blob: c0017d9d16e7a2ceb5aae0141ee260dd1ed2b6ec [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
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
Fabian Frederickdd4d9fe2014-08-08 14:23:44 -070010 * Simplified starting of init: Michael A. Griffith <grif@acm.org>
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 */
12
Andrew Mortonea676e82013-04-29 16:18:20 -070013#define DEBUG /* Enable initcall_debug */
14
Linus Torvalds1da177e2005-04-16 15:20:36 -070015#include <linux/types.h>
Paul Gortmaker8a293be2016-07-23 14:01:45 -040016#include <linux/extable.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/module.h>
18#include <linux/proc_fs.h>
Ingo Molnar5c2c5c52017-02-05 14:24:31 +010019#include <linux/binfmts.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/kernel.h>
21#include <linux/syscalls.h>
Ingo Molnar9b5609f2008-02-14 09:41:09 +010022#include <linux/stackprotector.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include <linux/string.h>
24#include <linux/ctype.h>
25#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/ioport.h>
27#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/initrd.h>
Mike Rapoport57c8a662018-10-30 15:09:49 -070029#include <linux/memblock.h>
Len Brown4a7a16d2009-06-12 20:42:08 -040030#include <linux/acpi.h>
Masami Hiramatsu7684b852020-01-11 01:03:44 +090031#include <linux/bootconfig.h>
Nicolas Pitre0c688612017-04-12 18:37:14 -040032#include <linux/console.h>
Ingo Molnar38b8d202017-02-08 18:51:31 +010033#include <linux/nmi.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/percpu.h>
35#include <linux/kmod.h>
Nick Piggindb64fe02008-10-18 20:27:03 -070036#include <linux/vmalloc.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/kernel_stat.h>
Rusty Russelld7cd5612006-12-07 02:14:08 +010038#include <linux/start_kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/security.h>
Jens Axboe3d442232008-06-26 11:21:34 +020040#include <linux/smp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041#include <linux/profile.h>
42#include <linux/rcupdate.h>
43#include <linux/moduleparam.h>
44#include <linux/kallsyms.h>
45#include <linux/writeback.h>
46#include <linux/cpu.h>
47#include <linux/cpuset.h>
Paul Menageddbcc7e2007-10-18 23:39:30 -070048#include <linux/cgroup.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/efi.h>
Thomas Gleixner906568c2007-02-16 01:28:01 -080050#include <linux/tick.h>
Frederic Weisbecker78634062017-10-27 04:42:28 +020051#include <linux/sched/isolation.h>
Andrew Morton6168a702007-02-17 21:22:39 -080052#include <linux/interrupt.h>
Shailabh Nagarc7572492006-07-14 00:24:40 -070053#include <linux/taskstats_kern.h>
Shailabh Nagarca74e922006-07-14 00:24:36 -070054#include <linux/delayacct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/unistd.h>
Alexey Dobriyan3ea056c2018-04-10 16:32:36 -070056#include <linux/utsname.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#include <linux/rmap.h>
58#include <linux/mempolicy.h>
59#include <linux/key.h>
Adrian Bunkb6cd0b72006-06-27 02:53:54 -070060#include <linux/buffer_head.h>
Joonsoo Kimeefa864b2014-12-12 16:55:46 -080061#include <linux/page_ext.h>
Ingo Molnar9a11b49a2006-07-03 00:24:33 -070062#include <linux/debug_locks.h>
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -070063#include <linux/debugobjects.h>
Ingo Molnarfbb9ce952006-07-03 00:24:50 -070064#include <linux/lockdep.h>
Catalin Marinas3c7b4e62009-06-11 13:22:39 +010065#include <linux/kmemleak.h>
Sukadev Bhattiprolu84d73782006-12-08 02:38:01 -080066#include <linux/pid_namespace.h>
Adrian Bunk1f217822006-12-19 13:01:28 -080067#include <linux/device.h>
Eric W. Biederman73c27992007-05-09 02:34:32 -070068#include <linux/kthread.h>
Adrian Bunke6fe6642007-11-09 22:39:39 +010069#include <linux/sched.h>
Ingo Molnar1777e462017-02-05 14:47:12 +010070#include <linux/sched/init.h>
Adrian Bunka1c9eea2008-02-06 01:36:44 -080071#include <linux/signal.h>
Akinobu Mita199f0ca2008-04-29 01:03:13 -070072#include <linux/idr.h>
Jason Wessel0b4b3822010-05-20 21:04:29 -050073#include <linux/kgdb.h>
Steven Rostedt68bf21a2008-08-14 15:45:08 -040074#include <linux/ftrace.h>
Arjan van de Ven22a9d642009-01-07 08:45:46 -080075#include <linux/async.h>
Feng Tang6ae69962009-08-14 15:13:46 -040076#include <linux/sfi.h>
Kay Sievers2b2af542009-04-30 15:23:42 +020077#include <linux/shmem_fs.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090078#include <linux/slab.h>
Peter Zijlstra24a24bb2010-11-17 23:17:33 +010079#include <linux/perf_event.h>
Al Viroa74fb732012-10-10 21:28:25 -040080#include <linux/ptrace.h>
Thomas Gleixneraa8c6242017-12-04 15:07:36 +010081#include <linux/pti.h>
Tejun Heobb813f42013-01-18 14:05:56 -080082#include <linux/blkdev.h>
83#include <linux/elevator.h>
Pavel Tatashin5d2a4e92018-07-19 16:55:41 -040084#include <linux/sched/clock.h>
Ingo Molnar29930022017-02-08 18:51:36 +010085#include <linux/sched/task.h>
Ingo Molnar68db0cf2017-02-08 18:51:37 +010086#include <linux/sched/task_stack.h>
Frederic Weisbecker65f382f2013-07-11 19:12:32 +020087#include <linux/context_tracking.h>
Theodore Ts'o47d06e52013-09-10 10:52:35 -040088#include <linux/random.h>
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -070089#include <linux/list.h>
Dmitry Kasatkinc9cd2ce2014-11-05 17:01:15 +020090#include <linux/integrity.h>
Al Viroe149ed22014-11-01 10:57:28 -040091#include <linux/proc_ns.h>
Toshi Kani0ddab1d2015-04-14 15:47:20 -070092#include <linux/io.h>
AKASHI Takahiro39290b32016-11-14 15:15:05 +090093#include <linux/cache.h>
Jinbum Park2959a5f2017-02-27 14:30:22 -080094#include <linux/rodata_test.h>
Josh Poimboeuf33352242018-02-20 11:37:51 -060095#include <linux/jump_label.h>
Mathieu Malaterreae67d582018-05-25 14:48:00 -070096#include <linux/mem_encrypt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
98#include <asm/io.h>
99#include <asm/bugs.h>
100#include <asm/setup.h>
Andi Kleena9401992005-07-28 21:15:30 -0700101#include <asm/sections.h>
Arjan van de Ven37b73c82006-01-06 00:12:01 -0800102#include <asm/cacheflush.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103
Steven Rostedt (VMware)4ee7c602018-03-23 10:18:03 -0400104#define CREATE_TRACE_POINTS
105#include <trace/events/initcall.h>
106
Sam Ravnborgaae5f662007-02-26 16:45:41 +0100107static int kernel_init(void *);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108
109extern void init_IRQ(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110extern void radix_tree_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111
Tejun Heo2ce802f2011-01-20 12:06:35 +0100112/*
113 * Debug helper: via this flag we know that we are in 'early bootup code'
114 * where only the boot processor is running with IRQ disabled. This means
115 * two things - IRQ must not be enabled before the flag is cleared and some
116 * operations which are not allowed with IRQ disabled are allowed while the
117 * flag is set.
118 */
119bool early_boot_irqs_disabled __read_mostly;
120
Paul E. McKenneya6826042009-02-25 18:03:42 -0800121enum system_states system_state __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122EXPORT_SYMBOL(system_state);
123
124/*
125 * Boot command-line arguments
126 */
127#define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
128#define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
129
130extern void time_init(void);
131/* Default late time init is NULL. archs can override this later. */
Jan Beulichd2e31922009-01-06 14:41:10 -0800132void (*__initdata late_time_init)(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800134/* Untouched command line saved by arch-specific code. */
135char __initdata boot_command_line[COMMAND_LINE_SIZE];
136/* Untouched saved command line (eg. for /proc) */
137char *saved_command_line;
138/* Command line for parameter parsing */
139static char *static_command_line;
Masami Hiramatsu51887d02020-01-11 01:04:43 +0900140/* Untouched extra command line */
141static char *extra_command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
143static char *execute_command;
Olof Johanssonffdfc402005-09-06 15:17:19 -0700144static char *ramdisk_execute_command;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
Jan Beulich8b3b2952007-07-15 23:41:07 -0700146/*
Hannes Frederic Sowac4b2c0c2013-10-19 21:48:53 +0200147 * Used to generate warnings if static_key manipulation functions are used
148 * before jump_label_init is called.
