute_command = NULL;
1210 prepare_namespace();
1211 }
1212
1213 /*
1214 * Ok, we have completed the initial bootup, and
1215 * we're essentially up and running. Get rid of the
1216 * initmem segments and start the user-mode stuff..
1217 *
1218 * rootfs is available now, try loading the public keys
1219 * and default modules
1220 */
1221
1222 integrity_load_keys();
1223}
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
91static int kernel_init(void *);
92
93extern void init_IRQ(void);
94extern void fork_init(unsigned long);
95extern void radix_tree_init(void);
96#ifndef CONFIG_DEBUG_RODATA
97static inline void mark_rodata_ro(void) { }
98#endif
99
100/*
101 * Debug helper: via this flag we know that we are in 'early bootup code'
102 * where only the boot processor is running with IRQ disabled. This means
103 * two things - IRQ must not be enabled before the flag is cleared and some
104 * operations which are not allowed with IRQ disabled are allowed while the
105 * flag is set.
106 */
107bool early_boot_irqs_disabled __read_mostly;
108
109enum system_states system_state __read_mostly;
110EXPORT_SYMBOL(system_state);
111
112/*
113 * Boot command-line arguments
114 */
115#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
116#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
117
118extern void time_init(void);
119/* Default late time init is NULL. archs can override this later. */
120void (*__initdata late_time_init)(void);
121
122/* Untouched command line saved by arch-specific code. */
123char __initdata boot_command_line[COMMAND_LINE_SIZE];
124/* Untouched saved command line (eg. for /proc) */
125char *saved_command_line;
126/* Command line for parameter parsing */
127static char *static_command_line;
128/* Command line for per-initcall parameter parsing */
129static char *initcall_command_line;
130
131static char *execute_command;
132static char *ramdisk_execute_command;
133
134/*
135 * Used to generate warnings if static_key manipulation functions are used
136 * before jump_label_init is called.
137 */
138bool static_key_initialized __read_mostly = false;
139EXPORT_SYMBOL_GPL(static_key_initialized);
140
141/*
142 * If set, this is an indication to the drivers that reset the underlying
143 * device before going ahead with the initialization otherwise driver might
144 * rely on the BIOS and skip the reset operation.
145 *
146 * This is useful if kernel is booting in an unreliable environment.
147 * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
148 * skipped and devices will be in unknown state.
149 */
150unsigned int reset_devices;
151EXPORT_SYMBOL(reset_devices);
152
153static int __init set_reset_devices(char *str)
154{
155 reset_devices = 1;
156 return 1;
157}
158
159__setup("reset_devices", set_reset_devices);
160
161static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
162const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
163static const char *panic_later, *panic_param;
164
165extern const struct obs_kernel_param __setup_start[], __setup_end[];
166
167static int __init obsolete_checksetup(char *line)
168{
169 const struct obs_kernel_param *p;
170 int had_early_param = 0;
171
172 p = __setup_start;
173 do {
174 int n = strlen(p->str);
175 if (parameqn(line, p->str, n)) {
176 if (p->early) {
177 /* Already done in parse_early_param?
178 * (Needs exact match on param part).
