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 #include <asm/sections.h>
  55 #include <asm/cacheflush.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 #ifndef CONFIG_DEBUG_RODATA
 104 static inline void mark_rodata_ro(void) { }
 105 #endif
 106
 107 #ifdef CONFIG_TC
 108 extern void tc_init(void);
 109 #endif
 110
 111 enum system_states system_state;
 112 EXPORT_SYMBOL(system_state);
 113
 114 /*
 115  * Boot command-line arguments
 116  */
 117 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
 118 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
 119
 120 extern void time_init(void);
 121 /* Default late time init is NULL. archs can override this later. */
 122 void (*late_time_init)(void);
 123 extern void softirq_init(void);
 124
 125 /* Untouched command line (eg. for /proc) saved by arch-specific code. */
 126 char saved_command_line[COMMAND_LINE_SIZE];
 127
 128 static char *execute_command;
 129 static char *ramdisk_execute_command;
 130
 131 /* Setup configured maximum number of CPUs to activate */
 132 static unsigned int max_cpus = NR_CPUS;
 133
 134 /*
 135  * Setup routine for controlling SMP activation
 136  *
 137  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
 138  * activation entirely (the MPS table probe still happens, though).
 139  *
 140  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
 141  * greater than 0, limits the maximum number of CPUs activated in
 142  * SMP mode to <NUM>.
 143  */
 144 static int __init nosmp(char *str)
 145 {
 146         max_cpus = 0;
 147         return 1;
 148 }
 149
 150 __setup("nosmp", nosmp);
 151
 152 static int __init maxcpus(char *str)
 153 {
 154         get_option(&str, &max_cpus);
 155         return 1;
 156 }
 157
 158 __setup("maxcpus=", maxcpus);
 159
 160 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 161 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 162 static const char *panic_later, *panic_param;
 163
 164 extern struct obs_kernel_param __setup_start[], __setup_end[];
 165
 166 static int __init obsolete_checksetup(char *line)
 167 {
 168         struct obs_kernel_param *p;
 169
 170         p = __setup_start;
 171         do {
 172                 int n = strlen(p->str);
 173                 if (!strncmp(line, p->str, n)) {
 174                         if (p->early) {
 175                                 /* Already done in parse_early_param?  (Needs
 176                                  * exact match on param part) */
 177                                 if (line[n] == '\0' || line[n] == '=')
 178                                         return 1;
 179                         } else if (!p->setup_func) {
 180                                 printk(KERN_WARNING "Parameter %s is obsolete,"
 181                                        " ignored\n", p->str);
 182                                 return 1;
 183                         } else if (p->setup_func(line + n))
 184                                 return 1;
 185                 }
 186                 p++;
 187         } while (p < __setup_end);
 188         return 0;
 189 }
 190
 191 /*
 192  * This should be approx 2 Bo*oMips to start (note initial shift), and will
 193  * still work even if initially too large, it will just take slightly longer
 194  */
 195 unsigned long loops_per_jiffy = (1<<12);
 196
 197 EXPORT_SYMBOL(loops_per_jiffy);
 198
 199 static int __init debug_kernel(char *str)
 200 {
 201         if (*str)
 202                 return 0;
 203         console_loglevel = 10;
 204         return 1;
 205 }
 206
 207 static int __init quiet_kernel(char *str)
 208 {
 209         if (*str)
 210                 return 0;
 211         console_loglevel = 4;
 212         return 1;
 213 }
 214
 215 __setup("debug", debug_kernel);
 216 __setup("quiet", quiet_kernel);
 217
 218 static int __init loglevel(char *str)
 219 {
 220         get_option(&str, &console_loglevel);
 221         return 1;
 222 }
 223
 224 __setup("loglevel=", loglevel);
 225
 226 /*
 227  * Unknown boot options get handed to init, unless they look like
 228  * failed parameters
 229  */
 230 static int __init unknown_bootoption(char *param, char *val)
 231 {
 232         /* Change NUL term back to "=", to make "param" the whole string. */
 233         if (val) {
 234                 /* param=val or param="val"? */
 235                 if (val == param+strlen(param)+1)
 236                         val[-1] = '=';
 237                 else if (val == param+strlen(param)+2) {
 238                         val[-2] = '=';
 239                         memmove(val-1, val, strlen(val)+1);
 240                         val--;
 241                 } else
 242                         BUG();
 243         }
 244
 245         /* Handle obsolete-style parameters */
 246         if (obsolete_checksetup(param))
 247                 return 0;
 248
 249         /*
 250          * Preemptive maintenance for "why didn't my mispelled command
 251          * line work?"
