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/vmalloc.h> 37 #include <linux/kernel_stat.h> 38 #include <linux/start_kernel.h> 39 #include <linux/security.h> 40 #include <linux/smp.h> 41 #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> 48 #include <linux/cgroup.h> 49 #include <linux/efi.h> 50 #include <linux/tick.h> 51 #include <linux/sched/isolation.h> 52 #include <linux/interrupt.h> 53 #include <linux/taskstats_kern.h> 54 #include <linux/delayacct.h> 55 #include <linux/unistd.h> 56 #include <linux/utsname.h> 57 #include <linux/rmap.h> 58 #include <linux/mempolicy.h> 59 #include <linux/key.h> 60 #include <linux/buffer_head.h> 61 #include <linux/page_ext.h> 62 #include <linux/debug_locks.h> 63 #include <linux/debugobjects.h> 64 #include <linux/lockdep.h> 65 #include <linux/kmemleak.h> 66 #include <linux/pid_namespace.h> 67 #include <linux/device.h> 68 #include <linux/kthread.h> 69 #include <linux/sched.h> 70 #include <linux/sched/init.h> 71 #include <linux/signal.h> 72 #include <linux/idr.h> 73 #include <linux/kgdb.h> 74 #include <linux/ftrace.h> 75 #include <linux/async.h> 76 #include <linux/sfi.h> 77 #include <linux/shmem_fs.h> 78 #include <linux/slab.h> 79 #include <linux/perf_event.h> 80 #include <linux/ptrace.h> 81 #include <linux/pti.h> 82 #include <linux/blkdev.h> 83 #include <linux/elevator.h> 84 #include <linux/sched/clock.h> 85 #include <linux/sched/task.h> 86 #include <linux/sched/task_stack.h> 87 #include <linux/context_tracking.h> 88 #include <linux/random.h> 89 #include <linux/list.h> 90 #include <linux/integrity.h> 91 #include <linux/proc_ns.h> 92 #include <linux/io.h> 93 #include <linux/cache.h> 94 #include <linux/rodata_test.h> 95 #include <linux/jump_label.h> 96 #include <linux/mem_encrypt.h> 97 98 #include <asm/io.h> 99 #include <asm/bugs.h> 100 #include <asm/setup.h> 101 #include <asm/sections.h> 102 #include <asm/cacheflush.h> 103 104 #define CREATE_TRACE_POINTS 105 #include <trace/events/initcall.h> 106 107 static int kernel_init(void *); 108 109 extern void init_IRQ(void); 110 extern void radix_tree_init(void); 111 112 /* 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 */ 119 bool early_boot_irqs_disabled __read_mostly; 120 121 enum system_states system_state __read_mostly; 122 EXPORT_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 129 130 extern void time_init(void); 131 /* Default late time init is NULL. archs can override this later. */ 132 void (*__initdata late_time_init)(void); 133 134 /* Untouched command line saved by arch-specific code. */ 135 char __initdata boot_command_line[COMMAND_LINE_SIZE]; 136 /* Untouched saved command line (eg. for /proc) */ 137 char *saved_command_line; 138 /* Command line for parameter parsing */ 139 static char *static_command_line; 140 /* Command line for per-initcall parameter parsing */ 141 static char *initcall_command_line; 142 143 static char *execute_command; 144 static char *ramdisk_execute_command; 145 146 /* 147 * Used to generate warnings if static_key manipulation functions are used 148 * before jump_label_init is called. 149 */ 150 bool static_key_initialized __read_mostly; 151 EXPORT_SYMBOL_GPL(static_key_initialized); 152 153 /* 154 * 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. 159 * For ex. kdump situation where previous kernel has crashed, BIOS has been 160 * skipped and devices will be in unknown state. 161 */ 162 unsigned int reset_devices; 163 EXPORT_SYMBOL(reset_devices); 164 165 static int __init set_reset_devices(char *str) 166 { 167 reset_devices = 1; 168 return 1; 169 } 170 171 __setup("reset_devices", set_reset_devices); 172 173 static const char *argv_init[MAX_INIT_ARGS+2] = { "init", NULL, }; 174 const char *envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, }; 175 static const char *panic_later, *panic_param; 176 177 extern const struct obs_kernel_param __setup_start[], __setup_end[]; 178 179 static bool __init obsolete_checksetup(char *line) 180 { 181 const struct obs_kernel_param *p; 182 bool had_early_param = false; 183 184 p = __setup_start; 185 do { 186 int n = strlen(p->str); 187 if (parameqn(line, p->str, n)) { 188 if (p->early) { 189 /* 