[PATCH] per-task-delay-accounting: setup
[deliverable/linux.git] / 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/delayacct.h>
45 #include <linux/unistd.h>
46 #include <linux/rmap.h>
47 #include <linux/mempolicy.h>
48 #include <linux/key.h>
49 #include <linux/unwind.h>
50 #include <linux/buffer_head.h>
51 #include <linux/debug_locks.h>
52 #include <linux/lockdep.h>
53
54 #include <asm/io.h>
55 #include <asm/bugs.h>
56 #include <asm/setup.h>
57 #include <asm/sections.h>
58 #include <asm/cacheflush.h>
59
60 #ifdef CONFIG_X86_LOCAL_APIC
61 #include <asm/smp.h>
62 #endif
63
64 /*
65 * This is one of the first .c files built. Error out early if we have compiler
66 * trouble.
67 *
68 * Versions of gcc older than that listed below may actually compile and link
69 * okay, but the end product can have subtle run time bugs. To avoid associated
70 * bogus bug reports, we flatly refuse to compile with a gcc that is known to be
71 * too old from the very beginning.
72 */
73 #if (__GNUC__ < 3) || (__GNUC__ == 3 && __GNUC_MINOR__ < 2)
74 #error Sorry, your GCC is too old. It builds incorrect kernels.
75 #endif
76
77 static int init(void *);
78
79 extern void init_IRQ(void);
80 extern void fork_init(unsigned long);
81 extern void mca_init(void);
82 extern void sbus_init(void);
83 extern void sysctl_init(void);
84 extern void signals_init(void);
85 extern void pidhash_init(void);
86 extern void pidmap_init(void);
87 extern void prio_tree_init(void);
88 extern void radix_tree_init(void);
89 extern void free_initmem(void);
90 extern void populate_rootfs(void);
91 extern void driver_init(void);
92 extern void prepare_namespace(void);
93 #ifdef CONFIG_ACPI
94 extern void acpi_early_init(void);
95 #else
96 static inline void acpi_early_init(void) { }
97 #endif
98 #ifndef CONFIG_DEBUG_RODATA
99 static inline void mark_rodata_ro(void) { }
100 #endif
101
102 #ifdef CONFIG_TC
103 extern void tc_init(void);
104 #endif
105
106 enum system_states system_state;
107 EXPORT_SYMBOL(system_state);
108
109 /*
110 * Boot command-line arguments
111 */
112 #define MAX_INIT_ARGS CONFIG_INIT_ENV_ARG_LIMIT
113 #define MAX_INIT_ENVS CONFIG_INIT_ENV_ARG_LIMIT
114
115 extern void time_init(void);
116 /* Default late time init is NULL. archs can override this later. */
117 void (*late_time_init)(void);
118 extern void softirq_init(void);
119
120 /* Untouched command line (eg. for /proc) saved by arch-specific code. */
121 char saved_command_line[COMMAND_LINE_SIZE];
122
123 static char *execute_command;
124 static char *ramdisk_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 static int __init rdinit_setup(char *str)
300 {
301 unsigned int i;
302
303 ramdisk_execute_command = str;
304 /* See "auto" comment in init_setup */
305 for (i = 1; i < MAX_INIT_ARGS; i++)
306 argv_init[i] = NULL;
307 return 1;
308 }
309 __setup("rdinit=", rdinit_setup);
310
311 #ifndef CONFIG_SMP
312
313 #ifdef CONFIG_X86_LOCAL_APIC
314 static void __init smp_init(void)
315 {
316 APIC_init_uniprocessor();
317 }
318 #else
319 #define smp_init() do { } while (0)
320 #endif
321
322 static inline void setup_per_cpu_areas(void) { }
323 static inline void smp_prepare_cpus(unsigned int maxcpus) { }
324
325 #else
326
327 #ifdef __GENERIC_PER_CPU
328 unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
329
330 EXPORT_SYMBOL(__per_cpu_offset);
331
332 static void __init setup_per_cpu_areas(void)
333 {
334 unsigned long size, i;
335 char *ptr;
336 unsigned long nr_possible_cpus = num_possible_cpus();
337
338 /* Copy section for each CPU (we discard the original) */
339 size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
340 #ifdef CONFIG_MODULES
341 if (size < PERCPU_ENOUGH_ROOM)
342 size = PERCPU_ENOUGH_ROOM;
343 #endif
344 ptr = alloc_bootmem(size * nr_possible_cpus);
