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/page_ext.h>
66#include <linux/debug_locks.h>
67#include <linux/debugobjects.h>
68#include <linux/lockdep.h>
69#include <linux/kmemleak.h>
70#include <linux/padata.h>
71#include <linux/pid_namespace.h>
72#include <linux/device/driver.h>
73#include <linux/kthread.h>
74#include <linux/sched.h>
75#include <linux/sched/init.h>
76#include <linux/signal.h>
77#include <linux/idr.h>
78#include <linux/kgdb.h>
79#include <linux/ftrace.h>
80#include <linux/async.h>
81#include <linux/shmem_fs.h>
82#include <linux/slab.h>
83#include <linux/perf_event.h>
84#include <linux/ptrace.h>
85#include <linux/pti.h>
86#include <linux/blkdev.h>
87#include <linux/sched/clock.h>
88#include <linux/sched/task.h>
89#include <linux/sched/task_stack.h>
90#include <linux/context_tracking.h>
91#include <linux/random.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/mem_encrypt.h>
100#include <linux/kcsan.h>
101#include <linux/init_syscalls.h>
102#include <linux/stackdepot.h>
103#include <linux/randomize_kstack.h>
104#include <net/net_namespace.h>
105
106#include <asm/io.h>
107#include <asm/bugs.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
119extern void init_IRQ(void);
120extern void radix_tree_init(void);
121extern void maple_tree_init(void);
122
123/*
124 * Debug helper: via this flag we know that we are in 'early bootup code'
125 * where only the boot processor is running with IRQ disabled. This means
126 * two things - IRQ must not be enabled before the flag is cleared and some
127 * operations which are not allowed with IRQ disabled are allowed while the
128 * flag is set.
129 */
130bool early_boot_irqs_disabled __read_mostly;
131
132enum system_states system_state __read_mostly;
133EXPORT_SYMBOL(system_state);
134
135/*
136 * Boot command-line arguments
137 */
138#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
139#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
140
141extern void time_init(void);
142/* Default late time init is NULL. archs can override this later. */
143void (*__initdata late_time_init)(void);
144
145/* Untouched command line saved by arch-specific code. */
146char __initdata boot_command_line[COMMAND_LINE_SIZE];
147/* Untouched saved command line (eg. for /proc) */
148char *saved_command_line __ro_after_init;
149unsigned int saved_command_line_len __ro_after_init;
150/* Command line for parameter parsing */
151static char *static_command_line;
152/* Untouched extra command line */
153static char *extra_command_line;
154/* Extra init arguments */
155static char *extra_init_args;
156
157#ifdef CONFIG_BOOT_CONFIG
158/* Is bootconfig on command line? */
159static bool bootconfig_found;
160static size_t initargs_offs;
161#else
162# define bootconfig_found false
163# define initargs_offs 0
164#endif
165
166static char *execute_command;
167static char *ramdisk_execute_command = "/init";
168
169/*
170 * Used to generate warnings if static_key manipulation functions are used
171 * before jump_label_init is called.
172 */
173bool static_key_initialized __read_mostly;
174EXPORT_SYMBOL_GPL(static_key_initialized);
175
176/*
177 * If set, this is an indication to the drivers that reset the underlying
178 * device before going ahead with the initialization otherwise driver might
179 * rely on the BIOS and skip the reset operation.
180 *
181 * This is useful if kernel is booting in an unreliable environment.
182 * For ex. kdump situation where previous kernel has crashed, BIOS has been
183 * skipped and devices will be in unknown state.
184 */
185unsigned int reset_devices;
186EXPORT_SYMBOL(reset_devices);
187
188static int __init set_reset_devices(char *str)
189{
190 reset_devices = 1;
191 return 1;
192}
193
194__setup("reset_devices", set_reset_devices);
195
196static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
197const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
198static const char *panic_later, *panic_param;
199
200extern const struct obs_kernel_param __setup_start[], __setup_end[];
201
202static bool __init obsolete_checksetup(char *line)
203{
204 const struct obs_kernel_param *p;
205 bool had_early_param = false;
206
207 p = __setup_start;
208 do {
209 int n = strlen(p->str);
210 if (parameqn(line, p->str, n)) {
211 if (p->early) {
212 /* Already done in parse_early_param?
213 * (Needs exact match on param part).
214 * Keep iterating, as we can have early
215 * params and __setups of same names 8( */
216 if (line[n] == '\0' || line[n] == '=')
217 had_early_param = true;
218 } else if (!p->setup_func) {
219 pr_warn("Parameter %s is obsolete, ignored\n",
220 p->str);
221 return true;
222 } else if (p->setup_func(line + n))
223 return true;
224 }
225 p++;
226 } while (p < __setup_end);
227
228 return had_early_param;
229}
230
231/*
232 * This should be approx 2 Bo*oMips to start (note initial shift), and will
233 * still work even if initially too large, it will just take slightly longer
234 */
235unsigned long loops_per_jiffy = (1<<12);
236EXPORT_SYMBOL(loops_per_jiffy);
237
238static int __init debug_kernel(char *str)
239{
240 console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
241 return 0;
242}
243
244static int __init quiet_kernel(char *str)
245{
246 console_loglevel = CONSOLE_LOGLEVEL_QUIET;
247 return 0;
248}
249
250early_param("debug", debug_kernel);
251early_param("quiet", quiet_kernel);
252
253static int __init loglevel(char *str)
254{
255 int newlevel;
256
257 /*
258 * Only update loglevel value when a correct setting was passed,
259 * to prevent blind crashes (when loglevel being set to 0) that
260 * are quite hard to debug
261 */
262 if (get_option(&str, &newlevel)) {
263 console_loglevel = newlevel;
264 return 0;
265 }
266
267 return -EINVAL;
268}
269
270early_param("loglevel", loglevel);
271
272#ifdef CONFIG_BLK_DEV_INITRD
273static void * __init get_boot_config_from_initrd(size_t *_size)
274{
275 u32 size, csum;
276 char *data;
277 u32 *hdr;
278 int i;
279
280 if (!initrd_end)
281 return NULL;
282
283 data = (char *)initrd_end - BOOTCONFIG_MAGIC_LEN;
284 /*
285 * Since Grub may align the size of initrd to 4, we must
286 * check the preceding 3 bytes as well.