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( */ 193 if (line[n] == '\0' || line[n] == '=') 194 had_early_param = true; 195 } else if (!p->setup_func) { 196 pr_warn("Parameter %s is obsolete, ignored\n", 197 p->str); 198 return true; 199 } else if (p->setup_func(line + n)) 200 return true; 201 } 202 p++; 203 } while (p < __setup_end); 204 205 return had_early_param; 206 } 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 */ 212 unsigned long loops_per_jiffy = (1<<12); 213 EXPORT_SYMBOL(loops_per_jiffy); 214 215 static int __init debug_kernel(char *str) 216 { 217 console_loglevel = CONSOLE_LOGLEVEL_DEBUG; 218 return 0; 219 } 220 221 static int __init quiet_kernel(char *str) 222 { 223 console_loglevel = CONSOLE_LOGLEVEL_QUIET; 224 return 0; 225 } 226 227 early_param("debug", debug_kernel); 228 early_param("quiet", quiet_kernel); 229 230 static int __init loglevel(char *str) 231 { 232 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; 245 } 246 247 early_param("loglevel", loglevel); 248 249 #ifdef CONFIG_BOOT_CONFIG 250 u32 boot_config_checksum(unsigned char *p, u32 size) 251 { 252 u32 ret = 0; 253 254 while (size--) 255 ret += *p++; 256 257 return ret; 258 } 259 260 static void __init setup_boot_config(void) 261 { 262 u32 size, csum; 263 char *data, *copy; 264 u32 *hdr; 265 266 if (!initrd_end) 267 return; 268 269 hdr = (u32 *)(initrd_end - 8); 270 size = hdr[0]; 271 csum = hdr[1]; 272 273 if (size >= XBC_DATA_MAX) 274 return; 275 276 data = ((void *)hdr) - size; 277 if ((unsigned long)data < initrd_start) 278 return; 279 280 if (boot_config_checksum((unsigned char *)data, size) != csum) 281 return; 282 283 copy = memblock_alloc(size + 1, SMP_CACHE_BYTES); 284 if (!copy) { 285 pr_err("Failed to allocate memory for boot config\n"); 286 return; 287 } 288 289 memcpy(copy, data, size); 290 copy[size] = '\0'; 291 292 if (xbc_init(copy) < 0) 293 pr_err("Failed to parse boot config\n"); 294 else 295 pr_info("Load boot config: %d bytes\n", size); 296 } 297 #else 298 #define setup_boot_config() do { } while (0) 299 #endif 300 301 /* Change NUL term back to "=", to make "param" the whole string. */ 302 static int __init repair_env_string(char *param, char *val, 303 const char *unused, void *arg) 304 { 305 if (val) { 306 /* param=val or param="val"? */ 307 if (val == param+strlen(param)+1) 308 val[-1] = '='; 309 else if (val == param+strlen(param)+2) { 310 val[-2] = '='; 311 memmove(val-1, val, strlen(val)+1); 312 val--; 313 } else 314 BUG(); 315 } 316 return 0; 317 } 318 319 /* Anything after -- gets handed straight to init. */ 320 static int __init set_init_arg(char *param, char *val, 321 const char *unused, void *arg) 322 { 323 unsigned int i; 324 325 if (panic_later) 326 return 0; 327 328 repair_env_string(param, val, unused, NULL); 329 330 for (i = 0; argv_init[i]; i++) { 331 if (i == MAX_INIT_ARGS) { 332 panic_later = "init"; 333 panic_param = param; 334 return 0; 335 } 336 } 337 argv_init[i] = param; 338 return 0; 339 } 340 341 /* 342 * Unknown boot options get handed to init, unless they look like 343 * unused parameters (modprobe will find them in /proc/cmdline). 344 */ 345 static int __init unknown_bootoption(char *param, char *val, 346 const char *unused, void *arg) 347 { 348 repair_env_string(param, val, unused, NULL); 349 350 /* Handle obsolete-style parameters */ 351 if (obsolete_checksetup(param)) 352 return 0; 353 354 /* Unused module parameter. */ 355 if (strchr(param, '.') && (!val || strchr(param, '.') < val)) 356 return 0; 357 358 if (panic_later) 359 return 0; 360 361 if (val) { 362 /* Environment option */ 363 unsigned int i; 364 for (i = 0; envp_init[i]; i++) { 365 if (i == MAX_INIT_ENVS) { 366 panic_later = "env"; 367 panic_param = param; 368 } 369 if (!strncmp(param, envp_init[i], val - param)) 370 break; 371 } 372 envp_init[i] = param; 373 } else { 374 /* Command line option */ 375 unsigned int i; 376 for (i = 0; argv_init[i]; i++) { 377 if (i == MAX_INIT_ARGS) { 378 panic_later = "init"; 379 panic_param = param; 380 } 381 } 382 argv_init[i] = param; 383 } 384 return 0; 385 } 386 387 static int __init init_setup(char *str) 388 { 389 unsigned int i; 390 391 execute_command = str; 392 /* 393 * In case LILO is going to boot us with default command line, 394 * it prepends "auto" before the whole cmdline which makes 395 * the shell think it should execute a script with such name. 396 * So we ignore all arguments entered _before_ init=... [MJ] 397 */ 398 for (i = 1; i < MAX_INIT_ARGS; i++) 399 argv_init[i] = NULL; 400 return 1; 401 } 402 __setup("init=", init_setup); 403 404 static int __init rdinit_setup(char *str) 405 { 406 unsigned int i; 407 408 ramdisk_execute_command = str; 409 /* See "auto" comment in init_setup */ 410 for (i = 1; i < MAX_INIT_ARGS; i++) 411 argv_init[i] = NULL; 412 return 1; 413 } 414 __setup("rdinit=", rdinit_setup); 415 416 #ifndef CONFIG_SMP 417 static const unsigned int setup_max_cpus = NR_CPUS; 418 static inline void setup_nr_cpu_ids(void) { } 419 static inline void smp_prepare_cpus(unsigned int maxcpus) { } 420 #endif 421 422 /* 423 * We need to store the untouched command line for future reference. 424 * We also need to store the touched command line since the parameter 425 * parsing is performed in place, and we should allow a component to 426 * store reference of name/value for future reference. 427 */ 428 static void __init setup_command_line(char *command_line) 429 { 430 size_t len = strlen(boot_command_line) + 1; 431 432 saved_command_line = memblock_alloc(len, SMP_CACHE_BYTES); 433 if (!saved_command_line) 434 panic("%s: Failed to allocate %zu bytes\n", __func__, len); 435 436 initcall_command_line = memblock_alloc(len, SMP_CACHE_BYTES); 437 if (!initcall_command_line) 438 panic("%s: Failed to allocate %zu bytes\n", __func__, len); 439 440 static_command_line = memblock_alloc(len, SMP_CACHE_BYTES); 441 if (!static_command_line) 442 panic("%s: Failed to allocate %zu bytes\n", __func__, len); 443 444 strcpy(saved_command_line, boot_command_line); 445 strcpy(static_command_line, command_line); 446 } 447 448 /* 449 * We need to finalize in a non-__init function or else race conditions 450 * between the root thread and the init thread may cause start_kernel to 451 * be reaped by free_initmem before the root thread has proceeded to 452 * cpu_idle. 453 * 454 * gcc-3.4 accidentally inlines this function, so use noinline. 455 */ 456 457 static __initdata DECLARE_COMPLETION(kthreadd_done); 458 459 noinline void __ref rest_init(void) 460 { 461 struct task_struct *tsk; 462 int pid; 463 464 rcu_scheduler_starting(); 465 /* 466 * We need to spawn init first so that it obtains pid 1, however 467 * the init task will end up wanting to create kthreads, which, if 468 * we schedule it before we create kthreadd, will OOPS. 469 */ 470 pid = kernel_thread(kernel_init, NULL, CLONE_FS); 471 /* 472 * Pin init on the boot CPU. Task migration is not properly working 473 * until sched_init_smp() has been run. It will set the allowed 474 * CPUs for init to the non isolated CPUs. 475 */ 476 rcu_read_lock(); 477 tsk = find_task_by_pid_ns(pid, &init_pid_ns); 478 set_cpus_allowed_ptr(tsk, cpumask_of(smp_processor_id())); 479 rcu_read_unlock(); 480 481 numa_default_policy(); 482 pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES); 483 rcu_read_lock(); 484 kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns); 485 rcu_read_unlock(); 486 487 /* 488 * Enable might_sleep() and smp_processor_id() checks. 489 * They cannot be enabled earlier because with CONFIG_PREEMPTION=y 490 * kernel_thread() would trigger might_sleep() splats. With 491 * CONFIG_PREEMPT_VOLUNTARY=y the init task might have scheduled 492 * already, but it's stuck on the kthreadd_done completion. 