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