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 fork_init(unsigned long);
  84 extern void mca_init(void);
  85 extern void sbus_init(void);
  86 extern void sysctl_init(void);
  87 extern void signals_init(void);
  88 extern void buffer_init(void);
  89 extern void pidhash_init(void);
  90 extern void pidmap_init(void);
  91 extern void prio_tree_init(void);
  92 extern void radix_tree_init(void);
  93 extern void free_initmem(void);
  94 extern void populate_rootfs(void);
  95 extern void driver_init(void);
  96 extern void prepare_namespace(void);
  97 #ifdef  CONFIG_ACPI
  98 extern void acpi_early_init(void);
  99 #else
 100 static inline void acpi_early_init(void) { }
 101 #endif
 102
 103 #ifdef CONFIG_TC
 104 extern void tc_init(void);
 105 #endif
 106
 107 enum system_states system_state;
 108 EXPORT_SYMBOL(system_state);
 109
 110 /*
 111  * Boot command-line arguments
 112  */
 113 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
 114 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
 115
 116 extern void time_init(void);
 117 /* Default late time init is NULL. archs can override this later. */
 118 void (*late_time_init)(void);
 119 extern void softirq_init(void);
 120
 121 /* Untouched command line (eg. for /proc) saved by arch-specific code. */
 122 char saved_command_line[COMMAND_LINE_SIZE];
 123
 124 static char *execute_command;
 125 static char *ramdisk_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 static int __init rdinit_setup(char *str)
 301 {
 302         unsigned int i;
 303
 304         ramdisk_execute_command = str;
 305         /* See "auto" comment in init_setup */
 306         for (i = 1; i < MAX_INIT_ARGS; i++)
 307                 argv_init[i] = NULL;
 308         return 1;
 309 }
 310 __setup("rdinit=", rdinit_setup);
 311
 312 extern void setup_arch(char **);
 313
 314 #ifndef CONFIG_SMP
 315
 316 #ifdef CONFIG_X86_LOCAL_APIC
 317 static void __init smp_init(void)
 318 {
 319         APIC_init_uniprocessor();
 320 }
 321 #else
 322 #define smp_init()      do { } while (0)
 323 #endif
 324
 325 static inline void setup_per_cpu_areas(void) { }
 326 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 327
 328 #else
 329
 330 #ifdef __GENERIC_PER_CPU
 331 unsigned long __per_cpu_offset[NR_CPUS];
 332
 333 EXPORT_SYMBOL(__per_cpu_offset);
 334
 335 static void __init setup_per_cpu_areas(void)
 336 {
 337         unsigned long size, i;
 338         char *ptr;
 339
 340         /* Copy section for each CPU (we discard the original) */
 341         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 342 #ifdef CONFIG_MODULES
 343         if (size < PERCPU_ENOUGH_ROOM)
 344                 size = PERCPU_ENOUGH_ROOM;
 345 #endif
 346
 347         ptr = alloc_bootmem(size * NR_CPUS);
 348
 349         for (i = 0; i < NR_CPUS; i++, ptr += size) {
 350                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 351                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 352         }
 353 }
 354 #endif /* !__GENERIC_PER_CPU */
 355
 356 /* Called by boot processor to activate the rest. */
 357 static void __init smp_init(void)
 358 {
 359         unsigned int i;
 360
 361         /* FIXME: This should be done in userspace --RR */
 362         for_each_present_cpu(i) {
 363                 if (num_online_cpus() >= max_cpus)
 364                         break;
 365                 if (!cpu_online(i))
 366                         cpu_up(i);
 367         }
 368
 369         /* Any cleanup work */
 370         printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
 371         smp_cpus_done(max_cpus);
 372 #if 0
 373         /* Get other processors into their bootup holding patterns. */
 374
 375         smp_commence();
 376 #endif
 377 }
 378
 379 #endif
 380
 381 /*
 382  * We need to finalize in a non-__init function or else race conditions
 383  * between the root thread and the init thread may cause start_kernel to
 384  * be reaped by free_initmem before the root thread has proceeded to
 385  * cpu_idle.
 386  *
 387  * gcc-3.4 accidentally inlines this function, so use noinline.
 388  */
 389
 390 static void noinline rest_init(void)
 391         __releases(kernel_lock)
 392 {
 393         kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
 394         numa_default_policy();
 395         unlock_kernel();
 396
 397         /*
 398          * The boot idle thread must execute schedule()
 399          * at least one to get things moving:
 400          */
 401         preempt_enable_no_resched();
 402         schedule();
 403         preempt_disable();
 404
 405         /* Call into cpu_idle with preempt disabled */
 406         cpu_idle();
 407 }
 408
 409 /* Check for early params. */
 410 static int __init do_early_param(char *param, char *val)
 411 {
 412         struct obs_kernel_param *p;
 413
 414         for (p = __setup_start; p < __setup_end; p++) {
 415                 if (p->early && strcmp(param, p->str) == 0) {
 416                         if (p->setup_func(val) != 0)
 417                                 printk(KERN_WARNING
 418                                        "Malformed early option '%s'\n", param);
 419                 }
 420         }
 421         /* We accept everything at this stage. */
 422         return 0;
 423 }
 424
 425 /* Arch code calls this early on, or if not, just before other parsing. */
 426 void __init parse_early_param(void)
 427 {
 428         static __initdata int done = 0;
 429         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 430
 431         if (done)
 432                 return;
 433
 434         /* All fall through to do_early_param. */
 435         strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
 436         parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
 437         done = 1;
 438 }
 439
 440 /*
 441  *      Activate the first processor.
