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 __KERNEL_SYSCALLS__
  13
  14 #include <linux/config.h>
  15 #include <linux/types.h>
  16 #include <linux/module.h>
  17 #include <linux/proc_fs.h>
  18 #include <linux/devfs_fs_kernel.h>
  19 #include <linux/kernel.h>
  20 #include <linux/syscalls.h>
  21 #include <linux/string.h>
  22 #include <linux/ctype.h>
  23 #include <linux/delay.h>
  24 #include <linux/utsname.h>
  25 #include <linux/ioport.h>
  26 #include <linux/init.h>
  27 #include <linux/smp_lock.h>
  28 #include <linux/initrd.h>
  29 #include <linux/hdreg.h>
  30 #include <linux/bootmem.h>
  31 #include <linux/tty.h>
  32 #include <linux/gfp.h>
  33 #include <linux/percpu.h>
  34 #include <linux/kmod.h>
  35 #include <linux/kernel_stat.h>
  36 #include <linux/security.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/efi.h>
  46 #include <linux/unistd.h>
  47 #include <linux/rmap.h>
  48 #include <linux/mempolicy.h>
  49 #include <linux/key.h>
  50
  51 #include <asm/io.h>
  52 #include <asm/bugs.h>
  53 #include <asm/setup.h>
  54 #include <asm/sections.h>
  55
  56 /*
  57  * This is one of the first .c files built. Error out early
  58  * if we have compiler trouble..
  59  */
  60 #if __GNUC__ == 2 && __GNUC_MINOR__ == 96
  61 #ifdef CONFIG_FRAME_POINTER
  62 #error This compiler cannot compile correctly with frame pointers enabled
  63 #endif
  64 #endif
  65
  66 #ifdef CONFIG_X86_LOCAL_APIC
  67 #include <asm/smp.h>
  68 #endif
  69
  70 /*
  71  * Versions of gcc older than that listed below may actually compile
  72  * and link okay, but the end product can have subtle run time bugs.
  73  * To avoid associated bogus bug reports, we flatly refuse to compile
  74  * with a gcc that is known to be too old from the very beginning.
  75  */
  76 #if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 95)
  77 #error Sorry, your GCC is too old. It builds incorrect kernels.
  78 #endif
  79
  80 static int init(void *);
  81
  82 extern void init_IRQ(void);
  83 extern void sock_init(void);
  84 extern void fork_init(unsigned long);
  85 extern void mca_init(void);
  86 extern void sbus_init(void);
  87 extern void sysctl_init(void);
  88 extern void signals_init(void);
  89 extern void buffer_init(void);
  90 extern void pidhash_init(void);
  91 extern void pidmap_init(void);
  92 extern void prio_tree_init(void);
  93 extern void radix_tree_init(void);
  94 extern void free_initmem(void);
  95 extern void populate_rootfs(void);
  96 extern void driver_init(void);
  97 extern void prepare_namespace(void);
  98 #ifdef  CONFIG_ACPI
  99 extern void acpi_early_init(void);
 100 #else
 101 static inline void acpi_early_init(void) { }
 102 #endif
 103
 104 #ifdef CONFIG_TC
 105 extern void tc_init(void);
 106 #endif
 107
 108 enum system_states system_state;
 109 EXPORT_SYMBOL(system_state);
 110
 111 /*
 112  * Boot command-line arguments
 113  */
 114 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
 115 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
 116
 117 extern void time_init(void);
 118 /* Default late time init is NULL. archs can override this later. */
 119 void (*late_time_init)(void);
 120 extern void softirq_init(void);
 121
 122 /* Untouched command line (eg. for /proc) saved by arch-specific code. */
 123 char saved_command_line[COMMAND_LINE_SIZE];
 124
 125 static char *execute_command;
 126
 127 /* Setup configured maximum number of CPUs to activate */
 128 static unsigned int max_cpus = NR_CPUS;
 129
 130 /*
 131  * Setup routine for controlling SMP activation
 132  *
 133  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
 134  * activation entirely (the MPS table probe still happens, though).
 135  *
 136  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
 137  * greater than 0, limits the maximum number of CPUs activated in
 138  * SMP mode to <NUM>.
