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/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/moduleloader.h>
92#include <linux/list.h>
93#include <linux/integrity.h>
94#include <linux/proc_ns.h>
95#include <linux/io.h>
96#include <linux/cache.h>
97#include <linux/rodata_test.h>
98#include <linux/jump_label.h>
99#include <linux/kcsan.h>
100#include <linux/init_syscalls.h>
101#include <linux/stackdepot.h>
102#include <linux/randomize_kstack.h>
103#include <linux/pidfs.h>
104#include <linux/ptdump.h>
105#include <net/net_namespace.h>
106
107#include <asm/io.h>
108#include <asm/setup.h>
109#include <asm/sections.h>
110#include <asm/cacheflush.h>
111
112#define CREATE_TRACE_POINTS
113#include <trace/events/initcall.h>
114
115#include <kunit/test.h>
116
117static int kernel_init(void *);
118
119/*
120 * Debug helper: via this flag we know that we are in 'early bootup code'
121 * where only the boot processor is running with IRQ disabled. This means
122 * two things - IRQ must not be enabled before the flag is cleared and some
123 * operations which are not allowed with IRQ disabled are allowed while the
124 * flag is set.
125 */
126bool early_boot_irqs_disabled __read_mostly;
127
128enum system_states system_state __read_mostly;
129EXPORT_SYMBOL(system_state);
130
131/*
132 * Boot command-line arguments
133 */
134#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
135#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
136
137/* Default late time init is NULL. archs can override this later. */
138void (*__initdata late_time_init)(void);
139
140/* Untouched command line saved by arch-specific code. */
141char __initdata boot_command_line[COMMAND_LINE_SIZE];
142/* Untouched saved command line (eg. for /proc) */
143char *saved_command_line __ro_after_init;
144unsigned int saved_command_line_len __ro_after_init;
145/* Command line for parameter parsing */
146static char *static_command_line;
147/* Untouched extra command line */
148static char *extra_command_line;
149/* Extra init arguments */
150static char *extra_init_args;
151
152#ifdef CONFIG_BOOT_CONFIG
153/* Is bootconfig on command line? */
154static bool bootconfig_found;
155static size_t initargs_offs;
156#else
157# define bootconfig_found false
158# define initargs_offs 0
159#endif
160
161static char *execute_command;
162static char *ramdisk_execute_command = "/init";
163
164/*
165 * Used to generate warnings if static_key manipulation functions are used
166 * before jump_label_init is called.
167 */
168bool static_key_initialized __read_mostly;
169EXPORT_SYMBOL_GPL(static_key_initialized);
170
171/*
172 * If set, this is an indication to the drivers that reset the underlying
173 * device before going ahead with the initialization otherwise driver might
174 * rely on the BIOS and skip the reset operation.
175 *
176 * This is useful if kernel is booting in an unreliable environment.
177 * For ex. kdump situation where previous kernel has crashed, BIOS has been
178 * skipped and devices will be in unknown state.
179 */
180unsigned int reset_devices;
181EXPORT_SYMBOL(reset_devices);
182
183static int __init set_reset_devices(char *str)
184{
185 reset_devices = 1;
186 return 1;
187}
188
189__setup("reset_devices", set_reset_devices);
190
191static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
192const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
193static const char *panic_later, *panic_param;
194
195static bool __init obsolete_checksetup(char *line)
196{
197 const struct obs_kernel_param *p;
198 bool had_early_param = false;
199
200 p = __setup_start;
201 do {
202 int n = strlen(p->str);
203 if (parameqn(line, p->str, n)) {
204 if (p->early) {
205 /* Already done in parse_early_param?
206 * (Needs exact match on param part).
207 * Keep iterating, as we can have early
208 * params and __setups of same names 8( */
209 if (line[n] == '\0' || line[n] == '=')
210 had_early_param = true;
211 } else if (!p->setup_func) {
212 pr_warn("Parameter %s is obsolete, ignored\n",
213 p->str);
214 return true;
215 } else if (p->setup_func(line + n))
216 return true;
217 }
218 p++;
219 } while (p < __setup_end);
220
221 return had_early_param;
222}
223
224/*
225 * This should be approx 2 Bo*oMips to start (note initial shift), and will
226 * still work even if initially too large, it will just take slightly longer
227 */
228unsigned long loops_per_jiffy = (1<<12);
229EXPORT_SYMBOL(loops_per_jiffy);
230
231static int __init debug_kernel(char *str)
232{
233 console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
234 return 0;
235}
236
237static int __init quiet_kernel(char *str)
238{
239 console_loglevel = CONSOLE_LOGLEVEL_QUIET;
240 return 0;
241}
242
243early_param("debug", debug_kernel);
244early_param("quiet", quiet_kernel);
245
246static int __init loglevel(char *str)
247{
248 int newlevel;
249
250 /*
251 * Only update loglevel value when a correct setting was passed,
252 * to prevent blind crashes (when loglevel being set to 0) that
253 * are quite hard to debug
254 */
255 if (get_option(&str, &newlevel)) {
256 console_loglevel = newlevel;
257 return 0;
258 }
259
260 return -EINVAL;
261}
262
263early_param("loglevel", loglevel);
264
265#ifdef CONFIG_BLK_DEV_INITRD
266static void * __init get_boot_config_from_initrd(size_t *_size)
267{
268 u32 size, csum;
269 char *data;
270 u32 *hdr;
271 int i;
272
273 if (!initrd_end)
274 return NULL;
275
276 data = (char *)initrd_end - BOOTCONFIG_MAGIC_LEN;
277 /*
278 * Since Grub may align the size of initrd to 4, we must
279 * check the preceding 3 bytes as well.
280 */
281 for (i = 0; i < 4; i++) {
282 if (!memcmp(data, BOOTCONFIG_MAGIC, BOOTCONFIG_MAGIC_LEN))
283 goto found;
284 data--;
285 }
286 return NULL;
287
288found:
289 hdr = (u32 *)(data - 8);
290 size = le32_to_cpu(hdr[0]);
291 csum = le32_to_cpu(hdr[1]);
292
293 data = ((void *)hdr) - size;
294 if ((unsigned long)data < initrd_start) {
295 pr_err("bootconfig size %d is greater than initrd size %ld\n",
296 size, initrd_end - initrd_start);
297 return NULL;
298 }
299
300 if (xbc_calc_checksum(data, size) != csum) {
301 pr_err("bootconfig checksum failed\n");
302 return NULL;
303 }
304
305 /* Remove bootconfig from initramfs/initrd */
306 initrd_end = (unsigned long)data;
307 if (_size)
308 *_size = size;
309
310 return data;
311}
312#else
313static void * __init get_boot_config_from_initrd(size_t *_size)
314{
315 return NULL;
316}
317#endif
318
319#ifdef CONFIG_BOOT_CONFIG
320
321static char xbc_namebuf[XBC_KEYLEN_MAX] __initdata;
322
323#define rest(dst, end) ((end) > (dst) ? (end) - (dst) : 0)
324
325static int __init xbc_snprint_cmdline(char *buf, size_t size,
326 struct xbc_node *root)
327{
328 struct xbc_node *knode, *vnode;
329 char *end = buf + size;
330 const char *val, *q;
331 int ret;
332
333 xbc_node_for_each_key_value(root, knode, val) {
334 ret = xbc_node_compose_key_after(root, knode,
335 xbc_namebuf, XBC_KEYLEN_MAX);
336 if (ret < 0)
337 return ret;
338
339 vnode = xbc_node_get_child(knode);
340 if (!vnode) {
341 ret = snprintf(buf, rest(buf, end), "%s ", xbc_namebuf);
342 if (ret < 0)
343 return ret;
344 buf += ret;
345 continue;
346 }
347 xbc_array_for_each_value(vnode, val) {
348 /*
349 * For prettier and more readable /proc/cmdline, only
350 * quote the value when necessary, i.e. when it contains
351 * whitespace.
