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 #include <net/sock.h>
  51
  52 #include <asm/io.h>
  53 #include <asm/bugs.h>
  54 #include <asm/setup.h>
  55 #include <asm/sections.h>
  56
  57 /*
  58  * This is one of the first .c files built. Error out early
  59  * if we have compiler trouble..
  60  */
  61 #if __GNUC__ == 2 && __GNUC_MINOR__ == 96
  62 #ifdef CONFIG_FRAME_POINTER
  63 #error This compiler cannot compile correctly with frame pointers enabled
  64 #endif
  65 #endif
  66
  67 /*
  68  * Versions of gcc older than that listed below may actually compile
  69  * and link okay, but the end product can have subtle run time bugs.
  70  * To avoid associated bogus bug reports, we flatly refuse to compile
  71  * with a gcc that is known to be too old from the very beginning.
  72  */
  73 #if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 95)
  74 #error Sorry, your GCC is too old. It builds incorrect kernels.
  75 #endif
  76
  77 static int init(void *);
  78
  79 extern void init_IRQ(void);
  80 extern void fork_init(unsigned long);
  81 extern void mca_init(void);
  82 extern void sbus_init(void);
  83 extern void sysctl_init(void);
  84 extern void signals_init(void);
  85 extern void buffer_init(void);
  86 extern void pidhash_init(void);
  87 extern void pidmap_init(void);
  88 extern void prio_tree_init(void);
  89 extern void radix_tree_init(void);
  90 extern void free_initmem(void);
  91 extern void populate_rootfs(void);
  92 extern void driver_init(void);
  93 extern void prepare_namespace(void);
  94 #ifdef  CONFIG_ACPI
  95 extern void acpi_early_init(void);
  96 #else
  97 static inline void acpi_early_init(void) { }
  98 #endif
  99
 100 #ifdef CONFIG_TC
 101 extern void tc_init(void);
 102 #endif
 103
 104 enum system_states system_state;
 105 EXPORT_SYMBOL(system_state);
 106
 107 /*
 108  * Boot command-line arguments
 109  */
 110 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
 111 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
 112
 113 extern void time_init(void);
 114 /* Default late time init is NULL. archs can override this later. */
 115 void (*late_time_init)(void);
 116 extern void softirq_init(void);
 117
 118 /* Untouched command line (eg. for /proc) saved by arch-specific code. */
 119 char saved_command_line[COMMAND_LINE_SIZE];
 120
 121 static char *execute_command;
 122 static char *ramdisk_execute_command;
 123
 124 /* Setup configured maximum number of CPUs to activate */
 125 static unsigned int max_cpus = NR_CPUS;
 126
 127 /*
 128  * Setup routine for controlling SMP activation
 129  *
 130  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
 131  * activation entirely (the MPS table probe still happens, though).
 132  *
 133  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
 134  * greater than 0, limits the maximum number of CPUs activated in
 135  * SMP mode to <NUM>.
 136  */
 137 static int __init nosmp(char *str)
 138 {
 139         max_cpus = 0;
 140         return 1;
 141 }
 142
 143 __setup("nosmp", nosmp);
 144
 145 static int __init maxcpus(char *str)
 146 {
 147         get_option(&str, &max_cpus);
 148         return 1;
 149 }
 150
 151 __setup("maxcpus=", maxcpus);
 152
 153 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 154 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 155 static const char *panic_later, *panic_param;
 156
 157 extern struct obs_kernel_param __setup_start[], __setup_end[];
 158
 159 static int __init obsolete_checksetup(char *line)
 160 {
 161         struct obs_kernel_param *p;
 162
 163         p = __setup_start;
 164         do {
 165                 int n = strlen(p->str);
 166                 if (!strncmp(line, p->str, n)) {
 167                         if (p->early) {
 168                                 /* Already done in parse_early_param?  (Needs
 169                                  * exact match on param part) */
 170                                 if (line[n] == '\0' || line[n] == '=')
 171                                         return 1;