 139  */
 140 static int __init nosmp(char *str)
 141 {
 142         max_cpus = 0;
 143         return 1;
 144 }
 145
 146 __setup("nosmp", nosmp);
 147
 148 static int __init maxcpus(char *str)
 149 {
 150         get_option(&str, &max_cpus);
 151         return 1;
 152 }
 153
 154 __setup("maxcpus=", maxcpus);
 155
 156 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 157 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 158 static const char *panic_later, *panic_param;
 159
 160 extern struct obs_kernel_param __setup_start[], __setup_end[];
 161
 162 static int __init obsolete_checksetup(char *line)
 163 {
 164         struct obs_kernel_param *p;
 165
 166         p = __setup_start;
 167         do {
 168                 int n = strlen(p->str);
 169                 if (!strncmp(line, p->str, n)) {
 170                         if (p->early) {
 171                                 /* Already done in parse_early_param?  (Needs
 172                                  * exact match on param part) */
 173                                 if (line[n] == '\0' || line[n] == '=')
 174                                         return 1;
 175                         } else if (!p->setup_func) {
 176                                 printk(KERN_WARNING "Parameter %s is obsolete,"
 177                                        " ignored\n", p->str);
 178                                 return 1;
 179                         } else if (p->setup_func(line + n))
 180                                 return 1;
 181                 }
 182                 p++;
 183         } while (p < __setup_end);
 184         return 0;
 185 }
 186
 187 /*
 188  * This should be approx 2 Bo*oMips to start (note initial shift), and will
 189  * still work even if initially too large, it will just take slightly longer
 190  */
 191 unsigned long loops_per_jiffy = (1<<12);
 192
 193 EXPORT_SYMBOL(loops_per_jiffy);
 194
 195 static int __init debug_kernel(char *str)
 196 {
 197         if (*str)
 198                 return 0;
 199         console_loglevel = 10;
 200         return 1;
 201 }
 202
 203 static int __init quiet_kernel(char *str)
 204 {
 205         if (*str)
 206                 return 0;
 207         console_loglevel = 4;
 208         return 1;
 209 }
 210
 211 __setup("debug", debug_kernel);
 212 __setup("quiet", quiet_kernel);
 213
 214 static int __init loglevel(char *str)
 215 {
 216         get_option(&str, &console_loglevel);
 217         return 1;
 218 }
 219
 220 __setup("loglevel=", loglevel);
 221
 222 /*
 223  * Unknown boot options get handed to init, unless they look like
 224  * failed parameters
 225  */
 226 static int __init unknown_bootoption(char *param, char *val)
 227 {
 228         /* Change NUL term back to "=", to make "param" the whole string. */
 229         if (val) {
 230                 /* param=val or param="val"? */
 231                 if (val == param+strlen(param)+1)
 232                         val[-1] = '=';
 233                 else if (val == param+strlen(param)+2) {
 234                         val[-2] = '=';
 235                         memmove(val-1, val, strlen(val)+1);
 236                         val--;
 237                 } else
 238                         BUG();
 239         }
 240
 241         /* Handle obsolete-style parameters */
 242         if (obsolete_checksetup(param))
 243                 return 0;
 244
 245         /*
 246          * Preemptive maintenance for "why didn't my mispelled command
 247          * line work?"
 248          */
 249         if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
 250                 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
 251                 return 0;
 252         }
 253
 254         if (panic_later)
 255                 return 0;
 256
 257         if (val) {
 258                 /* Environment option */
 259                 unsigned int i;
 260                 for (i = 0; envp_init[i]; i++) {
 261                         if (i == MAX_INIT_ENVS) {
 262                                 panic_later = "Too many boot env vars at `%s'";
 263                                 panic_param = param;
 264                         }
 265                         if (!strncmp(param, envp_init[i], val - param))
 266                                 break;
 267                 }
 268                 envp_init[i] = param;
 269         } else {
 270                 /* Command line option */
 271                 unsigned int i;
 272                 for (i = 0; argv_init[i]; i++) {
 273                         if (i == MAX_INIT_ARGS) {
 274                                 panic_later = "Too many boot init vars at `%s'";
 275                                 panic_param = param;
 276                         }
 277                 }
 278                 argv_init[i] = param;
 279         }
 280         return 0;
 281 }
 282
 283 static int __init init_setup(char *str)
 284 {
 285         unsigned int i;
 286
 287         execute_command = str;
 288         /*
 289          * In case LILO is going to boot us with default command line,
 290          * it prepends "auto" before the whole cmdline which makes
 291          * the shell think it should execute a script with such name.
