init/main.c

   1 /*
   2  *  linux/init/main.c
   3  *
   4  *  Copyright (C) 1991, 1992  Linus Torvalds
   5  *
   6  *  GK 2/5/95  -  Changed to support mounting root fs via NFS
   7  *  Added initrd & change_root: Werner Almesberger & Hans Lermen, Feb '96
   8  *  Moan early if gcc is old, avoiding bogus kernels - Paul Gortmaker, May '96
   9  *  Simplified starting of init:  Michael A. Griffith <grif@acm.org>
  10  */
  11
  12 #define DEBUG           /* Enable initcall_debug */
  13
  14 #include <linux/types.h>
  15 #include <linux/module.h>
  16 #include <linux/proc_fs.h>
  17 #include <linux/kernel.h>
  18 #include <linux/syscalls.h>
  19 #include <linux/stackprotector.h>
  20 #include <linux/string.h>
  21 #include <linux/ctype.h>
  22 #include <linux/delay.h>
  23 #include <linux/ioport.h>
  24 #include <linux/init.h>
  25 #include <linux/initrd.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/acpi.h>
  28 #include <linux/tty.h>
  29 #include <linux/percpu.h>
  30 #include <linux/kmod.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/kernel_stat.h>
  33 #include <linux/start_kernel.h>
  34 #include <linux/security.h>
  35 #include <linux/smp.h>
  36 #include <linux/profile.h>
  37 #include <linux/rcupdate.h>
  38 #include <linux/moduleparam.h>
  39 #include <linux/kallsyms.h>
  40 #include <linux/writeback.h>
  41 #include <linux/cpu.h>
  42 #include <linux/cpuset.h>
  43 #include <linux/cgroup.h>
  44 #include <linux/efi.h>
  45 #include <linux/tick.h>
  46 #include <linux/interrupt.h>
  47 #include <linux/taskstats_kern.h>
  48 #include <linux/delayacct.h>
  49 #include <linux/unistd.h>
  50 #include <linux/rmap.h>
  51 #include <linux/mempolicy.h>
  52 #include <linux/key.h>
  53 #include <linux/buffer_head.h>
  54 #include <linux/page_cgroup.h>
  55 #include <linux/debug_locks.h>
  56 #include <linux/debugobjects.h>
  57 #include <linux/lockdep.h>
  58 #include <linux/kmemleak.h>
  59 #include <linux/pid_namespace.h>
  60 #include <linux/device.h>
  61 #include <linux/kthread.h>
  62 #include <linux/sched.h>
  63 #include <linux/signal.h>
  64 #include <linux/idr.h>
  65 #include <linux/kgdb.h>
  66 #include <linux/ftrace.h>
  67 #include <linux/async.h>
  68 #include <linux/kmemcheck.h>
  69 #include <linux/sfi.h>
  70 #include <linux/shmem_fs.h>
  71 #include <linux/slab.h>
  72 #include <linux/perf_event.h>
  73 #include <linux/file.h>
  74 #include <linux/ptrace.h>
  75 #include <linux/blkdev.h>
  76 #include <linux/elevator.h>
  77 #include <linux/sched_clock.h>
  78 #include <linux/context_tracking.h>
  79 #include <linux/random.h>
  80
  81 #include <asm/io.h>
  82 #include <asm/bugs.h>
  83 #include <asm/setup.h>
  84 #include <asm/sections.h>
  85 #include <asm/cacheflush.h>
  86
  87 #ifdef CONFIG_X86_LOCAL_APIC
  88 #include <asm/smp.h>
  89 #endif
  90
  91 static int kernel_init(void *);
  92
  93 extern void init_IRQ(void);
  94 extern void fork_init(unsigned long);
  95 extern void mca_init(void);
  96 extern void sbus_init(void);
  97 extern void radix_tree_init(void);
  98 #ifndef CONFIG_DEBUG_RODATA
  99 static inline void mark_rodata_ro(void) { }
 100 #endif
 101
 102 #ifdef CONFIG_TC
 103 extern void tc_init(void);
 104 #endif
 105
 106 /*
 107  * Debug helper: via this flag we know that we are in 'early bootup code'
 108  * where only the boot processor is running with IRQ disabled.  This means
 109  * two things - IRQ must not be enabled before the flag is cleared and some
 110  * operations which are not allowed with IRQ disabled are allowed while the
 111  * flag is set.