 252          */
 253         if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
 254                 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
 255                 return 0;
 256         }
 257
 258         if (panic_later)
 259                 return 0;
 260
 261         if (val) {
 262                 /* Environment option */
 263                 unsigned int i;
 264                 for (i = 0; envp_init[i]; i++) {
 265                         if (i == MAX_INIT_ENVS) {
 266                                 panic_later = "Too many boot env vars at `%s'";
 267                                 panic_param = param;
 268                         }
 269                         if (!strncmp(param, envp_init[i], val - param))
 270                                 break;
 271                 }
 272                 envp_init[i] = param;
 273         } else {
 274                 /* Command line option */
 275                 unsigned int i;
 276                 for (i = 0; argv_init[i]; i++) {
 277                         if (i == MAX_INIT_ARGS) {
 278                                 panic_later = "Too many boot init vars at `%s'";
 279                                 panic_param = param;
 280                         }
 281                 }
 282                 argv_init[i] = param;
 283         }
 284         return 0;
 285 }
 286
 287 static int __init init_setup(char *str)
 288 {
 289         unsigned int i;
 290
 291         execute_command = str;
 292         /*
 293          * In case LILO is going to boot us with default command line,
 294          * it prepends "auto" before the whole cmdline which makes
 295          * the shell think it should execute a script with such name.
 296          * So we ignore all arguments entered _before_ init=... [MJ]
 297          */
 298         for (i = 1; i < MAX_INIT_ARGS; i++)
 299                 argv_init[i] = NULL;
 300         return 1;
 301 }
 302 __setup("init=", init_setup);
 303
 304 static int __init rdinit_setup(char *str)
 305 {
 306         unsigned int i;
 307
 308         ramdisk_execute_command = str;
 309         /* See "auto" comment in init_setup */
 310         for (i = 1; i < MAX_INIT_ARGS; i++)
 311                 argv_init[i] = NULL;
 312         return 1;
 313 }
 314 __setup("rdinit=", rdinit_setup);
 315
 316 extern void setup_arch(char **);
 317
 318 #ifndef CONFIG_SMP
 319
 320 #ifdef CONFIG_X86_LOCAL_APIC
 321 static void __init smp_init(void)
 322 {
 323         APIC_init_uniprocessor();
 324 }
 325 #else
 326 #define smp_init()      do { } while (0)
 327 #endif
 328
 329 static inline void setup_per_cpu_areas(void) { }
 330 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 331
 332 #else
 333
 334 #ifdef __GENERIC_PER_CPU
 335 unsigned long __per_cpu_offset[NR_CPUS];
 336
 337 EXPORT_SYMBOL(__per_cpu_offset);
 338
 339 static void __init setup_per_cpu_areas(void)
 340 {
 341         unsigned long size, i;
 342         char *ptr;
 343
 344         /* Copy section for each CPU (we discard the original) */
 345         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 346 #ifdef CONFIG_MODULES
 347         if (size < PERCPU_ENOUGH_ROOM)
 348                 size = PERCPU_ENOUGH_ROOM;
 349 #endif
 350
 351         ptr = alloc_bootmem(size * NR_CPUS);
 352
 353         for (i = 0; i < NR_CPUS; i++, ptr += size) {
 354                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 355                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 356         }
 357 }
 358 #endif /* !__GENERIC_PER_CPU */
 359
 360 /* Called by boot processor to activate the rest. */
 361 static void __init smp_init(void)
 362 {
 363         unsigned int i;
 364
 365         /* FIXME: This should be done in userspace --RR */
 366         for_each_present_cpu(i) {
 367                 if (num_online_cpus() >= max_cpus)
 368                         break;
 369                 if (!cpu_online(i))
 370                         cpu_up(i);
 371         }
 372
 373         /* Any cleanup work */
 374         printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
 375         smp_cpus_done(max_cpus);
 376 #if 0
 377         /* Get other processors into their bootup holding patterns. */
 378
 379         smp_commence();
 380 #endif
 381 }
 382
 383 #endif
 384
 385 /*
 386  * We need to finalize in a non-__init function or else race conditions
 387  * between the root thread and the init thread may cause start_kernel to
 388  * be reaped by free_initmem before the root thread has proceeded to
 389  * cpu_idle.
