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
 387         /*
 388          * The boot idle thread must execute schedule()
 389          * at least one to get things moving:
 390          */
 391         schedule();
 392
 393         cpu_idle();
 394 }
 395
 396 /* Check for early params. */
 397 static int __init do_early_param(char *param, char *val)
 398 {
 399         struct obs_kernel_param *p;
 400
 401         for (p = __setup_start; p < __setup_end; p++) {
 402                 if (p->early && strcmp(param, p->str) == 0) {
 403                         if (p->setup_func(val) != 0)
 404                                 printk(KERN_WARNING
 405                                        "Malformed early option '%s'\n", param);
 406                 }
 407         }
 408         /* We accept everything at this stage. */
 409         return 0;
 410 }
 411
 412 /* Arch code calls this early on, or if not, just before other parsing. */
 413 void __init parse_early_param(void)
 414 {
 415         static __initdata int done = 0;
 416         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 417
 418         if (done)
 419                 return;
 420
 421         /* All fall through to do_early_param. */
 422         strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
 423         parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
 424         done = 1;
 425 }
 426
 427 /*
 428  *      Activate the first processor.
 429  */
 430
 431 asmlinkage void __init start_kernel(void)
 432 {
 433         char * command_line;
 434         extern struct kernel_param __start___param[], __stop___param[];
 435 /*
 436  * Interrupts are still disabled. Do necessary setups, then
 437  * enable them
 438  */
 439         lock_kernel();
 440         page_address_init();
 441         printk(KERN_NOTICE);
 442         printk(linux_banner);
 443         setup_arch(&command_line);
 444         setup_per_cpu_areas();
 445
 446         /*
 447          * Mark the boot cpu "online" so that it can call console drivers in
 448          * printk() and can access its per-cpu storage.
 449          */
 450         smp_prepare_boot_cpu();
 451
 452         /*
 453          * Set up the scheduler prior starting any interrupts (such as the
 454          * timer interrupt). Full topology setup happens at smp_init()
 455          * time - but meanwhile we still have a functioning scheduler.
 456          */
 457         sched_init();
 458         /*
 459          * Disable preemption - early bootup scheduling is extremely
 460          * fragile until we cpu_idle() for the first time.
 461          */
 462         preempt_disable();
 463         build_all_zonelists();
 464         page_alloc_init();
 465         printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
 466         parse_early_param();
 467         parse_args("Booting kernel", command_line, __start___param,
 468                    __stop___param - __start___param,
 469                    &unknown_bootoption);
 470         sort_main_extable();
 471         trap_init();
 472         rcu_init();
 473         init_IRQ();
 474         pidhash_init();
 475         init_timers();
 476         softirq_init();
 477         time_init();
 478
 479         /*
 480          * HACK ALERT! This is early. We're enabling the console before
 481          * we've done PCI setups etc, and console_init() must be aware of
 482          * this. But we do want output early, in case something goes wrong.
 483          */
 484         console_init();
 485         if (panic_later)
 486                 panic(panic_later, panic_param);
 487         profile_init();
 488         local_irq_enable();
 489 #ifdef CONFIG_BLK_DEV_INITRD
 490         if (initrd_start && !initrd_below_start_ok &&
 491                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 492                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 493                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 494                 initrd_start = 0;
 495         }
 496 #endif
 497         vfs_caches_init_early();
 498         mem_init();
 499         kmem_cache_init();
 500         setup_per_cpu_pageset();
 501         numa_policy_init();
 502         if (late_time_init)
 503                 late_time_init();
 504         calibrate_delay();
 505         pidmap_init();
 506         pgtable_cache_init();
 507         prio_tree_init();
 508         anon_vma_init();
 509 #ifdef CONFIG_X86
 510         if (efi_enabled)
 511                 efi_enter_virtual_mode();
 512 #endif
 513         fork_init(num_physpages);
 514         proc_caches_init();
 515         buffer_init();
 516         unnamed_dev_init();
 517         key_init();
 518         security_init();
 519         vfs_caches_init(num_physpages);
 520         radix_tree_init();
 521         signals_init();
 522         /* rootfs populating might need page-writeback */
 523         page_writeback_init();
 524 #ifdef CONFIG_PROC_FS
 525         proc_root_init();
 526 #endif
 527         cpuset_init();
 528
 529         check_bugs();
 530
 531         acpi_early_init(); /* before LAPIC and SMP init */
 532
 533         /* Do the rest non-__init'ed, we're now alive */
 534         rest_init();
 535 }
 536
 537 static int __initdata initcall_debug;
 538
 539 static int __init initcall_debug_setup(char *str)
 540 {
 541         initcall_debug = 1;
 542         return 1;
 543 }
 544 __setup("initcall_debug", initcall_debug_setup);
 545
 546 struct task_struct *child_reaper = &init_task;
 547
 548 extern initcall_t __initcall_start[], __initcall_end[];
 549
 550 static void __init do_initcalls(void)
 551 {
 552         initcall_t *call;
 553         int count = preempt_count();
 554
 555         for (call = __initcall_start; call < __initcall_end; call++) {
 556                 char *msg;
 557
 558                 if (initcall_debug) {
 559                         printk(KERN_DEBUG "Calling initcall 0x%p", *call);
 560                         print_fn_descriptor_symbol(": %s()", (unsigned long) *call);
 561                         printk("\n");
 562                 }
 563
 564                 (*call)();
 565
 566                 msg = NULL;
 567                 if (preempt_count() != count) {
 568                         msg = "preemption imbalance";
 569                         preempt_count() = count;
 570                 }
 571                 if (irqs_disabled()) {
 572                         msg = "disabled interrupts";
 573                         local_irq_enable();
 574                 }
 575                 if (msg) {
 576                         printk(KERN_WARNING "error in initcall at 0x%p: "
 577                                 "returned with %s\n", *call, msg);
 578                 }
 579         }
 580
 581         /* Make sure there is no pending stuff from the initcall sequence */
 582         flush_scheduled_work();
 583 }
 584
 585 /*
 586  * Ok, the machine is now initialized. None of the devices
 587  * have been touched yet, but the CPU subsystem is up and
 588  * running, and memory and process management works.
