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/proc_fs.h>
  16 #include <linux/devfs_fs_kernel.h>
  17 #include <linux/kernel.h>
  18 #include <linux/unistd.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/kernel_stat.h>
  33 #include <linux/security.h>
  34 #include <linux/workqueue.h>
  35 #include <linux/profile.h>
  36 #include <linux/rcupdate.h>
  37 #include <linux/moduleparam.h>
  38 #include <linux/writeback.h>
  39 #include <linux/cpu.h>
  40
  41 #include <asm/io.h>
  42 #include <asm/bugs.h>
  43
  44 /*
  45  * This is one of the first .c files built. Error out early
  46  * if we have compiler trouble..
  47  */
  48 #if __GNUC__ == 2 && __GNUC_MINOR__ == 96
  49 #ifdef CONFIG_FRAME_POINTER
  50 #error This compiler cannot compile correctly with frame pointers enabled
  51 #endif
  52 #endif
  53
  54 #ifdef CONFIG_X86_LOCAL_APIC
  55 #include <asm/smp.h>
  56 #endif
  57
  58 /*
  59  * Versions of gcc older than that listed below may actually compile
  60  * and link okay, but the end product can have subtle run time bugs.
  61  * To avoid associated bogus bug reports, we flatly refuse to compile
  62  * with a gcc that is known to be too old from the very beginning.
  63  */
  64 #if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 95)
  65 #error Sorry, your GCC is too old. It builds incorrect kernels.
  66 #endif
  67
  68 extern char *linux_banner;
  69
  70 static int init(void *);
  71
  72 extern void init_IRQ(void);
  73 extern void sock_init(void);
  74 extern void fork_init(unsigned long);
  75 extern void mca_init(void);
  76 extern void sbus_init(void);
  77 extern void sysctl_init(void);
  78 extern void signals_init(void);
  79 extern void buffer_init(void);
  80 extern void pidhash_init(void);
  81 extern void pidmap_init(void);
  82 extern void pte_chain_init(void);
  83 extern void radix_tree_init(void);
  84 extern void free_initmem(void);
  85 extern void populate_rootfs(void);
  86 extern void driver_init(void);
  87
  88 #ifdef CONFIG_TC
  89 extern void tc_init(void);
  90 #endif
  91
  92 /*
  93  * Are we up and running (ie do we have all the infrastructure
  94  * set up)
  95  */
  96 int system_running = 0;
  97
  98 /*
  99  * Boot command-line arguments
 100  */
 101 #define MAX_INIT_ARGS 8
 102 #define MAX_INIT_ENVS 8
 103
 104 extern void time_init(void);
 105 /* Default late time init is NULL. archs can override this later. */
 106 void (*late_time_init)(void) = NULL;
 107 extern void softirq_init(void);
 108
 109 int rows, cols;
 110
 111 char *execute_command;
 112
 113 /* Setup configured maximum number of CPUs to activate */
 114 static unsigned int max_cpus = NR_CPUS;
 115
 116 /*
 117  * Setup routine for controlling SMP activation
 118  *
 119  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
 120  * activation entirely (the MPS table probe still happens, though).
 121  *
 122  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
 123  * greater than 0, limits the maximum number of CPUs activated in
 124  * SMP mode to <NUM>.