493 */ 494 system_state = SYSTEM_SCHEDULING; 495 496 complete(&kthreadd_done); 497 498 /* 499 * The boot idle thread must execute schedule() 500 * at least once to get things moving: 501 */ 502 schedule_preempt_disabled(); 503 /* Call into cpu_idle with preempt disabled */ 504 cpu_startup_entry(CPUHP_ONLINE); 505 } 506 507 /* Check for early params. */ 508 static int __init do_early_param(char *param, char *val, 509 const char *unused, void *arg) 510 { 511 const struct obs_kernel_param *p; 512 513 for (p = __setup_start; p < __setup_end; p++) { 514 if ((p->early && parameq(param, p->str)) || 515 (strcmp(param, "console") == 0 && 516 strcmp(p->str, "earlycon") == 0) 517 ) { 518 if (p->setup_func(val) != 0) 519 pr_warn("Malformed early option '%s'\n", param); 520 } 521 } 522 /* We accept everything at this stage. */ 523 return 0; 524 } 525 526 void __init parse_early_options(char *cmdline) 527 { 528 parse_args("early options", cmdline, NULL, 0, 0, 0, NULL, 529 do_early_param); 530 } 531 532 /* Arch code calls this early on, or if not, just before other parsing. */ 533 void __init parse_early_param(void) 534 { 535 static int done __initdata; 536 static char tmp_cmdline[COMMAND_LINE_SIZE] __initdata; 537 538 if (done) 539 return; 540 541 /* All fall through to do_early_param. */ 542 strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE); 543 parse_early_options(tmp_cmdline); 544 done = 1; 545 } 546 547 void __init __weak arch_post_acpi_subsys_init(void) { } 548 549 void __init __weak smp_setup_processor_id(void) 550 { 551 } 552 553 # if THREAD_SIZE >= PAGE_SIZE 554 void __init __weak thread_stack_cache_init(void) 555 { 556 } 557 #endif 558 559 void __init __weak mem_encrypt_init(void) { } 560 561 void __init __weak poking_init(void) { } 562 563 void __init __weak pgtable_cache_init(void) { } 564 565 bool initcall_debug; 566 core_param(initcall_debug, initcall_debug, bool, 0644); 567 568 #ifdef TRACEPOINTS_ENABLED 569 static void __init initcall_debug_enable(void); 570 #else 571 static inline void initcall_debug_enable(void) 572 { 573 } 574 #endif 575 576 /* Report memory auto-initialization states for this boot. */ 577 static void __init report_meminit(void) 578 { 579 const char *stack; 580 581 if (IS_ENABLED(CONFIG_INIT_STACK_ALL)) 582 stack = "all"; 583 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF_ALL)) 584 stack = "byref_all"; 585 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_BYREF)) 586 stack = "byref"; 587 else if (IS_ENABLED(CONFIG_GCC_PLUGIN_STRUCTLEAK_USER)) 588 stack = "__user"; 589 else 590 stack = "off"; 591 592 pr_info("mem auto-init: stack:%s, heap alloc:%s, heap free:%s\n", 593 stack, want_init_on_alloc(GFP_KERNEL) ? "on" : "off", 594 want_init_on_free() ? "on" : "off"); 595 if (want_init_on_free()) 596 pr_info("mem auto-init: clearing system memory may take some time...\n"); 597 } 598 599 /* 600 * Set up kernel memory allocators 601 */ 602 static void __init mm_init(void) 603 { 604 /* 605 * page_ext requires contiguous pages, 606 * bigger than MAX_ORDER unless SPARSEMEM. 607 */ 608 page_ext_init_flatmem(); 609 report_meminit(); 610 mem_init(); 611 kmem_cache_init(); 612 kmemleak_init(); 613 pgtable_init(); 614 debug_objects_mem_init(); 615 vmalloc_init(); 616 ioremap_huge_init(); 617 /* Should be run before the first non-init thread is created */ 618 init_espfix_bsp(); 619 /* Should be run after espfix64 is set up. */ 620 pti_init(); 621 } 622 623 void __init __weak arch_call_rest_init(void) 624 { 625 rest_init(); 626 } 627 628 asmlinkage __visible void __init start_kernel(void) 629 { 630 char *command_line; 631 char *after_dashes; 632 633 set_task_stack_end_magic(&init_task); 634 smp_setup_processor_id(); 635 debug_objects_early_init(); 636 637 cgroup_init_early(); 638 639 local_irq_disable(); 640 early_boot_irqs_disabled = true; 641 642 /* 643 * Interrupts are still disabled. Do necessary setups, then 644 * enable them. 645 */ 646 boot_cpu_init(); 647 page_address_init(); 648 pr_notice("%s", linux_banner); 649 early_security_init(); 650 setup_arch(&command_line); 651 setup_boot_config(); 652 setup_command_line(command_line); 653 setup_nr_cpu_ids(); 654 setup_per_cpu_areas(); 655 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */ 656 boot_cpu_hotplug_init(); 657 658 build_all_zonelists(NULL); 659 page_alloc_init(); 660 661 pr_notice("Kernel command line: %s\n", boot_command_line); 662 /* parameters may set static keys */ 663 jump_label_init(); 664 parse_early_param(); 665 after_dashes = parse_args("Booting kernel", 666 static_command_line, __start___param, 667 __stop___param - __start___param, 668 -1, -1, NULL, &unknown_bootoption); 669 if (!IS_ERR_OR_NULL(after_dashes)) 670 parse_args("Setting init args", after_dashes, NULL, 0, -1, -1, 671 NULL, set_init_arg); 672 673 /* 674 * These use large bootmem allocations and must precede 675 * kmem_cache_init() 676 */ 677 setup_log_buf(0); 678 vfs_caches_init_early(); 679 sort_main_extable(); 680 trap_init(); 681 mm_init(); 682 683 ftrace_init(); 684 685 /* trace_printk can be enabled here */ 686 early_trace_init(); 687 688 /* 689 * Set up the scheduler prior starting any interrupts (such as the 690 * timer interrupt). Full topology setup happens at smp_init() 691 * time - but meanwhile we still have a functioning scheduler. 