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/module.h>
  16 #include <linux/proc_fs.h>
  17 #include <linux/devfs_fs_kernel.h>
  18 #include <linux/kernel.h>
  19 #include <linux/unistd.h>
  20 #include <linux/string.h>
  21 #include <linux/ctype.h>
  22 #include <linux/delay.h>
  23 #include <linux/utsname.h>
  24 #include <linux/ioport.h>
  25 #include <linux/init.h>
  26 #include <linux/smp_lock.h>
  27 #include <linux/initrd.h>
  28 #include <linux/hdreg.h>
  29 #include <linux/bootmem.h>
  30 #include <linux/tty.h>
  31 #include <linux/gfp.h>
  32 #include <linux/percpu.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/writeback.h>
  40 #include <linux/cpu.h>
  41
  42 #include <asm/io.h>
  43 #include <asm/bugs.h>
  44
  45 /*
  46  * This is one of the first .c files built. Error out early
  47  * if we have compiler trouble..
  48  */
  49 #if __GNUC__ == 2 && __GNUC_MINOR__ == 96
  50 #ifdef CONFIG_FRAME_POINTER
  51 #error This compiler cannot compile correctly with frame pointers enabled
  52 #endif
  53 #endif
  54
  55 #ifdef CONFIG_X86_LOCAL_APIC
  56 #include <asm/smp.h>
  57 #endif
  58
  59 /*
  60  * Versions of gcc older than that listed below may actually compile
  61  * and link okay, but the end product can have subtle run time bugs.
  62  * To avoid associated bogus bug reports, we flatly refuse to compile
  63  * with a gcc that is known to be too old from the very beginning.
  64  */
  65 #if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 95)
  66 #error Sorry, your GCC is too old. It builds incorrect kernels.
  67 #endif
  68
  69 extern char *linux_banner;
  70
  71 static int init(void *);
  72
  73 extern void init_IRQ(void);
  74 extern void sock_init(void);
  75 extern void fork_init(unsigned long);
  76 extern void mca_init(void);
  77 extern void sbus_init(void);
  78 extern void sysctl_init(void);
  79 extern void signals_init(void);
  80 extern void buffer_init(void);
  81 extern void pidhash_init(void);
  82 extern void pidmap_init(void);
  83 extern void pte_chain_init(void);
  84 extern void radix_tree_init(void);
  85 extern void free_initmem(void);
  86 extern void populate_rootfs(void);
  87 extern void driver_init(void);
  88
  89 #ifdef CONFIG_TC
  90 extern void tc_init(void);
  91 #endif
  92
  93 /*
  94  * Are we up and running (ie do we have all the infrastructure
  95  * set up)
  96  */
  97 int system_running = 0;
  98
  99 /*
 100  * Boot command-line arguments
 101  */
 102 #define MAX_INIT_ARGS 8
 103 #define MAX_INIT_ENVS 8
 104
 105 extern void time_init(void);
 106 /* Default late time init is NULL. archs can override this later. */
 107 void (*late_time_init)(void) = NULL;
 108 extern void softirq_init(void);
 109
 110 int rows, cols;
 111
 112 char *execute_command;
 113
 114 /* Setup configured maximum number of CPUs to activate */
 115 static unsigned int max_cpus = NR_CPUS;
 116
 117 /*
 118  * Setup routine for controlling SMP activation
 119  *
 120  * Command-line option of "nosmp" or "maxcpus=0" will disable SMP
 121  * activation entirely (the MPS table probe still happens, though).
 122  *
 123  * Command-line option of "maxcpus=<NUM>", where <NUM> is an integer
 124  * greater than 0, limits the maximum number of CPUs activated in
 125  * SMP mode to <NUM>.
