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 #ifdef CONFIG_X86_LOCAL_APIC
  68 #include <asm/smp.h>
  69 #endif
  70
  71 /*
  72  * Versions of gcc older than that listed below may actually compile
  73  * and link okay, but the end product can have subtle run time bugs.
  74  * To avoid associated bogus bug reports, we flatly refuse to compile
  75  * with a gcc that is known to be too old from the very beginning.
  76  */
  77 #if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 95)
  78 #error Sorry, your GCC is too old. It builds incorrect kernels.
  79 #endif
  80
  81 static int init(void *);
  82
  83 extern void init_IRQ(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 static char *ramdisk_execute_command;
 127
 128 /* Setup configured maximum number of CPUs to activate */
 129 static unsigned int max_cpus = NR_CPUS;
 130
 131 /*
 132  * Setup routine for controlling SMP activation
 133  *
 134  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
 135  * activation entirely (the MPS table probe still happens, though).
 136  *
 137  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
 138  * greater than 0, limits the maximum number of CPUs activated in
 139  * SMP mode to <NUM>.
 140  */
 141 static int __init nosmp(char *str)
 142 {
 143         max_cpus = 0;
 144         return 1;
 145 }
 146
 147 __setup("nosmp", nosmp);
 148
 149 static int __init maxcpus(char *str)
 150 {
 151         get_option(&str, &max_cpus);
 152         return 1;
 153 }
 154
 155 __setup("maxcpus=", maxcpus);
 156
 157 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 158 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 159 static const char *panic_later, *panic_param;
 160
 161 extern struct obs_kernel_param __setup_start[], __setup_end[];
 162
 163 static int __init obsolete_checksetup(char *line)
 164 {
 165         struct obs_kernel_param *p;
 166
 167         p = __setup_start;
 168         do {
 169                 int n = strlen(p->str);
 170                 if (!strncmp(line, p->str, n)) {
 171                         if (p->early) {
 172                                 /* Already done in parse_early_param?  (Needs
 173                                  * exact match on param part) */
 174                                 if (line[n] == '\0' || line[n] == '=')
 175                                         return 1;
 176                         } else if (!p->setup_func) {
 177                                 printk(KERN_WARNING "Parameter %s is obsolete,"
 178                                        " ignored\n", p->str);
 179                                 return 1;
 180                         } else if (p->setup_func(line + n))
 181                                 return 1;
 182                 }
 183                 p++;
 184         } while (p < __setup_end);
 185         return 0;
 186 }
 187
 188 /*
 189  * This should be approx 2 Bo*oMips to start (note initial shift), and will
 190  * still work even if initially too large, it will just take slightly longer
 191  */
 192 unsigned long loops_per_jiffy = (1<<12);
 193
 194 EXPORT_SYMBOL(loops_per_jiffy);
 195
 196 static int __init debug_kernel(char *str)
 197 {
 198         if (*str)
 199                 return 0;
 200         console_loglevel = 10;
 201         return 1;
 202 }
 203
 204 static int __init quiet_kernel(char *str)
 205 {
 206         if (*str)
 207                 return 0;
 208         console_loglevel = 4;
 209         return 1;
 210 }
 211
 212 __setup("debug", debug_kernel);
 213 __setup("quiet", quiet_kernel);
 214
 215 static int __init loglevel(char *str)
 216 {
 217         get_option(&str, &console_loglevel);
 218         return 1;
 219 }
 220
 221 __setup("loglevel=", loglevel);
 222
 223 /*
 224  * Unknown boot options get handed to init, unless they look like
 225  * failed parameters
 226  */
 227 static int __init unknown_bootoption(char *param, char *val)
 228 {
 229         /* Change NUL term back to "=", to make "param" the whole string. */
 230         if (val) {
 231                 /* param=val or param="val"? */
 232                 if (val == param+strlen(param)+1)
 233                         val[-1] = '=';
 234                 else if (val == param+strlen(param)+2) {
 235                         val[-2] = '=';
 236                         memmove(val-1, val, strlen(val)+1);
 237                         val--;
 238                 } else
 239                         BUG();
 240         }
 241
 242         /* Handle obsolete-style parameters */
 243         if (obsolete_checksetup(param))
 244                 return 0;
 245
 246         /*
 247          * Preemptive maintenance for "why didn't my mispelled command
 248          * line work?"