 172                         } else if (!p->setup_func) {
 173                                 printk(KERN_WARNING "Parameter %s is obsolete,"
 174                                        " ignored\n", p->str);
 175                                 return 1;
 176                         } else if (p->setup_func(line + n))
 177                                 return 1;
 178                 }
 179                 p++;
 180         } while (p < __setup_end);
 181         return 0;
 182 }
 183
 184 /*
 185  * This should be approx 2 Bo*oMips to start (note initial shift), and will
 186  * still work even if initially too large, it will just take slightly longer
 187  */
 188 unsigned long loops_per_jiffy = (1<<12);
 189
 190 EXPORT_SYMBOL(loops_per_jiffy);
 191
 192 static int __init debug_kernel(char *str)
 193 {
 194         if (*str)
 195                 return 0;
 196         console_loglevel = 10;
 197         return 1;
 198 }
 199
 200 static int __init quiet_kernel(char *str)
 201 {
 202         if (*str)
 203                 return 0;
 204         console_loglevel = 4;
 205         return 1;
 206 }
 207
 208 __setup("debug", debug_kernel);
 209 __setup("quiet", quiet_kernel);
 210
 211 static int __init loglevel(char *str)
 212 {
 213         get_option(&str, &console_loglevel);
 214         return 1;
 215 }
 216
 217 __setup("loglevel=", loglevel);
 218
 219 /*
 220  * Unknown boot options get handed to init, unless they look like
 221  * failed parameters
 222  */
 223 static int __init unknown_bootoption(char *param, char *val)
 224 {
 225         /* Change NUL term back to "=", to make "param" the whole string. */
 226         if (val) {
 227                 /* param=val or param="val"? */
 228                 if (val == param+strlen(param)+1)
 229                         val[-1] = '=';
 230                 else if (val == param+strlen(param)+2) {
 231                         val[-2] = '=';
 232                         memmove(val-1, val, strlen(val)+1);
 233                         val--;
 234                 } else
 235                         BUG();
 236         }
 237
 238         /* Handle obsolete-style parameters */
 239         if (obsolete_checksetup(param))
 240                 return 0;
 241
 242         /*
 243          * Preemptive maintenance for "why didn't my mispelled command
 244          * line work?"
 245          */
 246         if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
 247                 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
 248                 return 0;
 249         }
 250
 251         if (panic_later)
 252                 return 0;
 253
 254         if (val) {
 255                 /* Environment option */
 256                 unsigned int i;
 257                 for (i = 0; envp_init[i]; i++) {
 258                         if (i == MAX_INIT_ENVS) {
 259                                 panic_later = "Too many boot env vars at `%s'";
 260                                 panic_param = param;
 261                         }
 262                         if (!strncmp(param, envp_init[i], val - param))
 263                                 break;
 264                 }
 265                 envp_init[i] = param;
 266         } else {
 267                 /* Command line option */
 268                 unsigned int i;
 269                 for (i = 0; argv_init[i]; i++) {
 270                         if (i == MAX_INIT_ARGS) {
 271                                 panic_later = "Too many boot init vars at `%s'";
 272                                 panic_param = param;
 273                         }
 274                 }
 275                 argv_init[i] = param;
 276         }
 277         return 0;
 278 }
 279
 280 static int __init init_setup(char *str)
 281 {
 282         unsigned int i;
 283
 284         execute_command = str;
 285         /*
 286          * In case LILO is going to boot us with default command line,
 287          * it prepends "auto" before the whole cmdline which makes
 288          * the shell think it should execute a script with such name.
