]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - kernel/panic.c
Merge tag 'dax-locking-for-4.7' of git://git.kernel.org/pub/scm/linux/kernel/git...
[mirror_ubuntu-artful-kernel.git] / kernel / panic.c
1 /*
2 * linux/kernel/panic.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 /*
8 * This function is used through-out the kernel (including mm and fs)
9 * to indicate a major problem.
10 */
11 #include <linux/debug_locks.h>
12 #include <linux/interrupt.h>
13 #include <linux/kmsg_dump.h>
14 #include <linux/kallsyms.h>
15 #include <linux/notifier.h>
16 #include <linux/module.h>
17 #include <linux/random.h>
18 #include <linux/ftrace.h>
19 #include <linux/reboot.h>
20 #include <linux/delay.h>
21 #include <linux/kexec.h>
22 #include <linux/sched.h>
23 #include <linux/sysrq.h>
24 #include <linux/init.h>
25 #include <linux/nmi.h>
26 #include <linux/console.h>
27 #include <linux/bug.h>
28
29 #define PANIC_TIMER_STEP 100
30 #define PANIC_BLINK_SPD 18
31
32 int panic_on_oops = CONFIG_PANIC_ON_OOPS_VALUE;
33 static unsigned long tainted_mask;
34 static int pause_on_oops;
35 static int pause_on_oops_flag;
36 static DEFINE_SPINLOCK(pause_on_oops_lock);
37 bool crash_kexec_post_notifiers;
38 int panic_on_warn __read_mostly;
39
40 int panic_timeout = CONFIG_PANIC_TIMEOUT;
41 EXPORT_SYMBOL_GPL(panic_timeout);
42
43 ATOMIC_NOTIFIER_HEAD(panic_notifier_list);
44
45 EXPORT_SYMBOL(panic_notifier_list);
46
47 static long no_blink(int state)
48 {
49 return 0;
50 }
51
52 /* Returns how long it waited in ms */
53 long (*panic_blink)(int state);
54 EXPORT_SYMBOL(panic_blink);
55
56 /*
57 * Stop ourself in panic -- architecture code may override this
58 */
59 void __weak panic_smp_self_stop(void)
60 {
61 while (1)
62 cpu_relax();
63 }
64
65 /*
66 * Stop ourselves in NMI context if another CPU has already panicked. Arch code
67 * may override this to prepare for crash dumping, e.g. save regs info.
68 */
69 void __weak nmi_panic_self_stop(struct pt_regs *regs)
70 {
71 panic_smp_self_stop();
72 }
73
74 atomic_t panic_cpu = ATOMIC_INIT(PANIC_CPU_INVALID);
75
76 /*
77 * A variant of panic() called from NMI context. We return if we've already
78 * panicked on this CPU. If another CPU already panicked, loop in
79 * nmi_panic_self_stop() which can provide architecture dependent code such
80 * as saving register state for crash dump.
81 */
82 void nmi_panic(struct pt_regs *regs, const char *msg)
83 {
84 int old_cpu, cpu;
85
86 cpu = raw_smp_processor_id();
87 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, cpu);
88
89 if (old_cpu == PANIC_CPU_INVALID)
90 panic("%s", msg);
91 else if (old_cpu != cpu)
92 nmi_panic_self_stop(regs);
93 }
94 EXPORT_SYMBOL(nmi_panic);
95
96 /**
97 * panic - halt the system
98 * @fmt: The text string to print
99 *
100 * Display a message, then perform cleanups.
101 *
102 * This function never returns.
103 */
104 void panic(const char *fmt, ...)
105 {
106 static char buf[1024];
107 va_list args;
108 long i, i_next = 0;
109 int state = 0;
110 int old_cpu, this_cpu;
111
112 /*
113 * Disable local interrupts. This will prevent panic_smp_self_stop
114 * from deadlocking the first cpu that invokes the panic, since
115 * there is nothing to prevent an interrupt handler (that runs
116 * after setting panic_cpu) from invoking panic() again.