 589  *
 590  * Now we can finally start doing some real work..
 591  */
 592 static void __init do_basic_setup(void)
 593 {
 594         /* drivers will send hotplug events */
 595         init_workqueues();
 596         usermodehelper_init();
 597         driver_init();
 598
 599 #ifdef CONFIG_SYSCTL
 600         sysctl_init();
 601 #endif
 602
 603         /* Networking initialization needs a process context */
 604         sock_init();
 605
 606         do_initcalls();
 607 }
 608
 609 static void do_pre_smp_initcalls(void)
 610 {
 611         extern int spawn_ksoftirqd(void);
 612 #ifdef CONFIG_SMP
 613         extern int migration_init(void);
 614
 615         migration_init();
 616 #endif
 617         spawn_ksoftirqd();
 618 }
 619
 620 static void run_init_process(char *init_filename)
 621 {
 622         argv_init[0] = init_filename;
 623         execve(init_filename, argv_init, envp_init);
 624 }
 625
 626 static inline void fixup_cpu_present_map(void)
 627 {
 628 #ifdef CONFIG_SMP
 629         int i;
 630
 631         /*
 632          * If arch is not hotplug ready and did not populate
 633          * cpu_present_map, just make cpu_present_map same as cpu_possible_map
 634          * for other cpu bringup code to function as normal. e.g smp_init() etc.
 635          */
 636         if (cpus_empty(cpu_present_map)) {
 637                 for_each_cpu(i) {
 638                         cpu_set(i, cpu_present_map);
 639                 }
 640         }
 641 #endif
 642 }
 643
 644 static int init(void * unused)
 645 {
 646         lock_kernel();
 647         /*
 648          * init can run on any cpu.
 649          */
 650         set_cpus_allowed(current, CPU_MASK_ALL);
 651         /*
 652          * Tell the world that we're going to be the grim
 653          * reaper of innocent orphaned children.
 654          *
 655          * We don't want people to have to make incorrect
 656          * assumptions about where in the task array this
 657          * can be found.
 658          */
 659         child_reaper = current;
 660
 661         /* Sets up cpus_possible() */
 662         smp_prepare_cpus(max_cpus);
 663
 664         do_pre_smp_initcalls();
 665
 666         fixup_cpu_present_map();
 667         smp_init();
 668         sched_init_smp();
 669
 670         cpuset_init_smp();
 671
 672         /*
 673          * Do this before initcalls, because some drivers want to access
 674          * firmware files.
 675          */
 676         populate_rootfs();
 677
 678         do_basic_setup();
 679
 680         /*
 681          * check if there is an early userspace init.  If yes, let it do all
 682          * the work
 683          */
 684         if (sys_access((const char __user *) "/init", 0) == 0)
 685                 execute_command = "/init";
 686         else
 687                 prepare_namespace();
 688
 689         /*
 690          * Ok, we have completed the initial bootup, and
 691          * we're essentially up and running. Get rid of the
 692          * initmem segments and start the user-mode stuff..
 693          */
 694         free_initmem();
 695         unlock_kernel();
 696         system_state = SYSTEM_RUNNING;
 697         numa_default_policy();
 698
 699         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 700                 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
 701
 702         (void) sys_dup(0);
 703         (void) sys_dup(0);
 704
 705         /*
 706          * We try each of these until one succeeds.
 707          *
 708          * The Bourne shell can be used instead of init if we are
 709          * trying to recover a really broken machine.
 710          */
 711
 712         if (execute_command)
 713                 run_init_process(execute_command);
 714
 715         run_init_process("/sbin/init");
 716         run_init_process("/etc/init");
 717         run_init_process("/bin/init");
 718         run_init_process("/bin/sh");
 719
 720         panic("No init found.  Try passing init= option to kernel.");
 721 }