 249          */
 250         if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
 251                 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
 252                 return 0;
 253         }
 254
 255         if (panic_later)
 256                 return 0;
 257
 258         if (val) {
 259                 /* Environment option */
 260                 unsigned int i;
 261                 for (i = 0; envp_init[i]; i++) {
 262                         if (i == MAX_INIT_ENVS) {
 263                                 panic_later = "Too many boot env vars at `%s'";
 264                                 panic_param = param;
 265                         }
 266                         if (!strncmp(param, envp_init[i], val - param))
 267                                 break;
 268                 }
 269                 envp_init[i] = param;
 270         } else {
 271                 /* Command line option */
 272                 unsigned int i;
 273                 for (i = 0; argv_init[i]; i++) {
 274                         if (i == MAX_INIT_ARGS) {
 275                                 panic_later = "Too many boot init vars at `%s'";
 276                                 panic_param = param;
 277                         }
 278                 }
 279                 argv_init[i] = param;
 280         }
 281         return 0;
 282 }
 283
 284 static int __init init_setup(char *str)
 285 {
 286         unsigned int i;
 287
 288         execute_command = str;
 289         /*
 290          * In case LILO is going to boot us with default command line,
 291          * it prepends "auto" before the whole cmdline which makes
 292          * the shell think it should execute a script with such name.
 293          * So we ignore all arguments entered _before_ init=... [MJ]
 294          */
 295         for (i = 1; i < MAX_INIT_ARGS; i++)
 296                 argv_init[i] = NULL;
 297         return 1;
 298 }
 299 __setup("init=", init_setup);
 300
 301 static int __init rdinit_setup(char *str)
 302 {
 303         unsigned int i;
 304
 305         ramdisk_execute_command = str;
 306         /* See "auto" comment in init_setup */
 307         for (i = 1; i < MAX_INIT_ARGS; i++)
 308                 argv_init[i] = NULL;
 309         return 1;
 310 }
 311 __setup("rdinit=", rdinit_setup);
 312
 313 extern void setup_arch(char **);
 314
 315 #ifndef CONFIG_SMP
 316
 317 #ifdef CONFIG_X86_LOCAL_APIC
 318 static void __init smp_init(void)
 319 {
 320         APIC_init_uniprocessor();
 321 }
 322 #else
 323 #define smp_init()      do { } while (0)
 324 #endif
 325
 326 static inline void setup_per_cpu_areas(void) { }
 327 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 328
 329 #else
 330
 331 #ifdef __GENERIC_PER_CPU
 332 unsigned long __per_cpu_offset[NR_CPUS];
 333
 334 EXPORT_SYMBOL(__per_cpu_offset);
 335
 336 static void __init setup_per_cpu_areas(void)
 337 {
 338         unsigned long size, i;
 339         char *ptr;
 340
 341         /* Copy section for each CPU (we discard the original) */
 342         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 343 #ifdef CONFIG_MODULES
 344         if (size < PERCPU_ENOUGH_ROOM)
 345                 size = PERCPU_ENOUGH_ROOM;
 346 #endif
 347
 348         ptr = alloc_bootmem(size * NR_CPUS);
 349
 350         for (i = 0; i < NR_CPUS; i++, ptr += size) {
 351                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 352                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 353         }
 354 }
 355 #endif /* !__GENERIC_PER_CPU */
 356
 357 /* Called by boot processor to activate the rest. */
 358 static void __init smp_init(void)
 359 {
 360         unsigned int i;
 361
 362         /* FIXME: This should be done in userspace --RR */
 363         for_each_present_cpu(i) {
 364                 if (num_online_cpus() >= max_cpus)
 365                         break;
 366                 if (!cpu_online(i))
 367                         cpu_up(i);
 368         }
 369
 370         /* Any cleanup work */
 371         printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
 372         smp_cpus_done(max_cpus);
 373 #if 0
 374         /* Get other processors into their bootup holding patterns. */
 375
 376         smp_commence();
 377 #endif
 378 }
 379
 380 #endif
 381
 382 /*
 383  * We need to finalize in a non-__init function or else race conditions
 384  * between the root thread and the init thread may cause start_kernel to
 385  * be reaped by free_initmem before the root thread has proceeded to
 386  * cpu_idle.