287 */
288 for (i = 0; i < 4; i++) {
289 if (!memcmp(data, BOOTCONFIG_MAGIC, BOOTCONFIG_MAGIC_LEN))
290 goto found;
291 data--;
292 }
293 return NULL;
294
295found:
296 hdr = (u32 *)(data - 8);
297 size = le32_to_cpu(hdr[0]);
298 csum = le32_to_cpu(hdr[1]);
299
300 data = ((void *)hdr) - size;
301 if ((unsigned long)data < initrd_start) {
302 pr_err("bootconfig size %d is greater than initrd size %ld\n",
303 size, initrd_end - initrd_start);
304 return NULL;
305 }
306
307 if (xbc_calc_checksum(data, size) != csum) {
308 pr_err("bootconfig checksum failed\n");
309 return NULL;
310 }
311
312 /* Remove bootconfig from initramfs/initrd */
313 initrd_end = (unsigned long)data;
314 if (_size)
315 *_size = size;
316
317 return data;
318}
319#else
320static void * __init get_boot_config_from_initrd(size_t *_size)
321{
322 return NULL;
323}
324#endif
325
326#ifdef CONFIG_BOOT_CONFIG
327
328static char xbc_namebuf[XBC_KEYLEN_MAX] __initdata;
329
330#define rest(dst, end) ((end) > (dst) ? (end) - (dst) : 0)
331
332static int __init xbc_snprint_cmdline(char *buf, size_t size,
333 struct xbc_node *root)
334{
335 struct xbc_node *knode, *vnode;
336 char *end = buf + size;
337 const char *val;
338 int ret;
339
340 xbc_node_for_each_key_value(root, knode, val) {
341 ret = xbc_node_compose_key_after(root, knode,
342 xbc_namebuf, XBC_KEYLEN_MAX);
343 if (ret < 0)
344 return ret;
345
346 vnode = xbc_node_get_child(knode);
347 if (!vnode) {
348 ret = snprintf(buf, rest(buf, end), "%s ", xbc_namebuf);
349 if (ret < 0)
350 return ret;
351 buf += ret;
352 continue;
353 }
354 xbc_array_for_each_value(vnode, val) {
355 ret = snprintf(buf, rest(buf, end), "%s=\"%s\" ",
356 xbc_namebuf, val);
357 if (ret < 0)
358 return ret;
359 buf += ret;
360 }
361 }
362
363 return buf - (end - size);
364}
365#undef rest
366
367/* Make an extra command line under given key word */
368static char * __init xbc_make_cmdline(const char *key)
369{
370 struct xbc_node *root;
371 char *new_cmdline;
372 int ret, len = 0;
373
374 root = xbc_find_node(key);
375 if (!root)
376 return NULL;
377
378 /* Count required buffer size */
379 len = xbc_snprint_cmdline(NULL, 0, root);
380 if (len <= 0)
381 return NULL;
382
383 new_cmdline = memblock_alloc(len + 1, SMP_CACHE_BYTES);
384 if (!new_cmdline) {
385 pr_err("Failed to allocate memory for extra kernel cmdline.\n");
386 return NULL;
387 }
388
389 ret = xbc_snprint_cmdline(new_cmdline, len + 1, root);
390 if (ret < 0 || ret > len) {
391 pr_err("Failed to print extra kernel cmdline.\n");
392 memblock_free(new_cmdline, len + 1);
393 return NULL;
394 }
395
396 return new_cmdline;
397}
398
399static int __init bootconfig_params(char *param, char *val,
400 const char *unused, void *arg)
401{
402 if (strcmp(param, "bootconfig") == 0) {
403 bootconfig_found = true;
404 }
405 return 0;
406}
407
408static int __init warn_bootconfig(char *str)
409{
410 /* The 'bootconfig' has been handled by bootconfig_params(). */
411 return 0;
412}
413
414static void __init setup_boot_config(void)
415{
416 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
417 const char *msg, *data;
418 int pos, ret;
419 size_t size;
420 char *err;
421
422 /* Cut out the bootconfig data even if we have no bootconfig option */
423 data = get_boot_config_from_initrd(&size);
424 /* If there is no bootconfig in initrd, try embedded one. */
425 if (!data)
426 data = xbc_get_embedded_bootconfig(&size);
427
428 strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
429 err = parse_args("bootconfig", tmp_cmdline, NULL, 0, 0, 0, NULL,
430 bootconfig_params);
431
432 if (IS_ERR(err) || !bootconfig_found)
433 return;
434
435 /* parse_args() stops at the next param of '--' and returns an address */
436 if (err)
437 initargs_offs = err - tmp_cmdline;
438
439 if (!data) {
440 pr_err("'bootconfig' found on command line, but no bootconfig found\n");
441 return;
442 }
443
444 if (size >= XBC_DATA_MAX) {
445 pr_err("bootconfig size %ld greater than max size %d\n",
446 (long)size, XBC_DATA_MAX);
447 return;
448 }
449
450 ret = xbc_init(data, size, &msg, &pos);
451 if (ret < 0) {
452 if (pos < 0)
453 pr_err("Failed to init bootconfig: %s.\n", msg);
454 else
455 pr_err("Failed to parse bootconfig: %s at %d.\n",
456 msg, pos);
457 } else {
458 xbc_get_info(&ret, NULL);
459 pr_info("Load bootconfig: %ld bytes %d nodes\n", (long)size, ret);
460 /* keys starting with "kernel." are passed via cmdline */
461 extra_command_line = xbc_make_cmdline("kernel");
462 /* Also, "init." keys are init arguments */
463 extra_init_args = xbc_make_cmdline("init");
464 }
465 return;
466}
467
468static void __init exit_boot_config(void)
469{
470 xbc_exit();
471}
472
473#else /* !CONFIG_BOOT_CONFIG */
474
475static void __init setup_boot_config(void)
476{
477 /* Remove bootconfig data from initrd */
478 get_boot_config_from_initrd(NULL);
479}
480
481static int __init warn_bootconfig(char *str)
482{
483 pr_warn("WARNING: 'bootconfig' found on the kernel command line but CONFIG_BOOT_CONFIG is not set.\n");
484 return 0;
485}
486
487#define exit_boot_config() do {} while (0)
488
489#endif /* CONFIG_BOOT_CONFIG */
490
491early_param("bootconfig", warn_bootconfig);
492
493/* Change NUL term back to "=", to make "param" the whole string. */
494static void __init repair_env_string(char *param, char *val)
495{
496 if (val) {
497 /* param=val or param="val"? */
498 if (val == param+strlen(param)+1)
499 val[-1] = '=';
500 else if (val == param+strlen(param)+2) {
501 val[-2] = '=';
502 memmove(val-1, val, strlen(val)+1);
503 } else
504 BUG();
505 }
506}
507
508/* Anything after -- gets handed straight to init. */
509static int __init set_init_arg(char *param, char *val,
510 const char *unused, void *arg)
511{
512 unsigned int i;
513
514 if (panic_later)
515 return 0;
516
517 repair_env_string(param, val);
518
519 for (i = 0; argv_init[i]; i++) {
520 if (i == MAX_INIT_ARGS) {
521 panic_later = "init";
522 panic_param = param;
523 return 0;
524 }
525 }
526 argv_init[i] = param;
527 return 0;
528}
529
530/*
531 * Unknown boot options get handed to init, unless they look like
532 * unused parameters (modprobe will find them in /proc/cmdline).