 112  */
 113 bool early_boot_irqs_disabled __read_mostly;
 114
 115 enum system_states system_state __read_mostly;
 116 EXPORT_SYMBOL(system_state);
 117
 118 /*
 119  * Boot command-line arguments
 120  */
 121 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
 122 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
 123
 124 extern void time_init(void);
 125 /* Default late time init is NULL. archs can override this later. */
 126 void (*__initdata late_time_init)(void);
 127
 128 /* Untouched command line saved by arch-specific code. */
 129 char __initdata boot_command_line[COMMAND_LINE_SIZE];
 130 /* Untouched saved command line (eg. for /proc) */
 131 char *saved_command_line;
 132 /* Command line for parameter parsing */
 133 static char *static_command_line;
 134
 135 static char *execute_command;
 136 static char *ramdisk_execute_command;
 137
 138 /*
 139  * Used to generate warnings if static_key manipulation functions are used
 140  * before jump_label_init is called.
 141  */
 142 bool static_key_initialized __read_mostly = false;
 143 EXPORT_SYMBOL_GPL(static_key_initialized);
 144
 145 /*
 146  * If set, this is an indication to the drivers that reset the underlying
 147  * device before going ahead with the initialization otherwise driver might
 148  * rely on the BIOS and skip the reset operation.
 149  *
 150  * This is useful if kernel is booting in an unreliable environment.
 151  * For ex. kdump situaiton where previous kernel has crashed, BIOS has been
 152  * skipped and devices will be in unknown state.
 153  */
 154 unsigned int reset_devices;
 155 EXPORT_SYMBOL(reset_devices);
 156
 157 static int __init set_reset_devices(char *str)
 158 {
 159         reset_devices = 1;
 160         return 1;
 161 }
 162
 163 __setup("reset_devices", set_reset_devices);
 164
 165 static const char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 166 const char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 167 static const char *panic_later, *panic_param;
 168
 169 extern const struct obs_kernel_param __setup_start[], __setup_end[];
 170
 171 static int __init obsolete_checksetup(char *line)
 172 {
 173         const struct obs_kernel_param *p;
 174         int had_early_param = 0;
 175
 176         p = __setup_start;
 177         do {
 178                 int n = strlen(p->str);
 179                 if (parameqn(line, p->str, n)) {
 180                         if (p->early) {
 181                                 /* Already done in parse_early_param?
 182                                  * (Needs exact match on param part).