 125  */
 126 static int __init nosmp(char *str)
 127 {
 128         max_cpus = 0;
 129         return 1;
 130 }
 131
 132 __setup("nosmp", nosmp);
 133
 134 static int __init maxcpus(char *str)
 135 {
 136         get_option(&str, &max_cpus);
 137         return 1;
 138 }
 139
 140 __setup("maxcpus=", maxcpus);
 141
 142 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 143 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 144
 145 __setup("profile=", profile_setup);
 146
 147 static int __init obsolete_checksetup(char *line)
 148 {
 149         struct obs_kernel_param *p;
 150         extern struct obs_kernel_param __setup_start, __setup_end;
 151
 152         p = &__setup_start;
 153         do {
 154                 int n = strlen(p->str);
 155                 if (!strncmp(line,p->str,n)) {
 156                         if (p->setup_func(line+n))
 157                                 return 1;
 158                 }
 159                 p++;
 160         } while (p < &__setup_end);
 161         return 0;
 162 }
 163
 164 /* this should be approx 2 Bo*oMips to start (note initial shift), and will
 165    still work even if initially too large, it will just take slightly longer */
 166 unsigned long loops_per_jiffy = (1<<12);
 167
 168 /* This is the number of bits of precision for the loops_per_jiffy.  Each
 169    bit takes on average 1.5/HZ seconds.  This (like the original) is a little
 170    better than 1% */
 171 #define LPS_PREC 8
 172
 173 void __init calibrate_delay(void)
 174 {
 175         unsigned long ticks, loopbit;
 176         int lps_precision = LPS_PREC;
 177
 178         loops_per_jiffy = (1<<12);
 179
 180         printk("Calibrating delay loop... ");
 181         while (loops_per_jiffy <<= 1) {
 182                 /* wait for "start of" clock tick */
 183                 ticks = jiffies;
 184                 while (ticks == jiffies)
 185                         /* nothing */;
 186                 /* Go .. */
 187                 ticks = jiffies;
 188                 __delay(loops_per_jiffy);
 189                 ticks = jiffies - ticks;
 190                 if (ticks)
 191                         break;
 192         }
 193
 194 /* Do a binary approximation to get loops_per_jiffy set to equal one clock
 195    (up to lps_precision bits) */
 196         loops_per_jiffy >>= 1;
 197         loopbit = loops_per_jiffy;
 198         while ( lps_precision-- && (loopbit >>= 1) ) {
 199                 loops_per_jiffy |= loopbit;
 200                 ticks = jiffies;
 201                 while (ticks == jiffies);
 202                 ticks = jiffies;
 203                 __delay(loops_per_jiffy);
 204                 if (jiffies != ticks)   /* longer than 1 tick */
 205                         loops_per_jiffy &= ~loopbit;
 206         }
 207
 208 /* Round the value and print it */
 209         printk("%lu.%02lu BogoMIPS\n",
 210                 loops_per_jiffy/(500000/HZ),
 211                 (loops_per_jiffy/(5000/HZ)) % 100);
 212 }
 213
 214 static int __init debug_kernel(char *str)
 215 {
 216         if (*str)
 217                 return 0;
 218         console_loglevel = 10;
 219         return 1;
 220 }
 221
 222 static int __init quiet_kernel(char *str)
 223 {
 224         if (*str)
 225                 return 0;
 226         console_loglevel = 4;
 227         return 1;
 228 }
 229
 230 __setup("debug", debug_kernel);
 231 __setup("quiet", quiet_kernel);
 232
 233 /* Unknown boot options get handed to init, unless they look like
 234    failed parameters */
 235 static int __init unknown_bootoption(char *param, char *val)
 236 {
 237         /* Change NUL term back to "=", to make "param" the whole string. */
 238         if (val)
 239                 val[-1] = '=';
 240
 241         /* Handle obsolete-style parameters */
 242         if (obsolete_checksetup(param))
 243                 return 0;
 244
 245         /* Preemptive maintenance for "why didn't my mispelled command
 246            line work?" */
 247         if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
 248                 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
 249                 return 0;
 250         }
 251
 252         if (val) {
 253                 /* Environment option */
 254                 unsigned int i;
 255                 for (i = 0; envp_init[i]; i++) {
 256                         if (i == MAX_INIT_ENVS)
 257                                 panic("Too many boot env vars at `%s'", param);
 258                 }
 259                 envp_init[i] = param;
 260         } else {
 261                 /* Command line option */
 262                 unsigned int i;
 263                 for (i = 0; argv_init[i]; i++) {
 264                         if (i == MAX_INIT_ARGS)
 265                                 panic("Too many boot init vars at `%s'",param);
 266                 }
 267                 argv_init[i] = param;
 268         }
 269         return 0;
 270 }
 271
 272 static int __init init_setup(char *str)
 273 {
 274         unsigned int i;
 275
 276         execute_command = str;
 277         /* In case LILO is going to boot us with default command line,
 278          * it prepends "auto" before the whole cmdline which makes
 279          * the shell think it should execute a script with such name.