 292          * So we ignore all arguments entered _before_ init=... [MJ]
 293          */
 294         for (i = 1; i < MAX_INIT_ARGS; i++)
 295                 argv_init[i] = NULL;
 296         return 1;
 297 }
 298 __setup("init=", init_setup);
 299
 300 extern void setup_arch(char **);
 301
 302 #ifndef CONFIG_SMP
 303
 304 #ifdef CONFIG_X86_LOCAL_APIC
 305 static void __init smp_init(void)
 306 {
 307         APIC_init_uniprocessor();
 308 }
 309 #else
 310 #define smp_init()      do { } while (0)
 311 #endif
 312
 313 static inline void setup_per_cpu_areas(void) { }
 314 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 315
 316 #else
 317
 318 #ifdef __GENERIC_PER_CPU
 319 unsigned long __per_cpu_offset[NR_CPUS];
 320
 321 EXPORT_SYMBOL(__per_cpu_offset);
 322
 323 static void __init setup_per_cpu_areas(void)
 324 {
 325         unsigned long size, i;
 326         char *ptr;
 327
 328         /* Copy section for each CPU (we discard the original) */
 329         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 330 #ifdef CONFIG_MODULES
 331         if (size < PERCPU_ENOUGH_ROOM)
 332                 size = PERCPU_ENOUGH_ROOM;
 333 #endif
 334
 335         ptr = alloc_bootmem(size * NR_CPUS);
 336
 337         for (i = 0; i < NR_CPUS; i++, ptr += size) {
 338                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 339                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 340         }
 341 }
 342 #endif /* !__GENERIC_PER_CPU */
 343
 344 /* Called by boot processor to activate the rest. */
 345 static void __init smp_init(void)
 346 {
 347         unsigned int i;
 348
 349         /* FIXME: This should be done in userspace --RR */
 350         for_each_present_cpu(i) {
 351                 if (num_online_cpus() >= max_cpus)
 352                         break;
 353                 if (!cpu_online(i))
 354                         cpu_up(i);
 355         }
 356
 357         /* Any cleanup work */
 358         printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
 359         smp_cpus_done(max_cpus);
 360 #if 0
 361         /* Get other processors into their bootup holding patterns. */
 362
 363         smp_commence();
 364 #endif
 365 }
 366
 367 #endif
 368
 369 /*
 370  * We need to finalize in a non-__init function or else race conditions
 371  * between the root thread and the init thread may cause start_kernel to
 372  * be reaped by free_initmem before the root thread has proceeded to
 373  * cpu_idle.
 374  *
 375  * gcc-3.4 accidentally inlines this function, so use noinline.
 376  */
 377
 378 static void noinline rest_init(void)
 379         __releases(kernel_lock)
 380 {
 381         kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
 382         numa_default_policy();
 383         unlock_kernel();
 384         preempt_enable_no_resched();
 385
 386         /*
 387          * The boot idle thread must execute schedule()
 388          * at least one to get things moving:
 389          */
 390         schedule();
 391
 392         cpu_idle();
 393 }
 394
 395 /* Check for early params. */
 396 static int __init do_early_param(char *param, char *val)
 397 {
 398         struct obs_kernel_param *p;
 399
 400         for (p = __setup_start; p < __setup_end; p++) {
 401                 if (p->early && strcmp(param, p->str) == 0) {
 402                         if (p->setup_func(val) != 0)
 403                                 printk(KERN_WARNING
 404                                        "Malformed early option '%s'\n", param);
 405                 }
 406         }
 407         /* We accept everything at this stage. */
 408         return 0;
 409 }
 410
 411 /* Arch code calls this early on, or if not, just before other parsing. */
 412 void __init parse_early_param(void)
 413 {
 414         static __initdata int done = 0;
 415         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 416
 417         if (done)
 418                 return;
 419
 420         /* All fall through to do_early_param. */
 421         strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
 422         parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
 423         done = 1;
 424 }
 425
 426 /*
 427  *      Activate the first processor.
