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// 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/bootconfig.h>
32#include <linux/console.h>
33#include <linux/nmi.h>
34#include <linux/percpu.h>
35#include <linux/kmod.h>
36#include <linux/kprobes.h>
37#include <linux/kmsan.h>
38#include <linux/vmalloc.h>
39#include <linux/kernel_stat.h>
40#include <linux/start_kernel.h>
41#include <linux/security.h>
42#include <linux/smp.h>
43#include <linux/profile.h>
44#include <linux/kfence.h>
45#include <linux/rcupdate.h>
46#include <linux/srcu.h>
47#include <linux/moduleparam.h>
48#include <linux/kallsyms.h>
49#include <linux/buildid.h>
50#include <linux/writeback.h>
51#include <linux/cpu.h>
52#include <linux/cpuset.h>
53#include <linux/cgroup.h>
54#include <linux/efi.h>
55#include <linux/tick.h>
56#include <linux/sched/isolation.h>
57#include <linux/interrupt.h>
58#include <linux/taskstats_kern.h>
59#include <linux/delayacct.h>
60#include <linux/unistd.h>
61#include <linux/utsname.h>
62#include <linux/rmap.h>
63#include <linux/mempolicy.h>
64#include <linux/key.h>
65#include <linux/debug_locks.h>
66#include <linux/debugobjects.h>
67#include <linux/lockdep.h>
68#include <linux/kmemleak.h>
69#include <linux/padata.h>
70#include <linux/pid_namespace.h>
71#include <linux/device/driver.h>
72#include <linux/kthread.h>
73#include <linux/sched.h>
74#include <linux/sched/init.h>
75#include <linux/signal.h>
76#include <linux/idr.h>
77#include <linux/kgdb.h>
78#include <linux/ftrace.h>
79#include <linux/async.h>
80#include <linux/shmem_fs.h>
81#include <linux/slab.h>
82#include <linux/perf_event.h>
83#include <linux/ptrace.h>
84#include <linux/pti.h>
85#include <linux/blkdev.h>
86#include <linux/sched/clock.h>
87#include <linux/sched/task.h>
88#include <linux/sched/task_stack.h>
89#include <linux/context_tracking.h>
90#include <linux/random.h>
91#include <linux/list.h>
92#include <linux/integrity.h>
93#include <linux/proc_ns.h>
94#include <linux/io.h>
95#include <linux/cache.h>
96#include <linux/rodata_test.h>
97#include <linux/jump_label.h>
98#include <linux/kcsan.h>
99#include <linux/init_syscalls.h>
100#include <linux/stackdepot.h>
101#include <linux/randomize_kstack.h>
102#include <net/net_namespace.h>
103
104#include <asm/io.h>
105#include <asm/setup.h>
106#include <asm/sections.h>
107#include <asm/cacheflush.h>
108
109#define CREATE_TRACE_POINTS
110#include <trace/events/initcall.h>
111
112#include <kunit/test.h>
113
114static int kernel_init(void *);
115
116/*
117 * Debug helper: via this flag we know that we are in 'early bootup code'
118 * where only the boot processor is running with IRQ disabled. This means
119 * two things - IRQ must not be enabled before the flag is cleared and some
120 * operations which are not allowed with IRQ disabled are allowed while the
121 * flag is set.
122 */
123bool early_boot_irqs_disabled __read_mostly;
124
125enum system_states system_state __read_mostly;
126EXPORT_SYMBOL(system_state);
127
128/*
129 * Boot command-line arguments
130 */
131#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
132#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
133
134/* Default late time init is NULL. archs can override this later. */
135void (*__initdata late_time_init)(void);
136
137/* Untouched command line saved by arch-specific code. */
138char __initdata boot_command_line[COMMAND_LINE_SIZE];
139/* Untouched saved command line (eg. for /proc) */
140char *saved_command_line __ro_after_init;
141unsigned int saved_command_line_len __ro_after_init;
142/* Command line for parameter parsing */
143static char *static_command_line;
144/* Untouched extra command line */
145static char *extra_command_line;
146/* Extra init arguments */
147static char *extra_init_args;
148
149#ifdef CONFIG_BOOT_CONFIG
150/* Is bootconfig on command line? */
151static bool bootconfig_found;
152static size_t initargs_offs;
153#else
154# define bootconfig_found false
155# define initargs_offs 0
156#endif
157
158static char *execute_command;
159static char *ramdisk_execute_command = "/init";
160
161/*
162 * Used to generate warnings if static_key manipulation functions are used
163 * before jump_label_init is called.
164 */
165bool static_key_initialized __read_mostly;
166EXPORT_SYMBOL_GPL(static_key_initialized);
167
168/*
169 * If set, this is an indication to the drivers that reset the underlying
170 * device before going ahead with the initialization otherwise driver might
171 * rely on the BIOS and skip the reset operation.
172 *
173 * This is useful if kernel is booting in an unreliable environment.
174 * For ex. kdump situation where previous kernel has crashed, BIOS has been
175 * skipped and devices will be in unknown state.
176 */
177unsigned int reset_devices;
178EXPORT_SYMBOL(reset_devices);
179
180static int __init set_reset_devices(char *str)
181{
182 reset_devices = 1;
183 return 1;
184}
185
186__setup("reset_devices", set_reset_devices);
187
188static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
189const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
190static const char *panic_later, *panic_param;
191
192static bool __init obsolete_checksetup(char *line)
193{
194 const struct obs_kernel_param *p;
195 bool had_early_param = false;
196
197 p = __setup_start;
198 do {
199 int n = strlen(p->str);
200 if (parameqn(line, p->str, n)) {
201 if (p->early) {
202 /* Already done in parse_early_param?
203 * (Needs exact match on param part).
