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/types.h>
  15 #include <linux/module.h>
  16 #include <linux/proc_fs.h>
  17 #include <linux/kernel.h>
  18 #include <linux/syscalls.h>
  19 #include <linux/string.h>
  20 #include <linux/ctype.h>
  21 #include <linux/delay.h>
  22 #include <linux/utsname.h>
  23 #include <linux/ioport.h>
  24 #include <linux/init.h>
  25 #include <linux/smp_lock.h>
  26 #include <linux/initrd.h>
  27 #include <linux/hdreg.h>
  28 #include <linux/bootmem.h>
  29 #include <linux/tty.h>
  30 #include <linux/gfp.h>
  31 #include <linux/percpu.h>
  32 #include <linux/kmod.h>
  33 #include <linux/kernel_stat.h>
  34 #include <linux/security.h>
  35 #include <linux/workqueue.h>
  36 #include <linux/profile.h>
  37 #include <linux/rcupdate.h>
  38 #include <linux/moduleparam.h>
  39 #include <linux/kallsyms.h>
  40 #include <linux/writeback.h>
  41 #include <linux/cpu.h>
  42 #include <linux/cpuset.h>
  43 #include <linux/efi.h>
  44 #include <linux/unistd.h>
  45 #include <linux/rmap.h>
  46 #include <linux/mempolicy.h>
  47 #include <linux/key.h>
  48 #include <linux/unwind.h>
  49 #include <linux/buffer_head.h>
  50
  51 #include <asm/io.h>
  52 #include <asm/bugs.h>
  53 #include <asm/setup.h>
  54 #include <asm/sections.h>
  55 #include <asm/cacheflush.h>
  56
  57 #ifdef CONFIG_X86_LOCAL_APIC
  58 #include <asm/smp.h>
  59 #endif
  60
  61 /*
  62  * This is one of the first .c files built. Error out early if we have compiler
  63  * trouble.
  64  *
  65  * Versions of gcc older than that listed below may actually compile and link
  66  * okay, but the end product can have subtle run time bugs.  To avoid associated
  67  * bogus bug reports, we flatly refuse to compile with a gcc that is known to be
  68  * too old from the very beginning.
  69  */
  70 #if (__GNUC__ < 3) || (__GNUC__ == 3 && __GNUC_MINOR__ < 2)
  71 #error Sorry, your GCC is too old. It builds incorrect kernels.
  72 #endif
  73
  74 static int init(void *);
  75
  76 extern void init_IRQ(void);
  77 extern void fork_init(unsigned long);
  78 extern void mca_init(void);
  79 extern void sbus_init(void);
  80 extern void sysctl_init(void);
  81 extern void signals_init(void);
  82 extern void pidhash_init(void);
  83 extern void pidmap_init(void);
  84 extern void prio_tree_init(void);
  85 extern void radix_tree_init(void);
  86 extern void free_initmem(void);
  87 extern void populate_rootfs(void);
  88 extern void driver_init(void);
  89 extern void prepare_namespace(void);
  90 #ifdef  CONFIG_ACPI
  91 extern void acpi_early_init(void);
  92 #else
  93 static inline void acpi_early_init(void) { }
  94 #endif
  95 #ifndef CONFIG_DEBUG_RODATA
  96 static inline void mark_rodata_ro(void) { }
  97 #endif
  98
  99 #ifdef CONFIG_TC
 100 extern void tc_init(void);
 101 #endif
 102
 103 enum system_states system_state;
 104 EXPORT_SYMBOL(system_state);
 105
 106 /*
 107  * Boot command-line arguments
 108  */
 109 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
 110 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
 111
 112 extern void time_init(void);
 113 /* Default late time init is NULL. archs can override this later. */
 114 void (*late_time_init)(void);
 115 extern void softirq_init(void);
 116
 117 /* Untouched command line (eg. for /proc) saved by arch-specific code. */
 118 char saved_command_line[COMMAND_LINE_SIZE];
 119
 120 static char *execute_command;
 121 static char *ramdisk_execute_command;
 122
 123 /* Setup configured maximum number of CPUs to activate */
 124 static unsigned int max_cpus = NR_CPUS;
 125
 126 /*
 127  * Setup routine for controlling SMP activation
 128  *
 129  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
 130  * activation entirely (the MPS table probe still happens, though).