352 */
353 q = strpbrk(val, " \t\r\n") ? "\"" : "";
354 ret = snprintf(buf, rest(buf, end), "%s=%s%s%s ",
355 xbc_namebuf, q, val, q);
356 if (ret < 0)
357 return ret;
358 buf += ret;
359 }
360 }
361
362 return buf - (end - size);
363}
364#undef rest
365
366/* Make an extra command line under given key word */
367static char * __init xbc_make_cmdline(const char *key)
368{
369 struct xbc_node *root;
370 char *new_cmdline;
371 int ret, len = 0;
372
373 root = xbc_find_node(key);
374 if (!root)
375 return NULL;
376
377 /* Count required buffer size */
378 len = xbc_snprint_cmdline(NULL, 0, root);
379 if (len <= 0)
380 return NULL;
381
382 new_cmdline = memblock_alloc(len + 1, SMP_CACHE_BYTES);
383 if (!new_cmdline) {
384 pr_err("Failed to allocate memory for extra kernel cmdline.\n");
385 return NULL;
386 }
387
388 ret = xbc_snprint_cmdline(new_cmdline, len + 1, root);
389 if (ret < 0 || ret > len) {
390 pr_err("Failed to print extra kernel cmdline.\n");
391 memblock_free(new_cmdline, len + 1);
392 return NULL;
393 }
394
395 return new_cmdline;
396}
397
398static int __init bootconfig_params(char *param, char *val,
399 const char *unused, void *arg)
400{
401 if (strcmp(param, "bootconfig") == 0) {
402 bootconfig_found = true;
403 }
404 return 0;
405}
406
407static int __init warn_bootconfig(char *str)
408{
409 /* The 'bootconfig' has been handled by bootconfig_params(). */
410 return 0;
411}
412
413static void __init setup_boot_config(void)
414{
415 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
416 const char *msg, *data;
417 int pos, ret;
418 size_t size;
419 char *err;
420
421 /* Cut out the bootconfig data even if we have no bootconfig option */
422 data = get_boot_config_from_initrd(&size);
423 /* If there is no bootconfig in initrd, try embedded one. */
424 if (!data)
425 data = xbc_get_embedded_bootconfig(&size);
426
427 strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
428 err = parse_args("bootconfig", tmp_cmdline, NULL, 0, 0, 0, NULL,
429 bootconfig_params);
430
431 if (IS_ERR(err) || !(bootconfig_found || IS_ENABLED(CONFIG_BOOT_CONFIG_FORCE)))
432 return;
433
434 /* parse_args() stops at the next param of '--' and returns an address */
435 if (err)
436 initargs_offs = err - tmp_cmdline;
437
438 if (!data) {
439 /* If user intended to use bootconfig, show an error level message */
440 if (bootconfig_found)
441 pr_err("'bootconfig' found on command line, but no bootconfig found\n");
442 else
443 pr_info("No bootconfig data provided, so skipping bootconfig");
444 return;
445 }
446
447 if (size >= XBC_DATA_MAX) {
448 pr_err("bootconfig size %ld greater than max size %d\n",
449 (long)size, XBC_DATA_MAX);
450 return;
451 }
452
453 ret = xbc_init(data, size, &msg, &pos);
454 if (ret < 0) {
455 if (pos < 0)
456 pr_err("Failed to init bootconfig: %s.\n", msg);
457 else
458 pr_err("Failed to parse bootconfig: %s at %d.\n",
459 msg, pos);
460 } else {
461 xbc_get_info(&ret, NULL);
462 pr_info("Load bootconfig: %ld bytes %d nodes\n", (long)size, ret);
463 /* keys starting with "kernel." are passed via cmdline */
464 extra_command_line = xbc_make_cmdline("kernel");
465 /* Also, "init." keys are init arguments */
466 extra_init_args = xbc_make_cmdline("init");
467 }
468 return;
469}
470
471static void __init exit_boot_config(void)
472{
473 xbc_exit();
474}
475
476#else /* !CONFIG_BOOT_CONFIG */
477
478static void __init setup_boot_config(void)
479{
480 /* Remove bootconfig data from initrd */
481 get_boot_config_from_initrd(NULL);
482}
483
484static int __init warn_bootconfig(char *str)
485{
486 pr_warn("WARNING: 'bootconfig' found on the kernel command line but CONFIG_BOOT_CONFIG is not set.\n");
487 return 0;
488}
489
490#define exit_boot_config() do {} while (0)
491
492#endif /* CONFIG_BOOT_CONFIG */
493
494early_param("bootconfig", warn_bootconfig);
495
496bool __init cmdline_has_extra_options(void)
497{
498 return extra_command_line || extra_init_args;
499}
500
501/* Change NUL term back to "=", to make "param" the whole string. */
502static void __init repair_env_string(char *param, char *val)
503{
504 if (val) {
505 /* param=val or param="val"? */
506 if (val == param+strlen(param)+1)
507 val[-1] = '=';
508 else if (val == param+strlen(param)+2) {
509 val[-2] = '=';
510 memmove(val-1, val, strlen(val)+1);
511 } else
512 BUG();
513 }
514}
515
516/* Anything after -- gets handed straight to init. */
517static int __init set_init_arg(char *param, char *val,
518 const char *unused, void *arg)
519{
520 unsigned int i;
521
522 if (panic_later)
523 return 0;
524
525 repair_env_string(param, val);
526
527 for (i = 0; argv_init[i]; i++) {
528 if (i == MAX_INIT_ARGS) {
529 panic_later = "init";
530 panic_param = param;
531 return 0;
532 }
533 }
534 argv_init[i] = param;
535 return 0;
536}
537
538/*
539 * Unknown boot options get handed to init, unless they look like
540 * unused parameters (modprobe will find them in /proc/cmdline).
541 */
542static int __init unknown_bootoption(char *param, char *val,
543 const char *unused, void *arg)
544{
545 size_t len = strlen(param);
546
547 /* Handle params aliased to sysctls */
548 if (sysctl_is_alias(param))
549 return 0;
550
551 repair_env_string(param, val);
552
553 /* Handle obsolete-style parameters */
554 if (obsolete_checksetup(param))
555 return 0;
556
557 /* Unused module parameter. */
558 if (strnchr(param, len, '.'))
559 return 0;
560
561 if (panic_later)
562 return 0;
563
564 if (val) {
565 /* Environment option */
566 unsigned int i;
567 for (i = 0; envp_init[i]; i++) {
568 if (i == MAX_INIT_ENVS) {
569 panic_later = "env";
570 panic_param = param;
571 }
572 if (!strncmp(param, envp_init[i], len+1))
573 break;
574 }
575 envp_init[i] = param;
576 } else {
577 /* Command line option */
578 unsigned int i;
579 for (i = 0; argv_init[i]; i++) {
580 if (i == MAX_INIT_ARGS) {
581 panic_later = "init";
582 panic_param = param;
583 }
584 }
585 argv_init[i] = param;
586 }
587 return 0;
588}
589
590static int __init init_setup(char *str)
591{
592 unsigned int i;
593
594 execute_command = str;
595 /*
596 * In case LILO is going to boot us with default command line,
597 * it prepends "auto" before the whole cmdline which makes
598 * the shell think it should execute a script with such name.
599 * So we ignore all arguments entered _before_ init=... [MJ]
600 */
601 for (i = 1; i < MAX_INIT_ARGS; i++)
602 argv_init[i] = NULL;
603 return 1;
604}
605__setup("init=", init_setup);
606
607static int __init rdinit_setup(char *str)
608{
609 unsigned int i;
610
611 ramdisk_execute_command = str;
612 /* See "auto" comment in init_setup */
613 for (i = 1; i < MAX_INIT_ARGS; i++)
614 argv_init[i] = NULL;
615 return 1;
616}
617__setup("rdinit=", rdinit_setup);
618
619#ifndef CONFIG_SMP
620static inline void setup_nr_cpu_ids(void) { }
621static inline void smp_prepare_cpus(unsigned int maxcpus) { }
622#endif
623
624/*
625 * We need to store the untouched command line for future reference.
626 * We also need to store the touched command line since the parameter
627 * parsing is performed in place, and we should allow a component to
628 * store reference of name/value for future reference.
629 */
630static void __init setup_command_line(char *command_line)
631{
632 size_t len, xlen = 0, ilen = 0;
633
634 if (extra_command_line)
635 xlen = strlen(extra_command_line);
636 if (extra_init_args) {
637 extra_init_args = strim(extra_init_args); /* remove trailing space */
638 ilen = strlen(extra_init_args) + 4; /* for " -- " */
639 }
640
641 len = xlen + strlen(boot_command_line) + ilen + 1;
642
643 saved_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
644 if (!saved_command_line)
645 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
646
647 len = xlen + strlen(command_line) + 1;
648
649 static_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
650 if (!static_command_line)
651 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
652
653 if (xlen) {
654 /*
655 * We have to put extra_command_line before boot command
656 * lines because there could be dashes (separator of init
657 * command line) in the command lines.
658 */
659 strcpy(saved_command_line, extra_command_line);
660 strcpy(static_command_line, extra_command_line);
661 }
662 strcpy(saved_command_line + xlen, boot_command_line);
663 strcpy(static_command_line + xlen, command_line);
664
665 if (ilen) {
666 /*
667 * Append supplemental init boot args to saved_command_line
668 * so that user can check what command line options passed
669 * to init.
670 * The order should always be
671 * " -- "[bootconfig init-param][cmdline init-param]
672 */
673 if (initargs_offs) {
674 len = xlen + initargs_offs;
675 strcpy(saved_command_line + len, extra_init_args);
676 len += ilen - 4; /* strlen(extra_init_args) */
677 strcpy(saved_command_line + len,
678 boot_command_line + initargs_offs - 1);
679 } else {
680 len = strlen(saved_command_line);
681 strcpy(saved_command_line + len, " -- ");
682 len += 4;
683 strcpy(saved_command_line + len, extra_init_args);
684 }
685 }
686
687 saved_command_line_len = strlen(saved_command_line);
688}
689
690/*
691 * We need to finalize in a non-__init function or else race conditions
692 * between the root thread and the init thread may cause start_kernel to
693 * be reaped by free_initmem before the root thread has proceeded to
694 * cpu_idle.