204 * Keep iterating, as we can have early
205 * params and __setups of same names 8( */
206 if (line[n] == '\0' || line[n] == '=')
207 had_early_param = true;
208 } else if (!p->setup_func) {
209 pr_warn("Parameter %s is obsolete, ignored\n",
210 p->str);
211 return true;
212 } else if (p->setup_func(line + n))
213 return true;
214 }
215 p++;
216 } while (p < __setup_end);
217
218 return had_early_param;
219}
220
221/*
222 * This should be approx 2 Bo*oMips to start (note initial shift), and will
223 * still work even if initially too large, it will just take slightly longer
224 */
225unsigned long loops_per_jiffy = (1<<12);
226EXPORT_SYMBOL(loops_per_jiffy);
227
228static int __init debug_kernel(char *str)
229{
230 console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
231 return 0;
232}
233
234static int __init quiet_kernel(char *str)
235{
236 console_loglevel = CONSOLE_LOGLEVEL_QUIET;
237 return 0;
238}
239
240early_param("debug", debug_kernel);
241early_param("quiet", quiet_kernel);
242
243static int __init loglevel(char *str)
244{
245 int newlevel;
246
247 /*
248 * Only update loglevel value when a correct setting was passed,
249 * to prevent blind crashes (when loglevel being set to 0) that
250 * are quite hard to debug
251 */
252 if (get_option(&str, &newlevel)) {
253 console_loglevel = newlevel;
254 return 0;
255 }
256
257 return -EINVAL;
258}
259
260early_param("loglevel", loglevel);
261
262#ifdef CONFIG_BLK_DEV_INITRD
263static void * __init get_boot_config_from_initrd(size_t *_size)
264{
265 u32 size, csum;
266 char *data;
267 u32 *hdr;
268 int i;
269
270 if (!initrd_end)
271 return NULL;
272
273 data = (char *)initrd_end - BOOTCONFIG_MAGIC_LEN;
274 /*
275 * Since Grub may align the size of initrd to 4, we must
276 * check the preceding 3 bytes as well.
277 */
278 for (i = 0; i < 4; i++) {
279 if (!memcmp(data, BOOTCONFIG_MAGIC, BOOTCONFIG_MAGIC_LEN))
280 goto found;
281 data--;
282 }
283 return NULL;
284
285found:
286 hdr = (u32 *)(data - 8);
287 size = le32_to_cpu(hdr[0]);
288 csum = le32_to_cpu(hdr[1]);
289
290 data = ((void *)hdr) - size;
291 if ((unsigned long)data < initrd_start) {
292 pr_err("bootconfig size %d is greater than initrd size %ld\n",
293 size, initrd_end - initrd_start);
294 return NULL;
295 }
296
297 if (xbc_calc_checksum(data, size) != csum) {
298 pr_err("bootconfig checksum failed\n");
299 return NULL;
300 }
301
302 /* Remove bootconfig from initramfs/initrd */
303 initrd_end = (unsigned long)data;
304 if (_size)
305 *_size = size;
306
307 return data;
308}
309#else
310static void * __init get_boot_config_from_initrd(size_t *_size)
311{
312 return NULL;
313}
314#endif
315
316#ifdef CONFIG_BOOT_CONFIG
317
318static char xbc_namebuf[XBC_KEYLEN_MAX] __initdata;
319
320#define rest(dst, end) ((end) > (dst) ? (end) - (dst) : 0)
321
322static int __init xbc_snprint_cmdline(char *buf, size_t size,
323 struct xbc_node *root)
324{
325 struct xbc_node *knode, *vnode;
326 char *end = buf + size;
327 const char *val;
328 int ret;
329
330 xbc_node_for_each_key_value(root, knode, val) {
331 ret = xbc_node_compose_key_after(root, knode,
332 xbc_namebuf, XBC_KEYLEN_MAX);
333 if (ret < 0)
334 return ret;
335
336 vnode = xbc_node_get_child(knode);
337 if (!vnode) {
338 ret = snprintf(buf, rest(buf, end), "%s ", xbc_namebuf);
339 if (ret < 0)
340 return ret;
341 buf += ret;
342 continue;
343 }
344 xbc_array_for_each_value(vnode, val) {
345 ret = snprintf(buf, rest(buf, end), "%s=\"%s\" ",
346 xbc_namebuf, val);
347 if (ret < 0)
348 return ret;
349 buf += ret;
350 }
351 }
352
353 return buf - (end - size);
354}
355#undef rest
356
357/* Make an extra command line under given key word */
358static char * __init xbc_make_cmdline(const char *key)
359{
360 struct xbc_node *root;
361 char *new_cmdline;
362 int ret, len = 0;
363
364 root = xbc_find_node(key);
365 if (!root)
366 return NULL;
367
368 /* Count required buffer size */
369 len = xbc_snprint_cmdline(NULL, 0, root);
370 if (len <= 0)
371 return NULL;
372
373 new_cmdline = memblock_alloc(len + 1, SMP_CACHE_BYTES);
374 if (!new_cmdline) {
375 pr_err("Failed to allocate memory for extra kernel cmdline.\n");
376 return NULL;
377 }
378
379 ret = xbc_snprint_cmdline(new_cmdline, len + 1, root);
380 if (ret < 0 || ret > len) {
381 pr_err("Failed to print extra kernel cmdline.\n");
382 memblock_free(new_cmdline, len + 1);
383 return NULL;
384 }
385
386 return new_cmdline;
387}
388
389static int __init bootconfig_params(char *param, char *val,
390 const char *unused, void *arg)
391{
392 if (strcmp(param, "bootconfig") == 0) {
393 bootconfig_found = true;
394 }
395 return 0;
396}
397
398static int __init warn_bootconfig(char *str)
399{
400 /* The 'bootconfig' has been handled by bootconfig_params(). */
401 return 0;
402}
403
404static void __init setup_boot_config(void)
405{
406 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
407 const char *msg, *data;
408 int pos, ret;
409 size_t size;
410 char *err;
411
412 /* Cut out the bootconfig data even if we have no bootconfig option */
413 data = get_boot_config_from_initrd(&size);
414 /* If there is no bootconfig in initrd, try embedded one. */