345
346 for_each_possible_cpu(i) {
347 __per_cpu_offset[i] = ptr - __per_cpu_start;
348 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
349 ptr += size;
350 }
351 }
352 #endif /* !__GENERIC_PER_CPU */
353
354 /* Called by boot processor to activate the rest. */
355 static void __init smp_init(void)
356 {
357 unsigned int i;
358
359 /* FIXME: This should be done in userspace --RR */
360 for_each_present_cpu(i) {
361 if (num_online_cpus() >= max_cpus)
362 break;
363 if (!cpu_online(i))
364 cpu_up(i);
365 }
366
367 /* Any cleanup work */
368 printk(KERN_INFO "Brought up %ld CPUs\n", (long)num_online_cpus());
369 smp_cpus_done(max_cpus);
370 #if 0
371 /* Get other processors into their bootup holding patterns. */
372
373 smp_commence();
374 #endif
375 }
376
377 #endif
378
379 /*
380 * We need to finalize in a non-__init function or else race conditions
381 * between the root thread and the init thread may cause start_kernel to
382 * be reaped by free_initmem before the root thread has proceeded to
383 * cpu_idle.
384 *
385 * gcc-3.4 accidentally inlines this function, so use noinline.
386 */
387
388 static void noinline rest_init(void)
389 __releases(kernel_lock)
390 {
391 kernel_thread(init, NULL, CLONE_FS | CLONE_SIGHAND);
392 numa_default_policy();
393 unlock_kernel();
394
395 /*
396 * The boot idle thread must execute schedule()
397 * at least one to get things moving:
398 */
399 preempt_enable_no_resched();
400 schedule();
401 preempt_disable();
402
403 /* Call into cpu_idle with preempt disabled */
404 cpu_idle();
405 }
406
407 /* Check for early params. */
408 static int __init do_early_param(char *param, char *val)
409 {
410 struct obs_kernel_param *p;
411
412 for (p = __setup_start; p < __setup_end; p++) {
413 if (p->early && strcmp(param, p->str) == 0) {
414 if (p->setup_func(val) != 0)
415 printk(KERN_WARNING
416 "Malformed early option '%s'\n", param);
417 }
418 }
419 /* We accept everything at this stage. */
420 return 0;
421 }
422
423 /* Arch code calls this early on, or if not, just before other parsing. */
424 void __init parse_early_param(void)
425 {
426 static __initdata int done = 0;
427 static __initdata char tmp_cmdline[COMMAND_LINE_SIZE];
428
429 if (done)
430 return;
431
432 /* All fall through to do_early_param. */
433 strlcpy(tmp_cmdline, saved_command_line, COMMAND_LINE_SIZE);
434 parse_args("early options", tmp_cmdline, NULL, 0, do_early_param);
435 done = 1;
436 }
437
438 /*
439 * Activate the first processor.
440 */
441
442 static void __init boot_cpu_init(void)
443 {
444 int cpu = smp_processor_id();
445 /* Mark the boot cpu "present", "online" etc for SMP and UP case */
446 cpu_set(cpu, cpu_online_map);
447 cpu_set(cpu, cpu_present_map);
448 cpu_set(cpu, cpu_possible_map);
449 }
450
451 void __init __attribute__((weak)) smp_setup_processor_id(void)
452 {
453 }
454
455 asmlinkage void __init start_kernel(void)
456 {
457 char * command_line;
458 extern struct kernel_param __start___param[], __stop___param[];
459
460 smp_setup_processor_id();
461
462 /*
463 * Need to run as early as possible, to initialize the
464 * lockdep hash:
465 */
466 lockdep_init();
467
468 local_irq_disable();
469 early_boot_irqs_off();
470 early_init_irq_lock_class();
471
472 /*
473 * Interrupts are still disabled. Do necessary setups, then
474 * enable them
475 */
476 lock_kernel();
477 boot_cpu_init();
478 page_address_init();
479 printk(KERN_NOTICE);
480 printk(linux_banner);
481 setup_arch(&command_line);
482 setup_per_cpu_areas();
483 smp_prepare_boot_cpu(); /* arch-specific boot-cpu hooks */
484
485 /*
486 * Set up the scheduler prior starting any interrupts (such as the
487 * timer interrupt). Full topology setup happens at smp_init()
488 * time - but meanwhile we still have a functioning scheduler.