 387  *
 388  * gcc-3.4 accidentally inlines this function, so use noinline.
 389  */
 390
 391 static void noinline rest_init(void)
 392         __releases(kernel_lock)
 393 {
 394         kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
 395         numa_default_policy();
 396         unlock_kernel();
 397
 398         /*
 399          * The boot idle thread must execute schedule()
 400          * at least one to get things moving:
 401          */
 402         preempt_enable_no_resched();
 403         schedule();
 404         preempt_disable();
 405
 406         /* Call into cpu_idle with preempt disabled */
 407         cpu_idle();
 408 }
 409
 410 /* Check for early params. */
 411 static int __init do_early_param(char *param, char *val)
 412 {
 413         struct obs_kernel_param *p;
 414
 415         for (p = __setup_start; p < __setup_end; p++) {
 416                 if (p->early && strcmp(param, p->str) == 0) {
 417                         if (p->setup_func(val) != 0)
 418                                 printk(KERN_WARNING
 419                                        "Malformed early option '%s'\n", param);
 420                 }
 421         }
 422         /* We accept everything at this stage. */
 423         return 0;
 424 }
 425
 426 /* Arch code calls this early on, or if not, just before other parsing. */
 427 void __init parse_early_param(void)
 428 {
 429         static __initdata int done = 0;
 430         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 431
 432         if (done)
 433                 return;
 434
 435         /* All fall through to do_early_param. */
 436         strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
 437         parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
 438         done = 1;
 439 }
 440
 441 /*
 442  *      Activate the first processor.
 443  */
 444
 445 asmlinkage void __init start_kernel(void)
 446 {
 447         char * command_line;
 448         extern struct kernel_param __start___param[], __stop___param[];
 449 /*
 450  * Interrupts are still disabled. Do necessary setups, then
 451  * enable them
 452  */
 453         lock_kernel();
 454         page_address_init();
 455         printk(KERN_NOTICE);
 456         printk(linux_banner);
 457         setup_arch(&command_line);
 458         setup_per_cpu_areas();
 459
 460         /*
 461          * Mark the boot cpu "online" so that it can call console drivers in
 462          * printk() and can access its per-cpu storage.
 463          */
 464         smp_prepare_boot_cpu();
 465
 466         /*
 467          * Set up the scheduler prior starting any interrupts (such as the
 468          * timer interrupt). Full topology setup happens at smp_init()
 469          * time - but meanwhile we still have a functioning scheduler.
 470          */
 471         sched_init();
 472         /*
 473          * Disable preemption - early bootup scheduling is extremely
 474          * fragile until we cpu_idle() for the first time.
 475          */
 476         preempt_disable();
 477         build_all_zonelists();
 478         page_alloc_init();
 479         printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
 480         parse_early_param();
 481         parse_args("Booting kernel", command_line, __start___param,
 482                    __stop___param - __start___param,
 483                    &unknown_bootoption);
 484         sort_main_extable();
 485         trap_init();
 486         rcu_init();
 487         init_IRQ();
 488         pidhash_init();
 489         init_timers();
 490         softirq_init();
 491         time_init();
 492
 493         /*
 494          * HACK ALERT! This is early. We're enabling the console before
 495          * we've done PCI setups etc, and console_init() must be aware of
 496          * this. But we do want output early, in case something goes wrong.