 183                                  * Keep iterating, as we can have early
 184                                  * params and __setups of same names 8( */
 185                                 if (line[n] == '\0' || line[n] == '=')
 186                                         had_early_param = 1;
 187                         } else if (!p->setup_func) {
 188                                 pr_warn("Parameter %s is obsolete, ignored\n",
 189                                         p->str);
 190                                 return 1;
 191                         } else if (p->setup_func(line + n))
 192                                 return 1;
 193                 }
 194                 p++;
 195         } while (p < __setup_end);
 196
 197         return had_early_param;
 198 }
 199
 200 /*
 201  * This should be approx 2 Bo*oMips to start (note initial shift), and will
 202  * still work even if initially too large, it will just take slightly longer
 203  */
 204 unsigned long loops_per_jiffy = (1<<12);
 205
 206 EXPORT_SYMBOL(loops_per_jiffy);
 207
 208 static int __init debug_kernel(char *str)
 209 {
 210         console_loglevel = 10;
 211         return 0;
 212 }
 213
 214 static int __init quiet_kernel(char *str)
 215 {
 216         console_loglevel = 4;
 217         return 0;
 218 }
 219
 220 early_param("debug", debug_kernel);
 221 early_param("quiet", quiet_kernel);
 222
 223 static int __init loglevel(char *str)
 224 {
 225         int newlevel;
 226
 227         /*
 228          * Only update loglevel value when a correct setting was passed,
 229          * to prevent blind crashes (when loglevel being set to 0) that
 230          * are quite hard to debug
 231          */
 232         if (get_option(&str, &newlevel)) {
 233                 console_loglevel = newlevel;
 234                 return 0;
 235         }
 236
 237         return -EINVAL;
 238 }
 239
 240 early_param("loglevel", loglevel);
 241
 242 /* Change NUL term back to "=", to make "param" the whole string. */
 243 static int __init repair_env_string(char *param, char *val, const char *unused)
 244 {
 245         if (val) {
 246                 /* param=val or param="val"? */
 247                 if (val == param+strlen(param)+1)
 248                         val[-1] = '=';
 249                 else if (val == param+strlen(param)+2) {
 250                         val[-2] = '=';
 251                         memmove(val-1, val, strlen(val)+1);
 252                         val--;
 253                 } else
 254                         BUG();
 255         }
 256         return 0;
 257 }
 258
 259 /*
 260  * Unknown boot options get handed to init, unless they look like
 261  * unused parameters (modprobe will find them in /proc/cmdline).
 262  */
 263 static int __init unknown_bootoption(char *param, char *val, const char *unused)
 264 {
 265         repair_env_string(param, val, unused);
 266
 267         /* Handle obsolete-style parameters */
 268         if (obsolete_checksetup(param))
 269                 return 0;
 270
 271         /* Unused module parameter. */
 272         if (strchr(param, '.') && (!val || strchr(param, '.') < val))
 273                 return 0;
 274
 275         if (panic_later)
 276                 return 0;
 277
 278         if (val) {
 279                 /* Environment option */
 280                 unsigned int i;
 281                 for (i = 0; envp_init[i]; i++) {
 282                         if (i == MAX_INIT_ENVS) {
 283                                 panic_later = "Too many boot env vars at `%s'";
 284                                 panic_param = param;
 285                         }
 286                         if (!strncmp(param, envp_init[i], val - param))
 287                                 break;
 288                 }
 289                 envp_init[i] = param;
 290         } else {
 291                 /* Command line option */
 292                 unsigned int i;
 293                 for (i = 0; argv_init[i]; i++) {
 294                         if (i == MAX_INIT_ARGS) {
 295                                 panic_later = "Too many boot init vars at `%s'";
 296                                 panic_param = param;
 297                         }
 298                 }
 299                 argv_init[i] = param;
 300         }
 301         return 0;
 302 }
 303
 304 static int __init init_setup(char *str)
 305 {
 306         unsigned int i;
 307
 308         execute_command = str;
 309         /*
 310          * In case LILO is going to boot us with default command line,
 311          * it prepends "auto" before the whole cmdline which makes
 312          * the shell think it should execute a script with such name.
 313          * So we ignore all arguments entered _before_ init=... [MJ]
 314          */
 315         for (i = 1; i < MAX_INIT_ARGS; i++)
 316                 argv_init[i] = NULL;
 317         return 1;
 318 }
 319 __setup("init=", init_setup);
 320
 321 static int __init rdinit_setup(char *str)
 322 {
 323         unsigned int i;
 324
 325         ramdisk_execute_command = str;
 326         /* See "auto" comment in init_setup */
 327         for (i = 1; i < MAX_INIT_ARGS; i++)
 328                 argv_init[i] = NULL;
 329         return 1;
 330 }
 331 __setup("rdinit=", rdinit_setup);
 332
 333 #ifndef CONFIG_SMP
 334 static const unsigned int setup_max_cpus = NR_CPUS;
 335 #ifdef CONFIG_X86_LOCAL_APIC
 336 static void __init smp_init(void)
 337 {
 338         APIC_init_uniprocessor();
 339 }
 340 #else
 341 #define smp_init()      do { } while (0)
 342 #endif
 343
 344 static inline void setup_nr_cpu_ids(void) { }
 345 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 346 #endif
 347
 348 /*
 349  * We need to store the untouched command line for future reference.