 442  */
 443
 444 asmlinkage void __init start_kernel(void)
 445 {
 446         char * command_line;
 447         extern struct kernel_param __start___param[], __stop___param[];
 448 /*
 449  * Interrupts are still disabled. Do necessary setups, then
 450  * enable them
 451  */
 452         lock_kernel();
 453         page_address_init();
 454         printk(KERN_NOTICE);
 455         printk(linux_banner);
 456         setup_arch(&command_line);
 457         setup_per_cpu_areas();
 458
 459         /*
 460          * Mark the boot cpu "online" so that it can call console drivers in
 461          * printk() and can access its per-cpu storage.
 462          */
 463         smp_prepare_boot_cpu();
 464
 465         /*
 466          * Set up the scheduler prior starting any interrupts (such as the
 467          * timer interrupt). Full topology setup happens at smp_init()
 468          * time - but meanwhile we still have a functioning scheduler.
 469          */
 470         sched_init();
 471         /*
 472          * Disable preemption - early bootup scheduling is extremely
 473          * fragile until we cpu_idle() for the first time.
 474          */
 475         preempt_disable();
 476         build_all_zonelists();
 477         page_alloc_init();
 478         printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
 479         parse_early_param();
 480         parse_args("Booting kernel", command_line, __start___param,
 481                    __stop___param - __start___param,
 482                    &unknown_bootoption);
 483         sort_main_extable();
 484         trap_init();
 485         rcu_init();
 486         init_IRQ();
 487         pidhash_init();
 488         init_timers();
 489         softirq_init();
 490         time_init();
 491
 492         /*
 493          * HACK ALERT! This is early. We're enabling the console before
 494          * we've done PCI setups etc, and console_init() must be aware of
 495          * this. But we do want output early, in case something goes wrong.
 496          */
 497         console_init();
 498         if (panic_later)
 499                 panic(panic_later, panic_param);
 500         profile_init();
 501         local_irq_enable();
 502 #ifdef CONFIG_BLK_DEV_INITRD
 503         if (initrd_start && !initrd_below_start_ok &&
 504                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 505                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 506                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 507                 initrd_start = 0;
 508         }
 509 #endif
 510         vfs_caches_init_early();
 511         mem_init();
 512         kmem_cache_init();
 513         setup_per_cpu_pageset();
 514         numa_policy_init();
 515         if (late_time_init)
 516                 late_time_init();
 517         calibrate_delay();
 518         pidmap_init();
 519         pgtable_cache_init();
 520         prio_tree_init();
 521         anon_vma_init();
 522 #ifdef CONFIG_X86
 523         if (efi_enabled)
 524                 efi_enter_virtual_mode();
 525 #endif
 526         fork_init(num_physpages);
 527         proc_caches_init();
 528         buffer_init();
 529         unnamed_dev_init();
 530         key_init();
 531         security_init();
 532         vfs_caches_init(num_physpages);
 533         radix_tree_init();
 534         signals_init();
 535         /* rootfs populating might need page-writeback */
 536         page_writeback_init();
 537 #ifdef CONFIG_PROC_FS
 538         proc_root_init();
 539 #endif
 540         cpuset_init();
 541
 542         check_bugs();
 543
 544         acpi_early_init(); /* before LAPIC and SMP init */
 545
 546         /* Do the rest non-__init'ed, we're now alive */
 547         rest_init();
 548 }
 549
 550 static int __initdata initcall_debug;
 551
 552 static int __init initcall_debug_setup(char *str)
 553 {
 554         initcall_debug = 1;
 555         return 1;
 556 }
 557 __setup("initcall_debug", initcall_debug_setup);
 558
 559 struct task_struct *child_reaper = &init_task;
 560
 561 extern initcall_t __initcall_start[], __initcall_end[];
 562
 563 static void __init do_initcalls(void)
 564 {
 565         initcall_t *call;
 566         int count = preempt_count();
 567
 568         for (call = __initcall_start; call < __initcall_end; call++) {
 569                 char *msg;
 570
 571                 if (initcall_debug) {
 572                         printk(KERN_DEBUG "Calling initcall 0x%p", *call);
 573                         print_fn_descriptor_symbol(": %s()", (unsigned long) *call);
 574                         printk("\n");
 575                 }
 576
 577                 (*call)();
 578
 579                 msg = NULL;
 580                 if (preempt_count() != count) {
 581                         msg = "preemption imbalance";
 582                         preempt_count() = count;
 583                 }
 584                 if (irqs_disabled()) {
 585                         msg = "disabled interrupts";
 586                         local_irq_enable();
 587                 }
 588                 if (msg) {
 589                         printk(KERN_WARNING "error in initcall at 0x%p: "
 590                                 "returned with %s\n", *call, msg);
 591                 }
 592         }
 593
 594         /* Make sure there is no pending stuff from the initcall sequence */
 595         flush_scheduled_work();
 596 }
 597
 598 /*
 599  * Ok, the machine is now initialized. None of the devices
 600  * have been touched yet, but the CPU subsystem is up and
 601  * running, and memory and process management works.