 428  */
 429
 430 asmlinkage void __init start_kernel(void)
 431 {
 432         char * command_line;
 433         extern struct kernel_param __start___param[], __stop___param[];
 434 /*
 435  * Interrupts are still disabled. Do necessary setups, then
 436  * enable them
 437  */
 438         lock_kernel();
 439         page_address_init();
 440         printk(KERN_NOTICE);
 441         printk(linux_banner);
 442         setup_arch(&command_line);
 443         setup_per_cpu_areas();
 444
 445         /*
 446          * Mark the boot cpu "online" so that it can call console drivers in
 447          * printk() and can access its per-cpu storage.
 448          */
 449         smp_prepare_boot_cpu();
 450
 451         /*
 452          * Set up the scheduler prior starting any interrupts (such as the
 453          * timer interrupt). Full topology setup happens at smp_init()
 454          * time - but meanwhile we still have a functioning scheduler.
 455          */
 456         sched_init();
 457         /*
 458          * Disable preemption - early bootup scheduling is extremely
 459          * fragile until we cpu_idle() for the first time.
 460          */
 461         preempt_disable();
 462         build_all_zonelists();
 463         page_alloc_init();
 464         printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
 465         parse_early_param();
 466         parse_args("Booting kernel", command_line, __start___param,
 467                    __stop___param - __start___param,
 468                    &unknown_bootoption);
 469         sort_main_extable();
 470         trap_init();
 471         rcu_init();
 472         init_IRQ();
 473         pidhash_init();
 474         init_timers();
 475         softirq_init();
 476         time_init();
 477
 478         /*
 479          * HACK ALERT! This is early. We're enabling the console before
 480          * we've done PCI setups etc, and console_init() must be aware of
 481          * this. But we do want output early, in case something goes wrong.
 482          */
 483         console_init();
 484         if (panic_later)
 485                 panic(panic_later, panic_param);
 486         profile_init();
 487         local_irq_enable();
 488 #ifdef CONFIG_BLK_DEV_INITRD
 489         if (initrd_start && !initrd_below_start_ok &&
 490                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 491                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 492                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 493                 initrd_start = 0;
 494         }
 495 #endif
 496         vfs_caches_init_early();
 497         mem_init();
 498         kmem_cache_init();
 499         setup_per_cpu_pageset();
 500         numa_policy_init();
 501         if (late_time_init)
 502                 late_time_init();
 503         calibrate_delay();
 504         pidmap_init();
 505         pgtable_cache_init();
 506         prio_tree_init();
 507         anon_vma_init();
 508 #ifdef CONFIG_X86
 509         if (efi_enabled)
 510                 efi_enter_virtual_mode();
 511 #endif
 512         fork_init(num_physpages);
 513         proc_caches_init();
 514         buffer_init();
 515         unnamed_dev_init();
 516         key_init();
 517         security_init();
 518         vfs_caches_init(num_physpages);
 519         radix_tree_init();
 520         signals_init();
 521         /* rootfs populating might need page-writeback */
 522         page_writeback_init();
 523 #ifdef CONFIG_PROC_FS
 524         proc_root_init();
 525 #endif
 526         cpuset_init();
 527
 528         check_bugs();
 529
 530         acpi_early_init(); /* before LAPIC and SMP init */
 531
 532         /* Do the rest non-__init'ed, we're now alive */
 533         rest_init();
 534 }
 535
 536 static int __initdata initcall_debug;
 537
 538 static int __init initcall_debug_setup(char *str)
 539 {
 540         initcall_debug = 1;
 541         return 1;
 542 }
 543 __setup("initcall_debug", initcall_debug_setup);
 544
 545 struct task_struct *child_reaper = &init_task;
 546
 547 extern initcall_t __initcall_start[], __initcall_end[];
 548
 549 static void __init do_initcalls(void)
 550 {
 551         initcall_t *call;
 552         int count = preempt_count();
 553
 554         for (call = __initcall_start; call < __initcall_end; call++) {
 555                 char *msg;
 556
 557                 if (initcall_debug) {
 558                         printk(KERN_DEBUG "Calling initcall 0x%p", *call);
 559                         print_fn_descriptor_symbol(": %s()", (unsigned long) *call);
 560                         printk("\n");
 561                 }
 562
 563                 (*call)();
 564
 565                 msg = NULL;
 566                 if (preempt_count() != count) {
 567                         msg = "preemption imbalance";
 568                         preempt_count() = count;
 569                 }
 570                 if (irqs_disabled()) {
 571                         msg = "disabled interrupts";
 572                         local_irq_enable();
 573                 }
 574                 if (msg) {
 575                         printk(KERN_WARNING "error in initcall at 0x%p: "
 576                                 "returned with %s\n", *call, msg);
 577                 }
 578         }
 579
 580         /* Make sure there is no pending stuff from the initcall sequence */
 581         flush_scheduled_work();
 582 }
 583
 584 /*
 585  * Ok, the machine is now initialized. None of the devices
 586  * have been touched yet, but the CPU subsystem is up and
 587  * running, and memory and process management works.