 497          */
 498         console_init();
 499         if (panic_later)
 500                 panic(panic_later, panic_param);
 501         profile_init();
 502         local_irq_enable();
 503 #ifdef CONFIG_BLK_DEV_INITRD
 504         if (initrd_start && !initrd_below_start_ok &&
 505                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 506                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 507                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 508                 initrd_start = 0;
 509         }
 510 #endif
 511         vfs_caches_init_early();
 512         mem_init();
 513         kmem_cache_init();
 514         setup_per_cpu_pageset();
 515         numa_policy_init();
 516         if (late_time_init)
 517                 late_time_init();
 518         calibrate_delay();
 519         pidmap_init();
 520         pgtable_cache_init();
 521         prio_tree_init();
 522         anon_vma_init();
 523 #ifdef CONFIG_X86
 524         if (efi_enabled)
 525                 efi_enter_virtual_mode();
 526 #endif
 527         fork_init(num_physpages);
 528         proc_caches_init();
 529         buffer_init();
 530         unnamed_dev_init();
 531         key_init();
 532         security_init();
 533         vfs_caches_init(num_physpages);
 534         radix_tree_init();
 535         signals_init();
 536         /* rootfs populating might need page-writeback */
 537         page_writeback_init();
 538 #ifdef CONFIG_PROC_FS
 539         proc_root_init();
 540 #endif
 541         cpuset_init();
 542
 543         check_bugs();
 544
 545         acpi_early_init(); /* before LAPIC and SMP init */
 546
 547         /* Do the rest non-__init'ed, we're now alive */
 548         rest_init();
 549 }
 550
 551 static int __initdata initcall_debug;
 552
 553 static int __init initcall_debug_setup(char *str)
 554 {
 555         initcall_debug = 1;
 556         return 1;
 557 }
 558 __setup("initcall_debug", initcall_debug_setup);
 559
 560 struct task_struct *child_reaper = &init_task;
 561
 562 extern initcall_t __initcall_start[], __initcall_end[];
 563
 564 static void __init do_initcalls(void)
 565 {
 566         initcall_t *call;
 567         int count = preempt_count();
 568
 569         for (call = __initcall_start; call < __initcall_end; call++) {
 570                 char *msg;
 571
 572                 if (initcall_debug) {
 573                         printk(KERN_DEBUG "Calling initcall 0x%p", *call);
 574                         print_fn_descriptor_symbol(": %s()", (unsigned long) *call);
 575                         printk("\n");
 576                 }
 577
 578                 (*call)();
 579
 580                 msg = NULL;
 581                 if (preempt_count() != count) {
 582                         msg = "preemption imbalance";
 583                         preempt_count() = count;
 584                 }
 585                 if (irqs_disabled()) {
 586                         msg = "disabled interrupts";
 587                         local_irq_enable();
 588                 }
 589                 if (msg) {
 590                         printk(KERN_WARNING "error in initcall at 0x%p: "
 591                                 "returned with %s\n", *call, msg);
 592                 }
 593         }
 594
 595         /* Make sure there is no pending stuff from the initcall sequence */
 596         flush_scheduled_work();
 597 }
 598
 599 /*
 600  * Ok, the machine is now initialized. None of the devices
 601  * have been touched yet, but the CPU subsystem is up and
 602  * running, and memory and process management works.
 603  *
 604  * Now we can finally start doing some real work..