533 */
534static int __init unknown_bootoption(char *param, char *val,
535 const char *unused, void *arg)
536{
537 size_t len = strlen(param);
538
539 repair_env_string(param, val);
540
541 /* Handle obsolete-style parameters */
542 if (obsolete_checksetup(param))
543 return 0;
544
545 /* Unused module parameter. */
546 if (strnchr(param, len, '.'))
547 return 0;
548
549 if (panic_later)
550 return 0;
551
552 if (val) {
553 /* Environment option */
554 unsigned int i;
555 for (i = 0; envp_init[i]; i++) {
556 if (i == MAX_INIT_ENVS) {
557 panic_later = "env";
558 panic_param = param;
559 }
560 if (!strncmp(param, envp_init[i], len+1))
561 break;
562 }
563 envp_init[i] = param;
564 } else {
565 /* Command line option */
566 unsigned int i;
567 for (i = 0; argv_init[i]; i++) {
568 if (i == MAX_INIT_ARGS) {
569 panic_later = "init";
570 panic_param = param;
571 }
572 }
573 argv_init[i] = param;
574 }
575 return 0;
576}
577
578static int __init init_setup(char *str)
579{
580 unsigned int i;
581
582 execute_command = str;
583 /*
584 * In case LILO is going to boot us with default command line,
585 * it prepends "auto" before the whole cmdline which makes
586 * the shell think it should execute a script with such name.
587 * So we ignore all arguments entered _before_ init=... [MJ]
588 */
589 for (i = 1; i < MAX_INIT_ARGS; i++)
590 argv_init[i] = NULL;
591 return 1;
592}
593__setup("init=", init_setup);
594
595static int __init rdinit_setup(char *str)
596{
597 unsigned int i;
598
599 ramdisk_execute_command = str;
600 /* See "auto" comment in init_setup */
601 for (i = 1; i < MAX_INIT_ARGS; i++)
602 argv_init[i] = NULL;
603 return 1;
604}
605__setup("rdinit=", rdinit_setup);
606
607#ifndef CONFIG_SMP
608static const unsigned int setup_max_cpus = NR_CPUS;
609static inline void setup_nr_cpu_ids(void) { }
610static inline void smp_prepare_cpus(unsigned int maxcpus) { }
611#endif
612
613/*
614 * We need to store the untouched command line for future reference.
615 * We also need to store the touched command line since the parameter
616 * parsing is performed in place, and we should allow a component to
617 * store reference of name/value for future reference.
618 */
619static void __init setup_command_line(char *command_line)
620{
621 size_t len, xlen = 0, ilen = 0;
622
623 if (extra_command_line)
624 xlen = strlen(extra_command_line);
625 if (extra_init_args)
626 ilen = strlen(extra_init_args) + 4; /* for " -- " */
627
628 len = xlen + strlen(boot_command_line) + 1;
629
630 saved_command_line = memblock_alloc(len + ilen, SMP_CACHE_BYTES);
631 if (!saved_command_line)
632 panic("%s: Failed to allocate %zu bytes\n", __func__, len + ilen);
633
634 static_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
635 if (!static_command_line)
636 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
637
638 if (xlen) {
639 /*
640 * We have to put extra_command_line before boot command
641 * lines because there could be dashes (separator of init
642 * command line) in the command lines.
643 */
644 strcpy(saved_command_line, extra_command_line);
645 strcpy(static_command_line, extra_command_line);
646 }
647 strcpy(saved_command_line + xlen, boot_command_line);
648 strcpy(static_command_line + xlen, command_line);
649
650 if (ilen) {
651 /*
652 * Append supplemental init boot args to saved_command_line
653 * so that user can check what command line options passed
654 * to init.
655 * The order should always be
656 * " -- "[bootconfig init-param][cmdline init-param]
657 */
658 if (initargs_offs) {
659 len = xlen + initargs_offs;
660 strcpy(saved_command_line + len, extra_init_args);
661 len += ilen - 4; /* strlen(extra_init_args) */
662 strcpy(saved_command_line + len,
663 boot_command_line + initargs_offs - 1);
664 } else {
665 len = strlen(saved_command_line);
666 strcpy(saved_command_line + len, " -- ");
667 len += 4;
668 strcpy(saved_command_line + len, extra_init_args);
669 }
670 }
671
672 saved_command_line_len = strlen(saved_command_line);
673}
674
675/*
676 * We need to finalize in a non-__init function or else race conditions
677 * between the root thread and the init thread may cause start_kernel to
678 * be reaped by free_initmem before the root thread has proceeded to
679 * cpu_idle.