692 */ 693 sched_init(); 694 /* 695 * Disable preemption - early bootup scheduling is extremely 696 * fragile until we cpu_idle() for the first time. 697 */ 698 preempt_disable(); 699 if (WARN(!irqs_disabled(), 700 "Interrupts were enabled *very* early, fixing it\n")) 701 local_irq_disable(); 702 radix_tree_init(); 703 704 /* 705 * Set up housekeeping before setting up workqueues to allow the unbound 706 * workqueue to take non-housekeeping into account. 707 */ 708 housekeeping_init(); 709 710 /* 711 * Allow workqueue creation and work item queueing/cancelling 712 * early. Work item execution depends on kthreads and starts after 713 * workqueue_init(). 714 */ 715 workqueue_init_early(); 716 717 rcu_init(); 718 719 /* Trace events are available after this */ 720 trace_init(); 721 722 if (initcall_debug) 723 initcall_debug_enable(); 724 725 context_tracking_init(); 726 /* init some links before init_ISA_irqs() */ 727 early_irq_init(); 728 init_IRQ(); 729 tick_init(); 730 rcu_init_nohz(); 731 init_timers(); 732 hrtimers_init(); 733 softirq_init(); 734 timekeeping_init(); 735 736 /* 737 * For best initial stack canary entropy, prepare it after: 738 * - setup_arch() for any UEFI RNG entropy and boot cmdline access 739 * - timekeeping_init() for ktime entropy used in rand_initialize() 740 * - rand_initialize() to get any arch-specific entropy like RDRAND 741 * - add_latent_entropy() to get any latent entropy 742 * - adding command line entropy 743 */ 744 rand_initialize(); 745 add_latent_entropy(); 746 add_device_randomness(command_line, strlen(command_line)); 747 boot_init_stack_canary(); 748 749 time_init(); 750 printk_safe_init(); 751 perf_event_init(); 752 profile_init(); 753 call_function_init(); 754 WARN(!irqs_disabled(), "Interrupts were enabled early\n"); 755 756 early_boot_irqs_disabled = false; 757 local_irq_enable(); 758 759 kmem_cache_init_late(); 760 761 /* 762 * HACK ALERT! This is early. We're enabling the console before 763 * we've done PCI setups etc, and console_init() must be aware of 764 * this. But we do want output early, in case something goes wrong. 765 */ 766 console_init(); 767 if (panic_later) 768 panic("Too many boot %s vars at `%s'", panic_later, 769 panic_param); 770 771 lockdep_init(); 772 773 /* 774 * Need to run this when irqs are enabled, because it wants 775 * to self-test [hard/soft]-irqs on/off lock inversion bugs 776 * too: 777 */ 778 locking_selftest(); 779 780 /* 781 * This needs to be called before any devices perform DMA 782 * operations that might use the SWIOTLB bounce buffers. It will 783 * mark the bounce buffers as decrypted so that their usage will 784 * not cause "plain-text" data to be decrypted when accessed. 785 */ 786 mem_encrypt_init(); 787 788 #ifdef CONFIG_BLK_DEV_INITRD 789 if (initrd_start && !initrd_below_start_ok && 790 page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) { 791 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n", 792 page_to_pfn(virt_to_page((void *)initrd_start)), 793 min_low_pfn); 794 initrd_start = 0; 795 } 796 #endif 797 setup_per_cpu_pageset(); 798 numa_policy_init(); 799 acpi_early_init(); 800 if (late_time_init) 801 late_time_init(); 802 sched_clock_init(); 803 calibrate_delay(); 804 pid_idr_init(); 805 anon_vma_init(); 806 #ifdef CONFIG_X86 807 if (efi_enabled(EFI_RUNTIME_SERVICES)) 808 efi_enter_virtual_mode(); 809 #endif 810 thread_stack_cache_init(); 811 cred_init(); 812 fork_init(); 813 proc_caches_init(); 814 uts_ns_init(); 815 buffer_init(); 816 key_init(); 817 security_init(); 818 dbg_late_init(); 819 vfs_caches_init(); 820 pagecache_init(); 821 signals_init(); 822 