695 *
696 * gcc-3.4 accidentally inlines this function, so use noinline.
697 */
698
699static __initdata DECLARE_COMPLETION(kthreadd_done);
700
701static noinline void __ref __noreturn rest_init(void)
702{
703 struct task_struct *tsk;
704 int pid;
705
706 rcu_scheduler_starting();
707 /*
708 * We need to spawn init first so that it obtains pid 1, however
709 * the init task will end up wanting to create kthreads, which, if
710 * we schedule it before we create kthreadd, will OOPS.
711 */
712 pid = user_mode_thread(kernel_init, NULL, CLONE_FS);
713 /*
714 * Pin init on the boot CPU. Task migration is not properly working
715 * until sched_init_smp() has been run. It will set the allowed
716 * CPUs for init to the non isolated CPUs.
717 */
718 rcu_read_lock();
719 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
720 tsk->flags |= PF_NO_SETAFFINITY;
721 set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id()));
722 rcu_read_unlock();
723
724 numa_default_policy();
725 pid = kernel_thread(kthreadd, NULL, NULL, CLONE_FS | CLONE_FILES);
726 rcu_read_lock();
727 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
728 rcu_read_unlock();
729
730 /*
731 * Enable might_sleep() and smp_processor_id() checks.
732 * They cannot be enabled earlier because with CONFIG_PREEMPTION=y
733 * kernel_thread() would trigger might_sleep() splats. With
734 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled
735 * already, but it's stuck on the kthreadd_done completion.
736 */
737 system_state = SYSTEM_SCHEDULING;
738
739 complete(&kthreadd_done);
740
741 /*
742 * The boot idle thread must execute schedule()
743 * at least once to get things moving:
744 */
745 schedule_preempt_disabled();
746 /* Call into cpu_idle with preempt disabled */
747 cpu_startup_entry(CPUHP_ONLINE);
748}
749
750/* Check for early params. */
751static int __init do_early_param(char *param, char *val,
752 const char *unused, void *arg)
753{
754 const struct obs_kernel_param *p;
755
756 for (p = __setup_start; p < __setup_end; p++) {
757 if (p->early && parameq(param, p->str)) {
758 if (p->setup_func(val) != 0)
759 pr_warn("Malformed early option '%s'\n", param);
760 }
761 }
762 /* We accept everything at this stage. */
763 return 0;
764}
765
766void __init parse_early_options(char *cmdline)
767{
768 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
769 do_early_param);
770}
771
772/* Arch code calls this early on, or if not, just before other parsing. */
773void __init parse_early_param(void)
774{
775 static int done __initdata;
776 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
777
778 if (done)
779 return;
780
781 /* All fall through to do_early_param. */
782 strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
783 parse_early_options(tmp_cmdline);
784 done = 1;
785}
786
787void __init __weak arch_post_acpi_subsys_init(void) { }
788
789void __init __weak smp_setup_processor_id(void)
790{
791}
792
793void __init __weak smp_prepare_boot_cpu(void)
794{
795}
796
797# if THREAD_SIZE >= PAGE_SIZE
798void __init __weak thread_stack_cache_init(void)
799{
800}
801#endif
802
803void __init __weak poking_init(void) { }
804
805void __init __weak pgtable_cache_init(void) { }
806
807void __init __weak trap_init(void) { }
808
809bool initcall_debug;
810core_param(initcall_debug, initcall_debug, bool, 0644);
811
812#ifdef TRACEPOINTS_ENABLED
813static void __init initcall_debug_enable(void);
814#else
815static inline void initcall_debug_enable(void)
816{
817}
818#endif
819
820#ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET
821DEFINE_STATIC_KEY_MAYBE_RO(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
822 randomize_kstack_offset);
823DEFINE_PER_CPU(u32, kstack_offset);
824
825static int __init early_randomize_kstack_offset(char *buf)
826{
827 int ret;
828 bool bool_result;
829
830 ret = kstrtobool(buf, &bool_result);
831 if (ret)
832 return ret;
833
834 if (bool_result)
835 static_branch_enable(&randomize_kstack_offset);
836 else
837 static_branch_disable(&randomize_kstack_offset);
838 return 0;
839}
840early_param("randomize_kstack_offset", early_randomize_kstack_offset);
841#endif
842
843static void __init print_unknown_bootoptions(void)
844{
845 char *unknown_options;
846 char *end;
847 const char *const *p;
848 size_t len;
849
850 if (panic_later || (!argv_init[1] && !envp_init[2]))
851 return;
852
853 /*
854 * Determine how many options we have to print out, plus a space
855 * before each
856 */
857 len = 1; /* null terminator */
858 for (p = &argv_init[1]; *p; p++) {
859 len++;
860 len += strlen(*p);
861 }
862 for (p = &envp_init[2]; *p; p++) {
863 len++;
864 len += strlen(*p);
865 }
866
867 unknown_options = memblock_alloc(len, SMP_CACHE_BYTES);
868 if (!unknown_options) {
869 pr_err("%s: Failed to allocate %zu bytes\n",
870 __func__, len);
871 return;
872 }
873 end = unknown_options;
874
875 for (p = &argv_init[1]; *p; p++)
876 end += sprintf(end, " %s", *p);
877 for (p = &envp_init[2]; *p; p++)
878 end += sprintf(end, " %s", *p);
879
880 /* Start at unknown_options[1] to skip the initial space */
881 pr_notice("Unknown kernel command line parameters \"%s\", will be passed to user space.\n",
882 &unknown_options[1]);
883 memblock_free(unknown_options, len);
884}
885
886static void __init early_numa_node_init(void)
887{
888#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
889#ifndef cpu_to_node
890 int cpu;
891
892 /* The early_cpu_to_node() should be ready here. */
893 for_each_possible_cpu(cpu)
894 set_cpu_numa_node(cpu, early_cpu_to_node(cpu));
895#endif
896#endif
897}
898
899asmlinkage __visible __init __no_sanitize_address __noreturn __no_stack_protector
900void start_kernel(void)
901{
902 char *command_line;
903 char *after_dashes;
904
905 set_task_stack_end_magic(&init_task);
906 smp_setup_processor_id();
907 debug_objects_early_init();
908 init_vmlinux_build_id();
909
910 cgroup_init_early();
911
912 local_irq_disable();
913 early_boot_irqs_disabled = true;
914
915 /*
916 * Interrupts are still disabled. Do necessary setups, then
917 * enable them.
918 */
919 boot_cpu_init();
920 page_address_init();
921 pr_notice("%s", linux_banner);
922 setup_arch(&command_line);
923 /* Static keys and static calls are needed by LSMs */
924 jump_label_init();
925 static_call_init();
926 early_security_init();
927 setup_boot_config();
928 setup_command_line(command_line);
929 setup_nr_cpu_ids();
930 setup_per_cpu_areas();
931 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
932 early_numa_node_init();
933 boot_cpu_hotplug_init();
934
935 pr_notice("Kernel command line: %s\n", saved_command_line);
936 /* parameters may set static keys */
937 parse_early_param();
938 after_dashes = parse_args("Booting kernel",
939 static_command_line, __start___param,
940 __stop___param - __start___param,
941 -1, -1, NULL, &unknown_bootoption);
942 print_unknown_bootoptions();
943 if (!IS_ERR_OR_NULL(after_dashes))
944 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
945 NULL, set_init_arg);
946 if (extra_init_args)
947 parse_args("Setting extra init args", extra_init_args,
948 NULL, 0, -1, -1, NULL, set_init_arg);
949
950 /* Architectural and non-timekeeping rng init, before allocator init */
951 random_init_early(command_line);
952
953 /*
954 * These use large bootmem allocations and must precede
955 * initalization of page allocator
956 */
957 setup_log_buf(0);
958 vfs_caches_init_early();
959 sort_main_extable();
960 trap_init();
961 mm_core_init();
962 poking_init();
963 ftrace_init();
964
965 /* trace_printk can be enabled here */
966 early_trace_init();
967
968 /*
969 * Set up the scheduler prior starting any interrupts (such as the
970 * timer interrupt). Full topology setup happens at smp_init()
971 * time - but meanwhile we still have a functioning scheduler.
972 */
973 sched_init();
974
975 if (WARN(!irqs_disabled(),
976 "Interrupts were enabled *very* early, fixing it\n"))
977 local_irq_disable();
978 radix_tree_init();
979 maple_tree_init();
980
981 /*
982 * Set up housekeeping before setting up workqueues to allow the unbound
983 * workqueue to take non-housekeeping into account.
984 */
985 housekeeping_init();
986
987 /*
988 * Allow workqueue creation and work item queueing/cancelling
989 * early. Work item execution depends on kthreads and starts after
990 * workqueue_init().
