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