117 */
118 local_irq_disable();
119
120 /*
121 * It's possible to come here directly from a panic-assertion and
122 * not have preempt disabled. Some functions called from here want
123 * preempt to be disabled. No point enabling it later though...
124 *
125 * Only one CPU is allowed to execute the panic code from here. For
126 * multiple parallel invocations of panic, all other CPUs either
127 * stop themself or will wait until they are stopped by the 1st CPU
128 * with smp_send_stop().
129 *
130 * `old_cpu == PANIC_CPU_INVALID' means this is the 1st CPU which
131 * comes here, so go ahead.
132 * `old_cpu == this_cpu' means we came from nmi_panic() which sets
133 * panic_cpu to this CPU. In this case, this is also the 1st CPU.
134 */
135 this_cpu = raw_smp_processor_id();
136 old_cpu = atomic_cmpxchg(&panic_cpu, PANIC_CPU_INVALID, this_cpu);
137
138 if (old_cpu != PANIC_CPU_INVALID && old_cpu != this_cpu)
139 panic_smp_self_stop();
140
141 console_verbose();
142 bust_spinlocks(1);
143 va_start(args, fmt);
144 vsnprintf(buf, sizeof(buf), fmt, args);
145 va_end(args);
146 pr_emerg("Kernel panic - not syncing: %s\n", buf);
147 #ifdef CONFIG_DEBUG_BUGVERBOSE
148 /*
149 * Avoid nested stack-dumping if a panic occurs during oops processing
150 */
151 if (!test_taint(TAINT_DIE) && oops_in_progress <= 1)
152 dump_stack();
153 #endif
154
155 /*
156 * If we have crashed and we have a crash kernel loaded let it handle
157 * everything else.
158 * If we want to run this after calling panic_notifiers, pass
159 * the "crash_kexec_post_notifiers" option to the kernel.
160 *
161 * Bypass the panic_cpu check and call __crash_kexec directly.
162 */
163 if (!crash_kexec_post_notifiers) {
164 printk_nmi_flush_on_panic();
165 __crash_kexec(NULL);
166 }
167
168 /*
169 * Note smp_send_stop is the usual smp shutdown function, which
170 * unfortunately means it may not be hardened to work in a panic
171 * situation.
172 */
173 smp_send_stop();
174
175 /*
176 * Run any panic handlers, including those that might need to
177 * add information to the kmsg dump output.
178 */
179 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
180
181 /* Call flush even twice. It tries harder with a single online CPU */
182 printk_nmi_flush_on_panic();
183 kmsg_dump(KMSG_DUMP_PANIC);
184
185 /*
186 * If you doubt kdump always works fine in any situation,
187 * "crash_kexec_post_notifiers" offers you a chance to run
188 * panic_notifiers and dumping kmsg before kdump.
189 * Note: since some panic_notifiers can make crashed kernel
190 * more unstable, it can increase risks of the kdump failure too.
191 *
192 * Bypass the panic_cpu check and call __crash_kexec directly.
193 */
194 if (crash_kexec_post_notifiers)
195 __crash_kexec(NULL);
196
197 bust_spinlocks(0);
198
199 /*
200 * We may have ended up stopping the CPU holding the lock (in
201 * smp_send_stop()) while still having some valuable data in the console
202 * buffer. Try to acquire the lock then release it regardless of the
203 * result. The release will also print the buffers out. Locks debug
204 * should be disabled to avoid reporting bad unlock balance when
205 * panic() is not being callled from OOPS.
206 */
207 debug_locks_off();
208 console_flush_on_panic();
209
210 if (!panic_blink)
211 panic_blink = no_blink;
212
213 if (panic_timeout > 0) {
214 /*
215 * Delay timeout seconds before rebooting the machine.
216 * We can't use the "normal" timers since we just panicked.
217 */
218 pr_emerg("Rebooting in %d seconds..", panic_timeout);
219
220 for (i = 0; i < panic_timeout * 1000; i += PANIC_TIMER_STEP) {
221 touch_nmi_watchdog();
222 if (i >= i_next) {
223 i += panic_blink(state ^= 1);
224 i_next = i + 3600 / PANIC_BLINK_SPD;
225 }
226 mdelay(PANIC_TIMER_STEP);
227 }
228 }
229 if (panic_timeout != 0) {
230 /*
231 * This will not be a clean reboot, with everything
232 * shutting down. But if there is a chance of
233 * rebooting the system it will be rebooted.