seq_file_init(); 823 proc_root_init(); 824 nsfs_init(); 825 cpuset_init(); 826 cgroup_init(); 827 taskstats_init_early(); 828 delayacct_init(); 829 830 poking_init(); 831 check_bugs(); 832 833 acpi_subsystem_init(); 834 arch_post_acpi_subsys_init(); 835 sfi_init_late(); 836 837 /* Do the rest non-__init'ed, we're now alive */ 838 arch_call_rest_init(); 839 } 840 841 /* Call all constructor functions linked into the kernel. */ 842 static void __init do_ctors(void) 843 { 844 #ifdef CONFIG_CONSTRUCTORS 845 ctor_fn_t *fn = (ctor_fn_t *) __ctors_start; 846 847 for (; fn < (ctor_fn_t *) __ctors_end; fn++) 848 (*fn)(); 849 #endif 850 } 851 852 #ifdef CONFIG_KALLSYMS 853 struct blacklist_entry { 854 struct list_head next; 855 char *buf; 856 }; 857 858 static __initdata_or_module LIST_HEAD(blacklisted_initcalls); 859 860 static int __init initcall_blacklist(char *str) 861 { 862 char *str_entry; 863 struct blacklist_entry *entry; 864 865 /* str argument is a comma-separated list of functions */ 866 do { 867 str_entry = strsep(&str, ","); 868 if (str_entry) { 869 pr_debug("blacklisting initcall %s\n", str_entry); 870 entry = memblock_alloc(sizeof(*entry), 871 SMP_CACHE_BYTES); 872 if (!entry) 873 panic("%s: Failed to allocate %zu bytes\n", 874 __func__, sizeof(*entry)); 875 entry->buf = memblock_alloc(strlen(str_entry) + 1, 876 SMP_CACHE_BYTES); 877 if (!entry->buf) 878 panic("%s: Failed to allocate %zu bytes\n", 879 __func__, strlen(str_entry) + 1); 880 strcpy(entry->buf, str_entry); 881 list_add(&entry->next, &blacklisted_initcalls); 882 } 883 } while (str_entry); 884 885 return 0; 886 } 887 888 static bool __init_or_module initcall_blacklisted(initcall_t fn) 889 { 890 struct blacklist_entry *entry; 891 char fn_name[KSYM_SYMBOL_LEN]; 892 unsigned long addr; 893 894 if (list_empty(&blacklisted_initcalls)) 895 return false; 896 897 addr = (unsigned long) dereference_function_descriptor(fn); 898 sprint_symbol_no_offset(fn_name, addr); 899 900 /* 901 * fn will be "function_name [module_name]" where [module_name] is not 902 * displayed for built-in init functions. Strip off the [module_name]. 903 */ 904 strreplace(fn_name, ' ', '\0'); 905 906 list_for_each_entry(entry, &blacklisted_initcalls, next) { 907 if (!strcmp(fn_name, entry->buf)) { 908 pr_debug("initcall %s blacklisted\n", fn_name); 909 return true; 910 } 911 } 912 913 return false; 914 } 915 #else 916 static int __init initcall_blacklist(char *str) 917 { 918 pr_warn("initcall_blacklist requires CONFIG_KALLSYMS\n"); 919 return 0; 920 } 921 922 static bool __init_or_module initcall_blacklisted(initcall_t fn) 923 { 924 return false; 925 } 926 #endif 927 __setup("initcall_blacklist=", initcall_blacklist); 928 929 static __init_or_module void 930 trace_initcall_start_cb(void *data, initcall_t fn) 931 { 932 ktime_t *calltime = (ktime_t *)data; 933 934 printk(KERN_DEBUG "calling %pS @ %i\n", fn, task_pid_nr(current)); 935 *calltime = ktime_get(); 936 } 937 938 static __init_or_module void 939 trace_initcall_finish_cb(void *data, initcall_t fn, int ret) 940 { 941 ktime_t *calltime = (ktime_t *)data; 942 ktime_t delta, rettime; 943 unsigned long long duration; 944 945 rettime = ktime_get(); 946 delta = ktime_sub(rettime, *calltime); 947 duration = (unsigned long long) ktime_to_ns(delta) >> 10; 948 printk(KERN_DEBUG "initcall %pS returned %d after %lld usecs\n", 949 fn, ret, duration); 950 } 951 952 static ktime_t initcall_calltime; 953 954 #ifdef TRACEPOINTS_ENABLED 955 static void __init initcall_debug_enable(void) 956 { 957 int ret; 958 959 ret = register_trace_initcall_start(trace_initcall_start_cb, 960 &initcall_calltime); 961 ret |= register_trace_initcall_finish(trace_initcall_finish_cb, 962 &initcall_calltime); 963 WARN(ret, "Failed to register initcall tracepoints\n"); 964 } 965 # define do_trace_initcall_start trace_initcall_start 966 # define do_trace_initcall_finish trace_initcall_finish 967 #else 968 static inline void do_trace_initcall_start(initcall_t fn) 969 { 970 if (!initcall_debug) 971 return; 972 trace_initcall_start_cb(&initcall_calltime, fn); 973 } 974 static inline void do_trace_initcall_finish(initcall_t fn, int ret) 975 { 976 if (!initcall_debug) 977 return; 978 trace_initcall_finish_cb(&initcall_calltime, fn, ret); 979 } 980 #endif /* !TRACEPOINTS_ENABLED */ 981 982 int __init_or_module do_one_initcall(initcall_t fn) 983 { 984 int count = preempt_count(); 985 char msgbuf[64]; 986 int ret; 987 988 if (initcall_blacklisted(fn)) 989 return -EPERM; 990 991 do_trace_initcall_start(fn); 992 ret = fn(); 993 do_trace_initcall_finish(fn, ret); 994 995 msgbuf[0] = 0; 996 997 if (preempt_count() != count) { 998 sprintf(msgbuf, "preemption imbalance "); 999 preempt_count_set(count); 1000 } 1001 if (irqs_disabled()) { 1002 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf)); 1003 local_irq_enable(); 1004 } 1005 WARN(msgbuf[0], "initcall %pS returned with %s\n", fn, msgbuf); 1006 1007 add_latent_entropy(); 1008 return ret; 1009 } 1010 1011 1012 extern initcall_entry_t __initcall_start[]; 1013 extern initcall_entry_t __initcall0_start[]; 1014 extern initcall_entry_t __initcall1_start[]; 1015 extern initcall_entry_t __initcall2_start[]; 1016 extern initcall_entry_t __initcall3_start[]; 1017 extern initcall_entry_t __initcall4_start[]; 1018 extern initcall_entry_t __initcall5_start[]; 1019 extern initcall_entry_t __initcall6_start[]; 1020 extern initcall_entry_t __initcall7_start[]; 1021 extern initcall_entry_t __initcall_end[]; 1022 1023 static initcall_entry_t *initcall_levels[] __initdata = { 1024 __initcall0_start, 1025 __initcall1_start, 1026 __initcall2_start, 1027 __initcall3_start, 1028 __initcall4_start, 1029 __initcall5_start, 1030 __initcall6_start, 1031 __initcall7_start, 1032 __initcall_end, 1033 }; 1034 1035 /* Keep these in sync with initcalls in include/linux/init.h */ 1036 static const char *initcall_level_names[] __initdata = { 1037 "pure", 1038 "core", 1039 "postcore", 1040 "arch", 1041 "subsys", 1042 "fs", 1043 "device", 1044 "late", 1045 }; 1046 1047 static void __init do_initcall_level(int level) 1048 { 1049 initcall_entry_t *fn; 1050 1051 strcpy(initcall_command_line, saved_command_line); 1052 parse_args(initcall_level_names[level], 1053 initcall_command_line, __start___param, 1054 __stop___param - __start___param, 1055 level, level, 1056 NULL, &repair_env_string); 1057 1058 trace_initcall_level(initcall_level_names[level]); 1059 for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++) 1060 do_one_initcall(initcall_from_entry(fn)); 1061 } 1062 1063 static void __init do_initcalls(void) 1064 { 1065 int level; 1066 1067 for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++) 1068 do_initcall_level(level); 1069 } 1070 1071 /* 1072 * Ok, the machine is now initialized. None of the devices 1073 * have been touched yet, but the CPU subsystem is up and 1074 * running, and memory and process management works. 1075 * 1076 * Now we can finally start doing some real work.. 1077 */ 1078 static void __init do_basic_setup(void) 1079 { 1080 cpuset_init_smp(); 1081 driver_init(); 1082 init_irq_proc(); 1083 do_ctors(); 1084 usermodehelper_enable(); 1085 do_initcalls(); 1086 } 1087 1088 static void __init do_pre_smp_initcalls(void) 1089 { 1090 initcall_entry_t *fn; 1091 1092 trace_initcall_level("early"); 1093 for (fn = __initcall_start; fn < __initcall0_start; fn++) 1094 do_one_initcall(initcall_from_entry(fn)); 1095 } 1096 1097 static int run_init_process(const char *init_filename) 1098 { 1099 argv_init[0] = init_filename; 1100 pr_info("Run %s as init process\n", init_filename); 1101 return do_execve(getname_kernel(init_filename), 1102 (const char __user *const __user *)argv_init, 1103 (const char __user *const __user *)envp_init); 1104 } 1105 1106 static int try_to_run_init_process(const char *init_filename) 1107 { 1108 int ret; 1109 1110 ret = run_init_process(init_filename); 1111 1112 if (ret && ret != -ENOENT) { 1113 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n", 1114 init_filename, ret); 1115 } 1116 1117 return ret; 1118 } 1119 1120 static noinline void __init kernel_init_freeable(void); 1121 1122 #if defined(CONFIG_STRICT_KERNEL_RWX) || defined(CONFIG_STRICT_MODULE_RWX) 1123 bool rodata_enabled __ro_after_init = true; 1124 static int __init set_debug_rodata(char *str) 1125 { 1126 return strtobool(str, &rodata_enabled); 1127 } 1128 __setup("rodata=", set_debug_rodata); 1129 #endif 1130 1131 #ifdef CONFIG_STRICT_KERNEL_RWX 1132 static void mark_readonly(void) 1133 { 1134 if (rodata_enabled) { 1135 /* 1136 * load_module() results in W+X mappings, which are cleaned 1137 * up with call_rcu(). Let's make sure that queued work is 1138 * flushed so that we don't hit false positives looking for 1139 * insecure pages which are W+X. 1140 */ 1141 rcu_barrier(); 1142 mark_rodata_ro(); 1143 rodata_test(); 1144 } else 1145 pr_info("Kernel memory protection disabled.