149 */
Fabian Frederickdd4d9fe2014-08-08 14:23:44 -0700150bool static_key_initialized __read_mostly;
Hannes Frederic Sowac4b2c0c2013-10-19 21:48:53 +0200151EXPORT_SYMBOL_GPL(static_key_initialized);
152
153/*
Jan Beulich8b3b2952007-07-15 23:41:07 -0700154 * If set, this is an indication to the drivers that reset the underlying
155 * device before going ahead with the initialization otherwise driver might
156 * rely on the BIOS and skip the reset operation.
157 *
158 * This is useful if kernel is booting in an unreliable environment.
Frans Klaverfc454fd2015-02-23 22:05:56 +0100159 * For ex. kdump situation where previous kernel has crashed, BIOS has been
Jan Beulich8b3b2952007-07-15 23:41:07 -0700160 * skipped and devices will be in unknown state.
161 */
162unsigned int reset_devices;
163EXPORT_SYMBOL(reset_devices);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Vivek Goyal7e962872006-09-27 01:50:44 -0700165static int __init set_reset_devices(char *str)
166{
167 reset_devices = 1;
168 return 1;
169}
170
171__setup("reset_devices", set_reset_devices);
172
Fabian Frederickdd4d9fe2014-08-08 14:23:44 -0700173static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
174const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175static const char *panic_later, *panic_param;
176
Rusty Russell914dcaa2010-08-11 23:04:18 -0600177extern const struct obs_kernel_param __setup_start[], __setup_end[];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
Yaowei Bai31c025b2016-01-20 14:59:27 -0800179static bool __init obsolete_checksetup(char *line)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
Rusty Russell914dcaa2010-08-11 23:04:18 -0600181 const struct obs_kernel_param *p;
Yaowei Bai31c025b2016-01-20 14:59:27 -0800182 bool had_early_param = false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184 p = __setup_start;
185 do {
186 int n = strlen(p->str);
Michal Schmidtb1e4d202011-10-10 00:03:37 +0200187 if (parameqn(line, p->str, n)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 if (p->early) {
Rusty Russell33df0d12006-09-26 10:52:32 +0200189 /* Already done in parse_early_param?
190 * (Needs exact match on param part).
191 * Keep iterating, as we can have early
192 * params and __setups of same names 8( */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 if (line[n] == '\0' || line[n] == '=')
Yaowei Bai31c025b2016-01-20 14:59:27 -0800194 had_early_param = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 } else if (!p->setup_func) {
Andrew Mortonea676e82013-04-29 16:18:20 -0700196 pr_warn("Parameter %s is obsolete, ignored\n",
197 p->str);
Yaowei Bai31c025b2016-01-20 14:59:27 -0800198 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 } else if (p->setup_func(line + n))
Yaowei Bai31c025b2016-01-20 14:59:27 -0800200 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 }
202 p++;
203 } while (p < __setup_end);
Rusty Russell33df0d12006-09-26 10:52:32 +0200204
205 return had_early_param;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206}
207
208/*
209 * This should be approx 2 Bo*oMips to start (note initial shift), and will
210 * still work even if initially too large, it will just take slightly longer
211 */
212unsigned long loops_per_jiffy = (1<<12);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213EXPORT_SYMBOL(loops_per_jiffy);
214
215static int __init debug_kernel(char *str)
216{
Borislav Petkova8fe19e2014-06-04 16:11:46 -0700217 console_loglevel = CONSOLE_LOGLEVEL_DEBUG;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800218 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
220
221static int __init quiet_kernel(char *str)
222{
Borislav Petkova8fe19e2014-06-04 16:11:46 -0700223 console_loglevel = CONSOLE_LOGLEVEL_QUIET;
Yinghai Luf6f21c812008-02-08 04:21:58 -0800224 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225}
226
Yinghai Luf6f21c812008-02-08 04:21:58 -0800227early_param("debug", debug_kernel);
228early_param("quiet", quiet_kernel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229
230static int __init loglevel(char *str)
231{
Alexander Sverdlin808bf292011-09-21 09:51:40 +0200232 int newlevel;
233
234 /*
235 * Only update loglevel value when a correct setting was passed,
236 * to prevent blind crashes (when loglevel being set to 0) that
237 * are quite hard to debug
238 */
239 if (get_option(&str, &newlevel)) {
240 console_loglevel = newlevel;
241 return 0;
242 }
243
244 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245}
246
Yinghai Luf6f21c812008-02-08 04:21:58 -0800247early_param("loglevel", loglevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Masami Hiramatsu7684b852020-01-11 01:03:44 +0900249#ifdef CONFIG_BOOT_CONFIG
Masami Hiramatsu51887d02020-01-11 01:04:43 +0900250
251char xbc_namebuf[XBC_KEYLEN_MAX] __initdata;
252
253#define rest(dst, end) ((end) > (dst) ? (end) - (dst) : 0)
254
255static int __init xbc_snprint_cmdline(char *buf, size_t size,
256 struct xbc_node *root)
257{
258 struct xbc_node *knode, *vnode;
259 char *end = buf + size;
260 char c = '\"';
261 const char *val;
262 int ret;
263
264 xbc_node_for_each_key_value(root, knode, val) {
265 ret = xbc_node_compose_key_after(root, knode,
266 xbc_namebuf, XBC_KEYLEN_MAX);
267 if (ret < 0)
268 return ret;
269
270 vnode = xbc_node_get_child(knode);
271 ret = snprintf(buf, rest(buf, end), "%s%c", xbc_namebuf,
272 vnode ? '=' : ' ');
273 if (ret < 0)
274 return ret;
275 buf += ret;
276 if (!vnode)
277 continue;
278
279 c = '\"';
280 xbc_array_for_each_value(vnode, val) {
281 ret = snprintf(buf, rest(buf, end), "%c%s", c, val);
282 if (ret < 0)
283 return ret;
284 buf += ret;
285 c = ',';
286 }
287 if (rest(buf, end) > 2)
288 strcpy(buf, "\" ");
289 buf += 2;
290 }
291
292 return buf - (end - size);
293}
294#undef rest
295
296/* Make an extra command line under given key word */
297static char * __init xbc_make_cmdline(const char *key)
298{
299 struct xbc_node *root;
300 char *new_cmdline;
301 int ret, len = 0;
302
303 root = xbc_find_node(key);
304 if (!root)
305 return NULL;
306
307 /* Count required buffer size */
308 len = xbc_snprint_cmdline(NULL, 0, root);
309 if (len <= 0)
310 return NULL;
311
312 new_cmdline = memblock_alloc(len + 1, SMP_CACHE_BYTES);
313 if (!new_cmdline) {
314 pr_err("Failed to allocate memory for extra kernel cmdline.\n");
315 return NULL;
316 }
317
318 ret = xbc_snprint_cmdline(new_cmdline, len + 1, root);
319 if (ret < 0 || ret > len) {
320 pr_err("Failed to print extra kernel cmdline.\n");
321 return NULL;
322 }
323
324 return new_cmdline;
325}
326
Masami Hiramatsu7684b852020-01-11 01:03:44 +0900327u32 boot_config_checksum(unsigned char *p, u32 size)
328{
329 u32 ret = 0;
330
331 while (size--)
332 ret += *p++;
333
334 return ret;
335}
336
337static void __init setup_boot_config(void)
338{
339 u32 size, csum;
340 char *data, *copy;
341 u32 *hdr;
342
343 if (!initrd_end)
344 return;
345
346 hdr = (u32 *)(initrd_end - 8);
347 size = hdr[0];
348 csum = hdr[1];
349
350 if (size >= XBC_DATA_MAX)
351 return;
352
353 data = ((void *)hdr) - size;
354 if ((unsigned long)data < initrd_start)
355 return;
356
357 if (boot_config_checksum((unsigned char *)data, size) != csum)
358 return;
359
360 copy = memblock_alloc(size + 1, SMP_CACHE_BYTES);
361 if (!copy) {
362 pr_err("Failed to allocate memory for boot config\n");
363 return;
364 }
365
366 memcpy(copy, data, size);
367 copy[size] = '\0';
368
369 if (xbc_init(copy) < 0)
370 pr_err("Failed to parse boot config\n");
Masami Hiramatsu51887d02020-01-11 01:04:43 +0900371 else {
Masami Hiramatsu7684b852020-01-11 01:03:44 +0900372 pr_info("Load boot config: %d bytes\n", size);
Masami Hiramatsu51887d02020-01-11 01:04:43 +0900373 /* keys starting with "kernel." are passed via cmdline */
374 extra_command_line = xbc_make_cmdline("kernel");
375 }
Masami Hiramatsu7684b852020-01-11 01:03:44 +0900376}
377#else
378#define setup_boot_config() do { } while (0)
379#endif
380
Chris Metcalfa99cd112012-04-06 12:53:50 -0400381/* Change NUL term back to "=", to make "param" the whole string. */
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700382static int __init repair_env_string(char *param, char *val,
383 const char *unused, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 if (val) {
386 /* param=val or param="val"? */
387 if (val == param+strlen(param)+1)
388 val[-1] = '=';
389 else if (val == param+strlen(param)+2) {
390 val[-2] = '=';
391 memmove(val-1, val, strlen(val)+1);
392 val--;
393 } else
394 BUG();
395 }
Chris Metcalfa99cd112012-04-06 12:53:50 -0400396 return 0;
397}
398
Rusty Russell51e158c2014-04-28 11:34:33 +0930399/* Anything after -- gets handed straight to init. */
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700400static int __init set_init_arg(char *param, char *val,
401 const char *unused, void *arg)
Rusty Russell51e158c2014-04-28 11:34:33 +0930402{
403 unsigned int i;
404
405 if (panic_later)
406 return 0;
407
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700408 repair_env_string(param, val, unused, NULL);
Rusty Russell51e158c2014-04-28 11:34:33 +0930409
410 for (i = 0; argv_init[i]; i++) {
411 if (i == MAX_INIT_ARGS) {
412 panic_later = "init";
413 panic_param = param;
414 return 0;
415 }
416 }
417 argv_init[i] = param;
418 return 0;
419}
420
Chris Metcalfa99cd112012-04-06 12:53:50 -0400421/*
422 * Unknown boot options get handed to init, unless they look like
423 * unused parameters (modprobe will find them in /proc/cmdline).