234 */
235 emergency_restart();
236 }
237 #ifdef __sparc__
238 {
239 extern int stop_a_enabled;
240 /* Make sure the user can actually press Stop-A (L1-A) */
241 stop_a_enabled = 1;
242 pr_emerg("Press Stop-A (L1-A) to return to the boot prom\n");
243 }
244 #endif
245 #if defined(CONFIG_S390)
246 {
247 unsigned long caller;
248
249 caller = (unsigned long)__builtin_return_address(0);
250 disabled_wait(caller);
251 }
252 #endif
253 pr_emerg("---[ end Kernel panic - not syncing: %s\n", buf);
254 local_irq_enable();
255 for (i = 0; ; i += PANIC_TIMER_STEP) {
256 touch_softlockup_watchdog();
257 if (i >= i_next) {
258 i += panic_blink(state ^= 1);
259 i_next = i + 3600 / PANIC_BLINK_SPD;
260 }
261 mdelay(PANIC_TIMER_STEP);
262 }
263 }
264
265 EXPORT_SYMBOL(panic);
266
267
268 struct tnt {
269 u8 bit;
270 char true;
271 char false;
272 };
273
274 static const struct tnt tnts[] = {
275 { TAINT_PROPRIETARY_MODULE, 'P', 'G' },
276 { TAINT_FORCED_MODULE, 'F', ' ' },
277 { TAINT_CPU_OUT_OF_SPEC, 'S', ' ' },
278 { TAINT_FORCED_RMMOD, 'R', ' ' },
279 { TAINT_MACHINE_CHECK, 'M', ' ' },
280 { TAINT_BAD_PAGE, 'B', ' ' },
281 { TAINT_USER, 'U', ' ' },
282 { TAINT_DIE, 'D', ' ' },
283 { TAINT_OVERRIDDEN_ACPI_TABLE, 'A', ' ' },
284 { TAINT_WARN, 'W', ' ' },
285 { TAINT_CRAP, 'C', ' ' },
286 { TAINT_FIRMWARE_WORKAROUND, 'I', ' ' },
287 { TAINT_OOT_MODULE, 'O', ' ' },
288 { TAINT_UNSIGNED_MODULE, 'E', ' ' },
289 { TAINT_SOFTLOCKUP, 'L', ' ' },
290 { TAINT_LIVEPATCH, 'K', ' ' },
291 };
292
293 /**
294 * print_tainted - return a string to represent the kernel taint state.
295 *
296 * 'P' - Proprietary module has been loaded.
297 * 'F' - Module has been forcibly loaded.
298 * 'S' - SMP with CPUs not designed for SMP.
299 * 'R' - User forced a module unload.
300 * 'M' - System experienced a machine check exception.
301 * 'B' - System has hit bad_page.
302 * 'U' - Userspace-defined naughtiness.
303 * 'D' - Kernel has oopsed before
304 * 'A' - ACPI table overridden.
305 * 'W' - Taint on warning.
306 * 'C' - modules from drivers/staging are loaded.
307 * 'I' - Working around severe firmware bug.
308 * 'O' - Out-of-tree module has been loaded.
309 * 'E' - Unsigned module has been loaded.
310 * 'L' - A soft lockup has previously occurred.
311 * 'K' - Kernel has been live patched.
312 *
313 * The string is overwritten by the next call to print_tainted().
314 */
315 const char *print_tainted(void)
316 {
317 static char buf[ARRAY_SIZE(tnts) + sizeof("Tainted: ")];
318
319 if (tainted_mask) {
320 char *s;
321 int i;
322
323 s = buf + sprintf(buf, "Tainted: ");
324 for (i = 0; i < ARRAY_SIZE(tnts); i++) {
325 const struct tnt *t = &tnts[i];
326 *s++ = test_bit(t->bit, &tainted_mask) ?