\n"); 1146 } 1147 #else 1148 static inline void mark_readonly(void) 1149 { 1150 pr_warn("This architecture does not have kernel memory protection.\n"); 1151 } 1152 #endif 1153 1154 void __weak free_initmem(void) 1155 { 1156 free_initmem_default(POISON_FREE_INITMEM); 1157 } 1158 1159 static int __ref kernel_init(void *unused) 1160 { 1161 int ret; 1162 1163 kernel_init_freeable(); 1164 /* need to finish all async __init code before freeing the memory */ 1165 async_synchronize_full(); 1166 ftrace_free_init_mem(); 1167 free_initmem(); 1168 mark_readonly(); 1169 1170 /* 1171 * Kernel mappings are now finalized - update the userspace page-table 1172 * to finalize PTI. 1173 */ 1174 pti_finalize(); 1175 1176 system_state = SYSTEM_RUNNING; 1177 numa_default_policy(); 1178 1179 rcu_end_inkernel_boot(); 1180 1181 if (ramdisk_execute_command) { 1182 ret = run_init_process(ramdisk_execute_command); 1183 if (!ret) 1184 return 0; 1185 pr_err("Failed to execute %s (error %d)\n", 1186 ramdisk_execute_command, ret); 1187 } 1188 1189 /* 1190 * We try each of these until one succeeds. 1191 * 1192 * The Bourne shell can be used instead of init if we are 1193 * trying to recover a really broken machine. 1194 */ 1195 if (execute_command) { 1196 ret = run_init_process(execute_command); 1197 if (!ret) 1198 return 0; 1199 panic("Requested init %s failed (error %d).", 1200 execute_command, ret); 1201 } 1202 if (!try_to_run_init_process("/sbin/init") || 1203 !try_to_run_init_process("/etc/init") || 1204 !try_to_run_init_process("/bin/init") || 1205 !try_to_run_init_process("/bin/sh")) 1206 return 0; 1207 1208 panic("No working init found. Try passing init= option to kernel. " 1209 "See Linux Documentation/admin-guide/init.rst for guidance."); 1210 } 1211 1212 void console_on_rootfs(void) 1213 { 1214 /* Open the /dev/console as stdin, this should never fail */ 1215 if (ksys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0) 1216 pr_err("Warning: unable to open an initial console.\n"); 1217 1218 /* create stdout/stderr */ 1219 (void) ksys_dup(0); 1220 (void) ksys_dup(0); 1221 } 1222 1223 static noinline void __init kernel_init_freeable(void) 1224 { 1225 /* 1226 * Wait until kthreadd is all set-up. 1227 */ 1228 wait_for_completion(&kthreadd_done); 1229 1230 /* Now the scheduler is fully set up and can do blocking allocations */ 1231 gfp_allowed_mask = __GFP_BITS_MASK; 1232 1233 /* 1234 * init can allocate pages on any node 1235 */ 1236 set_mems_allowed(node_states[N_MEMORY]); 1237 1238 cad_pid = task_pid(current); 1239 1240 smp_prepare_cpus(setup_max_cpus); 1241 1242 workqueue_init(); 1243 1244 init_mm_internals(); 1245 1246 do_pre_smp_initcalls(); 1247 lockup_detector_init(); 1248 1249 smp_init(); 1250 sched_init_smp(); 1251 1252 page_alloc_init_late(); 1253 /* Initialize page ext after all struct pages are initialized. */ 1254 page_ext_init(); 1255 1256 do_basic_setup(); 1257 1258 console_on_rootfs(); 1259 1260 /* 1261 * check if there is an early userspace init. If yes, let it do all 1262 * the work 1263 */ 1264 1265 if (!ramdisk_execute_command) 1266 ramdisk_execute_command = "/init"; 1267 1268 if (ksys_access((const char __user *) 1269 ramdisk_execute_command, 0) != 0) { 1270 ramdisk_execute_command = NULL; 1271 prepare_namespace(); 1272 } 1273 1274 /* 1275 * Ok, we have completed the initial bootup, and 1276 * we're essentially up and running. Get rid of the 1277 * initmem segments and start the user-mode stuff.. 1278 * 1279 * rootfs is available now, try loading the public keys 1280 * and default modules 1281 */ 1282 1283 integrity_load_keys(); 1284 } 1285