680 *
681 * gcc-3.4 accidentally inlines this function, so use noinline.
682 */
683
684static __initdata DECLARE_COMPLETION(kthreadd_done);
685
686noinline void __ref rest_init(void)
687{
688 struct task_struct *tsk;
689 int pid;
690
691 rcu_scheduler_starting();
692 /*
693 * We need to spawn init first so that it obtains pid 1, however
694 * the init task will end up wanting to create kthreads, which, if
695 * we schedule it before we create kthreadd, will OOPS.
696 */
697 pid = user_mode_thread(kernel_init, NULL, CLONE_FS);
698 /*
699 * Pin init on the boot CPU. Task migration is not properly working
700 * until sched_init_smp() has been run. It will set the allowed
701 * CPUs for init to the non isolated CPUs.
702 */
703 rcu_read_lock();
704 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
705 tsk->flags |= PF_NO_SETAFFINITY;
706 set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id()));
707 rcu_read_unlock();
708
709 numa_default_policy();
710 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
711 rcu_read_lock();
712 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
713 rcu_read_unlock();
714
715 /*
716 * Enable might_sleep() and smp_processor_id() checks.
717 * They cannot be enabled earlier because with CONFIG_PREEMPTION=y
718 * kernel_thread() would trigger might_sleep() splats. With
719 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled
720 * already, but it's stuck on the kthreadd_done completion.
721 */
722 system_state = SYSTEM_SCHEDULING;
723
724 complete(&kthreadd_done);
725
726 /*
727 * The boot idle thread must execute schedule()
728 * at least once to get things moving:
729 */
730 schedule_preempt_disabled();
731 /* Call into cpu_idle with preempt disabled */
732 cpu_startup_entry(CPUHP_ONLINE);
733}
734
735/* Check for early params. */
736static int __init do_early_param(char *param, char *val,
737 const char *unused, void *arg)
738{
739 const struct obs_kernel_param *p;
740
741 for (p = __setup_start; p < __setup_end; p++) {
742 if ((p->early && parameq(param, p->str)) ||
743 (strcmp(param, "console") == 0 &&
744 strcmp(p->str, "earlycon") == 0)
745 ) {
746 if (p->setup_func(val) != 0)
747 pr_warn("Malformed early option '%s'\n", param);
748 }
749 }
750 /* We accept everything at this stage. */
751 return 0;
752}
753
754void __init parse_early_options(char *cmdline)
755{
756 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
757 do_early_param);
758}
759
760/* Arch code calls this early on, or if not, just before other parsing. */
761void __init parse_early_param(void)
762{
763 static int done __initdata;
764 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
765
766 if (done)
767 return;
768
769 /* All fall through to do_early_param. */
770 strscpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
771 parse_early_options(tmp_cmdline);
772 done = 1;
773}
774
775void __init __weak arch_post_acpi_subsys_init(void) { }
776
777void __init __weak smp_setup_processor_id(void)
778{
779}
780
781# if THREAD_SIZE >= PAGE_SIZE
782void __init __weak thread_stack_cache_init(void)
783{
784}
785#endif
786
787void __init __weak mem_encrypt_init(void) { }
788
789void __init __weak poking_init(void) { }
790
791void __init __weak pgtable_cache_init(void) { }
792
793void __init __weak trap_init(void) { }
794
795bool initcall_debug;
796core_param(initcall_debug, initcall_debug, bool, 0644);
797
798#ifdef TRACEPOINTS_ENABLED
799static void __init initcall_debug_enable(void);
800#else
801static inline void initcall_debug_enable(void)
802{
803}
804#endif
805
806/* Report memory auto-initialization states for this boot. */
807static void __init report_meminit(void)
808{
809 const char *stack;
810
811 if (IS_ENABLED(CONFIG_INIT_STACK_ALL_PATTERN))
812 stack = "all(pattern)";
813 else if (IS_ENABLED(CONFIG_INIT_STACK_ALL_ZERO))
814 stack = "all(zero)";
815 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL))
816 stack = "byref_all(zero)";
817 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF))
818 stack = "byref(zero)";
819 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_USER))
820 stack = "__user(zero)";
821 else
822 stack = "off";
823
824 pr_info("mem auto-init: stack:%s, heap alloc:%s, heap free:%s\n",
825 stack, want_init_on_alloc(GFP_KERNEL) ? "on" : "off",
826 want_init_on_free() ? "on" : "off");
827 if (want_init_on_free())
828 pr_info("mem auto-init: clearing system memory may take some time...\n");
829}
830
831/*
832 * Set up kernel memory allocators
833 */
834static void __init mm_init(void)
835{
836 /*
837 * page_ext requires contiguous pages,
838 * bigger than MAX_ORDER unless SPARSEMEM.