489 */
490 sched_init();
491 /*
492 * Disable preemption - early bootup scheduling is extremely
493 * fragile until we cpu_idle() for the first time.
494 */
495 preempt_disable();
496 build_all_zonelists();
497 page_alloc_init();
498 printk(KERN_NOTICE "Kernel command line: %s\n", saved_command_line);
499 parse_early_param();
500 parse_args("Booting kernel", command_line, __start___param,
501 __stop___param - __start___param,
502 &unknown_bootoption);
503 sort_main_extable();
504 unwind_init();
505 trap_init();
506 rcu_init();
507 init_IRQ();
508 pidhash_init();
509 init_timers();
510 hrtimers_init();
511 softirq_init();
512 timekeeping_init();
513 time_init();
514 profile_init();
515 if (!irqs_disabled())
516 printk("start_kernel(): bug: interrupts were enabled early\n");
517 early_boot_irqs_on();
518 local_irq_enable();
519
520 /*
521 * HACK ALERT! This is early. We're enabling the console before
522 * we've done PCI setups etc, and console_init() must be aware of
523 * this. But we do want output early, in case something goes wrong.
524 */
525 console_init();
526 if (panic_later)
527 panic(panic_later, panic_param);
528
529 lockdep_info();
530
531 /*
532 * Need to run this when irqs are enabled, because it wants
533 * to self-test [hard/soft]-irqs on/off lock inversion bugs
534 * too:
535 */
536 locking_selftest();
537
538 #ifdef CONFIG_BLK_DEV_INITRD
539 if (initrd_start && !initrd_below_start_ok &&
540 initrd_start < min_low_pfn << PAGE_SHIFT) {
541 printk(KERN_CRIT "initrd overwritten (0x%08lx < 0x%08lx) - "
542 "disabling it.\n",initrd_start,min_low_pfn << PAGE_SHIFT);
543 initrd_start = 0;
544 }
545 #endif
546 vfs_caches_init_early();
547 cpuset_init_early();
548 mem_init();
549 kmem_cache_init();
550 setup_per_cpu_pageset();
551 numa_policy_init();
552 if (late_time_init)
553 late_time_init();
554 calibrate_delay();
555 pidmap_init();
556 pgtable_cache_init();
557 prio_tree_init();
558 anon_vma_init();
559 #ifdef CONFIG_X86
560 if (efi_enabled)
561 efi_enter_virtual_mode();
562 #endif
563 fork_init(num_physpages);
564 proc_caches_init();
565 buffer_init();
566 unnamed_dev_init();
567 key_init();
568 security_init();
569 vfs_caches_init(num_physpages);
570 radix_tree_init();
571 signals_init();
572 /* rootfs populating might need page-writeback */
573 page_writeback_init();
574 #ifdef CONFIG_PROC_FS
575 proc_root_init();
576 #endif
577 cpuset_init();
578 delayacct_init();
579
580 check_bugs();
581
582 acpi_early_init(); /* before LAPIC and SMP init */
583
584 /* Do the rest non-__init'ed, we're now alive */
585 rest_init();
586 }
587
588 static int __initdata initcall_debug;
589
590 static int __init initcall_debug_setup(char *str)
591 {
592 initcall_debug = 1;
593 return 1;
594 }
595 __setup("initcall_debug", initcall_debug_setup);
596
597 struct task_struct *child_reaper = &init_task;
598
599 extern initcall_t __initcall_start[], __initcall_end[];
600
601 static void __init do_initcalls(void)
602 {
603 initcall_t *call;
604 int count = preempt_count();
605
606 for (call = __initcall_start; call < __initcall_end; call++) {
607 char *msg = NULL;
608 char msgbuf[40];
609 int result;
610
611 if (initcall_debug) {
612 printk("Calling initcall 0x%p", *call);
613 print_fn_descriptor_symbol(": %s()",
614 (unsigned long) *call);
615 printk("\n");
616 }
617
618 result = (*call)();
619
620 if (result && result != -ENODEV && initcall_debug) {
621 sprintf(msgbuf, "error code %d", result);
622 msg = msgbuf;
623 }
624 if (preempt_count() != count) {
625 msg = "preemption imbalance";
626 preempt_count() = count;
627 }
628 if (irqs_disabled()) {
629 msg = "disabled interrupts";
630 local_irq_enable();
631 }
632 if (msg) {
633 printk(KERN_WARNING "initcall at 0x%p", *call);
634 print_fn_descriptor_symbol(": %s()",
635 (unsigned long) *call);
636 printk(": returned with %s\n", msg);
637 }
638 }
639
640 /* Make sure there is no pending stuff from the initcall sequence */
641 flush_scheduled_work();
642 }
643
644 /*
645 * Ok, the machine is now initialized. None of the devices
646 * have been touched yet, but the CPU subsystem is up and
647 * running, and memory and process management works.