 131  *
 132  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
 133  * greater than 0, limits the maximum number of CPUs activated in
 134  * SMP mode to <NUM>.
 135  */
 136 static int __init nosmp(char *str)
 137 {
 138         max_cpus = 0;
 139         return 1;
 140 }
 141
 142 __setup("nosmp", nosmp);
 143
 144 static int __init maxcpus(char *str)
 145 {
 146         get_option(&str, &max_cpus);
 147         return 1;
 148 }
 149
 150 __setup("maxcpus=", maxcpus);
 151
 152 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 153 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 154 static const char *panic_later, *panic_param;
 155
 156 extern struct obs_kernel_param __setup_start[], __setup_end[];
 157
 158 static int __init obsolete_checksetup(char *line)
 159 {
 160         struct obs_kernel_param *p;
 161
 162         p = __setup_start;
 163         do {
 164                 int n = strlen(p->str);
 165                 if (!strncmp(line, p->str, n)) {
 166                         if (p->early) {
 167                                 /* Already done in parse_early_param?  (Needs
 168                                  * exact match on param part) */
 169                                 if (line[n] == '\0' || line[n] == '=')
 170                                         return 1;
 171                         } else if (!p->setup_func) {
 172                                 printk(KERN_WARNING "Parameter %s is obsolete,"
 173                                        " ignored\n", p->str);
 174                                 return 1;
 175                         } else if (p->setup_func(line + n))
 176                                 return 1;
 177                 }
 178                 p++;
 179         } while (p < __setup_end);
 180         return 0;
 181 }
 182
 183 /*
 184  * This should be approx 2 Bo*oMips to start (note initial shift), and will
 185  * still work even if initially too large, it will just take slightly longer
 186  */
 187 unsigned long loops_per_jiffy = (1<<12);
 188
 189 EXPORT_SYMBOL(loops_per_jiffy);
 190
 191 static int __init debug_kernel(char *str)
 192 {
 193         if (*str)
 194                 return 0;
 195         console_loglevel = 10;
 196         return 1;
 197 }
 198
 199 static int __init quiet_kernel(char *str)
 200 {
 201         if (*str)
 202                 return 0;
 203         console_loglevel = 4;
 204         return 1;
 205 }
 206
 207 __setup("debug", debug_kernel);
 208 __setup("quiet", quiet_kernel);
 209
 210 static int __init loglevel(char *str)
 211 {
 212         get_option(&str, &console_loglevel);
 213         return 1;
 214 }
 215
 216 __setup("loglevel=", loglevel);
 217
 218 /*
 219  * Unknown boot options get handed to init, unless they look like
 220  * failed parameters
 221  */
 222 static int __init unknown_bootoption(char *param, char *val)
 223 {
 224         /* Change NUL term back to "=", to make "param" the whole string. */
 225         if (val) {
 226                 /* param=val or param="val"? */
 227                 if (val == param+strlen(param)+1)
 228                         val[-1] = '=';
 229                 else if (val == param+strlen(param)+2) {
 230                         val[-2] = '=';
 231                         memmove(val-1, val, strlen(val)+1);
 232                         val--;
 233                 } else
 234                         BUG();
 235         }
 236
 237         /* Handle obsolete-style parameters */
 238         if (obsolete_checksetup(param))
 239                 return 0;
 240
 241         /*
 242          * Preemptive maintenance for "why didn't my mispelled command
 243          * line work?"
 244          */
 245         if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
 246                 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
 247                 return 0;
 248         }
 249
 250         if (panic_later)
 251                 return 0;
 252
 253         if (val) {
 254                 /* Environment option */
 255                 unsigned int i;
 256                 for (i = 0; envp_init[i]; i++) {
 257                         if (i == MAX_INIT_ENVS) {
 258                                 panic_later = "Too many boot env vars at `%s'";
 259                                 panic_param = param;
 260                         }
 261                         if (!strncmp(param, envp_init[i], val - param))
 262                                 break;
 263                 }
 264                 envp_init[i] = param;
 265         } else {
 266                 /* Command line option */
 267                 unsigned int i;
 268                 for (i = 0; argv_init[i]; i++) {
 269                         if (i == MAX_INIT_ARGS) {
 270                                 panic_later = "Too many boot init vars at `%s'";
 271                                 panic_param = param;
 272                         }
 273                 }
 274                 argv_init[i] = param;
 275         }
 276         return 0;
 277 }
 278
 279 static int __init init_setup(char *str)
 280 {
 281         unsigned int i;
 282
 283         execute_command = str;
 284         /*
 285          * In case LILO is going to boot us with default command line,
 286          * it prepends "auto" before the whole cmdline which makes
 287          * the shell think it should execute a script with such name.