 289          * So we ignore all arguments entered _before_ init=... [MJ]
 290          */
 291         for (i = 1; i < MAX_INIT_ARGS; i++)
 292                 argv_init[i] = NULL;
 293         return 1;
 294 }
 295 __setup("init=", init_setup);
 296
 297 static int __init rdinit_setup(char *str)
 298 {
 299         unsigned int i;
 300
 301         ramdisk_execute_command = str;
 302         /* See "auto" comment in init_setup */
 303         for (i = 1; i < MAX_INIT_ARGS; i++)
 304                 argv_init[i] = NULL;
 305         return 1;
 306 }
 307 __setup("rdinit=", rdinit_setup);
 308
 309 extern void setup_arch(char **);
 310
 311 #ifndef CONFIG_SMP
 312
 313 #define smp_init()      do { } while (0)
 314
 315 static inline void setup_per_cpu_areas(void) { }
 316 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 317
 318 #else
 319
 320 #ifdef __GENERIC_PER_CPU
 321 unsigned long __per_cpu_offset[NR_CPUS];
 322
 323 EXPORT_SYMBOL(__per_cpu_offset);
 324
 325 static void __init setup_per_cpu_areas(void)
 326 {
 327         unsigned long size, i;
 328         char *ptr;
 329
 330         /* Copy section for each CPU (we discard the original) */
 331         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 332 #ifdef CONFIG_MODULES
 333         if (size < PERCPU_ENOUGH_ROOM)
 334                 size = PERCPU_ENOUGH_ROOM;
 335 #endif
 336
 337         ptr = alloc_bootmem(size * NR_CPUS);
 338
 339         for (i = 0; i < NR_CPUS; i++, ptr += size) {
 340                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 341                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 342         }
 343 }
 344 #endif /* !__GENERIC_PER_CPU */
 345
 346 /* Called by boot processor to activate the rest. */
 347 static void __init smp_init(void)
 348 {
 349         unsigned int i;
 350
 351         /* FIXME: This should be done in userspace --RR */
 352         for_each_present_cpu(i) {
 353                 if (num_online_cpus() >= max_cpus)
 354                         break;
 355                 if (!cpu_online(i))
 356                         cpu_up(i);
 357         }
 358
 359         /* Any cleanup work */
 360         printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
 361         smp_cpus_done(max_cpus);
 362 #if 0
 363         /* Get other processors into their bootup holding patterns. */
 364
 365         smp_commence();
 366 #endif
 367 }
 368
 369 #endif
 370
 371 /*
 372  * We need to finalize in a non-__init function or else race conditions
 373  * between the root thread and the init thread may cause start_kernel to
 374  * be reaped by free_initmem before the root thread has proceeded to
 375  * cpu_idle.
 376  *
 377  * gcc-3.4 accidentally inlines this function, so use noinline.
 378  */
 379
 380 static void noinline rest_init(void)
 381         __releases(kernel_lock)
 382 {
 383         kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
 384         numa_default_policy();
 385         unlock_kernel();
 386         preempt_enable_no_resched();
 387
 388         /*
 389          * The boot idle thread must execute schedule()
 390          * at least one to get things moving:
 391          */
 392         schedule();
 393
 394         cpu_idle();
 395 }
 396
 397 /* Check for early params. */
 398 static int __init do_early_param(char *param, char *val)
 399 {
 400         struct obs_kernel_param *p;
 401
 402         for (p = __setup_start; p < __setup_end; p++) {
 403                 if (p->early && strcmp(param, p->str) == 0) {
 404                         if (p->setup_func(val) != 0)
 405                                 printk(KERN_WARNING
 406                                        "Malformed early option '%s'\n", param);
 407                 }
 408         }
 409         /* We accept everything at this stage. */
 410         return 0;
 411 }
 412
 413 /* Arch code calls this early on, or if not, just before other parsing. */
 414 void __init parse_early_param(void)
 415 {
 416         static __initdata int done = 0;
 417         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 418
 419         if (done)
 420                 return;
 421
 422         /* All fall through to do_early_param. */
 423         strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
 424         parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
 425         done = 1;
 426 }
 427
 428 /*
 429  *      Activate the first processor.
 430  */
 431
 432 asmlinkage void __init start_kernel(void)
 433 {
 434         char * command_line;
 435         extern struct kernel_param __start___param[], __stop___param[];
 436 /*
 437  * Interrupts are still disabled. Do necessary setups, then
 438  * enable them
 439  */
 440         lock_kernel();
 441         page_address_init();
 442         printk(KERN_NOTICE);
 443         printk(linux_banner);
 444         setup_arch(&command_line);
 445         setup_per_cpu_areas();
 446
 447         /*
 448          * Mark the boot cpu "online" so that it can call console drivers in
 449          * printk() and can access its per-cpu storage.
 450          */
 451         smp_prepare_boot_cpu();
 452
 453         /*
 454          * Set up the scheduler prior starting any interrupts (such as the
 455          * timer interrupt). Full topology setup happens at smp_init()
 456          * time - but meanwhile we still have a functioning scheduler.