179 * Keep iterating, as we can have early
180 * params and __setups of same names 8( */
181 if (line[n] == '\0' || line[n] == '=')
182 had_early_param = 1;
183 } else if (!p->setup_func) {
184 pr_warn("Parameter %s is obsolete, ignored\n",
185 p->str);
186 return 1;
187 } else if (p->setup_func(line + n))
188 return 1;
189 }
190 p++;
191 } while (p < __setup_end);
192
193 return had_early_param;
194}
195
196/*
197 * This should be approx 2 Bo*oMips to start (note initial shift), and will
198 * still work even if initially too large, it will just take slightly longer
199 */
200unsigned long loops_per_jiffy = (1<<12);
201
202EXPORT_SYMBOL(loops_per_jiffy);
203
204static int __init debug_kernel(char *str)
205{
206 console_loglevel = 10;
207 return 0;
208}
209
210static int __init quiet_kernel(char *str)
211{
212 console_loglevel = 4;
213 return 0;
214}
215
216early_param("debug", debug_kernel);
217early_param("quiet", quiet_kernel);
218
219static int __init loglevel(char *str)
220{
221 int newlevel;
222
223 /*
224 * Only update loglevel value when a correct setting was passed,
225 * to prevent blind crashes (when loglevel being set to 0) that
226 * are quite hard to debug
227 */
228 if (get_option(&str, &newlevel)) {
229 console_loglevel = newlevel;
230 return 0;
231 }
232
233 return -EINVAL;
234}
235
236early_param("loglevel", loglevel);
237
238/* Change NUL term back to "=", to make "param" the whole string. */
239static int __init repair_env_string(char *param, char *val, const char *unused)
240{
241 if (val) {
242 /* param=val or param="val"? */
243 if (val == param+strlen(param)+1)
244 val[-1] = '=';
245 else if (val == param+strlen(param)+2) {
246 val[-2] = '=';
247 memmove(val-1, val, strlen(val)+1);
248 val--;
249 } else
250 BUG();
251 }
252 return 0;
253}
254
255/*
256 * Unknown boot options get handed to init, unless they look like
257 * unused parameters (modprobe will find them in /proc/cmdline).
258 */
259static int __init unknown_bootoption(char *param, char *val, const char *unused)
260{
261 repair_env_string(param, val, unused);
262
263 /* Handle obsolete-style parameters */
264 if (obsolete_checksetup(param))
265 return 0;
266
267 /* Unused module parameter. */
268 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
269 return 0;
270
271 if (panic_later)
272 return 0;
273
274 if (val) {
275 /* Environment option */
276 unsigned int i;
277 for (i = 0; envp_init[i]; i++) {
278 if (i == MAX_INIT_ENVS) {
279 panic_later = "env";
280 panic_param = param;
281 }
282 if (!strncmp(param, envp_init[i], val - param))
283 break;
284 }
285 envp_init[i] = param;
286 } else {
287 /* Command line option */
288 unsigned int i;
289 for (i = 0; argv_init[i]; i++) {
290 if (i == MAX_INIT_ARGS) {
291 panic_later = "init";
292 panic_param = param;
293 }
294 }
295 argv_init[i] = param;
296 }
297 return 0;
298}
299
300static int __init init_setup(char *str)
301{
302 unsigned int i;
303
304 execute_command = str;
305 /*
306 * In case LILO is going to boot us with default command line,
307 * it prepends "auto" before the whole cmdline which makes
308 * the shell think it should execute a script with such name.
309 * So we ignore all arguments entered _before_ init=... [MJ]
310 */
311 for (i = 1; i < MAX_INIT_ARGS; i++)
312 argv_init[i] = NULL;
313 return 1;
314}
315__setup("init=", init_setup);
316
317static int __init rdinit_setup(char *str)
318{
319 unsigned int i;
320
321 ramdisk_execute_command = str;
322 /* See "auto" comment in init_setup */
323 for (i = 1; i < MAX_INIT_ARGS; i++)
324 argv_init[i] = NULL;
325 return 1;
326}
327__setup("rdinit=", rdinit_setup);
328
329#ifndef CONFIG_SMP
330static const unsigned int setup_max_cpus = NR_CPUS;
331#ifdef CONFIG_X86_LOCAL_APIC
332static void __init smp_init(void)
333{
334 APIC_init_uniprocessor();
335}
336#else
337#define smp_init() do { } while (0)
338#endif
339
340static inline void setup_nr_cpu_ids(void) { }
341static inline void smp_prepare_cpus(unsigned int maxcpus) { }
342#endif
343
344/*
345 * We need to store the untouched command line for future reference.
346 * We also need to store the touched command line since the parameter
347 * parsing is performed in place, and we should allow a component to
348 * store reference of name/value for future reference.