424 */
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700425static int __init unknown_bootoption(char *param, char *val,
426 const char *unused, void *arg)
Chris Metcalfa99cd112012-04-06 12:53:50 -0400427{
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700428 repair_env_string(param, val, unused, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429
430 /* Handle obsolete-style parameters */
431 if (obsolete_checksetup(param))
432 return 0;
433
Rusty Russellf066a4f2009-12-01 14:56:44 +1030434 /* Unused module parameter. */
435 if (strchr(param, '.') && (!val || strchr(param, '.') < val))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437
438 if (panic_later)
439 return 0;
440
441 if (val) {
442 /* Environment option */
443 unsigned int i;
444 for (i = 0; envp_init[i]; i++) {
445 if (i == MAX_INIT_ENVS) {
Tetsuo Handa499a4582014-01-23 15:54:56 -0800446 panic_later = "env";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 panic_param = param;
448 }
449 if (!strncmp(param, envp_init[i], val - param))
450 break;
451 }
452 envp_init[i] = param;
453 } else {
454 /* Command line option */
455 unsigned int i;
456 for (i = 0; argv_init[i]; i++) {
457 if (i == MAX_INIT_ARGS) {
Tetsuo Handa499a4582014-01-23 15:54:56 -0800458 panic_later = "init";
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459 panic_param = param;
460 }
461 }
462 argv_init[i] = param;
463 }
464 return 0;
465}
466
467static int __init init_setup(char *str)
468{
469 unsigned int i;
470
471 execute_command = str;
472 /*
473 * In case LILO is going to boot us with default command line,
474 * it prepends "auto" before the whole cmdline which makes
475 * the shell think it should execute a script with such name.
476 * So we ignore all arguments entered _before_ init=... [MJ]
477 */
478 for (i = 1; i < MAX_INIT_ARGS; i++)
479 argv_init[i] = NULL;
480 return 1;
481}
482__setup("init=", init_setup);
483
Olof Johanssonffdfc402005-09-06 15:17:19 -0700484static int __init rdinit_setup(char *str)
485{
486 unsigned int i;
487
488 ramdisk_execute_command = str;
489 /* See "auto" comment in init_setup */
490 for (i = 1; i < MAX_INIT_ARGS; i++)
491 argv_init[i] = NULL;
492 return 1;
493}
494__setup("rdinit=", rdinit_setup);
495
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496#ifndef CONFIG_SMP
Amerigo Wang34db18a02011-03-22 16:34:06 -0700497static const unsigned int setup_max_cpus = NR_CPUS;
Mike Travise0982e92008-03-26 14:23:48 -0700498static inline void setup_nr_cpu_ids(void) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499static inline void smp_prepare_cpus(unsigned int maxcpus) { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500#endif
501
502/*
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800503 * We need to store the untouched command line for future reference.
504 * We also need to store the touched command line since the parameter
505 * parsing is performed in place, and we should allow a component to
506 * store reference of name/value for future reference.
507 */
508static void __init setup_command_line(char *command_line)
509{
Masami Hiramatsu51887d02020-01-11 01:04:43 +0900510 size_t len, xlen = 0;
511
512 if (extra_command_line)
513 xlen = strlen(extra_command_line);
514
515 len = xlen + strlen(boot_command_line) + 1;
Mike Rapoportf5c73102019-03-11 23:30:20 -0700516
517 saved_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
518 if (!saved_command_line)
519 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
520
Mike Rapoportf5c73102019-03-11 23:30:20 -0700521 static_command_line = memblock_alloc(len, SMP_CACHE_BYTES);
522 if (!static_command_line)
523 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
524
Masami Hiramatsu51887d02020-01-11 01:04:43 +0900525 if (xlen) {
526 /*
527 * We have to put extra_command_line before boot command
528 * lines because there could be dashes (separator of init
529 * command line) in the command lines.
530 */
531 strcpy(saved_command_line, extra_command_line);
532 strcpy(static_command_line, extra_command_line);
533 }
534 strcpy(saved_command_line + xlen, boot_command_line);
535 strcpy(static_command_line + xlen, command_line);
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800536}
537
538/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 * We need to finalize in a non-__init function or else race conditions
540 * between the root thread and the init thread may cause start_kernel to
541 * be reaped by free_initmem before the root thread has proceeded to
542 * cpu_idle.
543 *
544 * gcc-3.4 accidentally inlines this function, so use noinline.
545 */
546
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200547static __initdata DECLARE_COMPLETION(kthreadd_done);
548
Martin Schwidefsky53c99bd62018-08-31 10:42:24 +0200549noinline void __ref rest_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700550{
Thomas Gleixner8fb12152017-05-16 20:42:32 +0200551 struct task_struct *tsk;
Eric W. Biederman73c27992007-05-09 02:34:32 -0700552 int pid;
553
Paul E. McKenney7db905e2009-09-02 14:01:24 -0700554 rcu_scheduler_starting();
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200555 /*
Peter Zijlstra97158562010-06-30 10:37:11 +0200556 * We need to spawn init first so that it obtains pid 1, however
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200557 * the init task will end up wanting to create kthreads, which, if
558 * we schedule it before we create kthreadd, will OOPS.
559 */
Thomas Gleixner8fb12152017-05-16 20:42:32 +0200560 pid = kernel_thread(kernel_init, NULL, CLONE_FS);
561 /*
562 * Pin init on the boot CPU. Task migration is not properly working
563 * until sched_init_smp() has been run. It will set the allowed
564 * CPUs for init to the non isolated CPUs.
565 */
566 rcu_read_lock();
567 tsk = find_task_by_pid_ns(pid, &init_pid_ns);
568 set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id()));
569 rcu_read_unlock();
570
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571 numa_default_policy();
Eric W. Biederman73c27992007-05-09 02:34:32 -0700572 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800573 rcu_read_lock();
Pavel Emelyanov5cd20452008-04-30 00:54:24 -0700574 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
Paul E. McKenneyd11c5632010-02-22 17:04:50 -0800575 rcu_read_unlock();
Thomas Gleixner1c3c5ea2017-05-16 20:42:48 +0200576
577 /*
578 * Enable might_sleep() and smp_processor_id() checks.
Thomas Gleixnerc1a280b2019-07-26 23:19:37 +0200579 * They cannot be enabled earlier because with CONFIG_PREEMPTION=y
Thomas Gleixner1c3c5ea2017-05-16 20:42:48 +0200580 * kernel_thread() would trigger might_sleep() splats. With
581 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled
582 * already, but it's stuck on the kthreadd_done completion.