 390  *
 391  * gcc-3.4 accidentally inlines this function, so use noinline.
 392  */
 393
 394 static void noinline rest_init(void)
 395         __releases(kernel_lock)
 396 {
 397         kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
 398         numa_default_policy();
 399         unlock_kernel();
 400
 401         /*
 402          * The boot idle thread must execute schedule()
 403          * at least one to get things moving:
 404          */
 405         preempt_enable_no_resched();
 406         schedule();
 407         preempt_disable();
 408
 409         /* Call into cpu_idle with preempt disabled */
 410         cpu_idle();
 411 }
 412
 413 /* Check for early params. */
 414 static int __init do_early_param(char *param, char *val)
 415 {
 416         struct obs_kernel_param *p;
 417
 418         for (p = __setup_start; p < __setup_end; p++) {
 419                 if (p->early && strcmp(param, p->str) == 0) {
 420                         if (p->setup_func(val) != 0)
 421                                 printk(KERN_WARNING
 422                                        "Malformed early option '%s'\n", param);
 423                 }
 424         }
 425         /* We accept everything at this stage. */
 426         return 0;
 427 }
 428
 429 /* Arch code calls this early on, or if not, just before other parsing. */
 430 void __init parse_early_param(void)
 431 {
 432         static __initdata int done = 0;
 433         static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
 434
 435         if (done)
 436                 return;
 437
 438         /* All fall through to do_early_param. */
 439         strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
 440         parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
 441         done = 1;
 442 }
 443
 444 /*
 445  *      Activate the first processor.
 446  */
 447
 448 asmlinkage void __init start_kernel(void)
 449 {
 450         char * command_line;
 451         extern struct kernel_param __start___param[], __stop___param[];
 452 /*
 453  * Interrupts are still disabled. Do necessary setups, then
 454  * enable them
 455  */
 456         lock_kernel();
 457         page_address_init();
 458         printk(KERN_NOTICE);
 459         printk(linux_banner);
 460         setup_arch(&command_line);
 461         setup_per_cpu_areas();
 462
 463         /*
 464          * Mark the boot cpu "online" so that it can call console drivers in
 465          * printk() and can access its per-cpu storage.
 466          */
 467         smp_prepare_boot_cpu();
 468
 469         /*
 470          * Set up the scheduler prior starting any interrupts (such as the
 471          * timer interrupt). Full topology setup happens at smp_init()
 472          * time - but meanwhile we still have a functioning scheduler.
 473          */
 474         sched_init();
 475         /*
 476          * Disable preemption - early bootup scheduling is extremely
 477          * fragile until we cpu_idle() for the first time.
 478          */
 479         preempt_disable();
 480         build_all_zonelists();
 481         page_alloc_init();
 482         printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
 483         parse_early_param();
 484         parse_args("Booting kernel", command_line, __start___param,
 485                    __stop___param - __start___param,
 486                    &unknown_bootoption);
 487         sort_main_extable();
 488         trap_init();
 489         rcu_init();
 490         init_IRQ();
 491         pidhash_init();
 492         init_timers();
 493         softirq_init();
 494         time_init();
 495
 496         /*
 497          * HACK ALERT! This is early. We're enabling the console before
 498          * we've done PCI setups etc, and console_init() must be aware of
 499          * this. But we do want output early, in case something goes wrong.