415 if (!data)
416 data = xbc_get_embedded_bootconfig(&size);
417
418 strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
419 err = parse_args("bootconfig", tmp_cmdline, NULL, 0, 0, 0, NULL,
420 bootconfig_params);
421
422 if (IS_ERR(err) || !(bootconfig_found || IS_ENABLED(CONFIG_BOOT_CONFIG_FORCE)))
423 return;
424
425 /* parse_args() stops at the next param of '--' and returns an address */
426 if (err)
427 initargs_offs = err - tmp_cmdline;
428
429 if (!data) {
430 /* If user intended to use bootconfig, show an error level message */
431 if (bootconfig_found)
432 pr_err("'bootconfig' found on command line, but no bootconfig found\n");
433 else
434 pr_info("No bootconfig data provided, so skipping bootconfig");
435 return;
436 }
437
438 if (size >= XBC_DATA_MAX) {
439 pr_err("bootconfig size %ld greater than max size %d\n",
440 (long)size, XBC_DATA_MAX);
441 return;
442 }
443
444 ret = xbc_init(data, size, &msg, &pos);
445 if (ret < 0) {
446 if (pos < 0)
447 pr_err("Failed to init bootconfig: %s.\n", msg);
448 else
449 pr_err("Failed to parse bootconfig: %s at %d.\n",
450 msg, pos);
451 } else {
452 xbc_get_info(&ret, NULL);
453 pr_info("Load bootconfig: %ld bytes %d nodes\n", (long)size, ret);
454 /* keys starting with "kernel." are passed via cmdline */
455 extra_command_line = xbc_make_cmdline("kernel");
456 /* Also, "init." keys are init arguments */
457 extra_init_args = xbc_make_cmdline("init");
458 }
459 return;
460}
461
462static void __init exit_boot_config(void)
463{
464 xbc_exit();
465}
466
467#else /* !CONFIG_BOOT_CONFIG */
468
469static void __init setup_boot_config(void)
470{
471 /* Remove bootconfig data from initrd */
472 get_boot_config_from_initrd(NULL);
473}
474
475static int __init warn_bootconfig(char *str)
476{
477 pr_warn("WARNING: 'bootconfig' found on the kernel command line but CONFIG_BOOT_CONFIG is not set.\n");
478 return 0;
479}
480
481#define exit_boot_config() do {} while (0)
482
483#endif /* CONFIG_BOOT_CONFIG */
484
485early_param("bootconfig", warn_bootconfig);
486
487/* Change NUL term back to "=", to make "param" the whole string. */
488static void __init repair_env_string(char *param, char *val)
489{
490 if (val) {
491 /* param=val or param="val"? */
492 if (val == param+strlen(param)+1)
493 val[-1] = '=';
494 else if (val == param+strlen(param)+2) {
495 val[-2] = '=';
496 memmove(val-1, val, strlen(val)+1);
497 } else
498 BUG();
499 }
500}
501
502/* Anything after -- gets handed straight to init. */
503static int __init set_init_arg(char *param, char *val,
504 const char *unused, void *arg)
505{
506 unsigned int i;
507
508 if (panic_later)
509 return 0;
510
511 repair_env_string(param, val);
512
513 for (i = 0; argv_init[i]; i++) {
514 if (i == MAX_INIT_ARGS) {
515 panic_later = "init";
516 panic_param = param;
517 return 0;
518 }
519 }
520 argv_init[i] = param;
521 return 0;
522}
523
524/*
525 * Unknown boot options get handed to init, unless they look like
526 * unused parameters (modprobe will find them in /proc/cmdline).
527 */
528static int __init unknown_bootoption(char *param, char *val,
529 const char *unused, void *arg)
530{
531 size_t len = strlen(param);
532
533 /* Handle params aliased to sysctls */
534 if (sysctl_is_alias(param))
535 return 0;
536
537 repair_env_string(param, val);
538
539 /* Handle obsolete-style parameters */
540 if (obsolete_checksetup(param))
541 return 0;
542
543 /* Unused module parameter. */
544 if (strnchr(param, len, '.'))
545 return 0;
546
547 if (panic_later)
548 return 0;
549
550 if (val) {
551 /* Environment option */
552 unsigned int i;
553 for (i = 0; envp_init[i]; i++) {
554 if (i == MAX_INIT_ENVS) {
555 panic_later = "env";
556 panic_param = param;
557 }
558 if (!strncmp(param, envp_init[i], len+1))
559 break;
560 }
561 envp_init[i] = param;
562 } else {
563 /* Command line option */
564 unsigned int i;
565 for (i = 0; argv_init[i]; i++) {
566 if (i == MAX_INIT_ARGS) {
567 panic_later = "init";
568 panic_param = param;
569 }
570 }
571 argv_init[i] = param;
572 }
573 return 0;
574}
575
576static int __init init_setup(char *str)
577{
578 unsigned int i;
579
580 execute_command = str;
581 /*
582 * In case LILO is going to boot us with default command line,
583 * it prepends "auto" before the whole cmdline which makes
584 * the shell think it should execute a script with such name.
585 * So we ignore all arguments entered _before_ init=... [MJ]
586 */
587 for (i = 1; i < MAX_INIT_ARGS; i++)
588 argv_init[i] = NULL;
589 return 1;
590}
591__setup("init=", init_setup);
592
593static int __init rdinit_setup(char *str)
594{
595 unsigned int i;
596
597 ramdisk_execute_command = str;
598 /* See "auto" comment in init_setup */
599 for (i = 1; i < MAX_INIT_ARGS; i++)
600 argv_init[i] = NULL;
601 return 1;
602}
603__setup("rdinit=", rdinit_setup);
604
605#ifndef CONFIG_SMP
606static const unsigned int setup_max_cpus = NR_CPUS;
607static inline void setup_nr_cpu_ids(void) { }
608static inline void smp_prepare_cpus(unsigned int maxcpus) { }
609#endif
610
611/*
612 * We need to store the untouched command line for future reference.