583 */
584 system_state = SYSTEM_SCHEDULING;
585
Peter Zijlstrab433c3d2010-06-28 16:51:01 +0200586 complete(&kthreadd_done);
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200587
588 /*
589 * The boot idle thread must execute schedule()
Ingo Molnar1df21052007-07-09 18:51:58 +0200590 * at least once to get things moving:
Ingo Molnarf340c0d2005-06-28 16:40:42 +0200591 */
Thomas Gleixnerbd2f5532011-03-21 12:33:18 +0100592 schedule_preempt_disabled();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800593 /* Call into cpu_idle with preempt disabled */
Thomas Gleixnera1a04ec2013-03-21 22:49:34 +0100594 cpu_startup_entry(CPUHP_ONLINE);
Ingo Molnar1df21052007-07-09 18:51:58 +0200595}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596
597/* Check for early params. */
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700598static int __init do_early_param(char *param, char *val,
599 const char *unused, void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
Rusty Russell914dcaa2010-08-11 23:04:18 -0600601 const struct obs_kernel_param *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
603 for (p = __setup_start; p < __setup_end; p++) {
Michal Schmidtb1e4d202011-10-10 00:03:37 +0200604 if ((p->early && parameq(param, p->str)) ||
Yinghai Lu18a8bd92007-07-15 23:37:59 -0700605 (strcmp(param, "console") == 0 &&
606 strcmp(p->str, "earlycon") == 0)
607 ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 if (p->setup_func(val) != 0)
Andrew Mortonea676e82013-04-29 16:18:20 -0700609 pr_warn("Malformed early option '%s'\n", param);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 }
611 }
612 /* We accept everything at this stage. */
613 return 0;
614}
615
Magnus Damm13977092009-03-30 14:37:25 -0700616void __init parse_early_options(char *cmdline)
617{
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700618 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL,
619 do_early_param);
Magnus Damm13977092009-03-30 14:37:25 -0700620}
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622/* Arch code calls this early on, or if not, just before other parsing. */
623void __init parse_early_param(void)
624{
Fabian Frederickdd4d9fe2014-08-08 14:23:44 -0700625 static int done __initdata;
626 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
628 if (done)
629 return;
630
631 /* All fall through to do_early_param. */
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800632 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
Magnus Damm13977092009-03-30 14:37:25 -0700633 parse_early_options(tmp_cmdline);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 done = 1;
635}
636
Thomas Gleixnere7ff3a42016-12-09 19:29:10 +0100637void __init __weak arch_post_acpi_subsys_init(void) { }
638
Benjamin Herrenschmidt839ad622008-04-18 16:56:18 +1000639void __init __weak smp_setup_processor_id(void)
Andrew Morton033ab7f2006-06-30 01:55:50 -0700640{
641}
642
David Howellseded09c2012-11-02 13:20:42 +0000643# if THREAD_SIZE >= PAGE_SIZE
Linus Torvaldsb235bee2016-06-24 15:09:37 -0700644void __init __weak thread_stack_cache_init(void)
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000645{
646}
David Howellseded09c2012-11-02 13:20:42 +0000647#endif
Benjamin Herrenschmidt8c9843e2008-04-18 16:56:15 +1000648
Tom Lendackyc7753202017-07-17 16:10:21 -0500649void __init __weak mem_encrypt_init(void) { }
650
Nadav Amit4fc19702019-04-26 16:22:46 -0700651void __init __weak poking_init(void) { }
652
Mike Rapoport782de702019-09-23 15:35:31 -0700653void __init __weak pgtable_cache_init(void) { }
Nadav Amitcaa84132019-05-04 18:11:24 -0700654
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -0400655bool initcall_debug;
656core_param(initcall_debug, initcall_debug, bool, 0644);
Steven Rostedt (VMware)b0dc52f2018-04-06 09:24:25 -0400657
658#ifdef TRACEPOINTS_ENABLED
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -0400659static void __init initcall_debug_enable(void);
Steven Rostedt (VMware)b0dc52f2018-04-06 09:24:25 -0400660#else
661static inline void initcall_debug_enable(void)
662{
663}
664#endif
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -0400665
Alexander Potapenko23a5c8c2019-07-11 20:59:23 -0700666/* Report memory auto-initialization states for this boot. */
667static void __init report_meminit(void)
668{
669 const char *stack;
670
671 if (IS_ENABLED(CONFIG_INIT_STACK_ALL))
672 stack = "all";
673 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL))
674 stack = "byref_all";
675 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF))
676 stack = "byref";
677 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_USER))
678 stack = "__user";
679 else
680 stack = "off";
681
682 pr_info("mem auto-init: stack:%s, heap alloc:%s, heap free:%s\n",
683 stack, want_init_on_alloc(GFP_KERNEL) ? "on" : "off",
684 want_init_on_free() ? "on" : "off");
685 if (want_init_on_free())
686 pr_info("mem auto-init: clearing system memory may take some time...\n");
687}
688
Pekka Enberg444f4782009-06-11 18:29:06 +0300689/*
690 * Set up kernel memory allocators
691 */
692static void __init mm_init(void)
693{
Joonsoo Kimeefa864b2014-12-12 16:55:46 -0800694 /*
695 * page_ext requires contiguous pages,
696 * bigger than MAX_ORDER unless SPARSEMEM.
697 */
698 page_ext_init_flatmem();
Alexander Potapenko23a5c8c2019-07-11 20:59:23 -0700699 report_meminit();
Pekka Enberg444f4782009-06-11 18:29:06 +0300700 mem_init();
701 kmem_cache_init();
Catalin Marinasc5665862019-09-23 15:34:05 -0700702 kmemleak_init();
Kirill A. Shutemovb35f1812014-01-21 15:49:07 -0800703 pgtable_init();
Qian Caia9ee3a62018-12-28 00:32:32 -0800704 debug_objects_mem_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300705 vmalloc_init();
Toshi Kani0ddab1d2015-04-14 15:47:20 -0700706 ioremap_huge_init();
Thomas Gleixner613e3962017-12-17 10:56:29 +0100707 /* Should be run before the first non-init thread is created */
708 init_espfix_bsp();
Thomas Gleixneraa8c6242017-12-04 15:07:36 +0100709 /* Should be run after espfix64 is set up. */
710 pti_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300711}
712
Martin Schwidefsky53c99bd62018-08-31 10:42:24 +0200713void __init __weak arch_call_rest_init(void)
714{
715 rest_init();
716}
717
Andi Kleen722a9f92014-05-02 00:44:38 +0200718asmlinkage __visible void __init start_kernel(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719{
Fabian Frederickdd4d9fe2014-08-08 14:23:44 -0700720 char *command_line;
721 char *after_dashes;
Andrew Morton033ab7f2006-06-30 01:55:50 -0700722
Aaron Tomlind4311ff2014-09-12 14:16:17 +0100723 set_task_stack_end_magic(&init_task);
Ming Lei73839c52011-11-17 13:34:31 +0800724 smp_setup_processor_id();
Thomas Gleixner3ac7fe52008-04-30 00:55:01 -0700725 debug_objects_early_init();
Ingo Molnar42059422008-02-14 09:44:08 +0100726
Paul Menageddbcc7e2007-10-18 23:39:30 -0700727 cgroup_init_early();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700728
729 local_irq_disable();
Tejun Heo2ce802f2011-01-20 12:06:35 +0100730 early_boot_irqs_disabled = true;
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700731
Viresh Kumar1b3b3b42017-03-23 17:00:05 +0530732 /*
733 * Interrupts are still disabled. Do necessary setups, then
734 * enable them.
735 */
Stas Sergeev44fd2292006-03-23 02:59:44 -0800736 boot_cpu_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 page_address_init();
Andrew Mortonea676e82013-04-29 16:18:20 -0700738 pr_notice("%s", linux_banner);
Matthew Garrette6b1db92019-08-19 17:17:37 -0700739 early_security_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 setup_arch(&command_line);
Masami Hiramatsu7684b852020-01-11 01:03:44 +0900741 setup_boot_config();
Alon Bar-Lev30d7e0d2007-02-12 00:53:52 -0800742 setup_command_line(command_line);
Mike Travise0982e92008-03-26 14:23:48 -0700743 setup_nr_cpu_ids();
Tejun Heod6647bd2009-07-21 17:11:50 +0900744 setup_per_cpu_areas();
Stas Sergeev44fd2292006-03-23 02:59:44 -0800745 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
Linus Torvaldsb5b14042018-08-12 12:19:42 -0700746 boot_cpu_hotplug_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Michal Hocko72675e12017-09-06 16:20:24 -0700748 build_all_zonelists(NULL);
Pekka Enberg83b519e2009-06-10 19:40:04 +0300749 page_alloc_init();
750
Masami Hiramatsu51887d02020-01-11 01:04:43 +0900751 pr_notice("Kernel command line: %s\n", saved_command_line);
Dan Williams60411862019-04-18 17:50:44 -0700752 /* parameters may set static keys */
753 jump_label_init();
Pekka Enberg83b519e2009-06-10 19:40:04 +0300754 parse_early_param();
Rusty Russell51e158c2014-04-28 11:34:33 +0930755 after_dashes = parse_args("Booting kernel",
756 static_command_line, __start___param,
757 __stop___param - __start___param,
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700758 -1, -1, NULL, &unknown_bootoption);
Daniel Thompson3438cf52014-11-11 16:29:46 +1030759 if (!IS_ERR_OR_NULL(after_dashes))
Rusty Russell51e158c2014-04-28 11:34:33 +0930760 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1,
Luis R. Rodriguezecc86172015-03-30 16:20:03 -0700761 NULL, set_init_arg);
Jeremy Fitzhardinge97ce2c82011-10-12 16:17:54 -0700762
Pekka Enberg83b519e2009-06-10 19:40:04 +0300763 /*
764 * These use large bootmem allocations and must precede
765 * kmem_cache_init()
766 */
Mike Travis162a7e72011-05-24 17:13:20 -0700767 setup_log_buf(0);
Pekka Enberg83b519e2009-06-10 19:40:04 +0300768 vfs_caches_init_early();
769 sort_main_extable();
770 trap_init();
Pekka Enberg444f4782009-06-11 18:29:06 +0300771 mm_init();
KOSAKI Motohirode03c722011-05-24 17:12:15 -0700772
Steven Rostedt (VMware)f6317182017-03-03 13:43:34 -0500773 ftrace_init();
774
Steven Rostedt (VMware)e725c732017-03-03 13:37:33 -0500775 /* trace_printk can be enabled here */
776 early_trace_init();
777
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 /*
779 * Set up the scheduler prior starting any interrupts (such as the
780 * timer interrupt). Full topology setup happens at smp_init()
781 * time - but meanwhile we still have a functioning scheduler.