 126  */
 127 static int __init nosmp(char *str)
 128 {
 129         max_cpus = 0;
 130         return 1;
 131 }
 132
 133 __setup("nosmp", nosmp);
 134
 135 static int __init maxcpus(char *str)
 136 {
 137         get_option(&str, &max_cpus);
 138         return 1;
 139 }
 140
 141 __setup("maxcpus=", maxcpus);
 142
 143 static char * argv_init[MAX_INIT_ARGS+2] = { "init", NULL, };
 144 char * envp_init[MAX_INIT_ENVS+2] = { "HOME=/", "TERM=linux", NULL, };
 145
 146 __setup("profile=", profile_setup);
 147
 148 static int __init obsolete_checksetup(char *line)
 149 {
 150         struct obs_kernel_param *p;
 151         extern struct obs_kernel_param __setup_start, __setup_end;
 152
 153         p = &__setup_start;
 154         do {
 155                 int n = strlen(p->str);
 156                 if (!strncmp(line,p->str,n)) {
 157                         if (p->setup_func(line+n))
 158                                 return 1;
 159                 }
 160                 p++;
 161         } while (p < &__setup_end);
 162         return 0;
 163 }
 164
 165 /* this should be approx 2 Bo*oMips to start (note initial shift), and will
 166    still work even if initially too large, it will just take slightly longer */
 167 unsigned long loops_per_jiffy = (1<<12);
 168
 169 #ifndef __ia64__
 170 EXPORT_SYMBOL(loops_per_jiffy);
 171 #endif
 172
 173 /* This is the number of bits of precision for the loops_per_jiffy.  Each
 174    bit takes on average 1.5/HZ seconds.  This (like the original) is a little
 175    better than 1% */
 176 #define LPS_PREC 8
 177
 178 void __init calibrate_delay(void)
 179 {
 180         unsigned long ticks, loopbit;
 181         int lps_precision = LPS_PREC;
 182
 183         loops_per_jiffy = (1<<12);
 184
 185         printk("Calibrating delay loop... ");
 186         while (loops_per_jiffy <<= 1) {
 187                 /* wait for "start of" clock tick */
 188                 ticks = jiffies;
 189                 while (ticks == jiffies)
 190                         /* nothing */;
 191                 /* Go .. */
 192                 ticks = jiffies;
 193                 __delay(loops_per_jiffy);
 194                 ticks = jiffies - ticks;
 195                 if (ticks)
 196                         break;
 197         }
 198
 199 /* Do a binary approximation to get loops_per_jiffy set to equal one clock
 200    (up to lps_precision bits) */
 201         loops_per_jiffy >>= 1;
 202         loopbit = loops_per_jiffy;
 203         while ( lps_precision-- && (loopbit >>= 1) ) {
 204                 loops_per_jiffy |= loopbit;
 205                 ticks = jiffies;
 206                 while (ticks == jiffies);
 207                 ticks = jiffies;
 208                 __delay(loops_per_jiffy);
 209                 if (jiffies != ticks)   /* longer than 1 tick */
 210                         loops_per_jiffy &= ~loopbit;
 211         }
 212
 213 /* Round the value and print it */
 214         printk("%lu.%02lu BogoMIPS\n",
 215                 loops_per_jiffy/(500000/HZ),
 216                 (loops_per_jiffy/(5000/HZ)) % 100);
 217 }
 218
 219 static int __init debug_kernel(char *str)
 220 {
 221         if (*str)
 222                 return 0;
 223         console_loglevel = 10;
 224         return 1;
 225 }
 226
 227 static int __init quiet_kernel(char *str)
 228 {
 229         if (*str)
 230                 return 0;
 231         console_loglevel = 4;
 232         return 1;
 233 }
 234
 235 __setup("debug", debug_kernel);
 236 __setup("quiet", quiet_kernel);
 237
 238 /* Unknown boot options get handed to init, unless they look like
 239    failed parameters */
 240 static int __init unknown_bootoption(char *param, char *val)
 241 {
 242         /* Change NUL term back to "=", to make "param" the whole string. */
 243         if (val)
 244                 val[-1] = '=';
 245
 246         /* Handle obsolete-style parameters */
 247         if (obsolete_checksetup(param))
 248                 return 0;
 249
 250         /* Preemptive maintenance for "why didn't my mispelled command
 251            line work?" */
 252         if (strchr(param, '.') && (!val || strchr(param, '.') < val)) {
 253                 printk(KERN_ERR "Unknown boot option `%s': ignoring\n", param);
 254                 return 0;
 255         }
 256
 257         if (val) {
 258                 /* Environment option */
 259                 unsigned int i;
 260                 for (i = 0; envp_init[i]; i++) {
 261                         if (i == MAX_INIT_ENVS)
 262                                 panic("Too many boot env vars at `%s'", param);
 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("Too many boot init vars at `%s'",param);
 271                 }
 272                 argv_init[i] = param;
 273         }
 274         return 0;
 275 }
 276
 277 static int __init init_setup(char *str)
 278 {
 279         unsigned int i;
 280
 281         execute_command = str;
 282         /* In case LILO is going to boot us with default command line,
 283          * it prepends "auto" before the whole cmdline which makes
 284          * the shell think it should execute a script with such name.