991 */
992 workqueue_init_early();
993
994 rcu_init();
995
996 /* Trace events are available after this */
997 trace_init();
998
999 if (initcall_debug)
1000 initcall_debug_enable();
1001
1002 context_tracking_init();
1003 /* init some links before init_ISA_irqs() */
1004 early_irq_init();
1005 init_IRQ();
1006 tick_init();
1007 rcu_init_nohz();
1008 init_timers();
1009 srcu_init();
1010 hrtimers_init();
1011 softirq_init();
1012 timekeeping_init();
1013 time_init();
1014
1015 /* This must be after timekeeping is initialized */
1016 random_init();
1017
1018 /* These make use of the fully initialized rng */
1019 kfence_init();
1020 boot_init_stack_canary();
1021
1022 perf_event_init();
1023 profile_init();
1024 call_function_init();
1025 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
1026
1027 early_boot_irqs_disabled = false;
1028 local_irq_enable();
1029
1030 kmem_cache_init_late();
1031
1032 /*
1033 * HACK ALERT! This is early. We're enabling the console before
1034 * we've done PCI setups etc, and console_init() must be aware of
1035 * this. But we do want output early, in case something goes wrong.
1036 */
1037 console_init();
1038 if (panic_later)
1039 panic("Too many boot %s vars at `%s'", panic_later,
1040 panic_param);
1041
1042 lockdep_init();
1043
1044 /*
1045 * Need to run this when irqs are enabled, because it wants
1046 * to self-test [hard/soft]-irqs on/off lock inversion bugs
1047 * too:
1048 */
1049 locking_selftest();
1050
1051#ifdef CONFIG_BLK_DEV_INITRD
1052 if (initrd_start && !initrd_below_start_ok &&
1053 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
1054 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
1055 page_to_pfn(virt_to_page((void *)initrd_start)),
1056 min_low_pfn);
1057 initrd_start = 0;
1058 }
1059#endif
1060 setup_per_cpu_pageset();
1061 numa_policy_init();
1062 acpi_early_init();
1063 if (late_time_init)
1064 late_time_init();
1065 sched_clock_init();
1066 calibrate_delay();
1067
1068 arch_cpu_finalize_init();
1069
1070 pid_idr_init();
1071 anon_vma_init();
1072#ifdef CONFIG_X86
1073 if (efi_enabled(EFI_RUNTIME_SERVICES))
1074 efi_enter_virtual_mode();
1075#endif
1076 thread_stack_cache_init();
1077 cred_init();
1078 fork_init();
1079 proc_caches_init();
1080 uts_ns_init();
1081 key_init();
1082 security_init();
1083 dbg_late_init();
1084 net_ns_init();
1085 vfs_caches_init();
1086 pagecache_init();
1087 signals_init();
1088 seq_file_init();
1089 proc_root_init();
1090 nsfs_init();
1091 pidfs_init();
1092 cpuset_init();
1093 cgroup_init();
1094 taskstats_init_early();
1095 delayacct_init();
1096
1097 acpi_subsystem_init();
1098 arch_post_acpi_subsys_init();
1099 kcsan_init();
1100
1101 /* Do the rest non-__init'ed, we're now alive */
1102 rest_init();
1103
1104 /*
1105 * Avoid stack canaries in callers of boot_init_stack_canary for gcc-10
1106 * and older.
1107 */
1108#if !__has_attribute(__no_stack_protector__)
1109 prevent_tail_call_optimization();
1110#endif
1111}
1112
1113/* Call all constructor functions linked into the kernel. */
1114static void __init do_ctors(void)
1115{
1116/*
1117 * For UML, the constructors have already been called by the
1118 * normal setup code as it's just a normal ELF binary, so we
1119 * cannot do it again - but we do need CONFIG_CONSTRUCTORS
1120 * even on UML for modules.
1121 */
1122#if defined(CONFIG_CONSTRUCTORS) && !defined(CONFIG_UML)
1123 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
1124
1125 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
1126 (*fn)();
1127#endif
1128}
1129
1130#ifdef CONFIG_KALLSYMS
1131struct blacklist_entry {
1132 struct list_head next;
1133 char *buf;
1134};
1135
1136static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
1137
1138static int __init initcall_blacklist(char *str)
1139{
1140 char *str_entry;
1141 struct blacklist_entry *entry;
1142
1143 /* str argument is a comma-separated list of functions */
1144 do {
1145 str_entry = strsep(&str, ",");
1146 if (str_entry) {
1147 pr_debug("blacklisting initcall %s\n", str_entry);
1148 entry = memblock_alloc(sizeof(*entry),
1149 SMP_CACHE_BYTES);
1150 if (!entry)
1151 panic("%s: Failed to allocate %zu bytes\n",
1152 __func__, sizeof(*entry));
1153 entry->buf = memblock_alloc(strlen(str_entry) + 1,
1154 SMP_CACHE_BYTES);
1155 if (!entry->buf)
1156 panic("%s: Failed to allocate %zu bytes\n",
1157 __func__, strlen(str_entry) + 1);
1158 strcpy(entry->buf, str_entry);
1159 list_add(&entry->next, &blacklisted_initcalls);
1160 }
1161 } while (str_entry);
1162
1163 return 1;
1164}
1165
1166static bool __init_or_module initcall_blacklisted(initcall_t fn)
1167{
1168 struct blacklist_entry *entry;
1169 char fn_name[KSYM_SYMBOL_LEN];
1170 unsigned long addr;
1171
1172 if (list_empty(&blacklisted_initcalls))
1173 return false;
1174
1175 addr = (unsigned long) dereference_function_descriptor(fn);
1176 sprint_symbol_no_offset(fn_name, addr);
1177
1178 /*
1179 * fn will be "function_name [module_name]" where [module_name] is not
1180 * displayed for built-in init functions. Strip off the [module_name].
1181 */
1182 strreplace(fn_name, ' ', '\0');
1183
1184 list_for_each_entry(entry, &blacklisted_initcalls, next) {
1185 if (!strcmp(fn_name, entry->buf)) {
1186 pr_debug("initcall %s blacklisted\n", fn_name);
1187 return true;
1188 }
1189 }
1190
1191 return false;
1192}
1193#else
1194static int __init initcall_blacklist(char *str)
1195{
1196 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
1197 return 0;
1198}
1199
1200static bool __init_or_module initcall_blacklisted(initcall_t fn)
1201{
1202 return false;
1203}
1204#endif
1205__setup("initcall_blacklist=", initcall_blacklist);
1206
1207static __init_or_module void
1208trace_initcall_start_cb(void *data, initcall_t fn)
1209{
1210 ktime_t *calltime = data;
1211
1212 printk(KERN_DEBUG "calling %pS @ %i\n", fn, task_pid_nr(current));
1213 *calltime = ktime_get();
1214}
1215
1216static __init_or_module void
1217trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
1218{
1219 ktime_t rettime, *calltime = data;
1220
1221 rettime = ktime_get();
1222 printk(KERN_DEBUG "initcall %pS returned %d after %lld usecs\n",
1223 fn, ret, (unsigned long long)ktime_us_delta(rettime, *calltime));
1224}
1225
1226static ktime_t initcall_calltime;
1227
1228#ifdef TRACEPOINTS_ENABLED
1229static void __init initcall_debug_enable(void)
1230{
1231 int ret;
1232
1233 ret = register_trace_initcall_start(trace_initcall_start_cb,
1234 &initcall_calltime);
1235 ret |= register_trace_initcall_finish(trace_initcall_finish_cb,
1236 &initcall_calltime);
1237 WARN(ret, "Failed to register initcall tracepoints\n");
1238}
1239# define do_trace_initcall_start trace_initcall_start
1240# define do_trace_initcall_finish trace_initcall_finish
1241#else
1242static inline void do_trace_initcall_start(initcall_t fn)
1243{
1244 if (!initcall_debug)
1245 return;
1246 trace_initcall_start_cb(&initcall_calltime, fn);
1247}
1248static inline void do_trace_initcall_finish(initcall_t fn, int ret)
1249{
1250 if (!initcall_debug)
1251 return;
1252 trace_initcall_finish_cb(&initcall_calltime, fn, ret);
1253}
1254#endif /* !TRACEPOINTS_ENABLED */
1255
1256int __init_or_module do_one_initcall(initcall_t fn)
1257{
1258 int count = preempt_count();
1259 char msgbuf[64];
1260 int ret;
1261
1262 if (initcall_blacklisted(fn))
1263 return -EPERM;
1264
1265 do_trace_initcall_start(fn);
1266 ret = fn();
1267 do_trace_initcall_finish(fn, ret);
1268
1269 msgbuf[0] = 0;
1270
1271 if (preempt_count() != count) {
1272 sprintf(msgbuf, "preemption imbalance ");
1273 preempt_count_set(count);
1274 }
1275 if (irqs_disabled()) {
1276 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
1277 local_irq_enable();
1278 }
1279 WARN(msgbuf[0], "initcall %pS returned with %s\n", fn, msgbuf);
1280
1281 add_latent_entropy();
1282 return ret;
1283}
1284
1285
1286static initcall_entry_t *initcall_levels[] __initdata = {
1287 __initcall0_start,
1288 __initcall1_start,
1289 __initcall2_start,
1290 __initcall3_start,
1291 __initcall4_start,
1292 __initcall5_start,
1293 __initcall6_start,
1294 __initcall7_start,
1295 __initcall_end,
1296};
1297
1298/* Keep these in sync with initcalls in include/linux/init.h */
1299static const char *initcall_level_names[] __initdata = {
1300 "pure",
1301 "core",
1302 "postcore",
1303 "arch",
1304 "subsys",
1305 "fs",
1306 "device",
1307 "late",
1308};
1309
1310static int __init ignore_unknown_bootoption(char *param, char *val,
1311 const char *unused, void *arg)
1312{
1313 return 0;
1314}
1315
1316static void __init do_initcall_level(int level, char *command_line)
1317{
1318 initcall_entry_t *fn;
1319
1320 parse_args(initcall_level_names[level],
1321 command_line, __start___param,
1322 __stop___param - __start___param,
1323 level, level,
1324 NULL, ignore_unknown_bootoption);
1325
1326 trace_initcall_level(initcall_level_names[level]);
1327 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
1328 do_one_initcall(initcall_from_entry(fn));
1329}
1330
1331static void __init do_initcalls(void)
1332{
1333 int level;
1334 size_t len = saved_command_line_len + 1;
1335 char *command_line;
1336
1337 command_line = kzalloc(len, GFP_KERNEL);
1338 if (!command_line)
1339 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
1340
1341 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++) {
1342 /* Parser modifies command_line, restore it each time */
1343 strcpy(command_line, saved_command_line);
1344 do_initcall_level(level, command_line);
1345 }
1346
1347 kfree(command_line);
1348}
1349
1350/*
1351 * Ok, the machine is now initialized. None of the devices
1352 * have been touched yet, but the CPU subsystem is up and
1353 * running, and memory and process management works.