782 */
783 sched_init();
784 /*
785 * Disable preemption - early bootup scheduling is extremely
786 * fragile until we cpu_idle() for the first time.
787 */
788 preempt_disable();
Fabian Frederickdd4d9fe2014-08-08 14:23:44 -0700789 if (WARN(!irqs_disabled(),
790 "Interrupts were enabled *very* early, fixing it\n"))
Ard van Breemenc4a68302007-01-05 16:36:19 -0800791 local_irq_disable();
Matthew Wilcox0a835c42016-12-20 10:27:56 -0500792 radix_tree_init();
Tejun Heo3347fa02016-09-16 15:49:32 -0400793
794 /*
Tal Shorer7d229c62017-11-03 17:27:49 +0200795 * Set up housekeeping before setting up workqueues to allow the unbound
796 * workqueue to take non-housekeeping into account.
797 */
798 housekeeping_init();
799
800 /*
Tejun Heo3347fa02016-09-16 15:49:32 -0400801 * Allow workqueue creation and work item queueing/cancelling
802 * early. Work item execution depends on kthreads and starts after
803 * workqueue_init().
804 */
805 workqueue_init_early();
806
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 rcu_init();
Steven Rostedt (Red Hat)5f893b22014-12-12 20:05:10 -0500808
Steven Rostedt (VMware)e725c732017-03-03 13:37:33 -0500809 /* Trace events are available after this */
Steven Rostedt (Red Hat)5f893b22014-12-12 20:05:10 -0500810 trace_init();
811
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -0400812 if (initcall_debug)
813 initcall_debug_enable();
814
Frederic Weisbecker65f382f2013-07-11 19:12:32 +0200815 context_tracking_init();
Yinghai Lu0b8f1ef2008-12-05 18:58:31 -0800816 /* init some links before init_ISA_irqs() */
817 early_irq_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 init_IRQ();
Thomas Gleixnerad2b1352013-03-05 15:14:05 +0100819 tick_init();
Linus Torvaldsd6dd50e2014-10-13 15:44:12 +0200820 rcu_init_nohz();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 init_timers();
Thomas Gleixnerc0a31322006-01-09 20:52:32 -0800822 hrtimers_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 softirq_init();
john stultzad596172006-06-26 00:25:06 -0700824 timekeeping_init();
Kees Cookd5553522019-04-19 23:27:05 -0400825
826 /*
827 * For best initial stack canary entropy, prepare it after:
828 * - setup_arch() for any UEFI RNG entropy and boot cmdline access
829 * - timekeeping_init() for ktime entropy used in rand_initialize()
830 * - rand_initialize() to get any arch-specific entropy like RDRAND
831 * - add_latent_entropy() to get any latent entropy
832 * - adding command line entropy
833 */
834 rand_initialize();
835 add_latent_entropy();
836 add_device_randomness(command_line, strlen(command_line));
837 boot_init_stack_canary();
838
john stultz88fecaa2006-07-03 00:24:04 -0700839 time_init();
Sergey Senozhatskyf92bac32016-12-27 23:16:05 +0900840 printk_safe_init();
Stephane Eranian9e630202013-04-03 14:21:33 +0200841 perf_event_init();
Heiko Carstens93e02812006-07-03 00:24:24 -0700842 profile_init();
Takao Indohd8ad7d12011-03-29 12:35:04 -0400843 call_function_init();
Steven Rostedtf91eb622013-04-29 16:18:18 -0700844 WARN(!irqs_disabled(), "Interrupts were enabled early\n");
Joel Fernandes (Google)c3bc8fd2018-07-30 15:24:23 -0700845
Tejun Heo2ce802f2011-01-20 12:06:35 +0100846 early_boot_irqs_disabled = false;
Heiko Carstens93e02812006-07-03 00:24:24 -0700847 local_irq_enable();
Benjamin Herrenschmidtdcce2842009-06-18 13:24:12 +1000848
Pekka Enberg7e85ee02009-06-12 14:03:06 +0300849 kmem_cache_init_late();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
851 /*
852 * HACK ALERT! This is early. We're enabling the console before
853 * we've done PCI setups etc, and console_init() must be aware of
854 * this. But we do want output early, in case something goes wrong.
855 */
856 console_init();
857 if (panic_later)
Tetsuo Handa499a4582014-01-23 15:54:56 -0800858 panic("Too many boot %s vars at `%s'", panic_later,
859 panic_param);
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700860
Joel Fernandes (Google)c3bc8fd2018-07-30 15:24:23 -0700861 lockdep_init();
Ingo Molnarfbb9ce952006-07-03 00:24:50 -0700862
Ingo Molnar9a11b49a2006-07-03 00:24:33 -0700863 /*
864 * Need to run this when irqs are enabled, because it wants
865 * to self-test [hard/soft]-irqs on/off lock inversion bugs
866 * too:
867 */
868 locking_selftest();
869
Tom Lendackyc7753202017-07-17 16:10:21 -0500870 /*
871 * This needs to be called before any devices perform DMA
872 * operations that might use the SWIOTLB bounce buffers. It will
873 * mark the bounce buffers as decrypted so that their usage will
874 * not cause "plain-text" data to be decrypted when accessed.