 285          * So we ignore all arguments entered _before_ init=... [MJ]
 286          */
 287         for (i = 1; i < MAX_INIT_ARGS; i++)
 288                 argv_init[i] = NULL;
 289         return 1;
 290 }
 291 __setup("init=", init_setup);
 292
 293 extern void setup_arch(char **);
 294 extern void cpu_idle(void);
 295
 296 #ifndef CONFIG_SMP
 297
 298 #ifdef CONFIG_X86_LOCAL_APIC
 299 static void __init smp_init(void)
 300 {
 301         APIC_init_uniprocessor();
 302 }
 303 #else
 304 #define smp_init()      do { } while (0)
 305 #endif
 306
 307 static inline void setup_per_cpu_areas(void) { }
 308 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
 309
 310 #else
 311
 312 #ifdef __GENERIC_PER_CPU
 313 unsigned long __per_cpu_offset[NR_CPUS];
 314
 315 EXPORT_SYMBOL(__per_cpu_offset);
 316
 317 static void __init setup_per_cpu_areas(void)
 318 {
 319         unsigned long size, i;
 320         char *ptr;
 321         /* Created by linker magic */
 322         extern char __per_cpu_start[], __per_cpu_end[];
 323
 324         /* Copy section for each CPU (we discard the original) */
 325         size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
 326 #ifdef CONFIG_MODULES
 327         if (size < PERCPU_ENOUGH_ROOM)
 328                 size = PERCPU_ENOUGH_ROOM;
 329 #endif
 330
 331         ptr = alloc_bootmem(size * NR_CPUS);
 332
 333         for (i = 0; i < NR_CPUS; i++, ptr += size) {
 334                 __per_cpu_offset[i] = ptr - __per_cpu_start;
 335                 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
 336         }
 337 }
 338 #endif /* !__GENERIC_PER_CPU */
 339
 340 /* Called by boot processor to activate the rest. */
 341 static void __init smp_init(void)
 342 {
 343         unsigned int i;
 344
 345         /* FIXME: This should be done in userspace --RR */
 346         for (i = 0; i < NR_CPUS; i++) {
 347                 if (num_online_cpus() >= max_cpus)
 348                         break;
 349                 if (cpu_possible(i) && !cpu_online(i)) {
 350                         printk("Bringing up %i\n", i);
 351                         cpu_up(i);
 352                 }
 353         }
 354
 355         /* Any cleanup work */
 356         printk("CPUS done %u\n", max_cpus);
 357         smp_cpus_done(max_cpus);
 358 #if 0
 359         /* Get other processors into their bootup holding patterns. */
 360
 361         smp_threads_ready=1;
 362         smp_commence();
 363 #endif
 364 }
 365
 366 #endif
 367
 368 /*
 369  * We need to finalize in a non-__init function or else race conditions
 370  * between the root thread and the init thread may cause start_kernel to
 371  * be reaped by free_initmem before the root thread has proceeded to
 372  * cpu_idle.
 373  */
 374
 375 static void rest_init(void)
 376 {
 377         kernel_thread(init, NULL, CLONE_KERNEL);
 378         unlock_kernel();
 379         cpu_idle();
 380 }
 381
 382 /*
 383  *      Activate the first processor.
 384  */
 385
 386 asmlinkage void __init start_kernel(void)
 387 {
 388         char * command_line;
 389         extern char saved_command_line[];
 390         extern struct kernel_param __start___param[], __stop___param[];
 391 /*
 392  * Interrupts are still disabled. Do necessary setups, then
 393  * enable them
 394  */
 395         lock_kernel();
 396         printk(linux_banner);
 397         setup_arch(&command_line);
 398         setup_per_zone_pages_min();
 399         setup_per_cpu_areas();
 400
 401         /*
 402          * Mark the boot cpu "online" so that it can call console drivers in
 403          * printk() and can access its per-cpu storage.
 404          */
 405         smp_prepare_boot_cpu();
 406
 407         build_all_zonelists();
 408         page_alloc_init();
 409         printk("Kernel command line: %s\n", saved_command_line);
 410         parse_args("Booting kernel", command_line, __start___param,
 411                    __stop___param - __start___param,
 412                    &unknown_bootoption);
 413         trap_init();
 414         rcu_init();
 415         init_IRQ();
 416         pidhash_init();
 417         sched_init();
 418         softirq_init();
 419         time_init();
 420
 421         /*
 422          * HACK ALERT! This is early. We're enabling the console before
 423          * we've done PCI setups etc, and console_init() must be aware of
 424          * this. But we do want output early, in case something goes wrong.
 425          */
 426         console_init();
 427         profile_init();
 428         local_irq_enable();
 429 #ifdef CONFIG_BLK_DEV_INITRD
 430         if (initrd_start && !initrd_below_start_ok &&
 431                         initrd_start < min_low_pfn << PAGE_SHIFT) {
 432                 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
 433                     "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
 434                 initrd_start = 0;
 435         }
 436 #endif
 437         page_address_init();
 438         mem_init();
 439         kmem_cache_init();
 440         if (late_time_init)
 441                 late_time_init();
 442         calibrate_delay();
 443         pidmap_init();
 444         pgtable_cache_init();
 445         pte_chain_init();
 446         fork_init(num_physpages);
 447         proc_caches_init();
 448         buffer_init();
 449         security_scaffolding_startup();
 450         vfs_caches_init(num_physpages);
 451         radix_tree_init();
 452         signals_init();
 453         /* rootfs populating might need page-writeback */
 454         page_writeback_init();
 455         populate_rootfs();
 456 #ifdef CONFIG_PROC_FS
 457         proc_root_init();
 458 #endif
 459         check_bugs();
 460         printk("POSIX conformance testing by UNIFIX\n");
 461
 462         /*
 463          *      We count on the initial thread going ok
 464          *      Like idlers init is an unlocked kernel thread, which will
 465          *      make syscalls (and thus be locked).