 350  * We also need to store the touched command line since the parameter
 351  * parsing is performed in place, and we should allow a component to
 352  * store reference of name/value for future reference.
 353  */
 354 static void __init setup_command_line(char *command_line)
 355 {
 356         saved_command_line = alloc_bootmem(strlen (boot_command_line)+1);
 357         static_command_line = alloc_bootmem(strlen (command_line)+1);
 358         strcpy (saved_command_line, boot_command_line);
 359         strcpy (static_command_line, command_line);
 360 }
 361
 362 /*
 363  * We need to finalize in a non-__init function or else race conditions
 364  * between the root thread and the init thread may cause start_kernel to
 365  * be reaped by free_initmem before the root thread has proceeded to
 366  * cpu_idle.
 367  *
 368  * gcc-3.4 accidentally inlines this function, so use noinline.
 369  */
 370
 371 static __initdata DECLARE_COMPLETION(kthreadd_done);
 372
 373 static noinline void __init_refok rest_init(void)
 374 {
 375         int pid;
 376
 377         rcu_scheduler_starting();
 378         /*
 379          * We need to spawn init first so that it obtains pid 1, however
 380          * the init task will end up wanting to create kthreads, which, if
 381          * we schedule it before we create kthreadd, will OOPS.
 382          */
 383         kernel_thread(kernel_init, NULL, CLONE_FS | CLONE_SIGHAND);
 384         numa_default_policy();
 385         pid = kernel_thread(kthreadd, NULL, CLONE_FS | CLONE_FILES);
 386         rcu_read_lock();
 387         kthreadd_task = find_task_by_pid_ns(pid, &init_pid_ns);
 388         rcu_read_unlock();
 389         complete(&kthreadd_done);
 390
 391         /*
 392          * The boot idle thread must execute schedule()
 393          * at least once to get things moving:
 394          */
 395         init_idle_bootup_task(current);
 396         schedule_preempt_disabled();
 397         /* Call into cpu_idle with preempt disabled */
 398         cpu_startup_entry(CPUHP_ONLINE);
 399 }
 400
 401 /* Check for early params. */
 402 static int __init do_early_param(char *param, char *val, const char *unused)
 403 {
 404         const struct obs_kernel_param *p;
 405
 406         for (p = __setup_start; p < __setup_end; p++) {
 407                 if ((p->early && parameq(param, p->str)) ||
 408                     (strcmp(param, "console") == 0 &&
 409                      strcmp(p->str, "earlycon") == 0)
 410                 ) {
 411                         if (p->setup_func(val) != 0)
 412                                 pr_warn("Malformed early option '%s'\n", param);
 413                 }
 414         }
 415         /* We accept everything at this stage. */
 416         return 0;
 417 }
 418
 419 void __init parse_early_options(char *cmdline)
 420 {
 421         parse_args("early options", cmdline, NULL, 0, 0, 0, do_early_param);
 422 }
 423
 424 /* Arch code calls this early on, or if not, just before other parsing. */
 425 void __init parse_early_param(void)
 426 {
 427         static __initdata int done = 0;
 428         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 429
 430         if (done)
 431                 return;
 432
 433         /* All fall through to do_early_param. */
 434         strlcpy(tmp_cmdline, boot_command_line, COMMAND_LINE_SIZE);
 435         parse_early_options(tmp_cmdline);
 436         done = 1;
 437 }
 438
 439 /*
 440  *      Activate the first processor.