349 */
350static void __init setup_command_line(char *command_line)
351{
352 saved_command_line =
353 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
354 initcall_command_line =
355 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
356 static_command_line = memblock_virt_alloc(strlen(command_line) + 1, 0);
357 strcpy (saved_command_line, boot_command_line);
358 strcpy (static_command_line, command_line);
359}
360
361/*
362 * We need to finalize in a non-__init function or else race conditions
363 * between the root thread and the init thread may cause start_kernel to
364 * be reaped by free_initmem before the root thread has proceeded to
365 * cpu_idle.
366 *
367 * gcc-3.4 accidentally inlines this function, so use noinline.
368 */
369
370static __initdata DECLARE_COMPLETION(kthreadd_done);
371
372static noinline void __init_refok rest_init(void)
373{
374 int pid;
375
376 rcu_scheduler_starting();
377 /*
378 * We need to spawn init first so that it obtains pid 1, however
379 * the init task will end up wanting to create kthreads, which, if
380 * we schedule it before we create kthreadd, will OOPS.
381 */
382 kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
383 numa_default_policy();
384 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
385 rcu_read_lock();
386 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
387 rcu_read_unlock();
388 complete(&kthreadd_done);
389
390 /*
391 * The boot idle thread must execute schedule()
392 * at least once to get things moving:
393 */
394 init_idle_bootup_task(current);
395 schedule_preempt_disabled();
396 /* Call into cpu_idle with preempt disabled */
397 cpu_startup_entry(CPUHP_ONLINE);
398}
399
400/* Check for early params. */
401static int __init do_early_param(char *param, char *val, const char *unused)
402{
403 const struct obs_kernel_param *p;
404
405 for (p = __setup_start; p < __setup_end; p++) {
406 if ((p->early && parameq(param, p->str)) ||
407 (strcmp(param, "console") == 0 &&
408 strcmp(p->str, "earlycon") == 0)
409 ) {
410 if (p->setup_func(val) != 0)
411 pr_warn("Malformed early option '%s'\n", param);
412 }
413 }
414 /* We accept everything at this stage. */
415 return 0;
416}
417
418void __init parse_early_options(char *cmdline)
419{
420 parse_args("early options", cmdline, NULL, 0, 0, 0, do_early_param);
421}
422
423/* Arch code calls this early on, or if not, just before other parsing. */
424void __init parse_early_param(void)
425{
426 static __initdata int done = 0;
427 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
428
429 if (done)
430 return;
431
432 /* All fall through to do_early_param. */
433 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
434 parse_early_options(tmp_cmdline);
435 done = 1;
436}
437
438/*
439 * Activate the first processor.
440 */
441
442static void __init boot_cpu_init(void)
443{
444 int cpu = smp_processor_id();
445 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
446 set_cpu_online(cpu, true);
447 set_cpu_active(cpu, true);
448 set_cpu_present(cpu, true);
449 set_cpu_possible(cpu, true);
450}
451
452void __init __weak smp_setup_processor_id(void)
453{
454}
455
456# if THREAD_SIZE >= PAGE_SIZE
457void __init __weak thread_info_cache_init(void)
458{
459}
460#endif
461
462/*
463 * Set up kernel memory allocators
464 */
465static void __init mm_init(void)
466{
467 /*
468 * page_cgroup requires contiguous pages,
469 * bigger than MAX_ORDER unless SPARSEMEM.