648 *
649 * Now we can finally start doing some real work..
650 */
651 static void __init do_basic_setup(void)
652 {
653 /* drivers will send hotplug events */
654 init_workqueues();
655 usermodehelper_init();
656 driver_init();
657
658 #ifdef CONFIG_SYSCTL
659 sysctl_init();
660 #endif
661
662 do_initcalls();
663 }
664
665 static void do_pre_smp_initcalls(void)
666 {
667 extern int spawn_ksoftirqd(void);
668 #ifdef CONFIG_SMP
669 extern int migration_init(void);
670
671 migration_init();
672 #endif
673 spawn_ksoftirqd();
674 spawn_softlockup_task();
675 }
676
677 static void run_init_process(char *init_filename)
678 {
679 argv_init[0] = init_filename;
680 execve(init_filename, argv_init, envp_init);
681 }
682
683 static int init(void * unused)
684 {
685 lock_kernel();
686 /*
687 * init can run on any cpu.
688 */
689 set_cpus_allowed(current, CPU_MASK_ALL);
690 /*
691 * Tell the world that we're going to be the grim
692 * reaper of innocent orphaned children.
693 *
694 * We don't want people to have to make incorrect
695 * assumptions about where in the task array this
696 * can be found.
697 */
698 child_reaper = current;
699
700 smp_prepare_cpus(max_cpus);
701
702 do_pre_smp_initcalls();
703
704 smp_init();
705 sched_init_smp();
706
707 cpuset_init_smp();
708
709 /*
710 * Do this before initcalls, because some drivers want to access
711 * firmware files.
712 */
713 populate_rootfs();
714
715 do_basic_setup();
716
717 /*
718 * check if there is an early userspace init. If yes, let it do all
719 * the work
720 */
721
722 if (!ramdisk_execute_command)
723 ramdisk_execute_command = "/init";
724
725 if (sys_access((const char __user *) ramdisk_execute_command, 0) != 0) {
726 ramdisk_execute_command = NULL;
727 prepare_namespace();
728 }
729
730 /*
731 * Ok, we have completed the initial bootup, and
732 * we're essentially up and running. Get rid of the
733 * initmem segments and start the user-mode stuff..
734 */
735 free_initmem();
736 unlock_kernel();
737 mark_rodata_ro();
738 system_state = SYSTEM_RUNNING;
739 numa_default_policy();
740
741 if (sys_open((const char __user *) "/dev/console", O_RDWR, 0) < 0)
742 printk(KERN_WARNING "Warning: unable to open an initial console.\n");
743
744 (void) sys_dup(0);
745 (void) sys_dup(0);
746
747 if (ramdisk_execute_command) {
748 run_init_process(ramdisk_execute_command);
749 printk(KERN_WARNING "Failed to execute %s\n",
750 ramdisk_execute_command);
751 }
752
753 /*
754 * We try each of these until one succeeds.
755 *
756 * The Bourne shell can be used instead of init if we are
757 * trying to recover a really broken machine.
758 */
759 if (execute_command) {
760 run_init_process(execute_command);
761 printk(KERN_WARNING "Failed to execute %s. Attempting "
762 "defaults...\n", execute_command);
763 }
764 run_init_process("/sbin/init");
765 run_init_process("/etc/init");
766 run_init_process("/bin/init");
767 run_init_process("/bin/sh");
768
769 panic("No init found. Try passing init= option to kernel.");
770 }
This page took 0.055019 seconds and 5 git commands to generate.