 280          * So we ignore all arguments entered _before_ init=... [MJ]
 281          */
 282         for (i = 1; i < MAX_INIT_ARGS; i++)
 283                 argv_init[i] = NULL;
 284         return 1;
 285 }
 286 __setup("init=", init_setup);
 287
 288 extern void setup_arch(char **);
 289 extern void cpu_idle(void);
 290
 291 #ifndef CONFIG_SMP
 292
 293 #ifdef CONFIG_X86_LOCAL_APIC
 294 static void __init smp_init(void)
 295 {
 296         APIC_init_uniprocessor();
 297 }
 298 #else
 299 #define smp_init()      do { } while (0)
 300 #endif
 301
 302 static inline void setup_per_cpu_areas(void) { }
 303 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 304
 305 #else
 306
 307 #ifdef __GENERIC_PER_CPU
 308 unsigned long __per_cpu_offset[NR_CPUS];
 309
 310 static void __init setup_per_cpu_areas(void)
 311 {
 312         unsigned long size, i;
 313         char *ptr;
 314         /* Created by linker magic */
 315         extern char __per_cpu_start[], __per_cpu_end[];
 316
 317         /* Copy section for each CPU (we discard the original) */
 318         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 319 #ifdef CONFIG_MODULES
 320         if (size < PERCPU_ENOUGH_ROOM)
 321                 size = PERCPU_ENOUGH_ROOM;
 322 #endif
 323
 324         ptr = alloc_bootmem(size * NR_CPUS);
 325
 326         for (i = 0; i < NR_CPUS; i++, ptr += size) {
 327                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 328                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 329         }
 330 }
 331 #endif /* !__GENERIC_PER_CPU */
 332
 333 /* Called by boot processor to activate the rest. */
 334 static void __init smp_init(void)
 335 {
 336         unsigned int i;
 337
 338         /* FIXME: This should be done in userspace --RR */
 339         for (i = 0; i < NR_CPUS; i++) {
 340                 if (num_online_cpus() >= max_cpus)
 341                         break;
 342                 if (cpu_possible(i) && !cpu_online(i)) {
 343                         printk("Bringing up %i\n", i);
 344                         cpu_up(i);
 345                 }
 346         }
 347
 348         /* Any cleanup work */
 349         printk("CPUS done %u\n", max_cpus);
 350         smp_cpus_done(max_cpus);
 351 #if 0
 352         /* Get other processors into their bootup holding patterns. */
 353
 354         smp_threads_ready=1;
 355         smp_commence();
 356 #endif
 357 }
 358
 359 #endif
 360
 361 /*
 362  * We need to finalize in a non-__init function or else race conditions
 363  * between the root thread and the init thread may cause start_kernel to
 364  * be reaped by free_initmem before the root thread has proceeded to
 365  * cpu_idle.
 366  */
 367
 368 static void rest_init(void)
 369 {
 370         kernel_thread(init, NULL, CLONE_KERNEL);
 371         unlock_kernel();
 372         cpu_idle();
 373 }
 374
 375 /*
 376  *      Activate the first processor.
 377  */
 378
 379 asmlinkage void __init start_kernel(void)
 380 {
 381         char * command_line;
 382         extern char saved_command_line[];
 383         extern struct kernel_param __start___param[], __stop___param[];
 384 /*
 385  * Interrupts are still disabled. Do necessary setups, then
 386  * enable them
 387  */
 388         lock_kernel();
 389         printk(linux_banner);
 390         setup_arch(&command_line);
 391         setup_per_zone_pages_min();
 392         setup_per_cpu_areas();
 393
 394         /*
 395          * Mark the boot cpu "online" so that it can call console drivers in
 396          * printk() and can access its per-cpu storage.
 397          */
 398         smp_prepare_boot_cpu();
 399
 400         build_all_zonelists();
 401         page_alloc_init();
 402         printk("Kernel command line: %s\n", saved_command_line);
 403         parse_args("Booting kernel", command_line, __start___param,
 404                    __stop___param - __start___param,
 405                    &unknown_bootoption);
 406         trap_init();
 407         rcu_init();
 408         init_IRQ();
 409         pidhash_init();
 410         sched_init();
 411         softirq_init();
 412         time_init();
 413
 414         /*
 415          * HACK ALERT! This is early. We're enabling the console before
 416          * we've done PCI setups etc, and console_init() must be aware of
 417          * this. But we do want output early, in case something goes wrong.
 418          */
 419         console_init();
 420         profile_init();
 421         local_irq_enable();
 422 #ifdef CONFIG_BLK_DEV_INITRD
 423         if (initrd_start && !initrd_below_start_ok &&
 424                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 425                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 426                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 427                 initrd_start = 0;
 428         }
 429 #endif
 430         page_address_init();
 431         mem_init();
 432         kmem_cache_init();
 433         if (late_time_init)
 434                 late_time_init();
 435         calibrate_delay();
 436         pidmap_init();
 437         pgtable_cache_init();
 438         pte_chain_init();
 439         fork_init(num_physpages);
 440         proc_caches_init();
 441         buffer_init();
 442         security_scaffolding_startup();
 443         vfs_caches_init(num_physpages);
 444         radix_tree_init();
 445         signals_init();
 446         /* rootfs populating might need page-writeback */
 447         page_writeback_init();
 448         populate_rootfs();
 449 #ifdef CONFIG_PROC_FS
 450         proc_root_init();
 451 #endif
 452         check_bugs();
 453         printk("POSIX conformance testing by UNIFIX\n");
 454
 455         /*
 456          *      We count on the initial thread going ok
 457          *      Like idlers init is an unlocked kernel thread, which will
 458          *      make syscalls (and thus be locked).