 288          * So we ignore all arguments entered _before_ init=... [MJ]
 289          */
 290         for (i = 1; i < MAX_INIT_ARGS; i++)
 291                 argv_init[i] = NULL;
 292         return 1;
 293 }
 294 __setup("init=", init_setup);
 295
 296 static int __init rdinit_setup(char *str)
 297 {
 298         unsigned int i;
 299
 300         ramdisk_execute_command = str;
 301         /* See "auto" comment in init_setup */
 302         for (i = 1; i < MAX_INIT_ARGS; i++)
 303                 argv_init[i] = NULL;
 304         return 1;
 305 }
 306 __setup("rdinit=", rdinit_setup);
 307
 308 #ifndef CONFIG_SMP
 309
 310 #ifdef CONFIG_X86_LOCAL_APIC
 311 static void __init smp_init(void)
 312 {
 313         APIC_init_uniprocessor();
 314 }
 315 #else
 316 #define smp_init()      do { } while (0)
 317 #endif
 318
 319 static inline void setup_per_cpu_areas(void) { }
 320 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 321
 322 #else
 323
 324 #ifdef __GENERIC_PER_CPU
 325 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
 326
 327 EXPORT_SYMBOL(__per_cpu_offset);
 328
 329 static void __init setup_per_cpu_areas(void)
 330 {
 331         unsigned long size, i;
 332         char *ptr;
 333         unsigned long nr_possible_cpus = num_possible_cpus();
 334
 335         /* Copy section for each CPU (we discard the original) */
 336         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 337 #ifdef CONFIG_MODULES
 338         if (size < PERCPU_ENOUGH_ROOM)
 339                 size = PERCPU_ENOUGH_ROOM;
 340 #endif
 341         ptr = alloc_bootmem(size * nr_possible_cpus);
 342
 343         for_each_possible_cpu(i) {
 344                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 345                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 346                 ptr += size;
 347         }
 348 }
 349 #endif /* !__GENERIC_PER_CPU */
 350
 351 /* Called by boot processor to activate the rest. */
 352 static void __init smp_init(void)
 353 {
 354         unsigned int i;
 355
 356         /* FIXME: This should be done in userspace --RR */
 357         for_each_present_cpu(i) {
 358                 if (num_online_cpus() >= max_cpus)
 359                         break;
 360                 if (!cpu_online(i))
 361                         cpu_up(i);
 362         }
 363
 364         /* Any cleanup work */
 365         printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
 366         smp_cpus_done(max_cpus);
 367 #if 0
 368         /* Get other processors into their bootup holding patterns. */
 369
 370         smp_commence();
 371 #endif
 372 }
 373
 374 #endif
 375
 376 /*
 377  * We need to finalize in a non-__init function or else race conditions
 378  * between the root thread and the init thread may cause start_kernel to
 379  * be reaped by free_initmem before the root thread has proceeded to
 380  * cpu_idle.
 381  *
 382  * gcc-3.4 accidentally inlines this function, so use noinline.
 383  */
 384
 385 static void noinline rest_init(void)
 386         __releases(kernel_lock)
 387 {
 388         kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
 389         numa_default_policy();
 390         unlock_kernel();
 391
 392         /*
 393          * The boot idle thread must execute schedule()
 394          * at least one to get things moving:
 395          */
 396         preempt_enable_no_resched();
 397         schedule();
 398         preempt_disable();
 399
 400         /* Call into cpu_idle with preempt disabled */
 401         cpu_idle();
 402 }
 403
 404 /* Check for early params. */
 405 static int __init do_early_param(char *param, char *val)
 406 {
 407         struct obs_kernel_param *p;
 408
 409         for (p = __setup_start; p < __setup_end; p++) {
 410                 if (p->early && strcmp(param, p->str) == 0) {
 411                         if (p->setup_func(val) != 0)
 412                                 printk(KERN_WARNING
 413                                        "Malformed early option '%s'\n", param);
 414                 }
 415         }
 416         /* We accept everything at this stage. */
 417         return 0;
 418 }
 419
 420 /* Arch code calls this early on, or if not, just before other parsing. */
 421 void __init parse_early_param(void)
 422 {
 423         static __initdata int done = 0;
 424         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 425
 426         if (done)
 427                 return;
 428
 429         /* All fall through to do_early_param. */
 430         strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
 431         parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
 432         done = 1;
 433 }
 434
 435 /*
 436  *      Activate the first processor.