327 t->true : t->false;
328 }
329 *s = 0;
330 } else
331 snprintf(buf, sizeof(buf), "Not tainted");
332
333 return buf;
334 }
335
336 int test_taint(unsigned flag)
337 {
338 return test_bit(flag, &tainted_mask);
339 }
340 EXPORT_SYMBOL(test_taint);
341
342 unsigned long get_taint(void)
343 {
344 return tainted_mask;
345 }
346
347 /**
348 * add_taint: add a taint flag if not already set.
349 * @flag: one of the TAINT_* constants.
350 * @lockdep_ok: whether lock debugging is still OK.
351 *
352 * If something bad has gone wrong, you'll want @lockdebug_ok = false, but for
353 * some notewortht-but-not-corrupting cases, it can be set to true.
354 */
355 void add_taint(unsigned flag, enum lockdep_ok lockdep_ok)
356 {
357 if (lockdep_ok == LOCKDEP_NOW_UNRELIABLE && __debug_locks_off())
358 pr_warn("Disabling lock debugging due to kernel taint\n");
359
360 set_bit(flag, &tainted_mask);
361 }
362 EXPORT_SYMBOL(add_taint);
363
364 static void spin_msec(int msecs)
365 {
366 int i;
367
368 for (i = 0; i < msecs; i++) {
369 touch_nmi_watchdog();
370 mdelay(1);
371 }
372 }
373
374 /*
375 * It just happens that oops_enter() and oops_exit() are identically
376 * implemented...
377 */
378 static void do_oops_enter_exit(void)
379 {
380 unsigned long flags;
381 static int spin_counter;
382
383 if (!pause_on_oops)
384 return;
385
386 spin_lock_irqsave(&pause_on_oops_lock, flags);
387 if (pause_on_oops_flag == 0) {
388 /* This CPU may now print the oops message */
389 pause_on_oops_flag = 1;
390 } else {
391 /* We need to stall this CPU */
392 if (!spin_counter) {
393 /* This CPU gets to do the counting */
394 spin_counter = pause_on_oops;
395 do {
396 spin_unlock(&pause_on_oops_lock);
397 spin_msec(MSEC_PER_SEC);
398 spin_lock(&pause_on_oops_lock);
399 } while (--spin_counter);
400 pause_on_oops_flag = 0;
401 } else {
402 /* This CPU waits for a different one */
403 while (spin_counter) {
404 spin_unlock(&pause_on_oops_lock);
405 spin_msec(1);
406 spin_lock(&pause_on_oops_lock);
407 }
408 }
409 }
410 spin_unlock_irqrestore(&pause_on_oops_lock, flags);
411 }
412
413 /*
414 * Return true if the calling CPU is allowed to print oops-related info.
415 * This is a bit racy..
416 */
417 int oops_may_print(void)
418 {
419 return pause_on_oops_flag == 0;
420 }
421
422 /*
423 * Called when the architecture enters its oops handler, before it prints
424 * anything. If this is the first CPU to oops, and it's oopsing the first
425 * time then let it proceed.
426 *
427 * This is all enabled by the pause_on_oops kernel boot option. We do all
428 * this to ensure that oopses don't scroll off the screen. It has the
429 * side-effect of preventing later-oopsing CPUs from mucking up the display,
430 * too.
431 *
432 * It turns out that the CPU which is allowed to print ends up pausing for
433 * the right duration, whereas all the other CPUs pause for twice as long:
434 * once in oops_enter(), once in oops_exit().
435 */
436 void oops_enter(void)
437 {
438 tracing_off();
439 /* can't trust the integrity of the kernel anymore: */
440 debug_locks_off();
441 do_oops_enter_exit();
442 }
443
444 /*
445 * 64-bit random ID for oopses:
446 */
447 static u64 oops_id;
448
449 static int init_oops_id(void)
450 {
451 if (!oops_id)
452 get_random_bytes(&oops_id, sizeof(oops_id));
453 else
454 oops_id++;
455
456 return 0;
457 }
458 late_initcall(init_oops_id);
459
460 void print_oops_end_marker(void)
461 {
462 init_oops_id();
463 pr_warn("---[ end trace %016llx ]---\n", (unsigned long long)oops_id);
464 }
465
466 /*
467 * Called when the architecture exits its oops handler, after printing
468 * everything.