613 * We also need to store the touched command line since the parameter
614 * parsing is performed in place, and we should allow a component to
615 * store reference of name/value for future reference.
616 */
617static void __init setup_command_line(char *command_line)
618{
619 size_t len, xlen = 0, ilen = 0;
620
621 if (extra_command_line)
622 xlen = strlen(extra_command_line);
623 if (extra_init_args)
624 ilen = strlen(extra_init_args) + 4; /* for " -- " */
625
626 len = xlen + strlen(boot_command_line) + 1;
627
628 saved_command_line = memblock_alloc(len + ilen, SMP_CACHE_BYTES);
629 if (!saved_command_line)
630 panic("%s: Failed to allocate %zu bytes\n", __func__, len + ilen);
631
632 static_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
633 if (!static_command_line)
634 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
635
636 if (xlen) {
637 /*
638 * We have to put extra_command_line before boot command
639 * lines because there could be dashes (separator of init
640 * command line) in the command lines.
641 */
642 strcpy(saved_command_line, extra_command_line);
643 strcpy(static_command_line, extra_command_line);
644 }
645 strcpy(saved_command_line + xlen, boot_command_line);
646 strcpy(static_command_line + xlen, command_line);
647
648 if (ilen) {
649 /*
650 * Append supplemental init boot args to saved_command_line
651 * so that user can check what command line options passed
652 * to init.
653 * The order should always be
654 * " -- "[bootconfig init-param][cmdline init-param]
655 */
656 if (initargs_offs) {
657 len = xlen + initargs_offs;
658 strcpy(saved_command_line + len, extra_init_args);
659 len += ilen - 4; /* strlen(extra_init_args) */
660 strcpy(saved_command_line + len,
661 boot_command_line + initargs_offs - 1);
662 } else {
663 len = strlen(saved_command_line);
664 strcpy(saved_command_line + len, " -- ");
665 len += 4;
666 strcpy(saved_command_line + len, extra_init_args);
667 }
668 }
669
670 saved_command_line_len = strlen(saved_command_line);
671}
672
673/*
674 * We need to finalize in a non-__init function or else race conditions
675 * between the root thread and the init thread may cause start_kernel to
676 * be reaped by free_initmem before the root thread has proceeded to
677 * cpu_idle.
678 *
679 * gcc-3.4 accidentally inlines this function, so use noinline.
680 */
681
682static __initdata DECLARE_COMPLETION(kthreadd_done);
683
684noinline void __ref __noreturn rest_init(void)
685{
686 struct task_struct *tsk;
687 int pid;
688
689 rcu_scheduler_starting();
690 /*
691 * We need to spawn init first so that it obtains pid 1, however
692 * the init task will end up wanting to create kthreads, which, if
693 * we schedule it before we create kthreadd, will OOPS.
694 */
695 pid = user_mode_thread(kernel_init, NULL, CLONE_FS);
696 /*
697 * Pin init on the boot CPU. Task migration is not properly working
698 * until sched_init_smp() has been run. It will set the allowed
699 * CPUs for init to the non isolated CPUs.
700 */
701 rcu_read_lock();
702 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
703 tsk->flags |= PF_NO_SETAFFINITY;
704 set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id()));
705 rcu_read_unlock();
706
707 numa_default_policy();
708 pid = kernel_thread(kthreadd, NULL, NULL, CLONE_FS | CLONE_FILES);
709 rcu_read_lock();
710 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
711 rcu_read_unlock();
712
713 /*
714 * Enable might_sleep() and smp_processor_id() checks.
715 * They cannot be enabled earlier because with CONFIG_PREEMPTION=y
716 * kernel_thread() would trigger might_sleep() splats. With
717 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled
718 * already, but it's stuck on the kthreadd_done completion.
719 */
720 system_state = SYSTEM_SCHEDULING;
721
722 complete(&kthreadd_done);
723
724 /*
725 * The boot idle thread must execute schedule()
726 * at least once to get things moving:
727 */
728 schedule_preempt_disabled();
729 /* Call into cpu_idle with preempt disabled */
730 cpu_startup_entry(CPUHP_ONLINE);
731}
732
733/* Check for early params. */
734static int __init do_early_param(char *param, char *val,
735 const char *unused, void *arg)
736{
737 const struct obs_kernel_param *p;
738
739 for (p = __setup_start; p < __setup_end; p++) {
740 if ((p->early && parameq(param, p->str)) ||
741 (strcmp(param, "console") == 0 &&
742 strcmp(p->str, "earlycon") == 0)
743 ) {
744 if (p->setup_func(val) != 0)
745 pr_warn("Malformed early option '%s'\n", param);
746 }
747 }
748 /* We accept everything at this stage. */
749 return 0;
750}
751
752void __init parse_early_options(char *cmdline)
753{
754 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
755 do_early_param);
756}
757
758/* Arch code calls this early on, or if not, just before other parsing. */
759void __init parse_early_param(void)
760{
761 static int done __initdata;
762 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
763
764 if (done)
765 return;
766
767 /* All fall through to do_early_param. */
768 strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
769 parse_early_options(tmp_cmdline);
770 done = 1;
771}
772
773void __init __weak arch_post_acpi_subsys_init(void) { }
774
775void __init __weak smp_setup_processor_id(void)
776{
777}
778
779# if THREAD_SIZE >= PAGE_SIZE
780void __init __weak thread_stack_cache_init(void)
781{
782}
783#endif
784
785void __init __weak poking_init(void) { }
786
787void __init __weak pgtable_cache_init(void) { }
788
789void __init __weak trap_init(void) { }
790
791bool initcall_debug;
792core_param(initcall_debug, initcall_debug, bool, 0644);
793
794#ifdef TRACEPOINTS_ENABLED
795static void __init initcall_debug_enable(void);