 459          */
 460         init_idle(current, smp_processor_id());
 461
 462         /* Do the rest non-__init'ed, we're now alive */
 463         rest_init();
 464 }
 465
 466 int __initdata initcall_debug;
 467
 468 static int __init initcall_debug_setup(char *str)
 469 {
 470         initcall_debug = 1;
 471         return 1;
 472 }
 473 __setup("initcall_debug", initcall_debug_setup);
 474
 475 struct task_struct *child_reaper = &init_task;
 476
 477 extern initcall_t __initcall_start, __initcall_end;
 478
 479 static void __init do_initcalls(void)
 480 {
 481         initcall_t *call;
 482         int count = preempt_count();
 483
 484         for (call = &__initcall_start; call < &__initcall_end; call++) {
 485                 char *msg;
 486
 487                 if (initcall_debug)
 488                         printk("calling initcall 0x%p\n", *call);
 489
 490                 (*call)();
 491
 492                 msg = NULL;
 493                 if (preempt_count() != count) {
 494                         msg = "preemption imbalance";
 495                         preempt_count() = count;
 496                 }
 497                 if (irqs_disabled()) {
 498                         msg = "disabled interrupts";
 499                         local_irq_enable();
 500                 }
 501                 if (msg) {
 502                         printk("error in initcall at 0x%p: "
 503                                 "returned with %s\n", *call, msg);
 504                 }
 505         }
 506
 507         /* Make sure there is no pending stuff from the initcall sequence */
 508         flush_scheduled_work();
 509 }
 510
 511 /*
 512  * Ok, the machine is now initialized. None of the devices
 513  * have been touched yet, but the CPU subsystem is up and
 514  * running, and memory and process management works.
 515  *
 516  * Now we can finally start doing some real work..
 517  */
 518 static void __init do_basic_setup(void)
 519 {
 520         driver_init();
 521
 522 #ifdef CONFIG_SYSCTL
 523         sysctl_init();
 524 #endif
 525
 526         /* Networking initialization needs a process context */
 527         sock_init();
 528
 529         init_workqueues();
 530         do_initcalls();
 531 }
 532
 533 static void do_pre_smp_initcalls(void)
 534 {
 535         extern int spawn_ksoftirqd(void);
 536 #ifdef CONFIG_SMP
 537         extern int migration_init(void);
 538
 539         migration_init();
 540 #endif
 541         node_nr_running_init();
 542         spawn_ksoftirqd();
 543 }
 544
 545 static void run_init_process(char *init_filename)
 546 {
 547         argv_init[0] = init_filename;
 548         execve(init_filename, argv_init, envp_init);
 549 }
 550
 551 extern void prepare_namespace(void);
 552
 553 static int init(void * unused)
 554 {
 555         lock_kernel();
 556         /*
 557          * Tell the world that we're going to be the grim
 558          * reaper of innocent orphaned children.
 559          *
 560          * We don't want people to have to make incorrect
 561          * assumptions about where in the task array this
 562          * can be found.
 563          */
 564         child_reaper = current;
 565
 566         /* Sets up cpus_possible() */
 567         smp_prepare_cpus(max_cpus);
 568
 569         do_pre_smp_initcalls();
 570
 571         smp_init();
 572         do_basic_setup();
 573
 574         prepare_namespace();
 575
 576         /*
 577          * Ok, we have completed the initial bootup, and
 578          * we're essentially up and running. Get rid of the
 579          * initmem segments and start the user-mode stuff..
 580          */
 581         free_initmem();
 582         unlock_kernel();
 583         system_running = 1;
 584
 585         if (open("/dev/console", O_RDWR, 0) < 0)
 586                 printk("Warning: unable to open an initial console.\n");
 587
 588         (void) dup(0);
 589         (void) dup(0);
 590
 591         /*
 592          * We try each of these until one succeeds.
 593          *
 594          * The Bourne shell can be used instead of init if we are
 595          * trying to recover a really broken machine.
 596          */
 597
 598         if (execute_command)
 599                 run_init_process(execute_command);
 600
 601         run_init_process("/sbin/init");
 602         run_init_process("/etc/init");
 603         run_init_process("/bin/init");
 604         run_init_process("/bin/sh");
 605
 606         panic("No init found.  Try passing init= option to kernel.");
 607 }