470 */
471 page_cgroup_init_flatmem();
472 mem_init();
473 kmem_cache_init();
474 percpu_init_late();
475 pgtable_init();
476 vmalloc_init();
477}
478
479asmlinkage __visible void __init start_kernel(void)
480{
481 char * command_line;
482 extern const struct kernel_param __start___param[], __stop___param[];
483
484 /*
485 * Need to run as early as possible, to initialize the
486 * lockdep hash:
487 */
488 lockdep_init();
489 smp_setup_processor_id();
490 debug_objects_early_init();
491
492 /*
493 * Set up the the initial canary ASAP:
494 */
495 boot_init_stack_canary();
496
497 cgroup_init_early();
498
499 local_irq_disable();
500 early_boot_irqs_disabled = true;
501
502/*
503 * Interrupts are still disabled. Do necessary setups, then
504 * enable them
505 */
506 boot_cpu_init();
507 page_address_init();
508 pr_notice("%s", linux_banner);
509 setup_arch(&command_line);
510 mm_init_owner(&init_mm, &init_task);
511 mm_init_cpumask(&init_mm);
512 setup_command_line(command_line);
513 setup_nr_cpu_ids();
514 setup_per_cpu_areas();
515 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
516
517 build_all_zonelists(NULL, NULL);
518 page_alloc_init();
519
520 pr_notice("Kernel command line: %s\n", boot_command_line);
521 parse_early_param();
522 parse_args("Booting kernel", static_command_line, __start___param,
523 __stop___param - __start___param,
524 -1, -1, &unknown_bootoption);
525
526 jump_label_init();
527
528 /*
529 * These use large bootmem allocations and must precede
530 * kmem_cache_init()
531 */
532 setup_log_buf(0);
533 pidhash_init();
534 vfs_caches_init_early();
535 sort_main_extable();
536 trap_init();
537 mm_init();
538
539 /*
540 * Set up the scheduler prior starting any interrupts (such as the
541 * timer interrupt). Full topology setup happens at smp_init()
542 * time - but meanwhile we still have a functioning scheduler.
543 */
544 sched_init();
545 /*
546 * Disable preemption - early bootup scheduling is extremely
547 * fragile until we cpu_idle() for the first time.
548 */
549 preempt_disable();
550 if (WARN(!irqs_disabled(), "Interrupts were enabled *very* early, fixing it\n"))
551 local_irq_disable();
552 idr_init_cache();
553 rcu_init();
554 tick_nohz_init();
555 context_tracking_init();
556 radix_tree_init();
557 /* init some links before init_ISA_irqs() */
558 early_irq_init();
559 init_IRQ();
560 tick_init();
561 init_timers();
562 hrtimers_init();
563 softirq_init();
564 timekeeping_init();
565 time_init();
566 sched_clock_postinit();
567 perf_event_init();
568 profile_init();
569 call_function_init();
570 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
571 early_boot_irqs_disabled = false;
572 local_irq_enable();
573
574 kmem_cache_init_late();
575
576 /*
577 * HACK ALERT! This is early. We're enabling the console before
578 * we've done PCI setups etc, and console_init() must be aware of
579 * this. But we do want output early, in case something goes wrong.
580 */
581 console_init();
582 if (panic_later)
583 panic("Too many boot %s vars at `%s'", panic_later,
584 panic_param);
585
586 lockdep_info();
587
588 /*
589 * Need to run this when irqs are enabled, because it wants
590 * to self-test [hard/soft]-irqs on/off lock inversion bugs
591 * too:
592 */
593 locking_selftest();
594
595#ifdef CONFIG_BLK_DEV_INITRD
596 if (initrd_start && !initrd_below_start_ok &&
597 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
598 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
599 page_to_pfn(virt_to_page((void *)initrd_start)),
600 min_low_pfn);
601 initrd_start = 0;
602 }
603#endif
604 page_cgroup_init();
605 debug_objects_mem_init();
606 kmemleak_init();
607 setup_per_cpu_pageset();
608 numa_policy_init();
609 if (late_time_init)
610 late_time_init();
611 sched_clock_init();
612 calibrate_delay();
613 pidmap_init();
614 anon_vma_init();
615 acpi_early_init();
616#ifdef CONFIG_X86
617 if (efi_enabled(EFI_RUNTIME_SERVICES))
618 efi_enter_virtual_mode();
619#endif
620 thread_info_cache_init();
621 cred_init();
622 fork_init(totalram_pages);
623 proc_caches_init();
624 buffer_init();
625 key_init();
626 security_init();
627 dbg_late_init();
628 vfs_caches_init(totalram_pages);
629 signals_init();
630 /* rootfs populating might need page-writeback */
631 page_writeback_init();
632#ifdef CONFIG_PROC_FS
633 proc_root_init();