796#else
797static inline void initcall_debug_enable(void)
798{
799}
800#endif
801
802#ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET
803DEFINE_STATIC_KEY_MAYBE_RO(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
804 randomize_kstack_offset);
805DEFINE_PER_CPU(u32, kstack_offset);
806
807static int __init early_randomize_kstack_offset(char *buf)
808{
809 int ret;
810 bool bool_result;
811
812 ret = kstrtobool(buf, &bool_result);
813 if (ret)
814 return ret;
815
816 if (bool_result)
817 static_branch_enable(&randomize_kstack_offset);
818 else
819 static_branch_disable(&randomize_kstack_offset);
820 return 0;
821}
822early_param("randomize_kstack_offset", early_randomize_kstack_offset);
823#endif
824
825void __init __weak __noreturn arch_call_rest_init(void)
826{
827 rest_init();
828}
829
830static void __init print_unknown_bootoptions(void)
831{
832 char *unknown_options;
833 char *end;
834 const char *const *p;
835 size_t len;
836
837 if (panic_later || (!argv_init[1] && !envp_init[2]))
838 return;
839
840 /*
841 * Determine how many options we have to print out, plus a space
842 * before each
843 */
844 len = 1; /* null terminator */
845 for (p = &argv_init[1]; *p; p++) {
846 len++;
847 len += strlen(*p);
848 }
849 for (p = &envp_init[2]; *p; p++) {
850 len++;
851 len += strlen(*p);
852 }
853
854 unknown_options = memblock_alloc(len, SMP_CACHE_BYTES);
855 if (!unknown_options) {
856 pr_err("%s: Failed to allocate %zu bytes\n",
857 __func__, len);
858 return;
859 }
860 end = unknown_options;
861
862 for (p = &argv_init[1]; *p; p++)
863 end += sprintf(end, " %s", *p);
864 for (p = &envp_init[2]; *p; p++)
865 end += sprintf(end, " %s", *p);
866
867 /* Start at unknown_options[1] to skip the initial space */
868 pr_notice("Unknown kernel command line parameters \"%s\", will be passed to user space.\n",
869 &unknown_options[1]);
870 memblock_free(unknown_options, len);
871}
872
873asmlinkage __visible __init __no_sanitize_address __noreturn __no_stack_protector
874void start_kernel(void)
875{
876 char *command_line;
877 char *after_dashes;
878
879 set_task_stack_end_magic(&init_task);
880 smp_setup_processor_id();
881 debug_objects_early_init();
882 init_vmlinux_build_id();
883
884 cgroup_init_early();
885
886 local_irq_disable();
887 early_boot_irqs_disabled = true;
888
889 /*
890 * Interrupts are still disabled. Do necessary setups, then
891 * enable them.
892 */
893 boot_cpu_init();
894 page_address_init();
895 pr_notice("%s", linux_banner);
896 early_security_init();
897 setup_arch(&command_line);
898 setup_boot_config();
899 setup_command_line(command_line);
900 setup_nr_cpu_ids();
901 setup_per_cpu_areas();
902 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
903 boot_cpu_hotplug_init();
904
905 pr_notice("Kernel command line: %s\n", saved_command_line);
906 /* parameters may set static keys */
907 jump_label_init();
908 parse_early_param();
909 after_dashes = parse_args("Booting kernel",
910 static_command_line, __start___param,
911 __stop___param - __start___param,
912 -1, -1, NULL, &unknown_bootoption);
913 print_unknown_bootoptions();
914 if (!IS_ERR_OR_NULL(after_dashes))
915 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
916 NULL, set_init_arg);
917 if (extra_init_args)
918 parse_args("Setting extra init args", extra_init_args,
919 NULL, 0, -1, -1, NULL, set_init_arg);
920
921 /* Architectural and non-timekeeping rng init, before allocator init */
922 random_init_early(command_line);
923
924 /*
925 * These use large bootmem allocations and must precede
926 * initalization of page allocator
927 */
928 setup_log_buf(0);
929 vfs_caches_init_early();
930 sort_main_extable();
931 trap_init();
932 mm_core_init();
933 poking_init();
934 ftrace_init();
935
936 /* trace_printk can be enabled here */
937 early_trace_init();
938
939 /*
940 * Set up the scheduler prior starting any interrupts (such as the
941 * timer interrupt). Full topology setup happens at smp_init()
942 * time - but meanwhile we still have a functioning scheduler.
943 */
944 sched_init();
945
946 if (WARN(!irqs_disabled(),
947 "Interrupts were enabled *very* early, fixing it\n"))
948 local_irq_disable();
949 radix_tree_init();
950 maple_tree_init();
951
952 /*
953 * Set up housekeeping before setting up workqueues to allow the unbound
954 * workqueue to take non-housekeeping into account.
955 */
956 housekeeping_init();
957
958 /*
959 * Allow workqueue creation and work item queueing/cancelling
960 * early. Work item execution depends on kthreads and starts after
961 * workqueue_init().
962 */
963 workqueue_init_early();
964
965 rcu_init();
966
967 /* Trace events are available after this */
968 trace_init();
969
970 if (initcall_debug)
971 initcall_debug_enable();
972
973 context_tracking_init();
974 /* init some links before init_ISA_irqs() */
975 early_irq_init();
976 init_IRQ();
977 tick_init();
978 rcu_init_nohz();
979 init_timers();
980 srcu_init();
981 hrtimers_init();
982 softirq_init();
983 timekeeping_init();
984 time_init();
985
986 /* This must be after timekeeping is initialized */
987 random_init();
988
989 /* These make use of the fully initialized rng */
990 kfence_init();
991 boot_init_stack_canary();
992
993 perf_event_init();
994 profile_init();
995 call_function_init();
996 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
997
998 early_boot_irqs_disabled = false;
999 local_irq_enable();
1000
1001 kmem_cache_init_late();
1002
1003 /*
1004 * HACK ALERT! This is early. We're enabling the console before
1005 * we've done PCI setups etc, and console_init() must be aware of
1006 * this. But we do want output early, in case something goes wrong.