 457          */
 458         sched_init();
 459         /*
 460          * Disable preemption - early bootup scheduling is extremely
 461          * fragile until we cpu_idle() for the first time.
 462          */
 463         preempt_disable();
 464         build_all_zonelists();
 465         page_alloc_init();
 466         printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
 467         parse_early_param();
 468         parse_args("Booting kernel", command_line, __start___param,
 469                    __stop___param - __start___param,
 470                    &unknown_bootoption);
 471         sort_main_extable();
 472         trap_init();
 473         rcu_init();
 474         init_IRQ();
 475         pidhash_init();
 476         init_timers();
 477         softirq_init();
 478         time_init();
 479
 480         /*
 481          * HACK ALERT! This is early. We're enabling the console before
 482          * we've done PCI setups etc, and console_init() must be aware of
 483          * this. But we do want output early, in case something goes wrong.
 484          */
 485         console_init();
 486         if (panic_later)
 487                 panic(panic_later, panic_param);
 488         profile_init();
 489         local_irq_enable();
 490 #ifdef CONFIG_BLK_DEV_INITRD
 491         if (initrd_start && !initrd_below_start_ok &&
 492                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 493                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 494                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 495                 initrd_start = 0;
 496         }
 497 #endif
 498         vfs_caches_init_early();
 499         mem_init();
 500         kmem_cache_init();
 501         setup_per_cpu_pageset();
 502         numa_policy_init();
 503         if (late_time_init)
 504                 late_time_init();
 505         calibrate_delay();
 506         pidmap_init();
 507         pgtable_cache_init();
 508         prio_tree_init();
 509         anon_vma_init();
 510 #ifdef CONFIG_X86
 511         if (efi_enabled)
 512                 efi_enter_virtual_mode();
 513 #endif
 514         fork_init(num_physpages);
 515         proc_caches_init();
 516         buffer_init();
 517         unnamed_dev_init();
 518         key_init();
 519         security_init();
 520         vfs_caches_init(num_physpages);
 521         radix_tree_init();
 522         signals_init();
 523         /* rootfs populating might need page-writeback */
 524         page_writeback_init();
 525 #ifdef CONFIG_PROC_FS
 526         proc_root_init();
 527 #endif
 528         cpuset_init();
 529
 530         check_bugs();
 531
 532         acpi_early_init(); /* before LAPIC and SMP init */
 533
 534         /* Do the rest non-__init'ed, we're now alive */
 535         rest_init();
 536 }
 537
 538 static int __initdata initcall_debug;
 539
 540 static int __init initcall_debug_setup(char *str)
 541 {
 542         initcall_debug = 1;
 543         return 1;
 544 }
 545 __setup("initcall_debug", initcall_debug_setup);
 546
 547 struct task_struct *child_reaper = &init_task;
 548
 549 extern initcall_t __initcall_start[], __initcall_end[];
 550
 551 static void __init do_initcalls(void)
 552 {
 553         initcall_t *call;
 554         int count = preempt_count();
 555
 556         for (call = __initcall_start; call < __initcall_end; call++) {
 557                 char *msg;
 558
 559                 if (initcall_debug) {
 560                         printk(KERN_DEBUG "Calling initcall 0x%p", *call);
 561                         print_fn_descriptor_symbol(": %s()", (unsigned long) *call);
 562                         printk("\n");
 563                 }
 564
 565                 (*call)();
 566
 567                 msg = NULL;
 568                 if (preempt_count() != count) {
 569                         msg = "preemption imbalance";
 570                         preempt_count() = count;
 571                 }
 572                 if (irqs_disabled()) {
 573                         msg = "disabled interrupts";
 574                         local_irq_enable();
 575                 }
 576                 if (msg) {
 577                         printk(KERN_WARNING "error in initcall at 0x%p: "
 578                                 "returned with %s\n", *call, msg);
 579                 }
 580         }
 581
 582         /* Make sure there is no pending stuff from the initcall sequence */
 583         flush_scheduled_work();
 584 }
 585
 586 /*
 587  * Ok, the machine is now initialized. None of the devices
 588  * have been touched yet, but the CPU subsystem is up and
 589  * running, and memory and process management works.
 590  *
 591  * Now we can finally start doing some real work..