1354 *
1355 * Now we can finally start doing some real work..
1356 */
1357static void __init do_basic_setup(void)
1358{
1359 cpuset_init_smp();
1360 driver_init();
1361 init_irq_proc();
1362 do_ctors();
1363 do_initcalls();
1364}
1365
1366static void __init do_pre_smp_initcalls(void)
1367{
1368 initcall_entry_t *fn;
1369
1370 trace_initcall_level("early");
1371 for (fn = __initcall_start; fn < __initcall0_start; fn++)
1372 do_one_initcall(initcall_from_entry(fn));
1373}
1374
1375static int run_init_process(const char *init_filename)
1376{
1377 const char *const *p;
1378
1379 argv_init[0] = init_filename;
1380 pr_info("Run %s as init process\n", init_filename);
1381 pr_debug(" with arguments:\n");
1382 for (p = argv_init; *p; p++)
1383 pr_debug(" %s\n", *p);
1384 pr_debug(" with environment:\n");
1385 for (p = envp_init; *p; p++)
1386 pr_debug(" %s\n", *p);
1387 return kernel_execve(init_filename, argv_init, envp_init);
1388}
1389
1390static int try_to_run_init_process(const char *init_filename)
1391{
1392 int ret;
1393
1394 ret = run_init_process(init_filename);
1395
1396 if (ret && ret != -ENOENT) {
1397 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
1398 init_filename, ret);
1399 }
1400
1401 return ret;
1402}
1403
1404static noinline void __init kernel_init_freeable(void);
1405
1406#if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
1407bool rodata_enabled __ro_after_init = true;
1408
1409#ifndef arch_parse_debug_rodata
1410static inline bool arch_parse_debug_rodata(char *str) { return false; }
1411#endif
1412
1413static int __init set_debug_rodata(char *str)
1414{
1415 if (arch_parse_debug_rodata(str))
1416 return 0;
1417
1418 if (str && !strcmp(str, "on"))
1419 rodata_enabled = true;
1420 else if (str && !strcmp(str, "off"))
1421 rodata_enabled = false;
1422 else
1423 pr_warn("Invalid option string for rodata: '%s'\n", str);
1424 return 0;
1425}
1426early_param("rodata", set_debug_rodata);
1427#endif
1428
1429static void mark_readonly(void)
1430{
1431 if (IS_ENABLED(CONFIG_STRICT_KERNEL_RWX) && rodata_enabled) {
1432 /*
1433 * load_module() results in W+X mappings, which are cleaned
1434 * up with init_free_wq. Let's make sure that queued work is
1435 * flushed so that we don't hit false positives looking for
1436 * insecure pages which are W+X.
1437 */
1438 flush_module_init_free_work();
1439 jump_label_init_ro();
1440 mark_rodata_ro();
1441 debug_checkwx();
1442 rodata_test();
1443 } else if (IS_ENABLED(CONFIG_STRICT_KERNEL_RWX)) {
1444 pr_info("Kernel memory protection disabled.\n");
1445 } else if (IS_ENABLED(CONFIG_ARCH_HAS_STRICT_KERNEL_RWX)) {
1446 pr_warn("Kernel memory protection not selected by kernel config.\n");
1447 } else {
1448 pr_warn("This architecture does not have kernel memory protection.\n");
1449 }
1450}
1451
1452void __weak free_initmem(void)
1453{
1454 free_initmem_default(POISON_FREE_INITMEM);
1455}
1456
1457static int __ref kernel_init(void *unused)
1458{
1459 int ret;
1460
1461 /*
1462 * Wait until kthreadd is all set-up.
1463 */
1464 wait_for_completion(&kthreadd_done);
1465
1466 kernel_init_freeable();
1467 /* need to finish all async __init code before freeing the memory */
1468 async_synchronize_full();
1469
1470 system_state = SYSTEM_FREEING_INITMEM;
1471 kprobe_free_init_mem();
1472 ftrace_free_init_mem();
1473 kgdb_free_init_mem();
1474 exit_boot_config();
1475 free_initmem();
1476 mark_readonly();
1477
1478 /*
1479 * Kernel mappings are now finalized - update the userspace page-table
1480 * to finalize PTI.
1481 */
1482 pti_finalize();
1483
1484 system_state = SYSTEM_RUNNING;
1485 numa_default_policy();
1486
1487 rcu_end_inkernel_boot();
1488
1489 do_sysctl_args();
1490
1491 if (ramdisk_execute_command) {
1492 ret = run_init_process(ramdisk_execute_command);
1493 if (!ret)
1494 return 0;
1495 pr_err("Failed to execute %s (error %d)\n",
1496 ramdisk_execute_command, ret);
1497 }
1498
1499 /*
1500 * We try each of these until one succeeds.
1501 *
1502 * The Bourne shell can be used instead of init if we are
1503 * trying to recover a really broken machine.
1504 */
1505 if (execute_command) {
1506 ret = run_init_process(execute_command);
1507 if (!ret)
1508 return 0;
1509 panic("Requested init %s failed (error %d).",
1510 execute_command, ret);
1511 }
1512
1513 if (CONFIG_DEFAULT_INIT[0] != '\0') {
1514 ret = run_init_process(CONFIG_DEFAULT_INIT);
1515 if (ret)
1516 pr_err("Default init %s failed (error %d)\n",
1517 CONFIG_DEFAULT_INIT, ret);
1518 else
1519 return 0;
1520 }
1521
1522 if (!try_to_run_init_process("/sbin/init") ||
1523 !try_to_run_init_process("/etc/init") ||
1524 !try_to_run_init_process("/bin/init") ||
1525 !try_to_run_init_process("/bin/sh"))
1526 return 0;
1527
1528 panic("No working init found. Try passing init= option to kernel. "
1529 "See Linux Documentation/admin-guide/init.rst for guidance.");
1530}
1531
1532/* Open /dev/console, for stdin/stdout/stderr, this should never fail */
1533void __init console_on_rootfs(void)
1534{
1535 struct file *file = filp_open("/dev/console", O_RDWR, 0);
1536
1537 if (IS_ERR(file)) {
1538 pr_err("Warning: unable to open an initial console.\n");
1539 return;
1540 }
1541 init_dup(file);
1542 init_dup(file);
1543 init_dup(file);
1544 fput(file);
1545}
1546
1547static noinline void __init kernel_init_freeable(void)
1548{
1549 /* Now the scheduler is fully set up and can do blocking allocations */
1550 gfp_allowed_mask = __GFP_BITS_MASK;
1551
1552 /*
1553 * init can allocate pages on any node
1554 */
1555 set_mems_allowed(node_states[N_MEMORY]);
1556
1557 cad_pid = get_pid(task_pid(current));
1558
1559 smp_prepare_cpus(setup_max_cpus);
1560
1561 workqueue_init();
1562
1563 init_mm_internals();
1564
1565 rcu_init_tasks_generic();
1566 do_pre_smp_initcalls();
1567 lockup_detector_init();
1568
1569 smp_init();
1570 sched_init_smp();
1571
1572 workqueue_init_topology();
1573 async_init();
1574 padata_init();
1575 page_alloc_init_late();
1576
1577 do_basic_setup();
1578
1579 kunit_run_all_tests();
1580
1581 wait_for_initramfs();
1582 console_on_rootfs();
1583
1584 /*
1585 * check if there is an early userspace init. If yes, let it do all
1586 * the work
1587 */
1588 if (init_eaccess(ramdisk_execute_command) != 0) {
1589 ramdisk_execute_command = NULL;
1590 prepare_namespace();
1591 }
1592
1593 /*
1594 * Ok, we have completed the initial bootup, and
1595 * we're essentially up and running. Get rid of the
1596 * initmem segments and start the user-mode stuff..