 441  */
 442
 443 static void __init boot_cpu_init(void)
 444 {
 445         int cpu = smp_processor_id();
 446         /* Mark the boot cpu "present", "online" etc for SMP and UP case */
 447         set_cpu_online(cpu, true);
 448         set_cpu_active(cpu, true);
 449         set_cpu_present(cpu, true);
 450         set_cpu_possible(cpu, true);
 451 }
 452
 453 void __init __weak smp_setup_processor_id(void)
 454 {
 455 }
 456
 457 # if THREAD_SIZE >= PAGE_SIZE
 458 void __init __weak thread_info_cache_init(void)
 459 {
 460 }
 461 #endif
 462
 463 /*
 464  * Set up kernel memory allocators
 465  */
 466 static void __init mm_init(void)
 467 {
 468         /*
 469          * page_cgroup requires contiguous pages,
 470          * bigger than MAX_ORDER unless SPARSEMEM.
 471          */
 472         page_cgroup_init_flatmem();
 473         mem_init();
 474         kmem_cache_init();
 475         percpu_init_late();
 476         pgtable_cache_init();
 477         vmalloc_init();
 478 }
 479
 480 asmlinkage void __init start_kernel(void)
 481 {
 482         char * command_line;
 483         extern const struct kernel_param __start___param[], __stop___param[];
 484
 485         /*
 486          * Need to run as early as possible, to initialize the
 487          * lockdep hash:
 488          */
 489         lockdep_init();
 490         smp_setup_processor_id();
 491         debug_objects_early_init();
 492
 493         /*
 494          * Set up the the initial canary ASAP:
 495          */
 496         boot_init_stack_canary();
 497
 498         cgroup_init_early();
 499
 500         local_irq_disable();
 501         early_boot_irqs_disabled = true;
 502
 503 /*
 504  * Interrupts are still disabled. Do necessary setups, then
 505  * enable them
 506  */
 507         boot_cpu_init();
 508         page_address_init();
 509         pr_notice("%s", linux_banner);
 510         setup_arch(&command_line);
 511         mm_init_owner(&init_mm, &init_task);
 512         mm_init_cpumask(&init_mm);
 513         setup_command_line(command_line);
 514         setup_nr_cpu_ids();
 515         setup_per_cpu_areas();
 516         smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
 517
 518         build_all_zonelists(NULL, NULL);
 519         page_alloc_init();
 520
 521         pr_notice("Kernel command line: %s\n", boot_command_line);
 522         parse_early_param();
 523         parse_args("Booting kernel", static_command_line, __start___param,
 524                    __stop___param - __start___param,
 525                    -1, -1, &unknown_bootoption);
 526
 527         jump_label_init();
 528
 529         /*
 530          * These use large bootmem allocations and must precede
 531          * kmem_cache_init()
 532          */
 533         setup_log_buf(0);
 534         pidhash_init();
 535         vfs_caches_init_early();
 536         sort_main_extable();
 537         trap_init();
 538         mm_init();
 539
 540         /*
 541          * Set up the scheduler prior starting any interrupts (such as the
 542          * timer interrupt). Full topology setup happens at smp_init()
 543          * time - but meanwhile we still have a functioning scheduler.
 544          */
 545         sched_init();
 546         /*
 547          * Disable preemption - early bootup scheduling is extremely
 548          * fragile until we cpu_idle() for the first time.
 549          */
 550         preempt_disable();
 551         if (WARN(!irqs_disabled(), "Interrupts were enabled *very* early, fixing it\n"))
 552                 local_irq_disable();
 553         idr_init_cache();
 554         rcu_init();
 555         tick_nohz_init();
 556         context_tracking_init();
 557         radix_tree_init();
 558         /* init some links before init_ISA_irqs() */
 559         early_irq_init();
 560         init_IRQ();
 561         tick_init();
 562         init_timers();
 563         hrtimers_init();
 564         softirq_init();
 565         timekeeping_init();
 566         time_init();
 567         sched_clock_postinit();
 568         perf_event_init();
 569         profile_init();
 570         call_function_init();
 571         WARN(!irqs_disabled(), "Interrupts were enabled early\n");
 572         early_boot_irqs_disabled = false;
 573         local_irq_enable();
 574
 575         kmem_cache_init_late();
 576
 577         /*
 578          * HACK ALERT! This is early. We're enabling the console before
 579          * we've done PCI setups etc, and console_init() must be aware of
 580          * this. But we do want output early, in case something goes wrong.
