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/kernel.h>
  19 #include <linux/syscalls.h>
  20 #include <linux/string.h>
  21 #include <linux/ctype.h>
  22 #include <linux/delay.h>
  23 #include <linux/utsname.h>
  24 #include <linux/ioport.h>
  25 #include <linux/init.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/initrd.h>
  28 #include <linux/hdreg.h>
  29 #include <linux/bootmem.h>
  30 #include <linux/tty.h>
  31 #include <linux/gfp.h>
  32 #include <linux/percpu.h>
  33 #include <linux/kmod.h>
  34 #include <linux/kernel_stat.h>
  35 #include <linux/security.h>
  36 #include <linux/workqueue.h>
  37 #include <linux/profile.h>
  38 #include <linux/rcupdate.h>
  39 #include <linux/moduleparam.h>
  40 #include <linux/kallsyms.h>
  41 #include <linux/writeback.h>
  42 #include <linux/cpu.h>
  43 #include <linux/cpuset.h>
  44 #include <linux/efi.h>
  45 #include <linux/unistd.h>
  46 #include <linux/rmap.h>
  47 #include <linux/mempolicy.h>
  48 #include <linux/key.h>
  49 #include <linux/unwind.h>
  50 #include <linux/buffer_head.h>
  51
  52 #include <asm/io.h>
  53 #include <asm/bugs.h>
  54 #include <asm/setup.h>
  55 #include <asm/sections.h>
  56 #include <asm/cacheflush.h>
  57
  58 #ifdef CONFIG_X86_LOCAL_APIC
  59 #include <asm/smp.h>
  60 #endif
  61
  62 /*
  63  * This is one of the first .c files built. Error out early if we have compiler
  64  * trouble.
  65  *
  66  * Versions of gcc older than that listed below may actually compile and link
  67  * okay, but the end product can have subtle run time bugs.  To avoid associated
  68  * bogus bug reports, we flatly refuse to compile with a gcc that is known to be
  69  * too old from the very beginning.
  70  */
  71 #if (__GNUC__ < 3) || (__GNUC__ == 3 && __GNUC_MINOR__ < 2)
  72 #error Sorry, your GCC is too old. It builds incorrect kernels.
  73 #endif
  74
  75 static int init(void *);
  76
  77 extern void init_IRQ(void);
  78 extern void fork_init(unsigned long);
  79 extern void mca_init(void);
  80 extern void sbus_init(void);
  81 extern void sysctl_init(void);
  82 extern void signals_init(void);
  83 extern void pidhash_init(void);
  84 extern void pidmap_init(void);
  85 extern void prio_tree_init(void);
  86 extern void radix_tree_init(void);
  87 extern void free_initmem(void);
  88 extern void populate_rootfs(void);
  89 extern void driver_init(void);
  90 extern void prepare_namespace(void);
  91 #ifdef  CONFIG_ACPI
  92 extern void acpi_early_init(void);
  93 #else
  94 static inline void acpi_early_init(void) { }
  95 #endif
  96 #ifndef CONFIG_DEBUG_RODATA
  97 static inline void mark_rodata_ro(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 #ifndef CONFIG_SMP
 310
 311 #ifdef CONFIG_X86_LOCAL_APIC
 312 static void __init smp_init(void)
 313 {
 314         APIC_init_uniprocessor();
 315 }
 316 #else
 317 #define smp_init()      do { } while (0)
 318 #endif
 319
 320 static inline void setup_per_cpu_areas(void) { }
 321 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 322
 323 #else
 324
 325 #ifdef __GENERIC_PER_CPU
 326 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
 327
 328 EXPORT_SYMBOL(__per_cpu_offset);
 329
 330 static void __init setup_per_cpu_areas(void)
 331 {
 332         unsigned long size, i;
 333         char *ptr;
 334         unsigned long nr_possible_cpus = num_possible_cpus();
 335
 336         /* Copy section for each CPU (we discard the original) */
 337         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 338 #ifdef CONFIG_MODULES
 339         if (size < PERCPU_ENOUGH_ROOM)
 340                 size = PERCPU_ENOUGH_ROOM;
 341 #endif
 342         ptr = alloc_bootmem(size * nr_possible_cpus);
 343
 344         for_each_possible_cpu(i) {
 345                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 346                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 347                 ptr += size;
 348         }
 349 }
 350 #endif /* !__GENERIC_PER_CPU */
 351
 352 /* Called by boot processor to activate the rest. */
 353 static void __init smp_init(void)
 354 {
 355         unsigned int i;
 356
 357         /* FIXME: This should be done in userspace --RR */
 358         for_each_present_cpu(i) {
 359                 if (num_online_cpus() >= max_cpus)
 360                         break;
 361                 if (!cpu_online(i))
 362                         cpu_up(i);
 363         }
 364
 365         /* Any cleanup work */
 366         printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
 367         smp_cpus_done(max_cpus);
 368 #if 0
 369         /* Get other processors into their bootup holding patterns. */
 370
 371         smp_commence();
 372 #endif
 373 }
 374
 375 #endif
 376
 377 /*
 378  * We need to finalize in a non-__init function or else race conditions
 379  * between the root thread and the init thread may cause start_kernel to
 380  * be reaped by free_initmem before the root thread has proceeded to
 381  * cpu_idle.