839 */
840 page_ext_init_flatmem();
841 init_mem_debugging_and_hardening();
842 kfence_alloc_pool();
843 report_meminit();
844 kmsan_init_shadow();
845 stack_depot_early_init();
846 mem_init();
847 mem_init_print_info();
848 kmem_cache_init();
849 /*
850 * page_owner must be initialized after buddy is ready, and also after
851 * slab is ready so that stack_depot_init() works properly
852 */
853 page_ext_init_flatmem_late();
854 kmemleak_init();
855 pgtable_init();
856 debug_objects_mem_init();
857 vmalloc_init();
858 /* Should be run after vmap initialization */
859 if (early_page_ext_enabled())
860 page_ext_init();
861 /* Should be run before the first non-init thread is created */
862 init_espfix_bsp();
863 /* Should be run after espfix64 is set up. */
864 pti_init();
865 kmsan_init_runtime();
866 mm_cache_init();
867}
868
869#ifdef CONFIG_RANDOMIZE_KSTACK_OFFSET
870DEFINE_STATIC_KEY_MAYBE_RO(CONFIG_RANDOMIZE_KSTACK_OFFSET_DEFAULT,
871 randomize_kstack_offset);
872DEFINE_PER_CPU(u32, kstack_offset);
873
874static int __init early_randomize_kstack_offset(char *buf)
875{
876 int ret;
877 bool bool_result;
878
879 ret = kstrtobool(buf, &bool_result);
880 if (ret)
881 return ret;
882
883 if (bool_result)
884 static_branch_enable(&randomize_kstack_offset);
885 else
886 static_branch_disable(&randomize_kstack_offset);
887 return 0;
888}
889early_param("randomize_kstack_offset", early_randomize_kstack_offset);
890#endif
891
892void __init __weak arch_call_rest_init(void)
893{
894 rest_init();
895}
896
897static void __init print_unknown_bootoptions(void)
898{
899 char *unknown_options;
900 char *end;
901 const char *const *p;
902 size_t len;
903
904 if (panic_later || (!argv_init[1] && !envp_init[2]))
905 return;
906
907 /*
908 * Determine how many options we have to print out, plus a space
909 * before each
910 */
911 len = 1; /* null terminator */
912 for (p = &argv_init[1]; *p; p++) {
913 len++;
914 len += strlen(*p);
915 }
916 for (p = &envp_init[2]; *p; p++) {
917 len++;
918 len += strlen(*p);
919 }
920
921 unknown_options = memblock_alloc(len, SMP_CACHE_BYTES);
922 if (!unknown_options) {
923 pr_err("%s: Failed to allocate %zu bytes\n",
924 __func__, len);
925 return;
926 }
927 end = unknown_options;
928
929 for (p = &argv_init[1]; *p; p++)
930 end += sprintf(end, " %s", *p);
931 for (p = &envp_init[2]; *p; p++)
932 end += sprintf(end, " %s", *p);
933
934 /* Start at unknown_options[1] to skip the initial space */
935 pr_notice("Unknown kernel command line parameters \"%s\", will be passed to user space.\n",
936 &unknown_options[1]);
937 memblock_free(unknown_options, len);
938}
939
940asmlinkage __visible void __init __no_sanitize_address start_kernel(void)
941{
942 char *command_line;
943 char *after_dashes;
944
945 set_task_stack_end_magic(&init_task);
946 smp_setup_processor_id();
947 debug_objects_early_init();
948 init_vmlinux_build_id();
949
950 cgroup_init_early();
951
952 local_irq_disable();
953 early_boot_irqs_disabled = true;
954
955 /*
956 * Interrupts are still disabled. Do necessary setups, then
957 * enable them.
958 */
959 boot_cpu_init();
960 page_address_init();
961 pr_notice("%s", linux_banner);
962 early_security_init();
963 setup_arch(&command_line);
964 setup_boot_config();
965 setup_command_line(command_line);
966 setup_nr_cpu_ids();
967 setup_per_cpu_areas();
968 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
969 boot_cpu_hotplug_init();
970
971 build_all_zonelists(NULL);
972 page_alloc_init();
973
974 pr_notice("Kernel command line: %s\n", saved_command_line);
975 /* parameters may set static keys */
976 jump_label_init();
977 parse_early_param();
978 after_dashes = parse_args("Booting kernel",
979 static_command_line, __start___param,
980 __stop___param - __start___param,
981 -1, -1, NULL, &unknown_bootoption);
982 print_unknown_bootoptions();
983 if (!IS_ERR_OR_NULL(after_dashes))
984 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
985 NULL, set_init_arg);
986 if (extra_init_args)
987 parse_args("Setting extra init args", extra_init_args,
988 NULL, 0, -1, -1, NULL, set_init_arg);
989
990 /* Architectural and non-timekeeping rng init, before allocator init */
991 random_init_early(command_line);
992
993 /*
994 * These use large bootmem allocations and must precede
995 * kmem_cache_init()
996 */
997 setup_log_buf(0);
998 vfs_caches_init_early();
999 sort_main_extable();
1000 trap_init();
1001 mm_init();
1002 poking_init();
1003 ftrace_init();
1004
1005 /* trace_printk can be enabled here */
1006 early_trace_init();
1007
1008 /*
1009 * Set up the scheduler prior starting any interrupts (such as the
1010 * timer interrupt). Full topology setup happens at smp_init()
1011 * time - but meanwhile we still have a functioning scheduler.
1012 */
1013 sched_init();
1014
1015 if (WARN(!irqs_disabled(),
1016 "Interrupts were enabled *very* early, fixing it\n"))
1017 local_irq_disable();
1018 radix_tree_init();
1019 maple_tree_init();
1020
1021 /*
1022 * Set up housekeeping before setting up workqueues to allow the unbound
1023 * workqueue to take non-housekeeping into account.
1024 */
1025 housekeeping_init();
1026
1027 /*
1028 * Allow workqueue creation and work item queueing/cancelling
1029 * early. Work item execution depends on kthreads and starts after
1030 * workqueue_init().
1031 */
1032 workqueue_init_early();
1033
1034 rcu_init();
1035
1036 /* Trace events are available after this */
1037 trace_init();
1038
1039 if (initcall_debug)
1040 initcall_debug_enable();
1041
1042 context_tracking_init();
1043 /* init some links before init_ISA_irqs() */
1044 early_irq_init();
1045 init_IRQ();
1046 tick_init();
1047 rcu_init_nohz();
1048 init_timers();
1049 srcu_init();
1050 hrtimers_init();
1051 softirq_init();
1052 timekeeping_init();
1053 time_init();
1054
1055 /* This must be after timekeeping is initialized */
1056 random_init();
1057
1058 /* These make use of the fully initialized rng */
1059 kfence_init();
1060 boot_init_stack_canary();
1061
1062 perf_event_init();
1063 profile_init();
1064 call_function_init();
1065 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
1066
1067 early_boot_irqs_disabled = false;
1068 local_irq_enable();
1069
1070 kmem_cache_init_late();
1071
1072 /*
1073 * HACK ALERT! This is early. We're enabling the console before
1074 * we've done PCI setups etc, and console_init() must be aware of
1075 * this. But we do want output early, in case something goes wrong.