1597 *
1598 * rootfs is available now, try loading the public keys
1599 * and default modules
1600 */
1601
1602 integrity_load_keys();
1603}
1/*
2 * linux/init/main.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * GK 2/5/95 - Changed to support mounting root fs via NFS
7 * Added initrd & change_root: Werner Almesberger & Hans Lermen, Feb '96
8 * Moan early if gcc is old, avoiding bogus kernels - Paul Gortmaker, May '96
9 * Simplified starting of init: Michael A. Griffith <grif@acm.org>
10 */
11
12#define DEBUG /* Enable initcall_debug */
13
14#include <linux/types.h>
15#include <linux/extable.h>
16#include <linux/module.h>
17#include <linux/proc_fs.h>
18#include <linux/binfmts.h>
19#include <linux/kernel.h>
20#include <linux/syscalls.h>
21#include <linux/stackprotector.h>
22#include <linux/string.h>
23#include <linux/ctype.h>
24#include <linux/delay.h>
25#include <linux/ioport.h>
26#include <linux/init.h>
27#include <linux/initrd.h>
28#include <linux/bootmem.h>
29#include <linux/acpi.h>
30#include <linux/console.h>
31#include <linux/nmi.h>
32#include <linux/percpu.h>
33#include <linux/kmod.h>
34#include <linux/vmalloc.h>
35#include <linux/kernel_stat.h>
36#include <linux/start_kernel.h>
37#include <linux/security.h>
38#include <linux/smp.h>
39#include <linux/profile.h>
40#include <linux/rcupdate.h>
41#include <linux/moduleparam.h>
42#include <linux/kallsyms.h>
43#include <linux/writeback.h>
44#include <linux/cpu.h>
45#include <linux/cpuset.h>
46#include <linux/cgroup.h>
47#include <linux/efi.h>
48#include <linux/tick.h>
49#include <linux/sched/isolation.h>
50#include <linux/interrupt.h>
51#include <linux/taskstats_kern.h>
52#include <linux/delayacct.h>
53#include <linux/unistd.h>
54#include <linux/utsname.h>
55#include <linux/rmap.h>
56#include <linux/mempolicy.h>
57#include <linux/key.h>
58#include <linux/buffer_head.h>
59#include <linux/page_ext.h>
60#include <linux/debug_locks.h>
61#include <linux/debugobjects.h>
62#include <linux/lockdep.h>
63#include <linux/kmemleak.h>
64#include <linux/pid_namespace.h>
65#include <linux/device.h>
66#include <linux/kthread.h>
67#include <linux/sched.h>
68#include <linux/sched/init.h>
69#include <linux/signal.h>
70#include <linux/idr.h>
71#include <linux/kgdb.h>
72#include <linux/ftrace.h>
73#include <linux/async.h>
74#include <linux/sfi.h>
75#include <linux/shmem_fs.h>
76#include <linux/slab.h>
77#include <linux/perf_event.h>
78#include <linux/ptrace.h>
79#include <linux/pti.h>
80#include <linux/blkdev.h>
81#include <linux/elevator.h>
82#include <linux/sched_clock.h>
83#include <linux/sched/task.h>
84#include <linux/sched/task_stack.h>
85#include <linux/context_tracking.h>
86#include <linux/random.h>
87#include <linux/list.h>
88#include <linux/integrity.h>
89#include <linux/proc_ns.h>
90#include <linux/io.h>
91#include <linux/cache.h>
92#include <linux/rodata_test.h>
93#include <linux/jump_label.h>
94#include <linux/mem_encrypt.h>
95
96#include <asm/io.h>
97#include <asm/bugs.h>
98#include <asm/setup.h>
99#include <asm/sections.h>
100#include <asm/cacheflush.h>
101
102#define CREATE_TRACE_POINTS
103#include <trace/events/initcall.h>
104
105static int kernel_init(void *);
106
107extern void init_IRQ(void);
108extern void fork_init(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 saved_command_line =
378 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
379 initcall_command_line =
380 memblock_virt_alloc(strlen(boot_command_line) + 1, 0);
381 static_command_line = memblock_virt_alloc(strlen(command_line) + 1, 0);
382 strcpy(saved_command_line, boot_command_line);
383 strcpy(static_command_line, command_line);
384}
385
386/*
387 * We need to finalize in a non-__init function or else race conditions
388 * between the root thread and the init thread may cause start_kernel to
389 * be reaped by free_initmem before the root thread has proceeded to
390 * cpu_idle.
391 *
392 * gcc-3.4 accidentally inlines this function, so use noinline.
393 */
394
395static __initdata DECLARE_COMPLETION(kthreadd_done);
396
397static noinline void __ref rest_init(void)
398{
399 struct task_struct *tsk;
400 int pid;
401
402 rcu_scheduler_starting();
403 /*
404 * We need to spawn init first so that it obtains pid 1, however
405 * the init task will end up wanting to create kthreads, which, if
406 * we schedule it before we create kthreadd, will OOPS.
407 */
408 pid = kernel_thread(kernel_init, NULL, CLONE_FS);
409 /*
410 * Pin init on the boot CPU. Task migration is not properly working
411 * until sched_init_smp() has been run. It will set the allowed
412 * CPUs for init to the non isolated CPUs.
413 */
414 rcu_read_lock();
415 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
416 set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id()));
417 rcu_read_unlock();
418
419 numa_default_policy();
420 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
421 rcu_read_lock();
422 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
423 rcu_read_unlock();
424
425 /*
426 * Enable might_sleep() and smp_processor_id() checks.
427 * They cannot be enabled earlier because with CONFIG_PREEMPT=y
428 * kernel_thread() would trigger might_sleep() splats. With
429 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled
430 * already, but it's stuck on the kthreadd_done completion.
431 */
432 system_state = SYSTEM_SCHEDULING;
433
434 complete(&kthreadd_done);
435
436 /*
437 * The boot idle thread must execute schedule()
438 * at least once to get things moving:
439 */
440 schedule_preempt_disabled();
441 /* Call into cpu_idle with preempt disabled */
442 cpu_startup_entry(CPUHP_ONLINE);
443}
444
445/* Check for early params. */
446static int __init do_early_param(char *param, char *val,
447 const char *unused, void *arg)
448{
449 const struct obs_kernel_param *p;
450
451 for (p = __setup_start; p < __setup_end; p++) {
452 if ((p->early && parameq(param, p->str)) ||
453 (strcmp(param, "console") == 0 &&
454 strcmp(p->str, "earlycon") == 0)
455 ) {
456 if (p->setup_func(val) != 0)
457 pr_warn("Malformed early option '%s'\n", param);
458 }
459 }
460 /* We accept everything at this stage. */
461 return 0;
462}
463
464void __init parse_early_options(char *cmdline)
465{
466 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
467 do_early_param);
468}
469
470/* Arch code calls this early on, or if not, just before other parsing. */
471void __init parse_early_param(void)
472{
473 static int done __initdata;
474 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
475
476 if (done)
477 return;
478
479 /* All fall through to do_early_param. */
480 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
481 parse_early_options(tmp_cmdline);
482 done = 1;
483}
484
485void __init __weak arch_post_acpi_subsys_init(void) { }
486
487void __init __weak smp_setup_processor_id(void)
488{
489}
490
491# if THREAD_SIZE >= PAGE_SIZE
492void __init __weak thread_stack_cache_init(void)
493{
494}
495#endif
496
497void __init __weak mem_encrypt_init(void) { }
498
499bool initcall_debug;
500core_param(initcall_debug, initcall_debug, bool, 0644);
501
502#ifdef TRACEPOINTS_ENABLED
503static void __init initcall_debug_enable(void);
504#else
505static inline void initcall_debug_enable(void)
506{
507}
508#endif
509
510/*
511 * Set up kernel memory allocators
512 */
513static void __init mm_init(void)
514{
515 /*
516 * page_ext requires contiguous pages,
517 * bigger than MAX_ORDER unless SPARSEMEM.
518 */
519 page_ext_init_flatmem();
520 mem_init();
521 kmem_cache_init();
522 pgtable_init();
523 vmalloc_init();
524 ioremap_huge_init();
525 /* Should be run before the first non-init thread is created */
526 init_espfix_bsp();
527 /* Should be run after espfix64 is set up. */
528 pti_init();
529}
530
531asmlinkage __visible void __init start_kernel(void)
532{
533 char *command_line;
534 char *after_dashes;
535
536 set_task_stack_end_magic(&init_task);
537 smp_setup_processor_id();
538 debug_objects_early_init();
539
540 cgroup_init_early();
541
542 local_irq_disable();
543 early_boot_irqs_disabled = true;
544
545 /*
546 * Interrupts are still disabled. Do necessary setups, then
547 * enable them.