 382  *
 383  * gcc-3.4 accidentally inlines this function, so use noinline.
 384  */
 385
 386 static void noinline rest_init(void)
 387         __releases(kernel_lock)
 388 {
 389         kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
 390         numa_default_policy();
 391         unlock_kernel();
 392
 393         /*
 394          * The boot idle thread must execute schedule()
 395          * at least one to get things moving:
 396          */
 397         preempt_enable_no_resched();
 398         schedule();
 399         preempt_disable();
 400
 401         /* Call into cpu_idle with preempt disabled */
 402         cpu_idle();
 403 }
 404
 405 /* Check for early params. */
 406 static int __init do_early_param(char *param, char *val)
 407 {
 408         struct obs_kernel_param *p;
 409
 410         for (p = __setup_start; p < __setup_end; p++) {
 411                 if (p->early && strcmp(param, p->str) == 0) {
 412                         if (p->setup_func(val) != 0)
 413                                 printk(KERN_WARNING
 414                                        "Malformed early option '%s'\n", param);
 415                 }
 416         }
 417         /* We accept everything at this stage. */
 418         return 0;
 419 }
 420
 421 /* Arch code calls this early on, or if not, just before other parsing. */
 422 void __init parse_early_param(void)
 423 {
 424         static __initdata int done = 0;
 425         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 426
 427         if (done)
 428                 return;
 429
 430         /* All fall through to do_early_param. */
 431         strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
 432         parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
 433         done = 1;
 434 }
 435
 436 /*
 437  *      Activate the first processor.
 438  */
 439
 440 static void __init boot_cpu_init(void)
 441 {
 442         int cpu = smp_processor_id();
 443         /* Mark the boot cpu "present", "online" etc for SMP and UP case */
 444         cpu_set(cpu, cpu_online_map);
 445         cpu_set(cpu, cpu_present_map);
 446         cpu_set(cpu, cpu_possible_map);
 447 }
 448
 449 void __init __attribute__((weak)) smp_setup_processor_id(void)
 450 {
 451 }
 452
 453 asmlinkage void __init start_kernel(void)
 454 {
 455         char * command_line;
 456         extern struct kernel_param __start___param[], __stop___param[];
 457
 458         smp_setup_processor_id();
 459
 460 /*
 461  * Interrupts are still disabled. Do necessary setups, then
 462  * enable them
 463  */
 464         lock_kernel();
 465         boot_cpu_init();
 466         page_address_init();
 467         printk(KERN_NOTICE);
 468         printk(linux_banner);
 469         setup_arch(&command_line);
 470         setup_per_cpu_areas();
 471         smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
 472
 473         /*
 474          * Set up the scheduler prior starting any interrupts (such as the
 475          * timer interrupt). Full topology setup happens at smp_init()
 476          * time - but meanwhile we still have a functioning scheduler.
 477          */
 478         sched_init();
 479         /*
 480          * Disable preemption - early bootup scheduling is extremely
 481          * fragile until we cpu_idle() for the first time.
 482          */
 483         preempt_disable();
 484         build_all_zonelists();
 485         page_alloc_init();
 486         printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
 487         parse_early_param();
 488         parse_args("Booting kernel", command_line, __start___param,
 489                    __stop___param - __start___param,
 490                    &unknown_bootoption);
 491         sort_main_extable();
 492         unwind_init();
 493         trap_init();
 494         rcu_init();
 495         init_IRQ();
 496         pidhash_init();
 497         init_timers();
 498         hrtimers_init();
 499         softirq_init();
 500         time_init();
 501         timekeeping_init();
 502
 503         /*
 504          * HACK ALERT! This is early. We're enabling the console before
 505          * we've done PCI setups etc, and console_init() must be aware of
 506          * this. But we do want output early, in case something goes wrong.