 588  *
 589  * Now we can finally start doing some real work..
 590  */
 591 static void __init do_basic_setup(void)
 592 {
 593         /* drivers will send hotplug events */
 594         init_workqueues();
 595         usermodehelper_init();
 596         driver_init();
 597
 598 #ifdef CONFIG_SYSCTL
 599         sysctl_init();
 600 #endif
 601
 602         /* Networking initialization needs a process context */
 603         sock_init();
 604
 605         do_initcalls();
 606 }
 607
 608 static void do_pre_smp_initcalls(void)
 609 {
 610         extern int spawn_ksoftirqd(void);
 611 #ifdef CONFIG_SMP
 612         extern int migration_init(void);
 613
 614         migration_init();
 615 #endif
 616         spawn_ksoftirqd();
 617 }
 618
 619 static void run_init_process(char *init_filename)
 620 {
 621         argv_init[0] = init_filename;
 622         execve(init_filename, argv_init, envp_init);
 623 }
 624
 625 static inline void fixup_cpu_present_map(void)
 626 {
 627 #ifdef CONFIG_SMP
 628         int i;
 629
 630         /*
 631          * If arch is not hotplug ready and did not populate
 632          * cpu_present_map, just make cpu_present_map same as cpu_possible_map
 633          * for other cpu bringup code to function as normal. e.g smp_init() etc.
 634          */
 635         if (cpus_empty(cpu_present_map)) {
 636                 for_each_cpu(i) {
 637                         cpu_set(i, cpu_present_map);
 638                 }
 639         }
 640 #endif
 641 }
 642
 643 static int init(void * unused)
 644 {
 645         lock_kernel();
 646         /*
 647          * init can run on any cpu.
 648          */
 649         set_cpus_allowed(current, CPU_MASK_ALL);
 650         /*
 651          * Tell the world that we're going to be the grim
 652          * reaper of innocent orphaned children.
 653          *
 654          * We don't want people to have to make incorrect
 655          * assumptions about where in the task array this
 656          * can be found.
 657          */
 658         child_reaper = current;
 659
 660         /* Sets up cpus_possible() */
 661         smp_prepare_cpus(max_cpus);
 662
 663         do_pre_smp_initcalls();
 664
 665         fixup_cpu_present_map();
 666         smp_init();
 667         sched_init_smp();
 668
 669         cpuset_init_smp();
 670
 671         /*
 672          * Do this before initcalls, because some drivers want to access
 673          * firmware files.
 674          */
 675         populate_rootfs();
 676
 677         do_basic_setup();
 678
 679         /*
 680          * check if there is an early userspace init.  If yes, let it do all
 681          * the work
 682          */
 683         if (sys_access((const char __user *) "/init", 0) == 0)
 684                 execute_command = "/init";
 685         else
 686                 prepare_namespace();
 687
 688         /*
 689          * Ok, we have completed the initial bootup, and
 690          * we're essentially up and running. Get rid of the
 691          * initmem segments and start the user-mode stuff..
 692          */
 693         free_initmem();
 694         unlock_kernel();
 695         system_state = SYSTEM_RUNNING;
 696         numa_default_policy();
 697
 698         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 699                 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
 700
 701         (void) sys_dup(0);
 702         (void) sys_dup(0);
 703
 704         /*
 705          * We try each of these until one succeeds.
 706          *
 707          * The Bourne shell can be used instead of init if we are
 708          * trying to recover a really broken machine.
 709          */
 710
 711         if (execute_command)
 712                 run_init_process(execute_command);
 713
 714         run_init_process("/sbin/init");
 715         run_init_process("/etc/init");
 716         run_init_process("/bin/init");
 717         run_init_process("/bin/sh");
 718
 719         panic("No init found.  Try passing init= option to kernel.");
 720 }