875 */
876 mem_encrypt_init();
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878#ifdef CONFIG_BLK_DEV_INITRD
879 if (initrd_start && !initrd_below_start_ok &&
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700880 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
Andrew Mortonea676e82013-04-29 16:18:20 -0700881 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
Geert Uytterhoevenbd673c72008-07-29 22:33:36 -0700882 page_to_pfn(virt_to_page((void *)initrd_start)),
883 min_low_pfn);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 initrd_start = 0;
885 }
886#endif
Christoph Lametere7c8d5c2005-06-21 17:14:47 -0700887 setup_per_cpu_pageset();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 numa_policy_init();
Dou Liyang9c712062017-09-13 17:17:54 +0800889 acpi_early_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 if (late_time_init)
891 late_time_init();
Pavel Tatashin857baa82018-07-19 16:55:42 -0400892 sched_clock_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 calibrate_delay();
Gargi Sharma95846ec2017-11-17 15:30:30 -0800894 pid_idr_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 anon_vma_init();
Linus Torvalds11520e52012-12-15 15:15:24 -0800896#ifdef CONFIG_X86
Matt Fleming83e68182012-11-14 09:42:35 +0000897 if (efi_enabled(EFI_RUNTIME_SERVICES))
Linus Torvalds11520e52012-12-15 15:15:24 -0800898 efi_enter_virtual_mode();
899#endif
Linus Torvaldsb235bee2016-06-24 15:09:37 -0700900 thread_stack_cache_init();
David Howellsd84f4f92008-11-14 10:39:23 +1100901 cred_init();
Heinrich Schuchardtff691f62015-04-16 12:47:44 -0700902 fork_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 proc_caches_init();
Alexey Dobriyan3ea056c2018-04-10 16:32:36 -0700904 uts_ns_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 buffer_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 key_init();
907 security_init();
Jason Wessel0b4b3822010-05-20 21:04:29 -0500908 dbg_late_init();
Mel Gorman4248b0d2015-08-06 15:46:20 -0700909 vfs_caches_init();
Nicholas Piggin62906022016-12-25 13:00:30 +1000910 pagecache_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 signals_init();
Alexey Dobriyan0965232032018-04-10 16:34:45 -0700912 seq_file_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 proc_root_init();
Al Viroe149ed22014-11-01 10:57:28 -0400914 nsfs_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 cpuset_init();
Zefan Li695df212015-03-04 17:09:33 +0800916 cgroup_init();
Shailabh Nagarc7572492006-07-14 00:24:40 -0700917 taskstats_init_early();
Shailabh Nagarca74e922006-07-14 00:24:36 -0700918 delayacct_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Nadav Amit4fc19702019-04-26 16:22:46 -0700920 poking_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 check_bugs();
922
Rafael J. Wysockib064a8f2015-06-10 01:33:36 +0200923 acpi_subsystem_init();
Thomas Gleixnere7ff3a42016-12-09 19:29:10 +0100924 arch_post_acpi_subsys_init();
Feng Tang6ae69962009-08-14 15:13:46 -0400925 sfi_init_late();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926
927 /* Do the rest non-__init'ed, we're now alive */
Martin Schwidefsky53c99bd62018-08-31 10:42:24 +0200928 arch_call_rest_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929}
930
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700931/* Call all constructor functions linked into the kernel. */
932static void __init do_ctors(void)
933{
934#ifdef CONFIG_CONSTRUCTORS
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800935 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700936
H Hartley Sweeten196a15b2009-12-14 18:00:18 -0800937 for (; fn < (ctor_fn_t *) __ctors_end; fn++)
938 (*fn)();
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -0700939#endif
940}
941
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -0700942#ifdef CONFIG_KALLSYMS
943struct blacklist_entry {
944 struct list_head next;
945 char *buf;
946};
947
948static __initdata_or_module LIST_HEAD(blacklisted_initcalls);
949
950static int __init initcall_blacklist(char *str)
951{
952 char *str_entry;
953 struct blacklist_entry *entry;
954
955 /* str argument is a comma-separated list of functions */
956 do {
957 str_entry = strsep(&str, ",");
958 if (str_entry) {
959 pr_debug("blacklisting initcall %s\n", str_entry);
Mike Rapoport7e1c4e22018-10-30 15:09:57 -0700960 entry = memblock_alloc(sizeof(*entry),
961 SMP_CACHE_BYTES);
Mike Rapoportf5c73102019-03-11 23:30:20 -0700962 if (!entry)
963 panic("%s: Failed to allocate %zu bytes\n",
964 __func__, sizeof(*entry));
Mike Rapoport7e1c4e22018-10-30 15:09:57 -0700965 entry->buf = memblock_alloc(strlen(str_entry) + 1,
966 SMP_CACHE_BYTES);
Mike Rapoportf5c73102019-03-11 23:30:20 -0700967 if (!entry->buf)
968 panic("%s: Failed to allocate %zu bytes\n",
969 __func__, strlen(str_entry) + 1);
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -0700970 strcpy(entry->buf, str_entry);
971 list_add(&entry->next, &blacklisted_initcalls);
972 }
973 } while (str_entry);
974
975 return 0;
976}
977
978static bool __init_or_module initcall_blacklisted(initcall_t fn)
979{
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -0700980 struct blacklist_entry *entry;
Rasmus Villemoesc8cdd2b2016-05-20 17:04:27 -0700981 char fn_name[KSYM_SYMBOL_LEN];
Rasmus Villemoes0fd5ed82016-06-24 14:50:30 -0700982 unsigned long addr;
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -0700983
Rasmus Villemoesc8cdd2b2016-05-20 17:04:27 -0700984 if (list_empty(&blacklisted_initcalls))
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -0700985 return false;
986
Rasmus Villemoes0fd5ed82016-06-24 14:50:30 -0700987 addr = (unsigned long) dereference_function_descriptor(fn);
988 sprint_symbol_no_offset(fn_name, addr);
Rasmus Villemoesc8cdd2b2016-05-20 17:04:27 -0700989
Prarit Bhargava841c06d2016-08-02 14:07:15 -0700990 /*
991 * fn will be "function_name [module_name]" where [module_name] is not
992 * displayed for built-in init functions. Strip off the [module_name].
993 */
994 strreplace(fn_name, ' ', '\0');
995
Geliang Tange6fd1fb2016-03-15 14:52:46 -0700996 list_for_each_entry(entry, &blacklisted_initcalls, next) {
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -0700997 if (!strcmp(fn_name, entry->buf)) {
998 pr_debug("initcall %s blacklisted\n", fn_name);
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -0700999 return true;
1000 }
1001 }
1002
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -07001003 return false;
1004}
1005#else
1006static int __init initcall_blacklist(char *str)
1007{
1008 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n");
1009 return 0;
1010}
1011
1012static bool __init_or_module initcall_blacklisted(initcall_t fn)
1013{
1014 return false;
1015}
1016#endif
1017__setup("initcall_blacklist=", initcall_blacklist);
1018
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -04001019static __init_or_module void
1020trace_initcall_start_cb(void *data, initcall_t fn)
Linus Torvaldse0df1542008-05-15 18:14:01 -07001021{
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -04001022 ktime_t *calltime = (ktime_t *)data;
Linus Torvaldse0df1542008-05-15 18:14:01 -07001023
Sakari Ailusd75f7732019-03-25 21:32:28 +02001024 printk(KERN_DEBUG "calling %pS @ %i\n", fn, task_pid_nr(current));
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -04001025 *calltime = ktime_get();
1026}
1027
1028static __init_or_module void
1029trace_initcall_finish_cb(void *data, initcall_t fn, int ret)
1030{
1031 ktime_t *calltime = (ktime_t *)data;
1032 ktime_t delta, rettime;
1033 unsigned long long duration;
1034
Kevin Winchester22c5c032010-08-09 17:20:32 -07001035 rettime = ktime_get();
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -04001036 delta = ktime_sub(rettime, *calltime);
Kevin Winchester22c5c032010-08-09 17:20:32 -07001037 duration = (unsigned long long) ktime_to_ns(delta) >> 10;
Sakari Ailusd75f7732019-03-25 21:32:28 +02001038 printk(KERN_DEBUG "initcall %pS returned %d after %lld usecs\n",
Andrew Mortonea676e82013-04-29 16:18:20 -07001039 fn, ret, duration);
Kevin Winchester22c5c032010-08-09 17:20:32 -07001040}
1041
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -04001042static ktime_t initcall_calltime;
1043
Steven Rostedt (VMware)b0dc52f2018-04-06 09:24:25 -04001044#ifdef TRACEPOINTS_ENABLED
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -04001045static void __init initcall_debug_enable(void)
1046{
1047 int ret;
1048
1049 ret = register_trace_initcall_start(trace_initcall_start_cb,
1050 &initcall_calltime);
1051 ret |= register_trace_initcall_finish(trace_initcall_finish_cb,
1052 &initcall_calltime);
1053 WARN(ret, "Failed to register initcall tracepoints\n");
Kevin Winchester22c5c032010-08-09 17:20:32 -07001054}
Steven Rostedt (VMware)b0dc52f2018-04-06 09:24:25 -04001055# define do_trace_initcall_start trace_initcall_start
1056# define do_trace_initcall_finish trace_initcall_finish
1057#else
1058static inline void do_trace_initcall_start(initcall_t fn)
1059{
1060 if (!initcall_debug)
1061 return;