 507          */
 508         console_init();
 509         if (panic_later)
 510                 panic(panic_later, panic_param);
 511         profile_init();
 512         local_irq_enable();
 513 #ifdef CONFIG_BLK_DEV_INITRD
 514         if (initrd_start && !initrd_below_start_ok &&
 515                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 516                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 517                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 518                 initrd_start = 0;
 519         }
 520 #endif
 521         vfs_caches_init_early();
 522         cpuset_init_early();
 523         mem_init();
 524         kmem_cache_init();
 525         setup_per_cpu_pageset();
 526         numa_policy_init();
 527         if (late_time_init)
 528                 late_time_init();
 529         calibrate_delay();
 530         pidmap_init();
 531         pgtable_cache_init();
 532         prio_tree_init();
 533         anon_vma_init();
 534 #ifdef CONFIG_X86
 535         if (efi_enabled)
 536                 efi_enter_virtual_mode();
 537 #endif
 538         fork_init(num_physpages);
 539         proc_caches_init();
 540         buffer_init();
 541         unnamed_dev_init();
 542         key_init();
 543         security_init();
 544         vfs_caches_init(num_physpages);
 545         radix_tree_init();
 546         signals_init();
 547         /* rootfs populating might need page-writeback */
 548         page_writeback_init();
 549 #ifdef CONFIG_PROC_FS
 550         proc_root_init();
 551 #endif
 552         cpuset_init();
 553
 554         check_bugs();
 555
 556         acpi_early_init(); /* before LAPIC and SMP init */
 557
 558         /* Do the rest non-__init'ed, we're now alive */
 559         rest_init();
 560 }
 561
 562 static int __initdata initcall_debug;
 563
 564 static int __init initcall_debug_setup(char *str)
 565 {
 566         initcall_debug = 1;
 567         return 1;
 568 }
 569 __setup("initcall_debug", initcall_debug_setup);
 570
 571 struct task_struct *child_reaper = &init_task;
 572
 573 extern initcall_t __initcall_start[], __initcall_end[];
 574
 575 static void __init do_initcalls(void)
 576 {
 577         initcall_t *call;
 578         int count = preempt_count();
 579
 580         for (call = __initcall_start; call < __initcall_end; call++) {
 581                 char *msg = NULL;
 582                 char msgbuf[40];
 583                 int result;
 584
 585                 if (initcall_debug) {
 586                         printk("Calling initcall 0x%p", *call);
 587                         print_fn_descriptor_symbol(": %s()",
 588                                         (unsigned long) *call);
 589                         printk("\n");
 590                 }
 591
 592                 result = (*call)();
 593
 594                 if (result && result != -ENODEV && initcall_debug) {
 595                         sprintf(msgbuf, "error code %d", result);
 596                         msg = msgbuf;
 597                 }
 598                 if (preempt_count() != count) {
 599                         msg = "preemption imbalance";
 600                         preempt_count() = count;
 601                 }
 602                 if (irqs_disabled()) {
 603                         msg = "disabled interrupts";
 604                         local_irq_enable();
 605                 }
 606                 if (msg) {
 607                         printk(KERN_WARNING "initcall at 0x%p", *call);
 608                         print_fn_descriptor_symbol(": %s()",
 609                                         (unsigned long) *call);
 610                         printk(": returned with %s\n", msg);
 611                 }
 612         }
 613
 614         /* Make sure there is no pending stuff from the initcall sequence */
 615         flush_scheduled_work();
 616 }
 617
 618 /*
 619  * Ok, the machine is now initialized. None of the devices
 620  * have been touched yet, but the CPU subsystem is up and
 621  * running, and memory and process management works.
 622  *
 623  * Now we can finally start doing some real work..
 624  */
 625 static void __init do_basic_setup(void)
 626 {
 627         /* drivers will send hotplug events */
 628         init_workqueues();
 629         usermodehelper_init();
 630         driver_init();
 631
 632 #ifdef CONFIG_SYSCTL
 633         sysctl_init();
 634 #endif
 635
 636         do_initcalls();
 637 }
 638
 639 static void do_pre_smp_initcalls(void)
 640 {
 641         extern int spawn_ksoftirqd(void);
 642 #ifdef CONFIG_SMP
 643         extern int migration_init(void);
 644
 645         migration_init();
 646 #endif
 647         spawn_ksoftirqd();
 648         spawn_softlockup_task();
 649 }
 650
 651 static void run_init_process(char *init_filename)
 652 {
 653         argv_init[0] = init_filename;
 654         execve(init_filename, argv_init, envp_init);
 655 }
 656
 657 static int init(void * unused)
 658 {
 659         lock_kernel();
 660         /*
 661          * init can run on any cpu.
 662          */
 663         set_cpus_allowed(current, CPU_MASK_ALL);
 664         /*
 665          * Tell the world that we're going to be the grim
 666          * reaper of innocent orphaned children.
 667          *
 668          * We don't want people to have to make incorrect
 669          * assumptions about where in the task array this
 670          * can be found.
 671          */
 672         child_reaper = current;
 673
 674         smp_prepare_cpus(max_cpus);
 675
 676         do_pre_smp_initcalls();
 677
 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         mark_rodata_ro();
 712         system_state = SYSTEM_RUNNING;
 713         numa_default_policy();
 714
 715         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 716                 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
 717
 718         (void) sys_dup(0);
 719         (void) sys_dup(0);
 720
 721         if (ramdisk_execute_command) {
 722                 run_init_process(ramdisk_execute_command);
 723                 printk(KERN_WARNING "Failed to execute %s\n",
 724                                 ramdisk_execute_command);
 725         }
 726
 727         /*
 728          * We try each of these until one succeeds.
 729          *
 730          * The Bourne shell can be used instead of init if we are
 731          * trying to recover a really broken machine.
 732          */
 733         if (execute_command) {
 734                 run_init_process(execute_command);
 735                 printk(KERN_WARNING "Failed to execute %s.  Attempting "
 736                                         "defaults...\n", execute_command);
 737         }
 738         run_init_process("/sbin/init");
 739         run_init_process("/etc/init");
 740         run_init_process("/bin/init");
 741         run_init_process("/bin/sh");
 742
 743         panic("No init found.  Try passing init= option to kernel.");
 744 }