 605  */
 606 static void __init do_basic_setup(void)
 607 {
 608         /* drivers will send hotplug events */
 609         init_workqueues();
 610         usermodehelper_init();
 611         driver_init();
 612
 613 #ifdef CONFIG_SYSCTL
 614         sysctl_init();
 615 #endif
 616
 617         /* Networking initialization needs a process context */
 618         sock_init();
 619
 620         do_initcalls();
 621 }
 622
 623 static void do_pre_smp_initcalls(void)
 624 {
 625         extern int spawn_ksoftirqd(void);
 626 #ifdef CONFIG_SMP
 627         extern int migration_init(void);
 628
 629         migration_init();
 630 #endif
 631         spawn_ksoftirqd();
 632         spawn_softlockup_task();
 633 }
 634
 635 static void run_init_process(char *init_filename)
 636 {
 637         argv_init[0] = init_filename;
 638         execve(init_filename, argv_init, envp_init);
 639 }
 640
 641 static inline void fixup_cpu_present_map(void)
 642 {
 643 #ifdef CONFIG_SMP
 644         int i;
 645
 646         /*
 647          * If arch is not hotplug ready and did not populate
 648          * cpu_present_map, just make cpu_present_map same as cpu_possible_map
 649          * for other cpu bringup code to function as normal. e.g smp_init() etc.
 650          */
 651         if (cpus_empty(cpu_present_map)) {
 652                 for_each_cpu(i) {
 653                         cpu_set(i, cpu_present_map);
 654                 }
 655         }
 656 #endif
 657 }
 658
 659 static int init(void * unused)
 660 {
 661         lock_kernel();
 662         /*
 663          * init can run on any cpu.
 664          */
 665         set_cpus_allowed(current, CPU_MASK_ALL);
 666         /*
 667          * Tell the world that we're going to be the grim
 668          * reaper of innocent orphaned children.
 669          *
 670          * We don't want people to have to make incorrect
 671          * assumptions about where in the task array this
 672          * can be found.
 673          */
 674         child_reaper = current;
 675
 676         /* Sets up cpus_possible() */
 677         smp_prepare_cpus(max_cpus);
 678
 679         do_pre_smp_initcalls();
 680
 681         fixup_cpu_present_map();
 682         smp_init();
 683         sched_init_smp();
 684
 685         cpuset_init_smp();
 686
 687         /*
 688          * Do this before initcalls, because some drivers want to access
 689          * firmware files.
 690          */
 691         populate_rootfs();
 692
 693         do_basic_setup();
 694
 695         /*
 696          * check if there is an early userspace init.  If yes, let it do all
 697          * the work
 698          */
 699
 700         if (!ramdisk_execute_command)
 701                 ramdisk_execute_command = "/init";
 702
 703         if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
 704                 ramdisk_execute_command = NULL;
 705                 prepare_namespace();
 706         }
 707
 708         /*
 709          * Ok, we have completed the initial bootup, and
 710          * we're essentially up and running. Get rid of the
 711          * initmem segments and start the user-mode stuff..
 712          */
 713         free_initmem();
 714         unlock_kernel();
 715         system_state = SYSTEM_RUNNING;
 716         numa_default_policy();
 717
 718         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 719                 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
 720
 721         (void) sys_dup(0);
 722         (void) sys_dup(0);
 723
 724         if (ramdisk_execute_command) {
 725                 run_init_process(ramdisk_execute_command);
 726                 printk(KERN_WARNING "Failed to execute %s\n",
 727                                 ramdisk_execute_command);
 728         }
 729
 730         /*
 731          * We try each of these until one succeeds.
 732          *
 733          * The Bourne shell can be used instead of init if we are
 734          * trying to recover a really broken machine.
 735          */
 736         if (execute_command) {
 737                 run_init_process(execute_command);
 738                 printk(KERN_WARNING "Failed to execute %s.  Attempting "
 739                                         "defaults...\n", execute_command);
 740         }
 741         run_init_process("/sbin/init");
 742         run_init_process("/etc/init");
 743         run_init_process("/bin/init");
 744         run_init_process("/bin/sh");
 745
 746         panic("No init found.  Try passing init= option to kernel.");
 747 }