1062 trace_initcall_start_cb(&initcall_calltime, fn);
1063}
1064static inline void do_trace_initcall_finish(initcall_t fn, int ret)
1065{
1066 if (!initcall_debug)
1067 return;
1068 trace_initcall_finish_cb(&initcall_calltime, fn, ret);
1069}
1070#endif /* !TRACEPOINTS_ENABLED */
Kevin Winchester22c5c032010-08-09 17:20:32 -07001071
Kevin Winchestere4461272010-08-09 17:20:32 -07001072int __init_or_module do_one_initcall(initcall_t fn)
Kevin Winchester22c5c032010-08-09 17:20:32 -07001073{
1074 int count = preempt_count();
Steven Rostedtff1c8fa2013-07-03 15:05:37 -07001075 char msgbuf[64];
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -04001076 int ret;
Kevin Winchester22c5c032010-08-09 17:20:32 -07001077
Prarit Bhargava7b0b73d2014-06-04 16:12:17 -07001078 if (initcall_blacklisted(fn))
1079 return -EPERM;
1080
Steven Rostedt (VMware)b0dc52f2018-04-06 09:24:25 -04001081 do_trace_initcall_start(fn);
Steven Rostedt (VMware)4e379582018-03-26 13:31:07 -04001082 ret = fn();
Steven Rostedt (VMware)b0dc52f2018-04-06 09:24:25 -04001083 do_trace_initcall_finish(fn, ret);
Linus Torvaldse0df1542008-05-15 18:14:01 -07001084
1085 msgbuf[0] = 0;
1086
Linus Torvaldse0df1542008-05-15 18:14:01 -07001087 if (preempt_count() != count) {
Steven Rostedtbf5d7702013-05-01 13:35:51 -04001088 sprintf(msgbuf, "preemption imbalance ");
Peter Zijlstra4a2b4b22013-08-14 14:55:24 +02001089 preempt_count_set(count);
Linus Torvaldse0df1542008-05-15 18:14:01 -07001090 }
1091 if (irqs_disabled()) {
Cyrill Gorcunova76bfd02008-05-15 13:52:41 -07001092 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
Linus Torvaldse0df1542008-05-15 18:14:01 -07001093 local_irq_enable();
1094 }
Sakari Ailusd75f7732019-03-25 21:32:28 +02001095 WARN(msgbuf[0], "initcall %pS returned with %s\n", fn, msgbuf);
Arjan van de Ven59f94152008-07-30 12:49:02 -07001096
Emese Revfy38addce2016-06-20 20:41:19 +02001097 add_latent_entropy();
Américo Wang30dbb20e2010-05-26 18:57:53 +08001098 return ret;
Linus Torvaldse0df1542008-05-15 18:14:01 -07001099}
1100
1101
Ard Biesheuvel1b1eeca2018-08-21 21:56:13 -07001102extern initcall_entry_t __initcall_start[];
1103extern initcall_entry_t __initcall0_start[];
1104extern initcall_entry_t __initcall1_start[];
1105extern initcall_entry_t __initcall2_start[];
1106extern initcall_entry_t __initcall3_start[];
1107extern initcall_entry_t __initcall4_start[];
1108extern initcall_entry_t __initcall5_start[];
1109extern initcall_entry_t __initcall6_start[];
1110extern initcall_entry_t __initcall7_start[];
1111extern initcall_entry_t __initcall_end[];
Pawel Moll026cee02012-03-26 12:50:51 +10301112
Ard Biesheuvel1b1eeca2018-08-21 21:56:13 -07001113static initcall_entry_t *initcall_levels[] __initdata = {
Pawel Moll026cee02012-03-26 12:50:51 +10301114 __initcall0_start,
1115 __initcall1_start,
1116 __initcall2_start,
1117 __initcall3_start,
1118 __initcall4_start,
1119 __initcall5_start,
1120 __initcall6_start,
1121 __initcall7_start,
1122 __initcall_end,
1123};
1124
Jim Cromie96263d22012-06-14 16:00:59 -06001125/* Keep these in sync with initcalls in include/linux/init.h */
Alexey Dobriyan7c8f7192019-01-03 15:27:29 -08001126static const char *initcall_level_names[] __initdata = {
Steven Rostedt (VMware)a6fb6012018-03-22 19:28:54 -04001127 "pure",
Jim Cromie9fb48c72012-04-27 14:30:34 -06001128 "core",
1129 "postcore",
1130 "arch",
1131 "subsys",
1132 "fs",
1133 "device",
1134 "late",
Pawel Moll026cee02012-03-26 12:50:51 +10301135};
1136
Masami Hiramatsu0068c922020-01-11 01:04:31 +09001137static void __init do_initcall_level(int level, char *command_line)
Pawel Moll026cee02012-03-26 12:50:51 +10301138{
Ard Biesheuvel1b1eeca2018-08-21 21:56:13 -07001139 initcall_entry_t *fn;
Pawel Moll026cee02012-03-26 12:50:51 +10301140
Pawel Moll026cee02012-03-26 12:50:51 +10301141 parse_args(initcall_level_names[level],
Masami Hiramatsu0068c922020-01-11 01:04:31 +09001142 command_line, __start___param,
Pawel Moll026cee02012-03-26 12:50:51 +10301143 __stop___param - __start___param,
1144 level, level,
Luis R. Rodriguezecc86172015-03-30 16:20:03 -07001145 NULL, &repair_env_string);
Pawel Moll026cee02012-03-26 12:50:51 +10301146
Steven Rostedt (VMware)4ee7c602018-03-23 10:18:03 -04001147 trace_initcall_level(initcall_level_names[level]);
Pawel Moll026cee02012-03-26 12:50:51 +10301148 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
Ard Biesheuvel1b1eeca2018-08-21 21:56:13 -07001149 do_one_initcall(initcall_from_entry(fn));
Pawel Moll026cee02012-03-26 12:50:51 +10301150}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
1152static void __init do_initcalls(void)
1153{
Pawel Moll026cee02012-03-26 12:50:51 +10301154 int level;
Masami Hiramatsu0068c922020-01-11 01:04:31 +09001155 size_t len = strlen(saved_command_line) + 1;
1156 char *command_line;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157
Masami Hiramatsu0068c922020-01-11 01:04:31 +09001158 command_line = kzalloc(len, GFP_KERNEL);
1159 if (!command_line)
1160 panic("%s: Failed to allocate %zu bytes\n", __func__, len);
1161
1162 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++) {
1163 /* Parser modifies command_line, restore it each time */
1164 strcpy(command_line, saved_command_line);
1165 do_initcall_level(level, command_line);
1166 }
1167
1168 kfree(command_line);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169}
1170
1171/*
1172 * Ok, the machine is now initialized. None of the devices
1173 * have been touched yet, but the CPU subsystem is up and
1174 * running, and memory and process management works.
1175 *
1176 * Now we can finally start doing some real work..
1177 */
1178static void __init do_basic_setup(void)
1179{
Lai Jiangshan759ee092009-03-25 17:06:30 +08001180 cpuset_init_smp();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181 driver_init();
Eric W. Biedermanb04c3af2007-02-14 00:33:57 -08001182 init_irq_proc();
Peter Oberparleiterb99b87f2009-06-17 16:28:03 -07001183 do_ctors();
Linus Torvaldsd5767c52011-09-28 10:23:44 -07001184 usermodehelper_enable();
wangyanqingb0f84372011-09-29 15:09:40 +08001185 do_initcalls();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186}
1187
Eduard - Gabriel Munteanu7babe8d2008-07-25 19:45:11 -07001188static void __init do_pre_smp_initcalls(void)
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -07001189{
Ard Biesheuvel1b1eeca2018-08-21 21:56:13 -07001190 initcall_entry_t *fn;
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -07001191
Steven Rostedt (VMware)4ee7c602018-03-23 10:18:03 -04001192 trace_initcall_level("early");
Pawel Moll026cee02012-03-26 12:50:51 +10301193 for (fn = __initcall_start; fn < __initcall0_start; fn++)
Ard Biesheuvel1b1eeca2018-08-21 21:56:13 -07001194 do_one_initcall(initcall_from_entry(fn));
Eduard - Gabriel Munteanuc2147a52008-07-25 19:45:11 -07001195}
1196
Al Viroa74fb732012-10-10 21:28:25 -04001197static int run_init_process(const char *init_filename)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198{
1199 argv_init[0] = init_filename;
Paul Menzel3f5c15d2018-08-21 21:58:37 -07001200 pr_info("Run %s as init process\n", init_filename);
Linus Torvaldsc4ad8f92014-02-05 12:54:53 -08001201 return do_execve(getname_kernel(init_filename),
Al Viroae903ca2012-12-14 12:44:11 -05001202 (const char __user *const __user *)argv_init,
1203 (const char __user *const __user *)envp_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204}
1205
Michael Opdenackerba247622013-11-12 15:10:21 -08001206static int try_to_run_init_process(const char *init_filename)
1207{
1208 int ret;
1209
1210 ret = run_init_process(init_filename);
1211
1212 if (ret && ret != -ENOENT) {
1213 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
1214 init_filename, ret);
1215 }
1216
1217 return ret;
1218}
1219
Vineet Guptaf80b0c92012-12-21 12:25:44 +05301220static noinline void __init kernel_init_freeable(void);
Al Virod6b21232012-10-10 19:57:26 -04001221
Laura Abbott0f5bf6d2017-02-06 16:31:58 -08001222#if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX)
AKASHI Takahiro39290b32016-11-14 15:15:05 +09001223bool rodata_enabled __ro_after_init = true;
Kees Cookd2aa1ac2016-02-17 14:41:13 -08001224static int __init set_debug_rodata(char *str)
1225{
1226 return strtobool(str, &rodata_enabled);
1227}
1228__setup("rodata=", set_debug_rodata);
AKASHI Takahiro39290b32016-11-14 15:15:05 +09001229#endif
Kees Cookd2aa1ac2016-02-17 14:41:13 -08001230
Laura Abbott0f5bf6d2017-02-06 16:31:58 -08001231#ifdef CONFIG_STRICT_KERNEL_RWX
Kees Cookd2aa1ac2016-02-17 14:41:13 -08001232static void mark_readonly(void)
1233{
Jinbum Park2959a5f2017-02-27 14:30:22 -08001234 if (rodata_enabled) {
Jeffrey Hugoae646f02018-05-11 16:01:42 -07001235 /*
Paul E. McKenneyba180312018-11-06 18:58:01 -08001236 * load_module() results in W+X mappings, which are cleaned
1237 * up with call_rcu(). Let's make sure that queued work is
Jeffrey Hugoae646f02018-05-11 16:01:42 -07001238 * flushed so that we don't hit false positives looking for
1239 * insecure pages which are W+X.