1076 */
1077 console_init();
1078 if (panic_later)
1079 panic("Too many boot %s vars at `%s'", panic_later,
1080 panic_param);
1081
1082 lockdep_init();
1083
1084 /*
1085 * Need to run this when irqs are enabled, because it wants
1086 * to self-test [hard/soft]-irqs on/off lock inversion bugs
1087 * too:
1088 */
1089 locking_selftest();
1090
1091 /*
1092 * This needs to be called before any devices perform DMA
1093 * operations that might use the SWIOTLB bounce buffers. It will
1094 * mark the bounce buffers as decrypted so that their usage will
1095 * not cause "plain-text" data to be decrypted when accessed.
1096 */
1097 mem_encrypt_init();
1098
1099#ifdef CONFIG_BLK_DEV_INITRD
1100 if (initrd_start && !initrd_below_start_ok &&
1101 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
1102 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
1103 page_to_pfn(virt_to_page((void *)initrd_start)),
1104 min_low_pfn);
1105 initrd_start = 0;
1106 }
1107#endif
1108 setup_per_cpu_pageset();
1109 numa_policy_init();
1110 acpi_early_init();
1111 if (late_time_init)
1112 late_time_init();
1113 sched_clock_init();
1114 calibrate_delay();
1115 pid_idr_init();
1116 anon_vma_init();
1117#ifdef CONFIG_X86
1118 if (efi_enabled(EFI_RUNTIME_SERVICES))
1119 efi_enter_virtual_mode();
1120#endif
1121 thread_stack_cache_init();
1122 cred_init();
1123 fork_init();
1124 proc_caches_init();
1125 uts_ns_init();
1126 key_init();
1127 security_init();
1128 dbg_late_init();
1129 net_ns_init();
1130 vfs_caches_init();
1131 pagecache_init();
1132 signals_init();
1133 seq_file_init();
1134 proc_root_init();
1135 nsfs_init();
1136 cpuset_init();
1137 cgroup_init();
1138 taskstats_init_early();
1139 delayacct_init();
1140
1141 check_bugs();
1142
1143 acpi_subsystem_init();
1144 arch_post_acpi_subsys_init();
1145 kcsan_init();
1146
1147 /* Do the rest non-__init'ed, we're now alive */
1148 arch_call_rest_init();
1149
1150 prevent_tail_call_optimization();
1151}
1152
1153/* Call all constructor functions linked into the kernel. */
1154static void __init do_ctors(void)
1155{
1156/*
1157 * For UML, the constructors have already been called by the
1158 * normal setup code as it's just a normal ELF binary, so we
1159 * cannot do it again - but we do need CONFIG_CONSTRUCTORS
1160 * even on UML for modules.
1161 */
1162#if defined(CONFIG_CONSTRUCTORS) && !defined(CONFIG_UML)
1163 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
1164
1165 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
1166 (*fn)();
1167#endif
1168}
1169
1170#ifdef CONFIG_KALLSYMS
1171struct blacklist_entry {
1172 struct list_head next;
1173 char *buf;
1174};
1175
1176static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
1177
1178static int __init initcall_blacklist(char *str)
1179{
1180 char *str_entry;
1181 struct blacklist_entry *entry;
1182
1183 /* str argument is a comma-separated list of functions */
1184 do {
1185 str_entry = strsep(&str, ",");
1186 if (str_entry) {
1187 pr_debug("blacklisting initcall %s\n", str_entry);
1188 entry = memblock_alloc(sizeof(*entry),
1189 SMP_CACHE_BYTES);
1190 if (!entry)
1191 panic("%s: Failed to allocate %zu bytes\n",
1192 __func__, sizeof(*entry));
1193 entry->buf = memblock_alloc(strlen(str_entry) + 1,
1194 SMP_CACHE_BYTES);
1195 if (!entry->buf)
1196 panic("%s: Failed to allocate %zu bytes\n",
1197 __func__, strlen(str_entry) + 1);
1198 strcpy(entry->buf, str_entry);
1199 list_add(&entry->next, &blacklisted_initcalls);
1200 }
1201 } while (str_entry);
1202
1203 return 1;
1204}
1205
1206static bool __init_or_module initcall_blacklisted(initcall_t fn)
1207{
1208 struct blacklist_entry *entry;
1209 char fn_name[KSYM_SYMBOL_LEN];
1210 unsigned long addr;
1211
1212 if (list_empty(&blacklisted_initcalls))
1213 return false;
1214
1215 addr = (unsigned long) dereference_function_descriptor(fn);
1216 sprint_symbol_no_offset(fn_name, addr);
1217
1218 /*
1219 * fn will be "function_name [module_name]" where [module_name] is not
1220 * displayed for built-in init functions. Strip off the [module_name].