634#endif
635 cgroup_init();
636 cpuset_init();
637 taskstats_init_early();
638 delayacct_init();
639
640 check_bugs();
641
642 sfi_init_late();
643
644 if (efi_enabled(EFI_RUNTIME_SERVICES)) {
645 efi_late_init();
646 efi_free_boot_services();
647 }
648
649 ftrace_init();
650
651 /* Do the rest non-__init'ed, we're now alive */
652 rest_init();
653}
654
655/* Call all constructor functions linked into the kernel. */
656static void __init do_ctors(void)
657{
658#ifdef CONFIG_CONSTRUCTORS
659 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
660
661 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
662 (*fn)();
663#endif
664}
665
666bool initcall_debug;
667core_param(initcall_debug, initcall_debug, bool, 0644);
668
669static int __init_or_module do_one_initcall_debug(initcall_t fn)
670{
671 ktime_t calltime, delta, rettime;
672 unsigned long long duration;
673 int ret;
674
675 pr_debug("calling %pF @ %i\n", fn, task_pid_nr(current));
676 calltime = ktime_get();
677 ret = fn();
678 rettime = ktime_get();
679 delta = ktime_sub(rettime, calltime);
680 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
681 pr_debug("initcall %pF returned %d after %lld usecs\n",
682 fn, ret, duration);
683
684 return ret;
685}
686
687int __init_or_module do_one_initcall(initcall_t fn)
688{
689 int count = preempt_count();
690 int ret;
691 char msgbuf[64];
692
693 if (initcall_debug)
694 ret = do_one_initcall_debug(fn);
695 else
696 ret = fn();
697
698 msgbuf[0] = 0;
699
700 if (preempt_count() != count) {
701 sprintf(msgbuf, "preemption imbalance ");
702 preempt_count_set(count);
703 }
704 if (irqs_disabled()) {
705 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
706 local_irq_enable();
707 }
708 WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
709
710 return ret;
711}
712
713
714extern initcall_t __initcall_start[];
715extern initcall_t __initcall0_start[];
716extern initcall_t __initcall1_start[];
717extern initcall_t __initcall2_start[];
718extern initcall_t __initcall3_start[];
719extern initcall_t __initcall4_start[];
720extern initcall_t __initcall5_start[];
721extern initcall_t __initcall6_start[];
722extern initcall_t __initcall7_start[];
723extern initcall_t __initcall_end[];
724
725static initcall_t *initcall_levels[] __initdata = {
726 __initcall0_start,
727 __initcall1_start,
728 __initcall2_start,
729 __initcall3_start,
730 __initcall4_start,
731 __initcall5_start,
732 __initcall6_start,
733 __initcall7_start,
734 __initcall_end,
735};
736
737/* Keep these in sync with initcalls in include/linux/init.h */
738static char *initcall_level_names[] __initdata = {
739 "early",
740 "core",
741 "postcore",
742 "arch",
743 "subsys",
744 "fs",
745 "device",
746 "late",
747};
748
749static void __init do_initcall_level(int level)
750{
751 extern const struct kernel_param __start___param[], __stop___param[];
752 initcall_t *fn;
753
754 strcpy(initcall_command_line, saved_command_line);
755 parse_args(initcall_level_names[level],
756 initcall_command_line, __start___param,
757 __stop___param - __start___param,
758 level, level,
759 &repair_env_string);
760
761 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
762 do_one_initcall(*fn);
763}
764
765static void __init do_initcalls(void)
766{
767 int level;
768
769 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
770 do_initcall_level(level);
771}
772
773/*
774 * Ok, the machine is now initialized. None of the devices
775 * have been touched yet, but the CPU subsystem is up and
776 * running, and memory and process management works.
777 *
778 * Now we can finally start doing some real work..
779 */
780static void __init do_basic_setup(void)
781{
782 cpuset_init_smp();
783 usermodehelper_init();
784 shmem_init();
785 driver_init();
786 init_irq_proc();
787 do_ctors();
788 usermodehelper_enable();
789 do_initcalls();
790 random_int_secret_init();
791}
792
793static void __init do_pre_smp_initcalls(void)
794{
795 initcall_t *fn;
796
797 for (fn = __initcall_start; fn < __initcall0_start; fn++)
798 do_one_initcall(*fn);
799}
800
801/*
802 * This function requests modules which should be loaded by default and is
803 * called twice right after initrd is mounted and right before init is
804 * exec'd. If such modules are on either initrd or rootfs, they will be
805 * loaded before control is passed to userland.