1240 */
Paul E. McKenneyba180312018-11-06 18:58:01 -08001241 rcu_barrier();
Kees Cookd2aa1ac2016-02-17 14:41:13 -08001242 mark_rodata_ro();
Jinbum Park2959a5f2017-02-27 14:30:22 -08001243 rodata_test();
1244 } else
Kees Cookd2aa1ac2016-02-17 14:41:13 -08001245 pr_info("Kernel memory protection disabled.\n");
1246}
1247#else
1248static inline void mark_readonly(void)
1249{
1250 pr_warn("This architecture does not have kernel memory protection.\n");
1251}
1252#endif
1253
Mike Rapoport997aef62019-05-13 17:18:40 -07001254void __weak free_initmem(void)
1255{
Mike Rapoportf4039992019-05-13 17:18:46 -07001256 free_initmem_default(POISON_FREE_INITMEM);
Mike Rapoport997aef62019-05-13 17:18:40 -07001257}
1258
Al Virod6b21232012-10-10 19:57:26 -04001259static int __ref kernel_init(void *unused)
Vivek Goyalee5bfa62007-02-13 13:26:22 +01001260{
Michael Opdenackerba247622013-11-12 15:10:21 -08001261 int ret;
1262
Al Virod6b21232012-10-10 19:57:26 -04001263 kernel_init_freeable();
Arjan van de Ven22a9d642009-01-07 08:45:46 -08001264 /* need to finish all async __init code before freeing the memory */
1265 async_synchronize_full();
Steven Rostedt (VMware)b80f0f62017-04-03 12:57:35 -04001266 ftrace_free_init_mem();
Vivek Goyalee5bfa62007-02-13 13:26:22 +01001267 free_initmem();
Kees Cookd2aa1ac2016-02-17 14:41:13 -08001268 mark_readonly();
Joerg Roedelb9766902018-07-18 11:41:06 +02001269
1270 /*
1271 * Kernel mappings are now finalized - update the userspace page-table
1272 * to finalize PTI.
1273 */
1274 pti_finalize();
1275
Vivek Goyalee5bfa62007-02-13 13:26:22 +01001276 system_state = SYSTEM_RUNNING;
1277 numa_default_policy();
1278
Paul E. McKenney967dcb82015-11-25 16:52:36 -08001279 rcu_end_inkernel_boot();
1280
Vivek Goyalee5bfa62007-02-13 13:26:22 +01001281 if (ramdisk_execute_command) {
Michael Opdenackerba247622013-11-12 15:10:21 -08001282 ret = run_init_process(ramdisk_execute_command);
1283 if (!ret)
Al Viroa74fb732012-10-10 21:28:25 -04001284 return 0;
Michael Opdenackerba247622013-11-12 15:10:21 -08001285 pr_err("Failed to execute %s (error %d)\n",
1286 ramdisk_execute_command, ret);
Vivek Goyalee5bfa62007-02-13 13:26:22 +01001287 }
1288
1289 /*
1290 * We try each of these until one succeeds.
1291 *
1292 * The Bourne shell can be used instead of init if we are
1293 * trying to recover a really broken machine.
1294 */
1295 if (execute_command) {
Michael Opdenackerba247622013-11-12 15:10:21 -08001296 ret = run_init_process(execute_command);
1297 if (!ret)
Al Viroa74fb732012-10-10 21:28:25 -04001298 return 0;
Andy Lutomirski6ef45362014-12-10 15:52:19 -08001299 panic("Requested init %s failed (error %d).",
1300 execute_command, ret);
Vivek Goyalee5bfa62007-02-13 13:26:22 +01001301 }
Michael Opdenackerba247622013-11-12 15:10:21 -08001302 if (!try_to_run_init_process("/sbin/init") ||
1303 !try_to_run_init_process("/etc/init") ||
1304 !try_to_run_init_process("/bin/init") ||
1305 !try_to_run_init_process("/bin/sh"))
Al Viroa74fb732012-10-10 21:28:25 -04001306 return 0;
Vivek Goyalee5bfa62007-02-13 13:26:22 +01001307
Michael Opdenackerba247622013-11-12 15:10:21 -08001308 panic("No working init found. Try passing init= option to kernel. "
Mauro Carvalho Chehab8c27ceff32016-10-18 10:12:27 -02001309 "See Linux Documentation/admin-guide/init.rst for guidance.");
Vivek Goyalee5bfa62007-02-13 13:26:22 +01001310}
1311
Dominik Brodowskib49a7332018-10-23 16:00:10 +02001312void console_on_rootfs(void)
1313{
Dominik Brodowski74f1a292020-01-01 20:05:03 +01001314 /* Open the /dev/console as stdin, this should never fail */
1315 if (ksys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
1316 pr_err("Warning: unable to open an initial console.\n");
Dominik Brodowskib49a7332018-10-23 16:00:10 +02001317
Dominik Brodowski74f1a292020-01-01 20:05:03 +01001318 /* create stdout/stderr */
1319 (void) ksys_dup(0);
1320 (void) ksys_dup(0);
Dominik Brodowskib49a7332018-10-23 16:00:10 +02001321}
1322
Vineet Guptaf80b0c92012-12-21 12:25:44 +05301323static noinline void __init kernel_init_freeable(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324{
Peter Zijlstrab433c3d2010-06-28 16:51:01 +02001325 /*
1326 * Wait until kthreadd is all set-up.
1327 */
1328 wait_for_completion(&kthreadd_done);
Linus Torvalds31a67102012-05-21 12:52:42 -07001329
1330 /* Now the scheduler is fully set up and can do blocking allocations */
1331 gfp_allowed_mask = __GFP_BITS_MASK;
1332
Miao Xie58568d22009-06-16 15:31:49 -07001333 /*
1334 * init can allocate pages on any node
1335 */
Lai Jiangshan3c466d42012-12-12 13:51:40 -08001336 set_mems_allowed(node_states[N_MEMORY]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
Cedric Le Goater9ec52092006-10-02 02:19:00 -07001338 cad_pid = task_pid(current);
1339
Andi Kleenca74a6f2008-01-30 13:33:17 +01001340 smp_prepare_cpus(setup_max_cpus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
Tejun Heo3347fa02016-09-16 15:49:32 -04001342 workqueue_init();
1343
Michal Hocko597b7302017-03-31 15:11:47 -07001344 init_mm_internals();
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 do_pre_smp_initcalls();
Peter Zijlstra004417a2010-11-25 18:38:29 +01001347 lockup_detector_init();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 smp_init();
1350 sched_init_smp();
1351
Mel Gorman0e1cc952015-06-30 14:57:27 -07001352 page_alloc_init_late();
Qian Cai2f1ee092019-02-12 15:36:03 -08001353 /* Initialize page ext after all struct pages are initialized. */
1354 page_ext_init();
Mel Gorman0e1cc952015-06-30 14:57:27 -07001355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 do_basic_setup();
1357
Dominik Brodowskib49a7332018-10-23 16:00:10 +02001358 console_on_rootfs();
Eric W. Biederman2bd3a9972010-03-02 23:53:19 -08001359
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 /*
1361 * check if there is an early userspace init. If yes, let it do all
1362 * the work
1363 */
Olof Johanssonffdfc402005-09-06 15:17:19 -07001364
1365 if (!ramdisk_execute_command)
1366 ramdisk_execute_command = "/init";
1367
Dominik Brodowskicbfe20f2018-03-11 11:34:54 +01001368 if (ksys_access((const char __user *)
1369 ramdisk_execute_command, 0) != 0) {
Olof Johanssonffdfc402005-09-06 15:17:19 -07001370 ramdisk_execute_command = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 prepare_namespace();
Olof Johanssonffdfc402005-09-06 15:17:19 -07001372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
1374 /*
1375 * Ok, we have completed the initial bootup, and
1376 * we're essentially up and running. Get rid of the
1377 * initmem segments and start the user-mode stuff..
Dmitry Kasatkinc9cd2ce2014-11-05 17:01:15 +02001378 *
1379 * rootfs is available now, try loading the public keys
1380 * and default modules
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 */
Tejun Heobb813f42013-01-18 14:05:56 -08001382
Dmitry Kasatkinc9cd2ce2014-11-05 17:01:15 +02001383 integrity_load_keys();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384}