1007 */
1008 console_init();
1009 if (panic_later)
1010 panic("Too many boot %s vars at `%s'", panic_later,
1011 panic_param);
1012
1013 lockdep_init();
1014
1015 /*
1016 * Need to run this when irqs are enabled, because it wants
1017 * to self-test [hard/soft]-irqs on/off lock inversion bugs
1018 * too:
1019 */
1020 locking_selftest();
1021
1022#ifdef CONFIG_BLK_DEV_INITRD
1023 if (initrd_start && !initrd_below_start_ok &&
1024 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
1025 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
1026 page_to_pfn(virt_to_page((void *)initrd_start)),
1027 min_low_pfn);
1028 initrd_start = 0;
1029 }
1030#endif
1031 setup_per_cpu_pageset();
1032 numa_policy_init();
1033 acpi_early_init();
1034 if (late_time_init)
1035 late_time_init();
1036 sched_clock_init();
1037 calibrate_delay();
1038
1039 arch_cpu_finalize_init();
1040
1041 pid_idr_init();
1042 anon_vma_init();
1043#ifdef CONFIG_X86
1044 if (efi_enabled(EFI_RUNTIME_SERVICES))
1045 efi_enter_virtual_mode();
1046#endif
1047 thread_stack_cache_init();
1048 cred_init();
1049 fork_init();
1050 proc_caches_init();
1051 uts_ns_init();
1052 key_init();
1053 security_init();
1054 dbg_late_init();
1055 net_ns_init();
1056 vfs_caches_init();
1057 pagecache_init();
1058 signals_init();
1059 seq_file_init();
1060 proc_root_init();
1061 nsfs_init();
1062 cpuset_init();
1063 cgroup_init();
1064 taskstats_init_early();
1065 delayacct_init();
1066
1067 acpi_subsystem_init();
1068 arch_post_acpi_subsys_init();
1069 kcsan_init();
1070
1071 /* Do the rest non-__init'ed, we're now alive */
1072 arch_call_rest_init();
1073
1074 /*
1075 * Avoid stack canaries in callers of boot_init_stack_canary for gcc-10
1076 * and older.
1077 */
1078#if !__has_attribute(__no_stack_protector__)
1079 prevent_tail_call_optimization();
1080#endif
1081}
1082
1083/* Call all constructor functions linked into the kernel. */
1084static void __init do_ctors(void)
1085{
1086/*
1087 * For UML, the constructors have already been called by the
1088 * normal setup code as it's just a normal ELF binary, so we
1089 * cannot do it again - but we do need CONFIG_CONSTRUCTORS
1090 * even on UML for modules.
1091 */
1092#if defined(CONFIG_CONSTRUCTORS) && !defined(CONFIG_UML)
1093 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
1094
1095 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
1096 (*fn)();
1097#endif
1098}
1099
1100#ifdef CONFIG_KALLSYMS
1101struct blacklist_entry {
1102 struct list_head next;
1103 char *buf;
1104};
1105
1106static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
1107
1108static int __init initcall_blacklist(char *str)
1109{
1110 char *str_entry;
1111 struct blacklist_entry *entry;
1112
1113 /* str argument is a comma-separated list of functions */
1114 do {
1115 str_entry = strsep(&str, ",");
1116 if (str_entry) {
1117 pr_debug("blacklisting initcall %s\n", str_entry);
1118 entry = memblock_alloc(sizeof(*entry),
1119 SMP_CACHE_BYTES);
1120 if (!entry)
1121 panic("%s: Failed to allocate %zu bytes\n",
1122 __func__, sizeof(*entry));
1123 entry->buf = memblock_alloc(strlen(str_entry) + 1,
1124 SMP_CACHE_BYTES);
1125 if (!entry->buf)
1126 panic("%s: Failed to allocate %zu bytes\n",
1127 __func__, strlen(str_entry) + 1);
1128 strcpy(entry->buf, str_entry);
1129 list_add(&entry->next, &blacklisted_initcalls);
1130 }
1131 } while (str_entry);
1132
1133 return 1;
1134}
1135
1136static bool __init_or_module initcall_blacklisted(initcall_t fn)
1137{
1138 struct blacklist_entry *entry;
1139 char fn_name[KSYM_SYMBOL_LEN];
1140 unsigned long addr;
1141
1142 if (list_empty(&blacklisted_initcalls))
1143 return false;
1144
1145 addr = (unsigned long) dereference_function_descriptor(fn);
1146 sprint_symbol_no_offset(fn_name, addr);
1147
1148 /*
1149 * fn will be "function_name [module_name]" where [module_name] is not
1150 * displayed for built-in init functions. Strip off the [module_name].