 592  */
 593 static void __init do_basic_setup(void)
 594 {
 595         /* drivers will send hotplug events */
 596         init_workqueues();
 597         usermodehelper_init();
 598         driver_init();
 599
 600 #ifdef CONFIG_SYSCTL
 601         sysctl_init();
 602 #endif
 603
 604         /* Networking initialization needs a process context */
 605         sock_init();
 606
 607         do_initcalls();
 608 }
 609
 610 static void do_pre_smp_initcalls(void)
 611 {
 612         extern int spawn_ksoftirqd(void);
 613 #ifdef CONFIG_SMP
 614         extern int migration_init(void);
 615
 616         migration_init();
 617 #endif
 618         spawn_ksoftirqd();
 619         spawn_softlockup_task();
 620 }
 621
 622 static void run_init_process(char *init_filename)
 623 {
 624         argv_init[0] = init_filename;
 625         execve(init_filename, argv_init, envp_init);
 626 }
 627
 628 static inline void fixup_cpu_present_map(void)
 629 {
 630 #ifdef CONFIG_SMP
 631         int i;
 632
 633         /*
 634          * If arch is not hotplug ready and did not populate
 635          * cpu_present_map, just make cpu_present_map same as cpu_possible_map
 636          * for other cpu bringup code to function as normal. e.g smp_init() etc.
 637          */
 638         if (cpus_empty(cpu_present_map)) {
 639                 for_each_cpu(i) {
 640                         cpu_set(i, cpu_present_map);
 641                 }
 642         }
 643 #endif
 644 }
 645
 646 static int init(void * unused)
 647 {
 648         lock_kernel();
 649         /*
 650          * init can run on any cpu.
 651          */
 652         set_cpus_allowed(current, CPU_MASK_ALL);
 653         /*
 654          * Tell the world that we're going to be the grim
 655          * reaper of innocent orphaned children.
 656          *
 657          * We don't want people to have to make incorrect
 658          * assumptions about where in the task array this
 659          * can be found.
 660          */
 661         child_reaper = current;
 662
 663         /* Sets up cpus_possible() */
 664         smp_prepare_cpus(max_cpus);
 665
 666         do_pre_smp_initcalls();
 667
 668         fixup_cpu_present_map();
 669         smp_init();
 670         sched_init_smp();
 671
 672         cpuset_init_smp();
 673
 674         /*
 675          * Do this before initcalls, because some drivers want to access
 676          * firmware files.
 677          */
 678         populate_rootfs();
 679
 680         do_basic_setup();
 681
 682         /*
 683          * check if there is an early userspace init.  If yes, let it do all
 684          * the work
 685          */
 686
 687         if (!ramdisk_execute_command)
 688                 ramdisk_execute_command = "/init";
 689
 690         if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
 691                 ramdisk_execute_command = NULL;
 692                 prepare_namespace();
 693         }
 694
 695         /*
 696          * Ok, we have completed the initial bootup, and
 697          * we're essentially up and running. Get rid of the
 698          * initmem segments and start the user-mode stuff..
 699          */
 700         free_initmem();
 701         unlock_kernel();
 702         system_state = SYSTEM_RUNNING;
 703         numa_default_policy();
 704
 705         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 706                 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
 707
 708         (void) sys_dup(0);
 709         (void) sys_dup(0);
 710
 711         if (ramdisk_execute_command) {
 712                 run_init_process(ramdisk_execute_command);
 713                 printk(KERN_WARNING "Failed to execute %s\n",
 714                                 ramdisk_execute_command);
 715         }
 716
 717         /*
 718          * We try each of these until one succeeds.
 719          *
 720          * The Bourne shell can be used instead of init if we are
 721          * trying to recover a really broken machine.
 722          */
 723         if (execute_command) {
 724                 run_init_process(execute_command);
 725                 printk(KERN_WARNING "Failed to execute %s.  Attempting "
 726                                         "defaults...\n", execute_command);
 727         }
 728         run_init_process("/sbin/init");
 729         run_init_process("/etc/init");
 730         run_init_process("/bin/init");
 731         run_init_process("/bin/sh");
 732
 733         panic("No init found.  Try passing init= option to kernel.");
 734 }