 581          */
 582         console_init();
 583         if (panic_later)
 584                 panic(panic_later, panic_param);
 585
 586         lockdep_info();
 587
 588         /*
 589          * Need to run this when irqs are enabled, because it wants
 590          * to self-test [hard/soft]-irqs on/off lock inversion bugs
 591          * too:
 592          */
 593         locking_selftest();
 594
 595 #ifdef CONFIG_BLK_DEV_INITRD
 596         if (initrd_start && !initrd_below_start_ok &&
 597             page_to_pfn(virt_to_page((void *)initrd_start)) < min_low_pfn) {
 598                 pr_crit("initrd overwritten (0x%08lx < 0x%08lx) - disabling it.\n",
 599                     page_to_pfn(virt_to_page((void *)initrd_start)),
 600                     min_low_pfn);
 601                 initrd_start = 0;
 602         }
 603 #endif
 604         page_cgroup_init();
 605         debug_objects_mem_init();
 606         kmemleak_init();
 607         setup_per_cpu_pageset();
 608         numa_policy_init();
 609         if (late_time_init)
 610                 late_time_init();
 611         sched_clock_init();
 612         calibrate_delay();
 613         pidmap_init();
 614         anon_vma_init();
 615 #ifdef CONFIG_X86
 616         if (efi_enabled(EFI_RUNTIME_SERVICES))
 617                 efi_enter_virtual_mode();
 618 #endif
 619         thread_info_cache_init();
 620         cred_init();
 621         fork_init(totalram_pages);
 622         proc_caches_init();
 623         buffer_init();
 624         key_init();
 625         security_init();
 626         dbg_late_init();
 627         vfs_caches_init(totalram_pages);
 628         signals_init();
 629         /* rootfs populating might need page-writeback */
 630         page_writeback_init();
 631 #ifdef CONFIG_PROC_FS
 632         proc_root_init();
 633 #endif
 634         cgroup_init();
 635         cpuset_init();
 636         taskstats_init_early();
 637         delayacct_init();
 638
 639         check_bugs();
 640
 641         acpi_early_init(); /* before LAPIC and SMP init */
 642         sfi_init_late();
 643
 644         if (efi_enabled(EFI_RUNTIME_SERVICES)) {
 645                 efi_late_init();
 646                 efi_free_boot_services();
 647         }
 648
 649         ftrace_init();
 650
 651         /* Do the rest non-__init'ed, we're now alive */
 652         rest_init();
 653 }
 654
 655 /* Call all constructor functions linked into the kernel. */
 656 static void __init do_ctors(void)
 657 {
 658 #ifdef CONFIG_CONSTRUCTORS
 659         ctor_fn_t *fn = (ctor_fn_t *) __ctors_start;
 660
 661         for (; fn < (ctor_fn_t *) __ctors_end; fn++)
 662                 (*fn)();
 663 #endif
 664 }
 665
 666 bool initcall_debug;
 667 core_param(initcall_debug, initcall_debug, bool, 0644);
 668
 669 static int __init_or_module do_one_initcall_debug(initcall_t fn)
 670 {
 671         ktime_t calltime, delta, rettime;
 672         unsigned long long duration;
 673         int ret;
 674
 675         pr_debug("calling  %pF @ %i\n", fn, task_pid_nr(current));
 676         calltime = ktime_get();
 677         ret = fn();
 678         rettime = ktime_get();
 679         delta = ktime_sub(rettime, calltime);
 680         duration = (unsigned long long) ktime_to_ns(delta) >> 10;
 681         pr_debug("initcall %pF returned %d after %lld usecs\n",
 682                  fn, ret, duration);
 683
 684         return ret;
 685 }
 686
 687 int __init_or_module do_one_initcall(initcall_t fn)
 688 {
 689         int count = preempt_count();
 690         int ret;
 691         char msgbuf[64];
 692
 693         if (initcall_debug)
 694                 ret = do_one_initcall_debug(fn);