1221 */
1222 strreplace(fn_name, ' ', '\0');
1223
1224 list_for_each_entry(entry, &blacklisted_initcalls, next) {
1225 if (!strcmp(fn_name, entry->buf)) {
1226 pr_debug("initcall %s blacklisted\n", fn_name);
1227 return true;
1228 }
1229 }
1230
1231 return false;
1232}
1233#else
1234static int __init initcall_blacklist(char *str)
1235{
1236 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
1237 return 0;
1238}
1239
1240static bool __init_or_module initcall_blacklisted(initcall_t fn)
1241{
1242 return false;
1243}
1244#endif
1245__setup("initcall_blacklist=", initcall_blacklist);
1246
1247static __init_or_module void
1248trace_initcall_start_cb(void *data, initcall_t fn)
1249{
1250 ktime_t *calltime = data;
1251
1252 printk(KERN_DEBUG "calling %pS @ %i\n", fn, task_pid_nr(current));
1253 *calltime = ktime_get();
1254}
1255
1256static __init_or_module void
1257trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
1258{
1259 ktime_t rettime, *calltime = data;
1260
1261 rettime = ktime_get();
1262 printk(KERN_DEBUG "initcall %pS returned %d after %lld usecs\n",
1263 fn, ret, (unsigned long long)ktime_us_delta(rettime, *calltime));
1264}
1265
1266static ktime_t initcall_calltime;
1267
1268#ifdef TRACEPOINTS_ENABLED
1269static void __init initcall_debug_enable(void)
1270{
1271 int ret;
1272
1273 ret = register_trace_initcall_start(trace_initcall_start_cb,
1274 &initcall_calltime);
1275 ret |= register_trace_initcall_finish(trace_initcall_finish_cb,
1276 &initcall_calltime);
1277 WARN(ret, "Failed to register initcall tracepoints\n");
1278}
1279# define do_trace_initcall_start trace_initcall_start
1280# define do_trace_initcall_finish trace_initcall_finish
1281#else
1282static inline void do_trace_initcall_start(initcall_t fn)
1283{
1284 if (!initcall_debug)
1285 return;
1286 trace_initcall_start_cb(&initcall_calltime, fn);
1287}
1288static inline void do_trace_initcall_finish(initcall_t fn, int ret)
1289{
1290 if (!initcall_debug)
1291 return;
1292 trace_initcall_finish_cb(&initcall_calltime, fn, ret);
1293}
1294#endif /* !TRACEPOINTS_ENABLED */
1295
1296int __init_or_module do_one_initcall(initcall_t fn)
1297{
1298 int count = preempt_count();
1299 char msgbuf[64];
1300 int ret;
1301
1302 if (initcall_blacklisted(fn))
1303 return -EPERM;
1304
1305 do_trace_initcall_start(fn);
1306 ret = fn();
1307 do_trace_initcall_finish(fn, ret);
1308
1309 msgbuf[0] = 0;
1310
1311 if (preempt_count() != count) {
1312 sprintf(msgbuf, "preemption imbalance ");
1313 preempt_count_set(count);
1314 }
1315 if (irqs_disabled()) {
1316 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
1317 local_irq_enable();
1318 }
1319 WARN(msgbuf[0], "initcall %pS returned with %s\n", fn, msgbuf);
1320
1321 add_latent_entropy();
1322 return ret;
1323}
1324
1325
1326extern initcall_entry_t __initcall_start[];
1327extern initcall_entry_t __initcall0_start[];
1328extern initcall_entry_t __initcall1_start[];
1329extern initcall_entry_t __initcall2_start[];
1330extern initcall_entry_t __initcall3_start[];
1331extern initcall_entry_t __initcall4_start[];
1332extern initcall_entry_t __initcall5_start[];
1333extern initcall_entry_t __initcall6_start[];
1334extern initcall_entry_t __initcall7_start[];
1335extern initcall_entry_t __initcall_end[];
1336
1337static initcall_entry_t *initcall_levels[] __initdata = {
1338 __initcall0_start,
1339 __initcall1_start,
1340 __initcall2_start,
1341 __initcall3_start,
1342 __initcall4_start,
1343 __initcall5_start,
1344 __initcall6_start,
1345 __initcall7_start,
1346 __initcall_end,
1347};
1348
1349/* Keep these in sync with initcalls in include/linux/init.h */
1350static const char *initcall_level_names[] __initdata = {
1351 "pure",
1352 "core",
1353 "postcore",
1354 "arch",
1355 "subsys",
1356 "fs",
1357 "device",
1358 "late",
1359};
1360
1361static int __init ignore_unknown_bootoption(char *param, char *val,
1362 const char *unused, void *arg)
1363{
1364 return 0;
1365}
1366
1367static void __init do_initcall_level(int level, char *command_line)
1368{
1369 initcall_entry_t *fn;
1370
1371 parse_args(initcall_level_names[level],
1372 command_line, __start___param,
1373 __stop___param - __start___param,
1374 level, level,
1375 NULL, ignore_unknown_bootoption);
1376
1377 trace_initcall_level(initcall_level_names[level]);
1378 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
1379 do_one_initcall(initcall_from_entry(fn));
1380}
1381
1382static void __init do_initcalls(void)
1383{
1384 int level;
1385 size_t len = saved_command_line_len + 1;
1386 char *command_line;
1387
1388 command_line = kzalloc(len, GFP_KERNEL);
1389 if (!command_line)
1390 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
1391
1392 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++) {
1393 /* Parser modifies command_line, restore it each time */
1394 strcpy(command_line, saved_command_line);
1395 do_initcall_level(level, command_line);
1396 }
1397
1398 kfree(command_line);
1399}
1400
1401/*
1402 * Ok, the machine is now initialized. None of the devices
1403 * have been touched yet, but the CPU subsystem is up and
1404 * running, and memory and process management works.
1405 *
1406 * Now we can finally start doing some real work..
1407 */
1408static void __init do_basic_setup(void)
1409{
1410 cpuset_init_smp();
1411 driver_init();
1412 init_irq_proc();
1413 do_ctors();
1414 do_initcalls();
1415}
1416
1417static void __init do_pre_smp_initcalls(void)
1418{
1419 initcall_entry_t *fn;
1420
1421 trace_initcall_level("early");
1422 for (fn = __initcall_start; fn < __initcall0_start; fn++)
1423 do_one_initcall(initcall_from_entry(fn));
1424}
1425
1426static int run_init_process(const char *init_filename)
1427{
1428 const char *const *p;
1429
1430 argv_init[0] = init_filename;
1431 pr_info("Run %s as init process\n", init_filename);
1432 pr_debug(" with arguments:\n");
1433 for (p = argv_init; *p; p++)
1434 pr_debug(" %s\n", *p);
1435 pr_debug(" with environment:\n");
1436 for (p = envp_init; *p; p++)
1437 pr_debug(" %s\n", *p);
1438 return kernel_execve(init_filename, argv_init, envp_init);
1439}
1440
1441static int try_to_run_init_process(const char *init_filename)
1442{
1443 int ret;
1444
1445 ret = run_init_process(init_filename);
1446
1447 if (ret && ret != -ENOENT) {
1448 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
1449 init_filename, ret);
1450 }
1451
1452 return ret;
1453}
1454
1455static noinline void __init kernel_init_freeable(void);
1456
1457#if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
1458bool rodata_enabled __ro_after_init = true;
1459
1460#ifndef arch_parse_debug_rodata
1461static inline bool arch_parse_debug_rodata(char *str) { return false; }
1462#endif
1463
1464static int __init set_debug_rodata(char *str)
1465{
1466 if (arch_parse_debug_rodata(str))
1467 return 0;
1468
1469 if (str && !strcmp(str, "on"))
1470 rodata_enabled = true;
1471 else if (str && !strcmp(str, "off"))
1472 rodata_enabled = false;
1473 else
1474 pr_warn("Invalid option string for rodata: '%s'\n", str);
1475 return 0;
1476}
1477early_param("rodata", set_debug_rodata);
1478#endif
1479
1480#ifdef CONFIG_STRICT_KERNEL_RWX
1481static void mark_readonly(void)
1482{
1483 if (rodata_enabled) {
1484 /*
1485 * load_module() results in W+X mappings, which are cleaned
1486 * up with call_rcu(). Let's make sure that queued work is
1487 * flushed so that we don't hit false positives looking for
1488 * insecure pages which are W+X.