 500          */
 501         console_init();
 502         if (panic_later)
 503                 panic(panic_later, panic_param);
 504         profile_init();
 505         local_irq_enable();
 506 #ifdef CONFIG_BLK_DEV_INITRD
 507         if (initrd_start && !initrd_below_start_ok &&
 508                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 509                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 510                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 511                 initrd_start = 0;
 512         }
 513 #endif
 514         vfs_caches_init_early();
 515         mem_init();
 516         kmem_cache_init();
 517         setup_per_cpu_pageset();
 518         numa_policy_init();
 519         if (late_time_init)
 520                 late_time_init();
 521         calibrate_delay();
 522         pidmap_init();
 523         pgtable_cache_init();
 524         prio_tree_init();
 525         anon_vma_init();
 526 #ifdef CONFIG_X86
 527         if (efi_enabled)
 528                 efi_enter_virtual_mode();
 529 #endif
 530         fork_init(num_physpages);
 531         proc_caches_init();
 532         buffer_init();
 533         unnamed_dev_init();
 534         key_init();
 535         security_init();
 536         vfs_caches_init(num_physpages);
 537         radix_tree_init();
 538         signals_init();
 539         /* rootfs populating might need page-writeback */
 540         page_writeback_init();
 541 #ifdef CONFIG_PROC_FS
 542         proc_root_init();
 543 #endif
 544         cpuset_init();
 545
 546         check_bugs();
 547
 548         acpi_early_init(); /* before LAPIC and SMP init */
 549
 550         /* Do the rest non-__init'ed, we're now alive */
 551         rest_init();
 552 }
 553
 554 static int __initdata initcall_debug;
 555
 556 static int __init initcall_debug_setup(char *str)
 557 {
 558         initcall_debug = 1;
 559         return 1;
 560 }
 561 __setup("initcall_debug", initcall_debug_setup);
 562
 563 struct task_struct *child_reaper = &init_task;
 564
 565 extern initcall_t __initcall_start[], __initcall_end[];
 566
 567 static void __init do_initcalls(void)
 568 {
 569         initcall_t *call;
 570         int count = preempt_count();
 571
 572         for (call = __initcall_start; call < __initcall_end; call++) {
 573                 char *msg;
 574
 575                 if (initcall_debug) {
 576                         printk(KERN_DEBUG "Calling initcall 0x%p", *call);
 577                         print_fn_descriptor_symbol(": %s()", (unsigned long) *call);
 578                         printk("\n");
 579                 }
 580
 581                 (*call)();
 582
 583                 msg = NULL;
 584                 if (preempt_count() != count) {
 585                         msg = "preemption imbalance";
 586                         preempt_count() = count;
 587                 }
 588                 if (irqs_disabled()) {
 589                         msg = "disabled interrupts";
 590                         local_irq_enable();
 591                 }
 592                 if (msg) {
 593                         printk(KERN_WARNING "error in initcall at 0x%p: "
 594                                 "returned with %s\n", *call, msg);
 595                 }
 596         }
 597
 598         /* Make sure there is no pending stuff from the initcall sequence */
 599         flush_scheduled_work();
 600 }
 601
 602 /*
 603  * Ok, the machine is now initialized. None of the devices
 604  * have been touched yet, but the CPU subsystem is up and
 605  * running, and memory and process management works.
 606  *
 607  * Now we can finally start doing some real work..
 608  */
 609 static void __init do_basic_setup(void)
 610 {
 611         /* drivers will send hotplug events */
 612         init_workqueues();
 613         usermodehelper_init();
 614         driver_init();
 615
 616 #ifdef CONFIG_SYSCTL
 617         sysctl_init();
 618 #endif
 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         mark_rodata_ro();
 716         system_state = SYSTEM_RUNNING;
 717         numa_default_policy();
 718
 719         if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
 720                 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
 721
 722         (void) sys_dup(0);
 723         (void) sys_dup(0);
 724
 725         if (ramdisk_execute_command) {
 726                 run_init_process(ramdisk_execute_command);
 727                 printk(KERN_WARNING "Failed to execute %s\n",
 728                                 ramdisk_execute_command);
 729         }
 730
 731         /*
 732          * We try each of these until one succeeds.
 733          *
 734          * The Bourne shell can be used instead of init if we are
 735          * trying to recover a really broken machine.
 736          */
 737         if (execute_command) {
 738                 run_init_process(execute_command);
 739                 printk(KERN_WARNING "Failed to execute %s.  Attempting "
 740                                         "defaults...\n", execute_command);
 741         }
 742         run_init_process("/sbin/init");
 743         run_init_process("/etc/init");
 744         run_init_process("/bin/init");
 745         run_init_process("/bin/sh");
 746
 747         panic("No init found.  Try passing init= option to kernel.");
 748 }