 695         else
 696                 ret = fn();
 697
 698         msgbuf[0] = 0;
 699
 700         if (preempt_count() != count) {
 701                 sprintf(msgbuf, "preemption imbalance ");
 702                 preempt_count_set(count);
 703         }
 704         if (irqs_disabled()) {
 705                 strlcat(msgbuf, "disabled interrupts ", sizeof(msgbuf));
 706                 local_irq_enable();
 707         }
 708         WARN(msgbuf[0], "initcall %pF returned with %s\n", fn, msgbuf);
 709
 710         return ret;
 711 }
 712
 713
 714 extern initcall_t __initcall_start[];
 715 extern initcall_t __initcall0_start[];
 716 extern initcall_t __initcall1_start[];
 717 extern initcall_t __initcall2_start[];
 718 extern initcall_t __initcall3_start[];
 719 extern initcall_t __initcall4_start[];
 720 extern initcall_t __initcall5_start[];
 721 extern initcall_t __initcall6_start[];
 722 extern initcall_t __initcall7_start[];
 723 extern initcall_t __initcall_end[];
 724
 725 static initcall_t *initcall_levels[] __initdata = {
 726         __initcall0_start,
 727         __initcall1_start,
 728         __initcall2_start,
 729         __initcall3_start,
 730         __initcall4_start,
 731         __initcall5_start,
 732         __initcall6_start,
 733         __initcall7_start,
 734         __initcall_end,
 735 };
 736
 737 /* Keep these in sync with initcalls in include/linux/init.h */
 738 static char *initcall_level_names[] __initdata = {
 739         "early",
 740         "core",
 741         "postcore",
 742         "arch",
 743         "subsys",
 744         "fs",
 745         "device",
 746         "late",
 747 };
 748
 749 static void __init do_initcall_level(int level)
 750 {
 751         extern const struct kernel_param __start___param[], __stop___param[];
 752         initcall_t *fn;
 753
 754         strcpy(static_command_line, saved_command_line);
 755         parse_args(initcall_level_names[level],
 756                    static_command_line, __start___param,
 757                    __stop___param - __start___param,
 758                    level, level,
 759                    &repair_env_string);
 760
 761         for (fn = initcall_levels[level]; fn < initcall_levels[level+1]; fn++)
 762                 do_one_initcall(*fn);
 763 }
 764
 765 static void __init do_initcalls(void)
 766 {
 767         int level;
 768
 769         for (level = 0; level < ARRAY_SIZE(initcall_levels) - 1; level++)
 770                 do_initcall_level(level);
 771 }
 772
 773 /*
 774  * Ok, the machine is now initialized. None of the devices
 775  * have been touched yet, but the CPU subsystem is up and
 776  * running, and memory and process management works.
 777  *
 778  * Now we can finally start doing some real work..
 779  */
 780 static void __init do_basic_setup(void)
 781 {
 782         cpuset_init_smp();
 783         usermodehelper_init();
 784         shmem_init();
 785         driver_init();
 786         init_irq_proc();
 787         do_ctors();
 788         usermodehelper_enable();
 789         do_initcalls();
 790         random_int_secret_init();
 791 }
 792
 793 static void __init do_pre_smp_initcalls(void)
 794 {
 795         initcall_t *fn;
 796
 797         for (fn = __initcall_start; fn < __initcall0_start; fn++)
 798                 do_one_initcall(*fn);
 799 }
 800
 801 /*
 802  * This function requests modules which should be loaded by default and is
 803  * called twice right after initrd is mounted and right before init is
 804  * exec'd.  If such modules are on either initrd or rootfs, they will be
 805  * loaded before control is passed to userland.