548 */
549 boot_cpu_init();
550 page_address_init();
551 pr_notice("%s", linux_banner);
552 setup_arch(&command_line);
553 /*
554 * Set up the the initial canary and entropy after arch
555 * and after adding latent and command line entropy.
556 */
557 add_latent_entropy();
558 add_device_randomness(command_line, strlen(command_line));
559 boot_init_stack_canary();
560 mm_init_cpumask(&init_mm);
561 setup_command_line(command_line);
562 setup_nr_cpu_ids();
563 setup_per_cpu_areas();
564 boot_cpu_state_init();
565 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
566
567 build_all_zonelists(NULL);
568 page_alloc_init();
569
570 pr_notice("Kernel command line: %s\n", boot_command_line);
571 parse_early_param();
572 after_dashes = parse_args("Booting kernel",
573 static_command_line, __start___param,
574 __stop___param - __start___param,
575 -1, -1, NULL, &unknown_bootoption);
576 if (!IS_ERR_OR_NULL(after_dashes))
577 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
578 NULL, set_init_arg);
579
580 jump_label_init();
581
582 /*
583 * These use large bootmem allocations and must precede
584 * kmem_cache_init()
585 */
586 setup_log_buf(0);
587 vfs_caches_init_early();
588 sort_main_extable();
589 trap_init();
590 mm_init();
591
592 ftrace_init();
593
594 /* trace_printk can be enabled here */
595 early_trace_init();
596
597 /*
598 * Set up the scheduler prior starting any interrupts (such as the
599 * timer interrupt). Full topology setup happens at smp_init()
600 * time - but meanwhile we still have a functioning scheduler.
601 */
602 sched_init();
603 /*
604 * Disable preemption - early bootup scheduling is extremely
605 * fragile until we cpu_idle() for the first time.
606 */
607 preempt_disable();
608 if (WARN(!irqs_disabled(),
609 "Interrupts were enabled *very* early, fixing it\n"))
610 local_irq_disable();
611 radix_tree_init();
612
613 /*
614 * Set up housekeeping before setting up workqueues to allow the unbound
615 * workqueue to take non-housekeeping into account.
616 */
617 housekeeping_init();
618
619 /*
620 * Allow workqueue creation and work item queueing/cancelling
621 * early. Work item execution depends on kthreads and starts after
622 * workqueue_init().
623 */
624 workqueue_init_early();
625
626 rcu_init();
627
628 /* Trace events are available after this */
629 trace_init();
630
631 if (initcall_debug)
632 initcall_debug_enable();
633
634 context_tracking_init();
635 /* init some links before init_ISA_irqs() */
636 early_irq_init();
637 init_IRQ();
638 tick_init();
639 rcu_init_nohz();
640 init_timers();
641 hrtimers_init();
642 softirq_init();
643 timekeeping_init();
644 time_init();
645 sched_clock_postinit();
646 printk_safe_init();
647 perf_event_init();
648 profile_init();
649 call_function_init();
650 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
651 early_boot_irqs_disabled = false;
652 local_irq_enable();
653
654 kmem_cache_init_late();
655
656 /*
657 * HACK ALERT! This is early. We're enabling the console before
658 * we've done PCI setups etc, and console_init() must be aware of
659 * this. But we do want output early, in case something goes wrong.
660 */
661 console_init();
662 if (panic_later)
663 panic("Too many boot %s vars at `%s'", panic_later,
664 panic_param);
665
666 lockdep_info();
667
668 /*
669 * Need to run this when irqs are enabled, because it wants
670 * to self-test [hard/soft]-irqs on/off lock inversion bugs
671 * too:
672 */
673 locking_selftest();
674
675 /*
676 * This needs to be called before any devices perform DMA
677 * operations that might use the SWIOTLB bounce buffers. It will
678 * mark the bounce buffers as decrypted so that their usage will
679 * not cause "plain-text" data to be decrypted when accessed.
680 */
681 mem_encrypt_init();
682
683#ifdef CONFIG_BLK_DEV_INITRD
684 if (initrd_start && !initrd_below_start_ok &&
685 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
686 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
687 page_to_pfn(virt_to_page((void *)initrd_start)),
688 min_low_pfn);
689 initrd_start = 0;
690 }
691#endif
692 page_ext_init();
693 kmemleak_init();
694 debug_objects_mem_init();
695 setup_per_cpu_pageset();
696 numa_policy_init();
697 acpi_early_init();
698 if (late_time_init)
699 late_time_init();
700 calibrate_delay();
701 pid_idr_init();
702 anon_vma_init();
703#ifdef CONFIG_X86
704 if (efi_enabled(EFI_RUNTIME_SERVICES))
705 efi_enter_virtual_mode();
706#endif
707 thread_stack_cache_init();
708 cred_init();
709 fork_init();
710 proc_caches_init();
711 uts_ns_init();
712 buffer_init();
713 key_init();
714 security_init();
715 dbg_late_init();
716 vfs_caches_init();
717 pagecache_init();
718 signals_init();
719 seq_file_init();
720 proc_root_init();
721 nsfs_init();
722 cpuset_init();
723 cgroup_init();
724 taskstats_init_early();
725 delayacct_init();
726
727 check_bugs();
728
729 acpi_subsystem_init();
730 arch_post_acpi_subsys_init();
731 sfi_init_late();
732
733 if (efi_enabled(EFI_RUNTIME_SERVICES)) {
734 efi_free_boot_services();
735 }
736
737 /* Do the rest non-__init'ed, we're now alive */
738 rest_init();
739}
740
741/* Call all constructor functions linked into the kernel. */
742static void __init do_ctors(void)
743{
744#ifdef CONFIG_CONSTRUCTORS
745 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
746
747 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
748 (*fn)();
749#endif
750}
751
752#ifdef CONFIG_KALLSYMS
753struct blacklist_entry {
754 struct list_head next;
755 char *buf;
756};
757
758static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
759
760static int __init initcall_blacklist(char *str)
761{
762 char *str_entry;
763 struct blacklist_entry *entry;
764
765 /* str argument is a comma-separated list of functions */
766 do {
767 str_entry = strsep(&str, ",");
768 if (str_entry) {
769 pr_debug("blacklisting initcall %s\n", str_entry);
770 entry = alloc_bootmem(sizeof(*entry));
771 entry->buf = alloc_bootmem(strlen(str_entry) + 1);
772 strcpy(entry->buf, str_entry);
773 list_add(&entry->next, &blacklisted_initcalls);
774 }
775 } while (str_entry);
776
777 return 0;
778}
779
780static bool __init_or_module initcall_blacklisted(initcall_t fn)
781{
782 struct blacklist_entry *entry;
783 char fn_name[KSYM_SYMBOL_LEN];
784 unsigned long addr;
785
786 if (list_empty(&blacklisted_initcalls))
787 return false;
788
789 addr = (unsigned long) dereference_function_descriptor(fn);
790 sprint_symbol_no_offset(fn_name, addr);
791
792 /*
793 * fn will be "function_name [module_name]" where [module_name] is not
794 * displayed for built-in init functions. Strip off the [module_name].