1151 */
1152 strreplace(fn_name, ' ', '\0');
1153
1154 list_for_each_entry(entry, &blacklisted_initcalls, next) {
1155 if (!strcmp(fn_name, entry->buf)) {
1156 pr_debug("initcall %s blacklisted\n", fn_name);
1157 return true;
1158 }
1159 }
1160
1161 return false;
1162}
1163#else
1164static int __init initcall_blacklist(char *str)
1165{
1166 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
1167 return 0;
1168}
1169
1170static bool __init_or_module initcall_blacklisted(initcall_t fn)
1171{
1172 return false;
1173}
1174#endif
1175__setup("initcall_blacklist=", initcall_blacklist);
1176
1177static __init_or_module void
1178trace_initcall_start_cb(void *data, initcall_t fn)
1179{
1180 ktime_t *calltime = data;
1181
1182 printk(KERN_DEBUG "calling %pS @ %i\n", fn, task_pid_nr(current));
1183 *calltime = ktime_get();
1184}
1185
1186static __init_or_module void
1187trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
1188{
1189 ktime_t rettime, *calltime = data;
1190
1191 rettime = ktime_get();
1192 printk(KERN_DEBUG "initcall %pS returned %d after %lld usecs\n",
1193 fn, ret, (unsigned long long)ktime_us_delta(rettime, *calltime));
1194}
1195
1196static ktime_t initcall_calltime;
1197
1198#ifdef TRACEPOINTS_ENABLED
1199static void __init initcall_debug_enable(void)
1200{
1201 int ret;
1202
1203 ret = register_trace_initcall_start(trace_initcall_start_cb,
1204 &initcall_calltime);
1205 ret |= register_trace_initcall_finish(trace_initcall_finish_cb,
1206 &initcall_calltime);
1207 WARN(ret, "Failed to register initcall tracepoints\n");
1208}
1209# define do_trace_initcall_start trace_initcall_start
1210# define do_trace_initcall_finish trace_initcall_finish
1211#else
1212static inline void do_trace_initcall_start(initcall_t fn)
1213{
1214 if (!initcall_debug)
1215 return;
1216 trace_initcall_start_cb(&initcall_calltime, fn);
1217}
1218static inline void do_trace_initcall_finish(initcall_t fn, int ret)
1219{
1220 if (!initcall_debug)
1221 return;
1222 trace_initcall_finish_cb(&initcall_calltime, fn, ret);
1223}
1224#endif /* !TRACEPOINTS_ENABLED */
1225
1226int __init_or_module do_one_initcall(initcall_t fn)
1227{
1228 int count = preempt_count();
1229 char msgbuf[64];
1230 int ret;
1231
1232 if (initcall_blacklisted(fn))
1233 return -EPERM;
1234
1235 do_trace_initcall_start(fn);
1236 ret = fn();
1237 do_trace_initcall_finish(fn, ret);
1238
1239 msgbuf[0] = 0;
1240
1241 if (preempt_count() != count) {
1242 sprintf(msgbuf, "preemption imbalance ");
1243 preempt_count_set(count);
1244 }
1245 if (irqs_disabled()) {
1246 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
1247 local_irq_enable();
1248 }
1249 WARN(msgbuf[0], "initcall %pS returned with %s\n", fn, msgbuf);
1250
1251 add_latent_entropy();
1252 return ret;
1253}
1254
1255
1256static initcall_entry_t *initcall_levels[] __initdata = {
1257 __initcall0_start,
1258 __initcall1_start,
1259 __initcall2_start,
1260 __initcall3_start,
1261 __initcall4_start,
1262 __initcall5_start,
1263 __initcall6_start,
1264 __initcall7_start,
1265 __initcall_end,
1266};
1267
1268/* Keep these in sync with initcalls in include/linux/init.h */
1269static const char *initcall_level_names[] __initdata = {
1270 "pure",
1271 "core",
1272 "postcore",
1273 "arch",
1274 "subsys",
1275 "fs",
1276 "device",
1277 "late",
1278};
1279
1280static int __init ignore_unknown_bootoption(char *param, char *val,
1281 const char *unused, void *arg)
1282{
1283 return 0;
1284}
1285
1286static void __init do_initcall_level(int level, char *command_line)
1287{
1288 initcall_entry_t *fn;
1289
1290 parse_args(initcall_level_names[level],
1291 command_line, __start___param,
1292 __stop___param - __start___param,
1293 level, level,
1294 NULL, ignore_unknown_bootoption);
1295
1296 trace_initcall_level(initcall_level_names[level]);
1297 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
1298 do_one_initcall(initcall_from_entry(fn));
1299}
1300
1301static void __init do_initcalls(void)
1302{
1303 int level;
1304 size_t len = saved_command_line_len + 1;
1305 char *command_line;
1306
1307 command_line = kzalloc(len, GFP_KERNEL);
1308 if (!command_line)
1309 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
1310
1311 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++) {
1312 /* Parser modifies command_line, restore it each time */
1313 strcpy(command_line, saved_command_line);
1314 do_initcall_level(level, command_line);
1315 }
1316
1317 kfree(command_line);
1318}
1319
1320/*
1321 * Ok, the machine is now initialized. None of the devices
1322 * have been touched yet, but the CPU subsystem is up and
1323 * running, and memory and process management works.
1324 *
1325 * Now we can finally start doing some real work..
1326 */
1327static void __init do_basic_setup(void)
1328{
1329 cpuset_init_smp();
1330 driver_init();
1331 init_irq_proc();
1332 do_ctors();
1333 do_initcalls();
1334}
1335
1336static void __init do_pre_smp_initcalls(void)
1337{
1338 initcall_entry_t *fn;
1339
1340 trace_initcall_level("early");
1341 for (fn = __initcall_start; fn < __initcall0_start; fn++)
1342 do_one_initcall(initcall_from_entry(fn));
1343}
1344
1345static int run_init_process(const char *init_filename)
1346{
1347 const char *const *p;
1348
1349 argv_init[0] = init_filename;
1350 pr_info("Run %s as init process\n", init_filename);
1351 pr_debug(" with arguments:\n");
1352 for (p = argv_init; *p; p++)
1353 pr_debug(" %s\n", *p);
1354 pr_debug(" with environment:\n");
1355 for (p = envp_init; *p; p++)
1356 pr_debug(" %s\n", *p);
1357 return kernel_execve(init_filename, argv_init, envp_init);
1358}
1359
1360static int try_to_run_init_process(const char *init_filename)
1361{
1362 int ret;
1363
1364 ret = run_init_process(init_filename);
1365
1366 if (ret && ret != -ENOENT) {
1367 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
1368 init_filename, ret);
1369 }
1370
1371 return ret;
1372}
1373
1374static noinline void __init kernel_init_freeable(void);
1375
1376#if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
1377bool rodata_enabled __ro_after_init = true;
1378
1379#ifndef arch_parse_debug_rodata
1380static inline bool arch_parse_debug_rodata(char *str) { return false; }
1381#endif
1382
1383static int __init set_debug_rodata(char *str)
1384{
1385 if (arch_parse_debug_rodata(str))
1386 return 0;
1387
1388 if (str && !strcmp(str, "on"))
1389 rodata_enabled = true;
1390 else if (str && !strcmp(str, "off"))
1391 rodata_enabled = false;
1392 else
1393 pr_warn("Invalid option string for rodata: '%s'\n", str);
1394 return 0;
1395}
1396early_param("rodata", set_debug_rodata);
1397#endif
1398
1399#ifdef CONFIG_STRICT_KERNEL_RWX
1400static void mark_readonly(void)
1401{
1402 if (rodata_enabled) {
1403 /*
1404 * load_module() results in W+X mappings, which are cleaned
1405 * up with call_rcu(). Let's make sure that queued work is
1406 * flushed so that we don't hit false positives looking for
1407 * insecure pages which are W+X.