 806  */
 807 void __init load_default_modules(void)
 808 {
 809         load_default_elevator_module();
 810 }
 811
 812 static int run_init_process(const char *init_filename)
 813 {
 814         argv_init[0] = init_filename;
 815         return do_execve(init_filename,
 816                 (const char __user *const __user *)argv_init,
 817                 (const char __user *const __user *)envp_init);
 818 }
 819
 820 static int try_to_run_init_process(const char *init_filename)
 821 {
 822         int ret;
 823
 824         ret = run_init_process(init_filename);
 825
 826         if (ret && ret != -ENOENT) {
 827                 pr_err("Starting init: %s exists but couldn't execute it (error %d)\n",
 828                        init_filename, ret);
 829         }
 830
 831         return ret;
 832 }
 833
 834 static noinline void __init kernel_init_freeable(void);
 835
 836 static int __ref kernel_init(void *unused)
 837 {
 838         int ret;
 839
 840         kernel_init_freeable();
 841         /* need to finish all async __init code before freeing the memory */
 842         async_synchronize_full();
 843         free_initmem();
 844         mark_rodata_ro();
 845         system_state = SYSTEM_RUNNING;
 846         numa_default_policy();
 847
 848         flush_delayed_fput();
 849
 850         if (ramdisk_execute_command) {
 851                 ret = run_init_process(ramdisk_execute_command);
 852                 if (!ret)
 853                         return 0;
 854                 pr_err("Failed to execute %s (error %d)\n",
 855                        ramdisk_execute_command, ret);
 856         }
 857
 858         /*
 859          * We try each of these until one succeeds.
 860          *
 861          * The Bourne shell can be used instead of init if we are
 862          * trying to recover a really broken machine.
 863          */
 864         if (execute_command) {
 865                 ret = run_init_process(execute_command);
 866                 if (!ret)
 867                         return 0;
 868                 pr_err("Failed to execute %s (error %d).  Attempting defaults...\n",
 869                         execute_command, ret);
 870         }
 871         if (!try_to_run_init_process("/sbin/init") ||
 872             !try_to_run_init_process("/etc/init") ||
 873             !try_to_run_init_process("/bin/init") ||
 874             !try_to_run_init_process("/bin/sh"))
 875                 return 0;
 876
 877         panic("No working init found.  Try passing init= option to kernel. "
 878               "See Linux Documentation/init.txt for guidance.");
 879 }
 880
 881 static noinline void __init kernel_init_freeable(void)
 882 {
 883         /*
 884          * Wait until kthreadd is all set-up.
 885          */
 886         wait_for_completion(&kthreadd_done);
 887
 888         /* Now the scheduler is fully set up and can do blocking allocations */
 889         gfp_allowed_mask = __GFP_BITS_MASK;
 890
 891         /*
 892          * init can allocate pages on any node
 893          */
 894         set_mems_allowed(node_states[N_MEMORY]);
 895         /*
 896          * init can run on any cpu.
 897          */
 898         set_cpus_allowed_ptr(current, cpu_all_mask);
 899
 900         cad_pid = task_pid(current);
 901
 902         smp_prepare_cpus(setup_max_cpus);
 903
 904         do_pre_smp_initcalls();
 905         lockup_detector_init();
 906
 907         smp_init();
 908         sched_init_smp();
 909
 910         do_basic_setup();
 911
 912         /* Open the /dev/console on the rootfs, this should never fail */
 913         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 914                 pr_err("Warning: unable to open an initial console.\n");
 915
 916         (void) sys_dup(0);
 917         (void) sys_dup(0);
 918         /*
 919          * check if there is an early userspace init.  If yes, let it do all
 920          * the work
 921          */
 922
 923         if (!ramdisk_execute_command)
 924                 ramdisk_execute_command = "/init";
 925
 926         if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
 927                 ramdisk_execute_command = NULL;
 928                 prepare_namespace();
 929         }
 930
 931         /*
 932          * Ok, we have completed the initial bootup, and
 933          * we're essentially up and running. Get rid of the
 934          * initmem segments and start the user-mode stuff..
 935          */
 936
 937         /* rootfs is available now, try loading default modules */
 938         load_default_modules();
 939 }