 466          */
 467         init_idle(current, smp_processor_id());
 468
 469         /* Do the rest non-__init'ed, we're now alive */
 470         rest_init();
 471 }
 472
 473 int __initdata initcall_debug;
 474
 475 static int __init initcall_debug_setup(char *str)
 476 {
 477         initcall_debug = 1;
 478         return 1;
 479 }
 480 __setup("initcall_debug", initcall_debug_setup);
 481
 482 struct task_struct *child_reaper = &init_task;
 483
 484 extern initcall_t __initcall_start, __initcall_end;
 485
 486 static void __init do_initcalls(void)
 487 {
 488         initcall_t *call;
 489         int count = preempt_count();
 490
 491         for (call = &__initcall_start; call < &__initcall_end; call++) {
 492                 char *msg;
 493
 494                 if (initcall_debug)
 495                         printk("calling initcall 0x%p\n", *call);
 496
 497                 (*call)();
 498
 499                 msg = NULL;
 500                 if (preempt_count() != count) {
 501                         msg = "preemption imbalance";
 502                         preempt_count() = count;
 503                 }
 504                 if (irqs_disabled()) {
 505                         msg = "disabled interrupts";
 506                         local_irq_enable();
 507                 }
 508                 if (msg) {
 509                         printk("error in initcall at 0x%p: "
 510                                 "returned with %s\n", *call, msg);
 511                 }
 512         }
 513
 514         /* Make sure there is no pending stuff from the initcall sequence */
 515         flush_scheduled_work();
 516 }
 517
 518 /*
 519  * Ok, the machine is now initialized. None of the devices
 520  * have been touched yet, but the CPU subsystem is up and
 521  * running, and memory and process management works.
 522  *
 523  * Now we can finally start doing some real work..
 524  */
 525 static void __init do_basic_setup(void)
 526 {
 527         driver_init();
 528
 529 #ifdef CONFIG_SYSCTL
 530         sysctl_init();
 531 #endif
 532
 533         /* Networking initialization needs a process context */
 534         sock_init();
 535
 536         init_workqueues();
 537         do_initcalls();
 538 }
 539
 540 static void do_pre_smp_initcalls(void)
 541 {
 542         extern int spawn_ksoftirqd(void);
 543 #ifdef CONFIG_SMP
 544         extern int migration_init(void);
 545
 546         migration_init();
 547 #endif
 548         node_nr_running_init();
 549         spawn_ksoftirqd();
 550 }
 551
 552 static void run_init_process(char *init_filename)
 553 {
 554         argv_init[0] = init_filename;
 555         execve(init_filename, argv_init, envp_init);
 556 }
 557
 558 extern void prepare_namespace(void);
 559
 560 static int init(void * unused)
 561 {
 562         lock_kernel();
 563         /*
 564          * Tell the world that we're going to be the grim
 565          * reaper of innocent orphaned children.
 566          *
 567          * We don't want people to have to make incorrect
 568          * assumptions about where in the task array this
 569          * can be found.
 570          */
 571         child_reaper = current;
 572
 573         /* Sets up cpus_possible() */
 574         smp_prepare_cpus(max_cpus);
 575
 576         do_pre_smp_initcalls();
 577
 578         smp_init();
 579         do_basic_setup();
 580
 581         prepare_namespace();
 582
 583         /*
 584          * Ok, we have completed the initial bootup, and
 585          * we're essentially up and running. Get rid of the
 586          * initmem segments and start the user-mode stuff..
 587          */
 588         free_initmem();
 589         unlock_kernel();
 590         system_running = 1;
 591
 592         if (open("/dev/console", O_RDWR, 0) < 0)
 593                 printk("Warning: unable to open an initial console.\n");
 594
 595         (void) dup(0);
 596         (void) dup(0);
 597
 598         /*
 599          * We try each of these until one succeeds.
 600          *
 601          * The Bourne shell can be used instead of init if we are
 602          * trying to recover a really broken machine.
 603          */
 604
 605         if (execute_command)
 606                 run_init_process(execute_command);
 607
 608         run_init_process("/sbin/init");
 609         run_init_process("/etc/init");
 610         run_init_process("/bin/init");
 611         run_init_process("/bin/sh");
 612
 613         panic("No init found.  Try passing init= option to kernel.");
 614 }