1408 */
1409 rcu_barrier();
1410 mark_rodata_ro();
1411 rodata_test();
1412 } else
1413 pr_info("Kernel memory protection disabled.\n");
1414}
1415#elif defined(CONFIG_ARCH_HAS_STRICT_KERNEL_RWX)
1416static inline void mark_readonly(void)
1417{
1418 pr_warn("Kernel memory protection not selected by kernel config.\n");
1419}
1420#else
1421static inline void mark_readonly(void)
1422{
1423 pr_warn("This architecture does not have kernel memory protection.\n");
1424}
1425#endif
1426
1427void __weak free_initmem(void)
1428{
1429 free_initmem_default(POISON_FREE_INITMEM);
1430}
1431
1432static int __ref kernel_init(void *unused)
1433{
1434 int ret;
1435
1436 /*
1437 * Wait until kthreadd is all set-up.
1438 */
1439 wait_for_completion(&kthreadd_done);
1440
1441 kernel_init_freeable();
1442 /* need to finish all async __init code before freeing the memory */
1443 async_synchronize_full();
1444
1445 system_state = SYSTEM_FREEING_INITMEM;
1446 kprobe_free_init_mem();
1447 ftrace_free_init_mem();
1448 kgdb_free_init_mem();
1449 exit_boot_config();
1450 free_initmem();
1451 mark_readonly();
1452
1453 /*
1454 * Kernel mappings are now finalized - update the userspace page-table
1455 * to finalize PTI.
1456 */
1457 pti_finalize();
1458
1459 system_state = SYSTEM_RUNNING;
1460 numa_default_policy();
1461
1462 rcu_end_inkernel_boot();
1463
1464 do_sysctl_args();
1465
1466 if (ramdisk_execute_command) {
1467 ret = run_init_process(ramdisk_execute_command);
1468 if (!ret)
1469 return 0;
1470 pr_err("Failed to execute %s (error %d)\n",
1471 ramdisk_execute_command, ret);
1472 }
1473
1474 /*
1475 * We try each of these until one succeeds.
1476 *
1477 * The Bourne shell can be used instead of init if we are
1478 * trying to recover a really broken machine.
1479 */
1480 if (execute_command) {
1481 ret = run_init_process(execute_command);
1482 if (!ret)
1483 return 0;
1484 panic("Requested init %s failed (error %d).",
1485 execute_command, ret);
1486 }
1487
1488 if (CONFIG_DEFAULT_INIT[0] != '\0') {
1489 ret = run_init_process(CONFIG_DEFAULT_INIT);
1490 if (ret)
1491 pr_err("Default init %s failed (error %d)\n",
1492 CONFIG_DEFAULT_INIT, ret);
1493 else
1494 return 0;
1495 }
1496
1497 if (!try_to_run_init_process("/sbin/init") ||
1498 !try_to_run_init_process("/etc/init") ||
1499 !try_to_run_init_process("/bin/init") ||
1500 !try_to_run_init_process("/bin/sh"))
1501 return 0;
1502
1503 panic("No working init found. Try passing init= option to kernel. "
1504 "See Linux Documentation/admin-guide/init.rst for guidance.");
1505}
1506
1507/* Open /dev/console, for stdin/stdout/stderr, this should never fail */
1508void __init console_on_rootfs(void)
1509{
1510 struct file *file = filp_open("/dev/console", O_RDWR, 0);
1511
1512 if (IS_ERR(file)) {
1513 pr_err("Warning: unable to open an initial console.\n");
1514 return;
1515 }
1516 init_dup(file);
1517 init_dup(file);
1518 init_dup(file);
1519 fput(file);
1520}
1521
1522static noinline void __init kernel_init_freeable(void)
1523{
1524 /* Now the scheduler is fully set up and can do blocking allocations */
1525 gfp_allowed_mask = __GFP_BITS_MASK;
1526
1527 /*
1528 * init can allocate pages on any node
1529 */
1530 set_mems_allowed(node_states[N_MEMORY]);
1531
1532 cad_pid = get_pid(task_pid(current));
1533
1534 smp_prepare_cpus(setup_max_cpus);
1535
1536 workqueue_init();
1537
1538 init_mm_internals();
1539
1540 rcu_init_tasks_generic();
1541 do_pre_smp_initcalls();
1542 lockup_detector_init();
1543
1544 smp_init();
1545 sched_init_smp();
1546
1547 workqueue_init_topology();
1548 padata_init();
1549 page_alloc_init_late();
1550
1551 do_basic_setup();
1552
1553 kunit_run_all_tests();
1554
1555 wait_for_initramfs();
1556 console_on_rootfs();
1557
1558 /*
1559 * check if there is an early userspace init. If yes, let it do all
1560 * the work
1561 */
1562 if (init_eaccess(ramdisk_execute_command) != 0) {
1563 ramdisk_execute_command = NULL;
1564 prepare_namespace();
1565 }
1566
1567 /*
1568 * Ok, we have completed the initial bootup, and
1569 * we're essentially up and running. Get rid of the
1570 * initmem segments and start the user-mode stuff..
1571 *
1572 * rootfs is available now, try loading the public keys
1573 * and default modules
1574 */
1575
1576 integrity_load_keys();
1577}