 437  */
 438
 439 static void __init boot_cpu_init(void)
 440 {
 441         int cpu = smp_processor_id();
 442         /* Mark the boot cpu "present", "online" etc for SMP and UP case */
 443         cpu_set(cpu, cpu_online_map);
 444         cpu_set(cpu, cpu_present_map);
 445         cpu_set(cpu, cpu_possible_map);
 446 }
 447
 448 void __init __attribute__((weak)) smp_setup_processor_id(void)
 449 {
 450 }
 451
 452 asmlinkage void __init start_kernel(void)
 453 {
 454         char * command_line;
 455         extern struct kernel_param __start___param[], __stop___param[];
 456
 457         smp_setup_processor_id();
 458
 459 /*
 460  * Interrupts are still disabled. Do necessary setups, then
 461  * enable them
 462  */
 463         lock_kernel();
 464         boot_cpu_init();
 465         page_address_init();
 466         printk(KERN_NOTICE);
 467         printk(linux_banner);
 468         setup_arch(&command_line);
 469         setup_per_cpu_areas();
 470         smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
 471
 472         /*
 473          * Set up the scheduler prior starting any interrupts (such as the
 474          * timer interrupt). Full topology setup happens at smp_init()
 475          * time - but meanwhile we still have a functioning scheduler.
 476          */
 477         sched_init();
 478         /*
 479          * Disable preemption - early bootup scheduling is extremely
 480          * fragile until we cpu_idle() for the first time.
 481          */
 482         preempt_disable();
 483         build_all_zonelists();
 484         page_alloc_init();
 485         printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
 486         parse_early_param();
 487         parse_args("Booting kernel", command_line, __start___param,
 488                    __stop___param - __start___param,
 489                    &unknown_bootoption);
 490         sort_main_extable();
 491         unwind_init();
 492         trap_init();
 493         rcu_init();
 494         init_IRQ();
 495         pidhash_init();
 496         init_timers();
 497         hrtimers_init();
 498         softirq_init();
 499         time_init();
 500         timekeeping_init();
 501
 502         /*
 503          * HACK ALERT! This is early. We're enabling the console before
 504          * we've done PCI setups etc, and console_init() must be aware of
 505          * this. But we do want output early, in case something goes wrong.
 506          */
 507         console_init();
 508         if (panic_later)
 509                 panic(panic_later, panic_param);
 510         profile_init();
 511         local_irq_enable();
 512 #ifdef CONFIG_BLK_DEV_INITRD
 513         if (initrd_start && !initrd_below_start_ok &&
 514                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 515                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 516                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 517                 initrd_start = 0;
 518         }
 519 #endif
 520         vfs_caches_init_early();
 521         cpuset_init_early();
 522         mem_init();
 523         kmem_cache_init();
 524         setup_per_cpu_pageset();
 525         numa_policy_init();
 526         if (late_time_init)
 527                 late_time_init();
 528         calibrate_delay();
 529         pidmap_init();
 530         pgtable_cache_init();
 531         prio_tree_init();
 532         anon_vma_init();
 533 #ifdef CONFIG_X86
 534         if (efi_enabled)
 535                 efi_enter_virtual_mode();
 536 #endif
 537         fork_init(num_physpages);
 538         proc_caches_init();
 539         buffer_init();
 540         unnamed_dev_init();
 541         key_init();
 542         security_init();
 543         vfs_caches_init(num_physpages);
 544         radix_tree_init();
 545         signals_init();
 546         /* rootfs populating might need page-writeback */
 547         page_writeback_init();
 548 #ifdef CONFIG_PROC_FS
 549         proc_root_init();
 550 #endif
 551         cpuset_init();
 552
 553         check_bugs();
 554
 555         acpi_early_init(); /* before LAPIC and SMP init */
 556
 557         /* Do the rest non-__init'ed, we're now alive */
 558         rest_init();
 559 }
 560
 561 static int __initdata initcall_debug;
 562
 563 static int __init initcall_debug_setup(char *str)
 564 {
 565         initcall_debug = 1;
 566         return 1;
 567 }
 568 __setup("initcall_debug", initcall_debug_setup);
 569
 570 struct task_struct *child_reaper = &init_task;
 571
 572 extern initcall_t __initcall_start[], __initcall_end[];