 602  *
 603  * Now we can finally start doing some real work..
 604  */
 605 static void __init do_basic_setup(void)
 606 {
 607         /* drivers will send hotplug events */
 608         init_workqueues();
 609         usermodehelper_init();
 610         driver_init();
 611
 612 #ifdef CONFIG_SYSCTL
 613         sysctl_init();
 614 #endif
 615
 616         do_initcalls();
 617 }
 618
 619 static void do_pre_smp_initcalls(void)
 620 {
 621         extern int spawn_ksoftirqd(void);
 622 #ifdef CONFIG_SMP
 623         extern int migration_init(void);
 624
 625         migration_init();
 626 #endif
 627         spawn_ksoftirqd();
 628         spawn_softlockup_task();
 629 }
 630
 631 static void run_init_process(char *init_filename)
 632 {
 633         argv_init[0] = init_filename;
 634         execve(init_filename, argv_init, envp_init);
 635 }
 636
 637 static inline void fixup_cpu_present_map(void)
 638 {
 639 #ifdef CONFIG_SMP
 640         int i;
 641
 642         /*
 643          * If arch is not hotplug ready and did not populate
 644          * cpu_present_map, just make cpu_present_map same as cpu_possible_map
 645          * for other cpu bringup code to function as normal. e.g smp_init() etc.
 646          */
 647         if (cpus_empty(cpu_present_map)) {
 648                 for_each_cpu(i) {
 649                         cpu_set(i, cpu_present_map);
 650                 }
 651         }
 652 #endif
 653 }
 654
 655 static int init(void * unused)
 656 {
 657         lock_kernel();
 658         /*
 659          * init can run on any cpu.
 660          */
 661         set_cpus_allowed(current, CPU_MASK_ALL);
 662         /*
 663          * Tell the world that we're going to be the grim
 664          * reaper of innocent orphaned children.
 665          *
 666          * We don't want people to have to make incorrect
 667          * assumptions about where in the task array this
 668          * can be found.
 669          */
 670         child_reaper = current;
 671
 672         /* Sets up cpus_possible() */
 673         smp_prepare_cpus(max_cpus);
 674
 675         do_pre_smp_initcalls();
 676
 677         fixup_cpu_present_map();
 678         smp_init();
 679         sched_init_smp();
 680
 681         cpuset_init_smp();
 682
 683         /*
 684          * Do this before initcalls, because some drivers want to access
 685          * firmware files.
 686          */
 687         populate_rootfs();
 688
 689         do_basic_setup();
 690
 691         /*
 692          * check if there is an early userspace init.  If yes, let it do all
 693          * the work
 694          */
 695
 696         if (!ramdisk_execute_command)
 697                 ramdisk_execute_command = "/init";
 698
 699         if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
 700                 ramdisk_execute_command = NULL;
 701                 prepare_namespace();
 702         }
 703
 704         /*
 705          * Ok, we have completed the initial bootup, and
 706          * we're essentially up and running. Get rid of the
 707          * initmem segments and start the user-mode stuff..
 708          */
 709         free_initmem();
 710         unlock_kernel();
 711         system_state = SYSTEM_RUNNING;
 712         numa_default_policy();
 713
 714         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 715                 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
 716
 717         (void) sys_dup(0);
 718         (void) sys_dup(0);
 719
 720         if (ramdisk_execute_command) {
 721                 run_init_process(ramdisk_execute_command);
 722                 printk(KERN_WARNING "Failed to execute %s\n",
 723                                 ramdisk_execute_command);
 724         }
 725
 726         /*
 727          * We try each of these until one succeeds.
 728          *
 729          * The Bourne shell can be used instead of init if we are
 730          * trying to recover a really broken machine.
 731          */
 732         if (execute_command) {
 733                 run_init_process(execute_command);
 734                 printk(KERN_WARNING "Failed to execute %s.  Attempting "
 735                                         "defaults...\n", execute_command);
 736         }
 737         run_init_process("/sbin/init");
 738         run_init_process("/etc/init");
 739         run_init_process("/bin/init");
 740         run_init_process("/bin/sh");
 741
 742         panic("No init found.  Try passing init= option to kernel.");
 743 }