806 */
807void __init load_default_modules(void)
808{
809 load_default_elevator_module();
810}
811
812static int run_init_process(const char *init_filename)
813{
814 argv_init[0] = init_filename;
815 return do_execve(getname_kernel(init_filename),
816 (const char __user *const __user *)argv_init,
817 (const char __user *const __user *)envp_init);
818}
819
820static int try_to_run_init_process(const char *init_filename)
821{
822 int ret;
823
824 ret = run_init_process(init_filename);
825
826 if (ret && ret != -ENOENT) {
827 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
828 init_filename, ret);
829 }
830
831 return ret;
832}
833
834static noinline void __init kernel_init_freeable(void);
835
836static int __ref kernel_init(void *unused)
837{
838 int ret;
839
840 kernel_init_freeable();
841 /* need to finish all async __init code before freeing the memory */
842 async_synchronize_full();
843 free_initmem();
844 mark_rodata_ro();
845 system_state = SYSTEM_RUNNING;
846 numa_default_policy();
847
848 flush_delayed_fput();
849
850 if (ramdisk_execute_command) {
851 ret = run_init_process(ramdisk_execute_command);
852 if (!ret)
853 return 0;
854 pr_err("Failed to execute %s (error %d)\n",
855 ramdisk_execute_command, ret);
856 }
857
858 /*
859 * We try each of these until one succeeds.
860 *
861 * The Bourne shell can be used instead of init if we are
862 * trying to recover a really broken machine.
863 */
864 if (execute_command) {
865 ret = run_init_process(execute_command);
866 if (!ret)
867 return 0;
868 pr_err("Failed to execute %s (error %d). Attempting defaults...\n",
869 execute_command, ret);
870 }
871 if (!try_to_run_init_process("/sbin/init") ||
872 !try_to_run_init_process("/etc/init") ||
873 !try_to_run_init_process("/bin/init") ||
874 !try_to_run_init_process("/bin/sh"))
875 return 0;
876
877 panic("No working init found. Try passing init= option to kernel. "
878 "See Linux Documentation/init.txt for guidance.");
879}
880
881static noinline void __init kernel_init_freeable(void)
882{
883 /*
884 * Wait until kthreadd is all set-up.
885 */
886 wait_for_completion(&kthreadd_done);
887
888 /* Now the scheduler is fully set up and can do blocking allocations */
889 gfp_allowed_mask = __GFP_BITS_MASK;
890
891 /*
892 * init can allocate pages on any node
893 */
894 set_mems_allowed(node_states[N_MEMORY]);
895 /*
896 * init can run on any cpu.
897 */
898 set_cpus_allowed_ptr(current, cpu_all_mask);
899
900 cad_pid = task_pid(current);
901
902 smp_prepare_cpus(setup_max_cpus);
903
904 do_pre_smp_initcalls();
905 lockup_detector_init();
906
907 smp_init();
908 sched_init_smp();
909
910 do_basic_setup();
911
912 /* Open the /dev/console on the rootfs, this should never fail */
913 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
914 pr_err("Warning: unable to open an initial console.\n");
915
916 (void) sys_dup(0);
917 (void) sys_dup(0);
918 /*
919 * check if there is an early userspace init. If yes, let it do all
920 * the work
921 */
922
923 if (!ramdisk_execute_command)
924 ramdisk_execute_command = "/init";
925
926 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
927 ramdisk_execute_command = NULL;
928 prepare_namespace();
929 }
930
931 /*
932 * Ok, we have completed the initial bootup, and
933 * we're essentially up and running. Get rid of the
934 * initmem segments and start the user-mode stuff..
935 */
936
937 /* rootfs is available now, try loading default modules */
938 load_default_modules();
939}