795 */
796 strreplace(fn_name, ' ', '\0');
797
798 list_for_each_entry(entry, &blacklisted_initcalls, next) {
799 if (!strcmp(fn_name, entry->buf)) {
800 pr_debug("initcall %s blacklisted\n", fn_name);
801 return true;
802 }
803 }
804
805 return false;
806}
807#else
808static int __init initcall_blacklist(char *str)
809{
810 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
811 return 0;
812}
813
814static bool __init_or_module initcall_blacklisted(initcall_t fn)
815{
816 return false;
817}
818#endif
819__setup("initcall_blacklist=", initcall_blacklist);
820
821static __init_or_module void
822trace_initcall_start_cb(void *data, initcall_t fn)
823{
824 ktime_t *calltime = (ktime_t *)data;
825
826 printk(KERN_DEBUG "calling %pF @ %i\n", fn, task_pid_nr(current));
827 *calltime = ktime_get();
828}
829
830static __init_or_module void
831trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
832{
833 ktime_t *calltime = (ktime_t *)data;
834 ktime_t delta, rettime;
835 unsigned long long duration;
836
837 rettime = ktime_get();
838 delta = ktime_sub(rettime, *calltime);
839 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
840 printk(KERN_DEBUG "initcall %pF returned %d after %lld usecs\n",
841 fn, ret, duration);
842}
843
844static ktime_t initcall_calltime;
845
846#ifdef TRACEPOINTS_ENABLED
847static void __init initcall_debug_enable(void)
848{
849 int ret;
850
851 ret = register_trace_initcall_start(trace_initcall_start_cb,
852 &initcall_calltime);
853 ret |= register_trace_initcall_finish(trace_initcall_finish_cb,
854 &initcall_calltime);
855 WARN(ret, "Failed to register initcall tracepoints\n");
856}
857# define do_trace_initcall_start trace_initcall_start
858# define do_trace_initcall_finish trace_initcall_finish
859#else
860static inline void do_trace_initcall_start(initcall_t fn)
861{
862 if (!initcall_debug)
863 return;
864 trace_initcall_start_cb(&initcall_calltime, fn);
865}
866static inline void do_trace_initcall_finish(initcall_t fn, int ret)
867{
868 if (!initcall_debug)
869 return;
870 trace_initcall_finish_cb(&initcall_calltime, fn, ret);
871}
872#endif /* !TRACEPOINTS_ENABLED */
873
874int __init_or_module do_one_initcall(initcall_t fn)
875{
876 int count = preempt_count();
877 char msgbuf[64];
878 int ret;
879
880 if (initcall_blacklisted(fn))
881 return -EPERM;
882
883 do_trace_initcall_start(fn);
884 ret = fn();
885 do_trace_initcall_finish(fn, ret);
886
887 msgbuf[0] = 0;
888
889 if (preempt_count() != count) {
890 sprintf(msgbuf, "preemption imbalance ");
891 preempt_count_set(count);
892 }
893 if (irqs_disabled()) {
894 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
895 local_irq_enable();
896 }
897 WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
898
899 add_latent_entropy();
900 return ret;
901}
902
903
904extern initcall_t __initcall_start[];
905extern initcall_t __initcall0_start[];
906extern initcall_t __initcall1_start[];
907extern initcall_t __initcall2_start[];
908extern initcall_t __initcall3_start[];
909extern initcall_t __initcall4_start[];
910extern initcall_t __initcall5_start[];
911extern initcall_t __initcall6_start[];
912extern initcall_t __initcall7_start[];
913extern initcall_t __initcall_end[];
914
915static initcall_t *initcall_levels[] __initdata = {
916 __initcall0_start,
917 __initcall1_start,
918 __initcall2_start,
919 __initcall3_start,
920 __initcall4_start,
921 __initcall5_start,
922 __initcall6_start,
923 __initcall7_start,
924 __initcall_end,
925};
926
927/* Keep these in sync with initcalls in include/linux/init.h */
928static char *initcall_level_names[] __initdata = {
929 "pure",
930 "core",
931 "postcore",
932 "arch",
933 "subsys",
934 "fs",
935 "device",
936 "late",
937};
938
939static void __init do_initcall_level(int level)
940{
941 initcall_t *fn;
942
943 strcpy(initcall_command_line, saved_command_line);
944 parse_args(initcall_level_names[level],
945 initcall_command_line, __start___param,
946 __stop___param - __start___param,
947 level, level,
948 NULL, &repair_env_string);
949
950 trace_initcall_level(initcall_level_names[level]);
951 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
952 do_one_initcall(*fn);
953}
954
955static void __init do_initcalls(void)
956{
957 int level;
958
959 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
960 do_initcall_level(level);
961}
962
963/*
964 * Ok, the machine is now initialized. None of the devices
965 * have been touched yet, but the CPU subsystem is up and
966 * running, and memory and process management works.
967 *
968 * Now we can finally start doing some real work..
969 */
970static void __init do_basic_setup(void)
971{
972 cpuset_init_smp();
973 shmem_init();
974 driver_init();
975 init_irq_proc();
976 do_ctors();
977 usermodehelper_enable();
978 do_initcalls();
979}
980
981static void __init do_pre_smp_initcalls(void)
982{
983 initcall_t *fn;
984
985 trace_initcall_level("early");
986 for (fn = __initcall_start; fn < __initcall0_start; fn++)
987 do_one_initcall(*fn);
988}
989
990/*
991 * This function requests modules which should be loaded by default and is
992 * called twice right after initrd is mounted and right before init is
993 * exec'd. If such modules are on either initrd or rootfs, they will be
994 * loaded before control is passed to userland.
995 */
996void __init load_default_modules(void)
997{
998 load_default_elevator_module();
999}
1000
1001static int run_init_process(const char *init_filename)
1002{
1003 argv_init[0] = init_filename;
1004 return do_execve(getname_kernel(init_filename),
1005 (const char __user *const __user *)argv_init,
1006 (const char __user *const __user *)envp_init);
1007}
1008
1009static int try_to_run_init_process(const char *init_filename)
1010{
1011 int ret;
1012
1013 ret = run_init_process(init_filename);
1014
1015 if (ret && ret != -ENOENT) {
1016 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
1017 init_filename, ret);
1018 }
1019
1020 return ret;
1021}
1022
1023static noinline void __init kernel_init_freeable(void);
1024
1025#if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
1026bool rodata_enabled __ro_after_init = true;
1027static int __init set_debug_rodata(char *str)
1028{
1029 return strtobool(str, &rodata_enabled);
1030}
1031__setup("rodata=", set_debug_rodata);
1032#endif
1033
1034#ifdef CONFIG_STRICT_KERNEL_RWX
1035static void mark_readonly(void)
1036{
1037 if (rodata_enabled) {
1038 /*
1039 * load_module() results in W+X mappings, which are cleaned up
1040 * with call_rcu_sched(). Let's make sure that queued work is
1041 * flushed so that we don't hit false positives looking for
1042 * insecure pages which are W+X.
1043 */
1044 rcu_barrier_sched();
1045 mark_rodata_ro();
1046 rodata_test();
1047 } else
1048 pr_info("Kernel memory protection disabled.\n");
1049}
1050#else
1051static inline void mark_readonly(void)
1052{
1053 pr_warn("This architecture does not have kernel memory protection.\n");
1054}
1055#endif
1056
1057static int __ref kernel_init(void *unused)
1058{
1059 int ret;
1060
1061 kernel_init_freeable();
1062 /* need to finish all async __init code before freeing the memory */
1063 async_synchronize_full();
1064 ftrace_free_init_mem();
1065 jump_label_invalidate_initmem();
1066 free_initmem();
1067 mark_readonly();
1068 system_state = SYSTEM_RUNNING;
1069 numa_default_policy();
1070
1071 rcu_end_inkernel_boot();
1072
1073 if (ramdisk_execute_command) {
1074 ret = run_init_process(ramdisk_execute_command);
1075 if (!ret)
1076 return 0;
1077 pr_err("Failed to execute %s (error %d)\n",
1078 ramdisk_execute_command, ret);
1079 }
1080
1081 /*
1082 * We try each of these until one succeeds.
1083 *
1084 * The Bourne shell can be used instead of init if we are
1085 * trying to recover a really broken machine.
1086 */
1087 if (execute_command) {
1088 ret = run_init_process(execute_command);
1089 if (!ret)
1090 return 0;
1091 panic("Requested init %s failed (error %d).",
1092 execute_command, ret);
1093 }
1094 if (!try_to_run_init_process("/sbin/init") ||
1095 !try_to_run_init_process("/etc/init") ||
1096 !try_to_run_init_process("/bin/init") ||
1097 !try_to_run_init_process("/bin/sh"))
1098 return 0;
1099
1100 panic("No working init found. Try passing init= option to kernel. "
1101 "See Linux Documentation/admin-guide/init.rst for guidance.");
1102}
1103
1104static noinline void __init kernel_init_freeable(void)
1105{
1106 /*
1107 * Wait until kthreadd is all set-up.
1108 */
1109 wait_for_completion(&kthreadd_done);
1110
1111 /* Now the scheduler is fully set up and can do blocking allocations */
1112 gfp_allowed_mask = __GFP_BITS_MASK;
1113
1114 /*
1115 * init can allocate pages on any node
1116 */
1117 set_mems_allowed(node_states[N_MEMORY]);
1118
1119 cad_pid = task_pid(current);
1120
1121 smp_prepare_cpus(setup_max_cpus);
1122
1123 workqueue_init();
1124
1125 init_mm_internals();
1126
1127 do_pre_smp_initcalls();
1128 lockup_detector_init();
1129
1130 smp_init();
1131 sched_init_smp();
1132
1133 page_alloc_init_late();
1134
1135 do_basic_setup();
1136
1137 /* Open the /dev/console on the rootfs, this should never fail */
1138 if (ksys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
1139 pr_err("Warning: unable to open an initial console.\n");
1140
1141 (void) ksys_dup(0);
1142 (void) ksys_dup(0);
1143 /*
1144 * check if there is an early userspace init. If yes, let it do all
1145 * the work
1146 */
1147
1148 if (!ramdisk_execute_command)
1149 ramdisk_execute_command = "/init";
1150
1151 if (ksys_access((const char __user *)
1152 ramdisk_execute_command, 0) != 0) {
1153 ramdisk_execute_command = NULL;
1154 prepare_namespace();
1155 }
1156
1157 /*
1158 * Ok, we have completed the initial bootup, and
1159 * we're essentially up and running. Get rid of the
1160 * initmem segments and start the user-mode stuff..
1161 *
1162 * rootfs is available now, try loading the public keys
1163 * and default modules
1164 */
1165
1166 integrity_load_keys();
1167 load_default_modules();
1168}