469 */
470 void oops_exit(void)
471 {
472 do_oops_enter_exit();
473 print_oops_end_marker();
474 kmsg_dump(KMSG_DUMP_OOPS);
475 }
476
477 struct warn_args {
478 const char *fmt;
479 va_list args;
480 };
481
482 void __warn(const char *file, int line, void *caller, unsigned taint,
483 struct pt_regs *regs, struct warn_args *args)
484 {
485 disable_trace_on_warning();
486
487 pr_warn("------------[ cut here ]------------\n");
488
489 if (file)
490 pr_warn("WARNING: CPU: %d PID: %d at %s:%d %pS\n",
491 raw_smp_processor_id(), current->pid, file, line,
492 caller);
493 else
494 pr_warn("WARNING: CPU: %d PID: %d at %pS\n",
495 raw_smp_processor_id(), current->pid, caller);
496
497 if (args)
498 vprintk(args->fmt, args->args);
499
500 if (panic_on_warn) {
501 /*
502 * This thread may hit another WARN() in the panic path.
503 * Resetting this prevents additional WARN() from panicking the
504 * system on this thread. Other threads are blocked by the
505 * panic_mutex in panic().
506 */
507 panic_on_warn = 0;
508 panic("panic_on_warn set ...\n");
509 }
510
511 print_modules();
512
513 if (regs)
514 show_regs(regs);
515 else
516 dump_stack();
517
518 print_oops_end_marker();
519
520 /* Just a warning, don't kill lockdep. */
521 add_taint(taint, LOCKDEP_STILL_OK);
522 }
523
524 #ifdef WANT_WARN_ON_SLOWPATH
525 void warn_slowpath_fmt(const char *file, int line, const char *fmt, ...)
526 {
527 struct warn_args args;
528
529 args.fmt = fmt;
530 va_start(args.args, fmt);
531 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL,
532 &args);
533 va_end(args.args);
534 }
535 EXPORT_SYMBOL(warn_slowpath_fmt);
536
537 void warn_slowpath_fmt_taint(const char *file, int line,
538 unsigned taint, const char *fmt, ...)
539 {
540 struct warn_args args;
541
542 args.fmt = fmt;
543 va_start(args.args, fmt);
544 __warn(file, line, __builtin_return_address(0), taint, NULL, &args);
545 va_end(args.args);
546 }
547 EXPORT_SYMBOL(warn_slowpath_fmt_taint);
548
549 void warn_slowpath_null(const char *file, int line)
550 {
551 __warn(file, line, __builtin_return_address(0), TAINT_WARN, NULL, NULL);
552 }
553 EXPORT_SYMBOL(warn_slowpath_null);
554 #endif
555
556 #ifdef CONFIG_CC_STACKPROTECTOR
557
558 /*
559 * Called when gcc's -fstack-protector feature is used, and
560 * gcc detects corruption of the on-stack canary value
561 */
562 __visible void __stack_chk_fail(void)
563 {
564 panic("stack-protector: Kernel stack is corrupted in: %p\n",
565 __builtin_return_address(0));
566 }
567 EXPORT_SYMBOL(__stack_chk_fail);
568
569 #endif
570
571 core_param(panic, panic_timeout, int, 0644);
572 core_param(pause_on_oops, pause_on_oops, int, 0644);
573 core_param(panic_on_warn, panic_on_warn, int, 0644);
574
575 static int __init setup_crash_kexec_post_notifiers(char *s)
576 {
577 crash_kexec_post_notifiers = true;
578 return 0;
579 }
580 early_param("crash_kexec_post_notifiers", setup_crash_kexec_post_notifiers);
581
582 static int __init oops_setup(char *s)
583 {
584 if (!s)
585 return -EINVAL;
586 if (!strcmp(s, "panic"))
587 panic_on_oops = 1;
588 return 0;
589 }
590 early_param("oops", oops_setup);