 573
 574 static void __init do_initcalls(void)
 575 {
 576         initcall_t *call;
 577         int count = preempt_count();
 578
 579         for (call = __initcall_start; call < __initcall_end; call++) {
 580                 char *msg = NULL;
 581                 char msgbuf[40];
 582                 int result;
 583
 584                 if (initcall_debug) {
 585                         printk("Calling initcall 0x%p", *call);
 586                         print_fn_descriptor_symbol(": %s()",
 587                                         (unsigned long) *call);
 588                         printk("\n");
 589                 }
 590
 591                 result = (*call)();
 592
 593                 if (result && result != -ENODEV && initcall_debug) {
 594                         sprintf(msgbuf, "error code %d", result);
 595                         msg = msgbuf;
 596                 }
 597                 if (preempt_count() != count) {
 598                         msg = "preemption imbalance";
 599                         preempt_count() = count;
 600                 }
 601                 if (irqs_disabled()) {
 602                         msg = "disabled interrupts";
 603                         local_irq_enable();
 604                 }
 605                 if (msg) {
 606                         printk(KERN_WARNING "initcall at 0x%p", *call);
 607                         print_fn_descriptor_symbol(": %s()",
 608                                         (unsigned long) *call);
 609                         printk(": returned with %s\n", msg);
 610                 }
 611         }
 612
 613         /* Make sure there is no pending stuff from the initcall sequence */
 614         flush_scheduled_work();
 615 }
 616
 617 /*
 618  * Ok, the machine is now initialized. None of the devices
 619  * have been touched yet, but the CPU subsystem is up and
 620  * running, and memory and process management works.
 621  *
 622  * Now we can finally start doing some real work..
 623  */
 624 static void __init do_basic_setup(void)
 625 {
 626         /* drivers will send hotplug events */
 627         init_workqueues();
 628         usermodehelper_init();
 629         driver_init();
 630
 631 #ifdef CONFIG_SYSCTL
 632         sysctl_init();
 633 #endif
 634
 635         do_initcalls();
 636 }
 637
 638 static void do_pre_smp_initcalls(void)
 639 {
 640         extern int spawn_ksoftirqd(void);
 641 #ifdef CONFIG_SMP
 642         extern int migration_init(void);
 643
 644         migration_init();
 645 #endif
 646         spawn_ksoftirqd();
 647         spawn_softlockup_task();
 648 }
 649
 650 static void run_init_process(char *init_filename)
 651 {
 652         argv_init[0] = init_filename;
 653         execve(init_filename, argv_init, envp_init);
 654 }
 655
 656 static int init(void * unused)
 657 {
 658         lock_kernel();
 659         /*
 660          * init can run on any cpu.
 661          */
 662         set_cpus_allowed(current, CPU_MASK_ALL);
 663         /*
 664          * Tell the world that we're going to be the grim
 665          * reaper of innocent orphaned children.
 666          *
 667          * We don't want people to have to make incorrect
 668          * assumptions about where in the task array this
 669          * can be found.
 670          */
 671         child_reaper = current;
 672
 673         smp_prepare_cpus(max_cpus);
 674
 675         do_pre_smp_initcalls();
 676
 677         smp_init();
 678         sched_init_smp();
 679
 680         cpuset_init_smp();
 681
 682         /*
 683          * Do this before initcalls, because some drivers want to access
 684          * firmware files.
 685          */
 686         populate_rootfs();
 687
 688         do_basic_setup();
 689
 690         /*
 691          * check if there is an early userspace init.  If yes, let it do all
 692          * the work
 693          */
 694
 695         if (!ramdisk_execute_command)
 696                 ramdisk_execute_command = "/init";
 697
 698         if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
 699                 ramdisk_execute_command = NULL;
 700                 prepare_namespace();
 701         }
 702
 703         /*
 704          * Ok, we have completed the initial bootup, and
 705          * we're essentially up and running. Get rid of the
 706          * initmem segments and start the user-mode stuff..
 707          */
 708         free_initmem();
 709         unlock_kernel();
 710         mark_rodata_ro();
 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 }