1489 */
1490 rcu_barrier();
1491 mark_rodata_ro();
1492 rodata_test();
1493 } else
1494 pr_info("Kernel memory protection disabled.\n");
1495}
1496#elif defined(CONFIG_ARCH_HAS_STRICT_KERNEL_RWX)
1497static inline void mark_readonly(void)
1498{
1499 pr_warn("Kernel memory protection not selected by kernel config.\n");
1500}
1501#else
1502static inline void mark_readonly(void)
1503{
1504 pr_warn("This architecture does not have kernel memory protection.\n");
1505}
1506#endif
1507
1508void __weak free_initmem(void)
1509{
1510 free_initmem_default(POISON_FREE_INITMEM);
1511}
1512
1513static int __ref kernel_init(void *unused)
1514{
1515 int ret;
1516
1517 /*
1518 * Wait until kthreadd is all set-up.
1519 */
1520 wait_for_completion(&kthreadd_done);
1521
1522 kernel_init_freeable();
1523 /* need to finish all async __init code before freeing the memory */
1524 async_synchronize_full();
1525
1526 system_state = SYSTEM_FREEING_INITMEM;
1527 kprobe_free_init_mem();
1528 ftrace_free_init_mem();
1529 kgdb_free_init_mem();
1530 exit_boot_config();
1531 free_initmem();
1532 mark_readonly();
1533
1534 /*
1535 * Kernel mappings are now finalized - update the userspace page-table
1536 * to finalize PTI.
1537 */
1538 pti_finalize();
1539
1540 system_state = SYSTEM_RUNNING;
1541 numa_default_policy();
1542
1543 rcu_end_inkernel_boot();
1544
1545 do_sysctl_args();
1546
1547 if (ramdisk_execute_command) {
1548 ret = run_init_process(ramdisk_execute_command);
1549 if (!ret)
1550 return 0;
1551 pr_err("Failed to execute %s (error %d)\n",
1552 ramdisk_execute_command, ret);
1553 }
1554
1555 /*
1556 * We try each of these until one succeeds.
1557 *
1558 * The Bourne shell can be used instead of init if we are
1559 * trying to recover a really broken machine.
1560 */
1561 if (execute_command) {
1562 ret = run_init_process(execute_command);
1563 if (!ret)
1564 return 0;
1565 panic("Requested init %s failed (error %d).",
1566 execute_command, ret);
1567 }
1568
1569 if (CONFIG_DEFAULT_INIT[0] != '\0') {
1570 ret = run_init_process(CONFIG_DEFAULT_INIT);
1571 if (ret)
1572 pr_err("Default init %s failed (error %d)\n",
1573 CONFIG_DEFAULT_INIT, ret);
1574 else
1575 return 0;
1576 }
1577
1578 if (!try_to_run_init_process("/sbin/init") ||
1579 !try_to_run_init_process("/etc/init") ||
1580 !try_to_run_init_process("/bin/init") ||
1581 !try_to_run_init_process("/bin/sh"))
1582 return 0;
1583
1584 panic("No working init found. Try passing init= option to kernel. "
1585 "See Linux Documentation/admin-guide/init.rst for guidance.");
1586}
1587
1588/* Open /dev/console, for stdin/stdout/stderr, this should never fail */
1589void __init console_on_rootfs(void)
1590{
1591 struct file *file = filp_open("/dev/console", O_RDWR, 0);
1592
1593 if (IS_ERR(file)) {
1594 pr_err("Warning: unable to open an initial console.\n");
1595 return;
1596 }
1597 init_dup(file);
1598 init_dup(file);
1599 init_dup(file);
1600 fput(file);
1601}
1602
1603static noinline void __init kernel_init_freeable(void)
1604{
1605 /* Now the scheduler is fully set up and can do blocking allocations */
1606 gfp_allowed_mask = __GFP_BITS_MASK;
1607
1608 /*
1609 * init can allocate pages on any node
1610 */
1611 set_mems_allowed(node_states[N_MEMORY]);
1612
1613 cad_pid = get_pid(task_pid(current));
1614
1615 smp_prepare_cpus(setup_max_cpus);
1616
1617 workqueue_init();
1618
1619 init_mm_internals();
1620
1621 rcu_init_tasks_generic();
1622 do_pre_smp_initcalls();
1623 lockup_detector_init();
1624
1625 smp_init();
1626 sched_init_smp();
1627
1628 padata_init();
1629 page_alloc_init_late();
1630 /* Initialize page ext after all struct pages are initialized. */
1631 if (!early_page_ext_enabled())
1632 page_ext_init();
1633
1634 do_basic_setup();
1635
1636 kunit_run_all_tests();
1637
1638 wait_for_initramfs();
1639 console_on_rootfs();
1640
1641 /*
1642 * check if there is an early userspace init. If yes, let it do all
1643 * the work
1644 */
1645 if (init_eaccess(ramdisk_execute_command) != 0) {
1646 ramdisk_execute_command = NULL;
1647 prepare_namespace();
1648 }
1649
1650 /*
1651 * Ok, we have completed the initial bootup, and
1652 * we're essentially up and running. Get rid of the
1653 * initmem segments and start the user-mode stuff..
1654 *
1655 * rootfs is available now, try loading the public keys
1656 * and default modules
1657 */
1658
1659 integrity_load_keys();
1660}
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}